Classifying hyperspectral remote sensing images across different scenes has recently emerged as a significant challenge. When only historical labeled images (source domain, SD) are available, it is crucial to leverage these images effectively to train a model with strong generalization ability that can be directly applied to classify unseen samples (target domain, TD). To address these challenges, this paper proposes a novel single-domain generalization (SDG) network, termed the domain-aware adversarial domain augmentation network (DADAnet) for cross-scene hyperspectral image classification (HSIC). DADAnet involves two stages: adversarial domain augmentation (ADA) and task-specific training. ADA employs a progressive adversarial generation strategy to construct an augmented domain (AD). To enhance variability in both spatial and spectral dimensions, a domain-aware spatial-spectral mask (DSSM) encoder is constructed to increase the diversity of the generated adversarial samples. Furthermore, a two-level contrastive loss (TCC) is designed and incorporated into the ADA to ensure both the diversity and effectiveness of AD samples. Finally, DADAnet performs supervised learning jointly on the SD and AD during the task-specific training stage. Experimental results on two public hyperspectral image datasets and a new Hangzhouwan (HZW) dataset demonstrate that the proposed DADAnet outperforms existing domain adaptation (DA) and domain generalization (DG) methods, achieving overall accuracies of 80.69%, 63.75%, and 87.61% on three datasets, respectively.

Change detection (CD) in hyperspectral images (HSIs) has become an increasingly vital research field in remote sensing. Over the past few years, the adoption of deep learning approaches, particularly convolutional neural network (CNN) and transformer-based architectures have significantly advanced performance in this field. While these models effectively capture spectral-spatial features, they may also introduce redundant or irrelevant spatial information, potentially degrading the accuracy of HSI CD. To address this challenge, a center-pixel and gated mechanism-based attention network (CGMNet) is proposed for HSI CD, leveraging the central pixel's significance to enhance accuracy and robustness. First, a gated-based center spatial attention (GCSA) module is designed to emphasize spatial relationships surrounding the central pixel. By incorporating gating mechanisms, GCSA selectively enhances relevant spatial features while suppressing irrelevant information. Second, a gated-based spectral attention (GSA) module is proposed to dynamically highlight the most significant spectral features, ensuring an effective spectral representation. Finally, a global transform fusion (GTF) module is proposed to capture global contextual information and to fuse it with the extracted spatial and spectral features. Moreover, we introduce a novel benchmark dataset, named the Hangzhou Bay (HZB), specifically designed to advance coastal remote sensing research. Experimental evaluations conducted on three publicly available datasets, as well as the HZB dataset, show that our CGMNet consistently outperforms some state-of-the-art methods in the HSI CD task. The source code of the proposed CGMNet, along with the HZB dataset, will be made publicly available at https://github.com/creativeXin/CGMNet.

Due to the complex acquisition environment and scarcity of labels, domain adaptation (DA) techniques are widely applied to cross-scenario hyperspectral image (HSI) classification to achieve more precise labeling. Many existing approaches mainly rely on convolutional neural networks (CNNs) to capture local spatial contextual relationships, supplemented by graph convolutional networks (GCNs) for long-range modeling. However, GCNs usually require full batch training and fixed initial graph structures, which significantly limits the exploration of topological structures. To address this, a multi-contrastive and dynamic topological matching network (MCDTM) is introduced to accomplish cross-domain HSI classification. Unlike fixed graph construction methods, mini-batches of samples are utilized to construct dynamic subgraphs within the source and target domains, respectively, with locally extracted features from CNNs serving as the basis for graph construction. More importantly, as the model is optimized and the domain gap narrows, the dynamic graph structure is adaptively adjusted according to the evolving samples, thereby boosting the accuracy of the graph and enhancing the discriminative power of the model. Moreover, the integration of weighted multi-positive contrastive learning and graph matching achieves distribution alignment and graph alignment, enhancing the model's capacity to distinguish and align complex patterns in HSI. Experimental results in three tasks show that the MCDTM surpasses several advanced DA methods, achieving impressive accuracies of 80.10%, 70.01%, and 94.17% on the Houston, HyRank, and YC-YC tasks, thereby showcasing its superior performance.

Recently, contrastive learning (CL) has shown excellent performance in hyperspectral image (HSI) classification. However, existing CL based methods face two specific challenges. (1) Multi-view samples inevitably introduce ambiguity and uncertainty due to data augmentation operations. Traditional contrastive learning methods fail to effectively model these dynamic ambiguous features, resulting in a lack of robustness in the feature learning process. (2) Existing CL based methods primarily learns feature representations by pulling positive samples closer and pushing negative samples apart. But, they lack structured modeling of intra-class feature compactness and inter-class feature separability. To address these challenges, we propose a fuzzy triplet contrastive learning (FTCL) method for HSI classification. For the first challenge, we propose a multi-view fuzzy neighborhood learning (MFNL) module. This module effectively models the ambiguity among multi-view samples through fuzzy membership calculation, multi-view fuzzy weight matrix generation, and weighted feature aggregation, significantly enhancing the robustness and stability of feature representations. To tackle the second challenge, we design a triplet feature discriminative (TFD) classifier, which improves intra-class compactness by minimizing the distance between anchor samples and positive samples, while enhancing inter-class separability by maximizing the distance between anchor samples and negative samples. This enables precise modeling of intra-class compactness and inter-class separability. The proposed method is evaluated on four HSI datasets, and the experimental results demonstrate that the proposed method outperforms the state-of-the-art methods.

Recently, deep neural networks have demonstrated outstanding performance in hyperspectral image (HSI) change detection (CD), especially when there is sufficient labeled samples. However, the labels of HSI are difficult to obtain, and acquiring enough labels to train deep network is a great challenge in practice. Therefore, to mitigate the effect of insufficient labels in detection results, this paper proposes a cross-scene few-shot learning (FSL) network based on intra-inter domain contrastive alignment (CAFSL) for HSI-CD, which combines contrastive learning (CL) and FSL into a unified framework, aiming to achieve better detection results using only a few labeled samples. Specifically, we perform cross-scene FSL using a pair of dual-phase images from a very high-resolution image (VHRI) as the source domain and a pair of HSI as the target domain. Then, an intra-domain supervised contrastive learning (INSCL) module is designed to enhance the compactness within classes and widen the discrimination between classes by maximizing the feature similarity of intra-class and minimizing the feature similarity of inter-class. Finally, a cross-domain contrastive alignment (CRCA) module is proposed to align the features of source and target domains, which mitigates the effect of domain migration problems caused by different data types. Experiments on three HSI benchmark datasets reveal that the CAFSL algorithm outperforms current advanced algorithms based on deep learning and FSL while with limited labeled samples.

Cloud mask underpins accurate precipitation nowcasting, which in turn is vital for understanding the hydrological cycle, supporting disaster prevention, solar energy forecasting and transportation. However, cloud mask nowcasting remains challenging because meteorological data exhibit irregular temporal and spatial variations, including fine-scale structures, and often suffer from highly skewed precipitation intensity distributions. Existing methods struggle to capture complex spatiotemporal dynamics and preserve fine-scale structures due to limitations in handling sparse data from numerical weather prediction (NWP) model. To address these issues, we propose an asymmetric dual-branch non-causal mamba U-Net (ADNM-UNet) featuring three key components: (1) The Asymmetric Dual-branch Non-causal Mamba (ADNM) implements a novel asymmetric bidirectional modeling framework that resolves directional bias in conventional Mamba architectures. This design preserves precise cloud boundary delineation while capturing long-range spatiotemporal dependencies in sparse data from NWP. (2) The Multi-scale Attention Enhancement Module (MAEM) enhances discriminative feature representation and suppresses spectral redundancy through anisotropic convolution kernels and hybrid pooling. This mechanism significantly improves edge retention in precipitation systems while attenuating atmospheric noise interference. (3) Complementing these advancements, the Wavelet Decomposition and Fusion Module (WDFM) maintains cloud contour integrity across scales through multiresolution decomposition. Extensive experiments demonstrate that ADNM-UNet outperforms existing methods across all metrics, achieving 27.96% improvement in CSI and 22.89% in HSS for metrics over the best performing baseline models at high intensity scenarios. Our project is open source and available on GitHub at: https://github.com/kanyu369/ADNM-UNet.

Recently, domain adaptation (DA) methods based on contrastive learning are widely used to solve the cross-scene classification problem. However, existing contrastive learning methods only focus on source domain or target domain features, or do not adequately consider the interaction of domain information, thus the learned domain-invariant features still have large discrepancies. To address this problem, we propose a novel domain fusion contrastive learning (DFCL) framework for cross-scene hyperspectral image (HSI) classification. DFCL uses an interdomain and intradomain dual-domain fusion strategy at the feature level, which introduces domain information as a noise interference term for sample enhancement. With the interference of domain information, same category samples are pulled closer and different categories samples are pushed further apart to learn more discriminative features. In addition, we construct an intermediate domain through the source and target domains and define a feature space loss that measures domain discrepancy by feature similarity and label similarity. Finally, a progressive selection strategy based on prototype learning is proposed to select high-confidence pseudolabels for DFCL. Experiments on three HSI cross-scene datasets show that the proposed method is superior to existing DA methods.

Recently, Mamba has made significant strides in sequence modeling, with its global receptive field, dynamic weighting strategy and linear growth in computational complexity. In remote sensing (RS) change detection (CD), several studies have demonstrated that Mambas leverage a unique scanning mechanism to traverse images from various directions, showcasing excellent long-range modeling capabilities. However, as the network depth increases, Mamba often struggle to retain shallow textures and local features effectively. In particular, modern RS images frequently capture complex surface scenes, including seasonal climate variations and densely built environments, making local contextual details crucial for effective CD. Therefore, a difference enhancement and inter-scale interactive fusion Mamba (DEIF-Mamba) is proposed to alleviate the issue. This entire network framework integrates CNN and Mamba, utilizing CNN to capture local feature information, while Mamba employs a cross-scanning mechanism to integrate global information. To address the interference caused by mixed texture features and the missed detection of subtle changes in complex scenes, a differential feature enhancement module (DFEM) is proposed to enrich local contextual details and improve feature representation. In addition, we propose an inter-scale interactive fusion (ISIF) strategy to fully utilize the cross-scale interactive information and minimize information redundancy. Extensive experiments on four CD datasets demonstrate that the proposed DEIF-Mamba achieves an average F1 of 85.87%, and shows superior performance compared with other state-of-the-art (SOTA) methods. Code will be available online (https://github.com/Jyl199904/DEIF-Mamba).

Hyperspectral image (HSI) change detection (CD) plays a crucial role in remote sensing observation. It leverages the abundant spectral and spatial information in bi-temporal HSIs to identify subtle Earth surface changes. Most current deep-learning-based HSI CD methods primarily utilize convolutional neural networks or transformers to extract features from bi-temporal images. However, these methods lack an effective attention mechanism to enhance differential features. In addition, they do not fully leverage the aggregation relationship between the features of bi-temporal images to extract interaction features. To address these challenges, we propose a novel adaptive information weighting and synchronized enhancement network (AIWSEN) for HSI CD. This network employs the information entropy to capture change features specific to the CD task and enhances bi-temporal interaction features. Specifically, an adaptive information weighting attention module (AIWAM) leverages the maximum discrete entropy theorem to capture the difference information. A dual-time synchronic change enhancing module (DSCEM) is designed to extract features by interactively aggregating features from bi-temporal HSIs to enhance difference features. A bi-temporal image feature selection and fusion module (BFSFM) is constructed to filter out important features using forget and update gates. Experimental results on three HSI CD datasets demonstrate that the proposed AIWSEN method outperforms several state-of-the-art methods. The source code of the proposed AIWSEN will be released athttps://github.com/creativeXin/AIWSEN.

In recent years, graph neural networks (GNNs) have been increasingly applied to hyperspectral image change detection (HSI-CD). However, existing methods propagate features by weighting nodes and edges, which may weaken the original node characteristics and spectral differences. Moreover, current methods typically treat bitemporal features in isolation and lack inter-temporal semantic modeling. To address the above issues, this paper proposes a graph contrastive learning and multi-graph attention fusion network (GCLMA), which leverages contrastive learning to extract discriminative features and enhances semantic relevance through graph attention-based interactions across multiple graphs. The proposed GCLMA method mainly consists of four modules, i.e., spatial-spectral graph construction (SSGC), contrastive feature learning ResGCN (CFLR), differential feature enhancement (DFE), and multi-graph interactive fusion (MGIF). The SSGC module is first designed to construct a well-defined graph structure by adaptively integrating spatial proximity and spectral similarity. Then, the CFLR module utilizes contrastive learning to extract discriminative feature representations that are sensitive to subtle changes. Next, the DFE module amplifies significant spectral variations to emphasize critical change regions. Finally, the MGIF module employs the graph attention mechanism to interactively fuse enhanced difference graph features with bitemporal graph features, thereby facilitating semantic modeling. Experimental results on three HSI-CD datasets show that the proposed GCLMA outperforms most existing state-of-the-art methods. The source code of the proposed method will be released at https://github.com/YYYYYJJJJJJ/GCLMA.

Hypergraph neural networks (HGNNs) have garnered considerable attention in hyperspectral image (HSI) classification for their ability to model higher-order nonlinear relationships. However, most existing HGNN-based HSI classification methods adopt simple hypergraph structures, which fail to fully leverage low-order and high-order information within HSIs. Moreover, these methods are constrained by rigid single-scale superpixel segmentations, which fail to fully capture rich spectral-spatial features of HSIs. To overcome these limitations, this paper proposes a progressive hybrid-order hypergraph framework (PHHF) that establishes a graph-hypergraph collaborative learning paradigm to learn multi-scale hybrid-order features across hierarchical graphs. Specifically, by synergistically modeling multiple structural information, we develop a progressive hypergraph neural network framework to enhance the extraction of hierarchical representations of HSIs. Second, at each level, a novel hybrid-order hyperedge generation strategy is designed to enhance spectral-spatial consistency across scales and overcome the limitations of existing hypergraph methods. Finally, a novel heterogeneous kernel-based convolution (HetConv) is introduced to enhance pixel-level feature extraction for the PHHF. Compared to conventional convolution, HetConv offers richer spatial-spectral details with lower computational cost. Extensive experiments on widely-used HSI datasets demonstrate that our proposed method achieves state-of-the-art HSI classification performance.

In recent years, few-shot learning (FSL) has made significant progress in hyperspectral image classification (HSIC) by transferring meta-knowledge from a source domain with sufficient labeled samples to a target domain with limited labeled samples. However, existing FSL methods face two key challenges in coastal wetland HSIC applications, i.e., prototype instability due to limited labeled samples and domain shift due to the domain’s distribution difference. These challenges are further exacerbated by the unique characteristics of coastal wetland environments, which have complex land cover classes and subtle spectral differences between land cover classes. To address these limitations, we propose a language-guided and similarity-aware network (LGSAnet) for few-shot coastal wetland HSIC in this article. The network mainly consists of two modules: language-guided prototype alignment (LGPA) and similarity-aware prototype calibration (SAPC). The LGPA module uses linguistic features to guide the learning of visual features, enabling the model to obtain visual representations enriched with linguistic prior knowledge. This guided alignment can learn more accurate and stable visual prototype representations, thereby improving the accuracy of the model. To mitigate domain shift, an SAPC module is constructed to refine the target domain prototypes and minimize domain-specific differences using the domain’s semantic similarity. The experimental results on three coastal wetland hyperspectral datasets demonstrate that the proposed LGSAnet outperforms existing state-of-the-art FSL methods.

Deep learning has advanced hyperspectral image classification (HSIC), but label scarcity remains a significant challenge. Traditional unimodal methods usually produce unstable prototypes, while multimodal methods suffer from cross-modal semantic misalignment. Moreover, standard metrics fail to capture high-order feature correlations, further limiting the discriminative ability of the model. To address these issues, we propose a query-oriented dynamic multimodal alignment (QODMA) method, which integrates visual-textual guidance with dual-distance metric learning for few-shot HSIC. Specifically, a query-oriented dynamic attention (QODA) module is designed to bridge the modality gap by aligning visual and textual features through query-driven attention interactions. A bidirectional attention mechanism is constructed to employ contrastive learning to enhance intra-class compactness. Additionally, a dual-distance metric learning (DML) module that combines the Euclidean distance and Brownian distance covariance (BDC) metrics is employed to refine the feature space representation, thereby enhancing the discriminative ability of the model. To mitigate domain shift, an inter-domain structural consistency loss (IDSCL) is constructed. Experimental results on four public hyperspectral data sets demonstrate that the proposed QODMA outperforms state-of-the-art few-shot classification methods.

In recent years, marine oil spills have occurred frequently, causing serious damage to the marine ecological environment. Hyperspectral images (HSIs) can provide rich spectral and spatial information, and have broad development prospects in marine oil spill detection. This article proposes a hyperspectral marine oil spill detection network, HMOSDN, that integrates improved superpixel segmentation and a mixed attention mechanism (MAM). First, to deal with the extensive clutter and diverse morphology of marine oil spill areas in HSIs, we propose a new superpixel segmentation algorithm based on improved simple linear iterative clustering (ISLIC), which achieves preliminary extraction of spatial features and reduces spatial noise via a Gaussian filter and a pixel intensity smoothing technique (PIST). Then, to further fuse spectral and spatial features and strengthen the feature mining and utilization of fused information, we design a spectral-spatial feature extraction network with an MAM, MAM-SSFEN, which adds a spectral attention module and a spatial attention module, further improving the performance of the deep feature extraction network for oil spill detection. Experiments on the hyperspectral oil spill database (HOSD) demonstrate that our proposed method, HMOSDN, outperforms several other detection techniques regarding area under the curve (AUC) and recall evaluation metrics.

Precipitation nowcasting is a challenging task in the context of global climate variability. However, existing radar echo or numerical weather prediction data methods lack deep modeling between echograms at different time points and have difficulty in accurately capturing irregular variations and small-scale features of precipitable clouds. To address these challenges, we propose for the first time a U-Net short-term precipitation prediction network based on vision Mamba technology for the precipitation nowcasting mission, named Mamba-UNet. Specifically, Mamba-UNet includes two core modules: the dual-branch Mamba fusion module and the multiscale spatiotemporal attention module. Finally, we propose a loss function namely dynamic quantile weighted loss to address the problem of imbalanced precipitation intensity distribution. To validate the capacity of the proposed method, the experiments were conducted on an analysis dataset of the local analysis and prediction system model in a specific region of East China. The experimental results show that our proposed Mamba-UNet has the best overall performance.

Since the information across all bands of the cloud-contaminated region is missing, thick cloud removal for remote sensing images (RSIs) is still a challenging problem. Recently, the availability of rich spatial–spectral–temporal information for multitemporal RSIs provides the possibility for addressing the thick cloud removal problem. However, existing methods explore the holistic redundancy of multitemporal RSIs and neglect the important semantic clue of multitemporal images. In this letter, we propose a superpixel-oriented thick cloud removal (STORM) model for multitemporal images, where the multitemporal superpixel as the generic unit allows us to exploit redundancy with semantic clue in a low-rank optimization problem. To harness the resultant irregular fourth-order tensor (i.e., multitemporal superpixels) in the optimization problem, we cleverly introduce the weighted tensor to transform the irregular tensor into the regular tensor, which naturally leads to a standard low-rank tensor optimization problem. To tackle the tensor optimization problem, we develop a proximal alternating minimization (PAM)-based algorithm. Extensive simulated and real experiments on multitemporal RSIs acquired by Sentinel-2 and Landsat-8 satellites demonstrate the superior performance of the proposed method over the comparison methods.

Road extraction from high-resolution remote sensing images is a challenging task due to the presence of disturbing features and the diversity of road representations. To overcome these problems, deep neural networks-based methods have been recently used to improve the speed and accuracy of road extraction. In this letter, we propose a simple yet effective method for feature extraction with context fusion and self-learning sampling, which we call dual feature fusion (DFF). Moreover, we point out that the DFF method is functionally similar to the downsampling and upsampling structure. From this, we propose a network with a dense feature skip connect structure (DFC-UNet) to extract the roads from remote sensing images. The complexity of the high-dimensional features of the U-shaped structure is also analyzed, and the redundant features are suppressed through the equivalent replacement of the DFF block. Aiming at the unbalanced characteristics of samples and the topological characteristics of the road network, we then propose a comprehensive loss function based on dynamic weighting to strengthen the learning of the road network. Experimental results on the Massachusetts road dataset, the DeepGlobe dataset, and the CHN6_CUG dataset confirm the effectiveness of the proposed method.

Hyperspectral image classification (HIC) has attracted considerable attention in the last two decades, and significant progress has been made. However, the small sample size problem of HIC is still challenging. This letter presents a two-stage HIC approach that achieves high classification accuracies with few labeled samples. For a given hyperspectral image, the spatial features are first extracted by local binary patterns (LBP). Spatial features and spectral features for each pixel are then stacked into feature vectors. These vectors are fed into SVM to finish the first classification stage. Based on the preliminary classification results, a superpixel segmentation method is introduced for selecting some superpixels which include training samples and all test pixels assigned to some class. These selected superpixels with their labels obtained by SVM are then added to training samples. According to the enlarged training sample set, Random Multi-Graph (RMG) is finally utilized to classify the remaining samples. Experimental results on three benchmark HSI datasets demonstrate that the proposed LBP and RMG-based two-stage method (LBP-RMG2) significantly outperforms several state-of-the-art algorithms with a few labeled samples.

Band selection aims at selecting a subset of representative bands from original hyperspectral images (HSIs) to alleviate data redundancy. There are at least two issues existing in previous methods. First, most of them ignore global or local structural information without considering both two aspects. Second, the high-order correlations among spectral bands are not explored during learning. In this paper, we propose a tensorial global-local graph self-representation (TGSR) method for hyperspectral band selection. Specifically, we segment the HSI into diverse superpixels to show the inherent spectral-spatial structures. Based on the generated superpixels, we learn the global and local graphs to explore complex structural information from global pixels and local regions. To alleviate the computational burden, a transformation is designed for easy graph convolution of global graph and pixel spectral matrix. With global and local knowledge, we formulate a global-local graph self-representation model to conduct band correlation learning in a self-weighted manner. To explore the high-order correlations among bands, we reorganize the self-representation coefficient matrices into a tensor with low-rank constraint. We design an alternating optimization algorithm to solve the proposed model. The most representative band is selected from each band subset by performing spectral clustering on the constructed affinity matrix. Experiments on HSI datasets verify the effectiveness of our method over the state-of-the-art methods. The source code is released athttps://github.com/ZhangYongshan/TGSR.

Recently, the adversarial domain adaptation (ADA) methods have been widely investigated and applied in cross-domain hyperspectral image (HSI) classification. However, most ADA algorithms aim to align the cross-domain distribution without focusing on the class separability of the aligned target features and the information of samples within the domain. To address these issues, a new ADA framework based on calibrated prototype and dynamic instance convolution (CPDIC) is proposed in this paper for cross domain HSI classification. The CPDIC is composed of a generator, a calibrated discriminator and a classifier. The generator includes a static 3D convolutional network (SCN) and a dynamic instance convolutional network (DICN), where the SCN is used to extract coarse-grained features of HSI and the DICN can extract sample-specific fine-grained features using instance convolutions generated from dynamic instance convolution kernel generation (DCKG) module. As for the generator, the static and dynamic interactive feature extraction network extracts robust domain-invariant features with discriminability. The calibrated discriminator aligns the marginal distribution between domains and calibrate the predicted pseudo labels of target domain. For classification, a calibrated prototype loss (CPL) is introduced to align the class distribution across domains. The results of three cross-domain HSI classification tasks show that the proposed CPDIC outperforms existing unsupervised domain adaptation (UDA) algorithms.

Contrastive learning achieves good performance on hyperspectral image classification (HSIC), but its application on cross-scene classification is still challenging due to domain shift. The emergence of domain adaptation (DA) techniques can reduce domain discrepancy and transfer a model between two domains. Recently, instance-level contrast adaptation methods can connect two related domains, and domain-invariant features are extracted. However, it is sensitive to noisy samples and only learns low-level discriminative features. To solve these problems, a novel domain invariant and compact prototype contrast adaptation (DIC-proCA) framework is proposed for HSIC. About the proposed DIC-proCA, the prototype is introduced into the contrastive learning framework, which serves as a representative embedding of semantically similar samples, has class representativeness and can alleviate the negative impact of outliers. Taking into account the class representativeness of the prototype and the discriminability of the sample itself, a bidirectional inter-domain instance-to-prototype contrastive loss is proposed. It explicitly expresses feature relationships between categories in different domains, and then extracts domain-invariant features. Meanwhile, the mining of compact discriminative features within the target domain is facilitated by instance-level contrastive learning after data augmentation. In addition, the strategy of label smoothing promotes the clusters in the domain to be more compact and evenly separated, making the model more generalizable. Three cross-scene HSIC tasks demonstrate that the proposed DIC-proCA exhibits superior performance compared to some advanced DA algorithms.

The hyperspectral imaging instrument makes a trade-off by sacrificing spatial resolution to achieve high spectral resolution. This compromise leads to a low signal-to-noise ratio, and hyperspectral images (HSIs) are often heavily contaminated with mixed noise, which is an inherent challenge. Previous research has achieved satisfactory results for natural image denoising; hyperspectral denoising has remained a formidable task. In this article, we introduce an innovative method called the multistage hybrid-denoising network for satellite hyperspectral images (SUC-MSDN). SUC-MSDN initially decomposes the noisy HSI into multiple scales and constructs a multistage denoising network by analyzing the spatial spectrum texture distribution characteristics of noise signals. Instead of simply stacking the output results from each scale, SUC-MSDN uses the denoising results from the low-scale network as prior knowledge for the high-scale denoising network to more accurately remove the final noise components. Extensive experimental datasets are used to validate the performance of SUC-MSDN. Experimental results show that SUC-MSDN outperforms benchmark methods and significantly enhances the accuracy of land cover mapping.

This paper proposes a novel deep learning-based framework for hyperspectral band selection, named Cross-Scene Self-Representative Network (CSSRnet). The proposed method leverages the rich labels of the source domain (SD) to guide the band selection in the target domain (TD). To our knowledge, CSSRnet is the first deep learning-based solution for cross-scene hyperspectral band selection. First, the CSSRnet employs contextual attention mechanism to capture the latent features of SD and TD. It combines the self-attention mechanism with convolutional operations to capture static and dynamic contextual information. Then, the self-representative layer provides the self-representative coefficient of SD and TD. Subsequently, the maximum mean difference is utilized to align the self-representative coefficients of both SD and TD. To enhance the representativeness and precision of these coefficients, we introduce different tasks for the SD and TD branches. Finally, a suitable band subset is selected based on a ranking method that evaluates each band’s importance by considering its self-representative coefficient matrix. Experiments are carried out to assess the efficacy of CSSRnet. These experiments focus on evaluating classification accuracy across various cross-scene datasets, the utility of cross-scene concepts, and the practical application in coastal wetland. Experimental results confirm the effectiveness of CSSRnet.

Hyperspectral image (HSI) change detection (CD) has gained significant attention in the field of remote sensing. Current CD methods typically extract features based on spatial or spectral correlations between bitemporal HSIs, which often overlook the difference information, leading to a decrease in CD accuracy. Furthermore, these algorithms do not fully consider the alignment of features between images across different channel and spatial dimensions. To tackle these issues, we propose a novel approach called the differential information-enhanced feature exchange network (DIEFEN) for HSI CD, which leverages the difference information between images and enhances the alignment of bitemporal image features to improve CD accuracy. Specifically, an enhanced differential multihead attention (EDMA) module is proposed to utilize difference information to guide the feature aggregation of bitemporal images, effectively highlighting changing pixels and suppressing unchanging pixels. A feature focus and long-range attention (FFLA) module is designed to extract local and global features, and a channel-spatial interaction (CSI) module is constructed to align features and mitigate the impact of noise. Experimental results on three HSI CD datasets demonstrate that the proposed DIEFEN method outperforms several state-of-the-art methods. The source code of the proposed DIEFEN is released athttps://github.com/creativeXin/DIEFEN.

Currently, deep learning methods represented by convolutional neural networks (CNNs) or Transformers are of great interest in hyperspectral image (HSI) classification. And recent works show that hybrid models using CNN and Transformer modules are expected to achieve better performance than when they are used alone. However, these hybrid models applied to HSI classification consider the combination of 2D CNN and Transformer, which makes the models have high computational complexity. And the information of multiple spectral dimensions different from ordinary RGB images has not been fully excavated. Based on this, we propose DBCTNet, a double branch Convolution-Transformer network. Specifically, a MSpeFE module is used for multiscale spectral feature extraction at the early stage of the proposed network. Then a ConvTE block is designed to improve the original Transformer encoder, where a Conv spectral projection unit and a convolutional multihead self-attention (CMHSA) unit are proposed to extract spatial and global spectral features. A double branch module is further built based on 3D CNN and ConvTE. This module can fully integrate spatial and local-global spectral features, while also having low computational complexity. Experiment results on four public datasets, Pavia University, Houston, WHU-Hi-LongKou and HuangHeKou, show that DBCTNet achieves satisfactory performance with a small number of parameters and relatively excellent efficiency compared to nine other networks. The implement of DBCTNet will be available publicly at https://github.com/xurui-joei/DBCTNet.

Graph convolutional networks (GCNs) have attracted significant attention in hyperspectral image (HSI) change detection (CD) due to their capability to perform shape-adaptive convolutions and capture complex patterns within HSIs. Existing GCN-based methods typically preprocess bitemporal HSIs into graphs using a specific superpixel segmentation. However, this preprocessing step limits the modeling of spatial topologies to a fixed scale. Besides, these methods do not consider distribution shifts between bitemporal HSIs. To overcome these limitations, this article proposes a fine–coarse–fine progressive graph framework with distribution alignment (FCFDA) to learn progressive features across multilevel graphs for HSI-CD. Specifically, for each bitemporal HSI, we generate multiple hierarchical segmentations ranging from fine to coarse by gradually merging neighboring superpixels and subsequently transforming these segmentations into multilevel graphs. Second, instead of simply concatenating features from different hierarchies, FCFDA integrates them progressively from fine to coarse and then back to fine, generating subtle features tailored to the pixel-wise CD task. Finally, an effective distribution alignment (DA) method is designed to align the feature space of the bitemporal HSIs, thus mitigating the adverse effects of distribution shifts. Experiments conducted on real HSI-CD datasets demonstrate the effectiveness and superiority of the FCFDA.

Graph neural networks (GNNs) have garnered significant attention in hyperspectral image (HSI) change detection (CD). However, existing GNN-based methods extract features by aggregating neighborhood information, which is essentially a low-pass Laplacian smoothing operation and tends to diminish change information between bitemporal HSIs. In addition, these methods rely on fixed hand-crafted graphs, and thus cannot capture complex structures of HSIs well. To address these deficiencies, this paper develops a Sharpening-Aware Graph Network (SAGN) for achieving high-quality HSI CD. Firstly, to counteract the weakening of differences caused by Laplacian smoothing, this paper proposes a novel Laplacian sharpening-based graph convolution (LSGC) module to accentuate change information between bitemporal HSIs. Secondly, instead of using “similarity graphs”, this paper constructs untied “difference graphs” for bitemporal HSIs to model dissimilarities between changed pixels and their neighbors. The SAGN can dynamically update graph structures across all layers, aiming to further maximize the divergence. Finally, a joint loss function, incorporating modified cross-entropy loss and contrastive loss, is devised to enhance inter-class discrimination of learned features and alleviate the issues stemming from imbalanced labeled samples. Experiments on various HSI CD datasets demonstrate the effectiveness and superiority of the proposed SAGN.

Very limited training samples pose significant challenges for hyperspectral image (HSI) classification. To address this issue, small-sample learning methods based on classical machine learning or deep learning offer promising solutions. In this article, a novel two-stage learning-based small-sample classification framework is proposed for HSIs, termed edge-preserving features-based smooth ordering (EPFSO). In the proposed EPFSO, a self-training approach and two screening mechanisms are designed to iteratively learn newly labeled samples from a vast pool of unlabeled samples, thereby enhancing classification accuracies by incorporating these additional samples into the training set. The preprocessing step involves using edge-preserving filters to extract key features and generate low-dimensional feature images. Subsequently, all samples are ordered based on spectral similarity and spatial proximity, resulting in a smooth 1-D signal. In the case of limited labeled samples, a specialized self-training approach based on linear interpolation is utilized to iteratively learn newly labeled samples from unlabeled samples. This process continues until no further labeled samples are introduced, enabling gradual improvement in classification performance. In addition, two screening mechanisms are designed into the self-training process to strike a balance between the reliability and quantity of newly labeled samples. Finally, once a sufficient number of training samples are available, a majority voting mechanism is employed to efficiently classify the remaining samples. Experimental results on three open HSI datasets demonstrate that the proposed EPFSO framework outperforms several state-of-the-art methods, including six deep learning approaches. This validates the attractiveness of using EPFSO to address the challenges associated with limited labeled samples.

In recent years, few-shot learning (FSL) has made significant progress in hyperspectral image classification (HSIC) by transferring metaknowledge from a labeled source domain to a target domain with very limited labeled samples. Considering that natural images have rich spatial texture information, heterogeneous FSL (HFSL) by using natural images as the source domain and hyperspectral image (HSI) as the target domain has shown excellent performance. However, some problems also exist in the HFSL, such as poor generalization ability from natural images to HSIs, prototype instability due to limited labeled samples, and domain shift between different types of images. To address these problems, we propose a mutual information enhancement FSL (MIEFSL) method for HSIC, which mainly contains three modules, i.e., mutual information enhancement (MIE), intradomain prototype rectification (IPR), and interdomain distribution alignment (IDA). In order to improve the generalization ability of the network and preserve the raw data information as much as possible, an MIE module is designed to maximize the mutual information (MI) between the support set samples and their corresponding masked samples. To stabilize the prototypes, an IPR module is constructed through a distribution expansion strategy. In addition, to alleviate domain shifts between different types of images, an IDA is performed between source and target domains. Experimental results demonstrate that the proposed MIEFSL outperforms existing state-of-the-art FSL methods and achieves the overall accuracy (OA) of 78.34%, 90.31%, and 91.72% on Indian Pines (IP), University of Pavia (UP), and Salinas (SA) in the case of only five labeled samples, respectively.

Hyperspectral image (HSI) classification is an essential task in remote sensing with substantial practical significance. However, most existing convolutional neural network (CNN)-based classification methods focus only on local spatial features while neglecting global spectral dependencies. Meanwhile, Transformer-based methods exhibit robust capabilities for global spectral feature modeling but struggle to extract local spatial features effectively. To fully exploit the local spatial feature extraction capabilities of CNN-based networks and the global spectral feature extraction capabilities of Transformer-based networks, this paper proposes a dual-branch convolutional Transformer method with efficient interactive self-attention for hyperspectral image classification, namely the dual-branch convolutional Transformer network (DCTN), which can aggregate local and global spatial-spectral features fully. Specifically, DCTN includes two core modules: the spatial-spectral fusion projection module and the efficient interactive self-attention module. The former utilizes 3D convolution with adaptive pooling and 2D group convolution with residual connection to parallel extract fused and grouped spatial-spectral features, respectively. The latter performs efficient interactive self-attention across height, width and spectral dimensions, enabling deep fusion of spatial-spectral features. Extensive experiments on three real HSI datasets demonstrate that the proposed DCTN method outperforms existing classification methods, yielding state-of-the-art classification performance. The code is available at https://github.com/AllFever/DeepHyperX-DCTN for reproducibility.

Deep learning plays a crucial role in hyperspectral image (HSI) classification, with the Transformer being highly favored by researchers due to its exceptional ability to model long-range dependencies. However, the Transformer necessitates a substantial amount of labeled training samples to train its numerous parameters, exacerbating the challenge of training an effective HSI classification Transformer model, particularly given the inherent scarcity of HSI data. Therefore, we propose a novel method for HSI classification, termed multiscale spatial-spectral masked Transformer with local contrastive learning (MSMT-LCL). This method consists of two stages: self-supervised pretraining and supervised fine-tuning. Initially, we utilize the multiscale augmented feature mapping module (MAFM) to project original HSI data into two mixed-scale feature maps, which are then separately fed into two masked Transformer branches for reconstruction. To facilitate the model in learning the dependency relationships between central pixel land-cover information and neighboring land cover, we introduce a novel mask strategy based on center-patch. Furthermore, in the pretraining stage, we integrate local contrastive learning (LCL) to enable the model to focus on local center information at varying scales. Upon completion of pretraining, the network undergoes fine-tuning to obtain feature maps at two different scales. Subsequently, we devise a novel adaptive multiscale feature fusion module (AMFM) to adaptively aggregate these two features and produce the final classification results. Extensive experiments on three real datasets demonstrate the superiority of our proposed MSMT-LCL method over several state-of-the-art HSI classification methods.

The convolutional sparse coding (CSC) can learn shift-invariant convolution kernels. In deep convolutional neural networks, it takes a lot of time to train the convolution kernels. In this letter, a deep two-stage CSC network (DTCSCNet) is proposed, which can be used to simultaneously extract spatial features and spectral features from hyperspectral image (HSI) without back propagation and fine-tuning process, thus saving a lot of time. Furthermore, to further improve the performance of the network, we incorporate multiscale information. After deep feature extraction using DTCSCNet, we further investigate the classification performance of different classifiers on the extracted features. Experimental results show that the proposed method can obtain better classification performance compared with some closely related HSI classification methods.

In recent years, with the continuous development of deep learning (DL), neural networks have demonstrated good results in large-sample hyperspectral image (HSI) classification. However, in practice, labels are often limited. In order to use fewer labeled samples without degrading the classification performance, this letter proposes a new semi-supervised classification method named prototype and active learning network (PALN), which integrates DL, active learning (AL) and prototype learning (PL) into a framework. After training the DL network with a small number of available labels, samples with high uncertainty are selected by AL to assign true labels, while samples more similar with prototypes are chosen by PL with their pseudo labels, and all selected samples are appended to the training set for the next training. Compared with existing classification methods, our method achieves good performance on two hyperspectral datasets.

Deep learning methods face significant challenges in practical cross-scene classification tasks of hyperspectral images, primarily due to the difficulty of acquiring labels and the issue of inconsistent distribution caused by spectral drift. To tackle the above issues, we propose a deep dynamic adaptation network based on joint correlation alignment (DDAN-JCA) for cross-scene hyperspectral image classification. First, the dual-channel residual network (DCRN) and the attention mechanism module (AMM) are employed to extract spatial-spectral joint features from both source domain and target domain. Then, the method of correlation alignment (CORAL) is employed to minimize the marginal distribution discrepancy between two domains and further reduce the conditional distribution discrepancy of each class. Finally, a dynamic distribution adaptation strategy is used to dynamically adjust the importance of marginal distribution and conditional distribution by using a balance factor. DDAN-JCA can achieve unsupervised classification without using target labels. The performance of DDAN-JCA has been validated using three hyperspectral datasets, and the experimental results demonstrate that DDAN-JCA significantly enhances classification accuracy and exhibits greater robustness compared to state-of-the-art methods.

Deep learning has emerged as a powerful method for hyperspectral image (HSI) classification. However, a significant prerequisite for HSI classification using deep learning is enough labeled samples, which is both time-consuming and labor-intensive. Yet, labeled samples are essential for training deep learning models. This paper proposes an HSI classification method based on the self-supervised learning of spectral masking (SSLSM). The method mainly includes two steps: self-supervised pre-training and fine-tuning. First, considering the rich spectral information of HSI, we propose masked spectral reconstruction as the pretext task. The unmasked data is input into the encoder and decoder sequentially, which are composed of a multi-layer transformer, for feature learning for masked spectral reconstruction. Second, we use reference samples to fine-tune the network, and the encoder and decoder are innovatively cascaded for deep semantic feature extraction, which can further improve the ability of feature extraction in the downstream classification tasks. Experiment results show that, compared with other methods, the SSLSM obtains the highest classification accuracy of 96.52%, 97.03%, and 96.70% on the Indian Pines dataset, Pavia University dataset, and Yancheng Wetlands dataset, respectively. Our method can also be applied to other HSI datasets, and the codes will be available from https://github.com/CIRSM-GRoup/2023-TGRS-SSLSM.

Deep learning has been extensively used for hyperspectral image (HSI) classification with significant success, but the classification of high-dimensional HSI datasets with a limited amount of labeled samples is still a great challenge. Few-shot learning (FSL) has shown excellent performance in solving small-sample classification problems. However, most of the existing FSL methods usually suffer from the prototype instability and domain shift. In order to address these problems, this paper proposes a category-specific prototype self-refinement contrastive learning (CPSRCL) method for cross-domain FSL of HSIs. Our method uses a supervised contrastive learning (SCL) strategy to promote intra-class compactness and inter-class dispersion of features in the metric space. To stabilize and refine the prototypes of the support set, a category-specific prototype self-refinement (CSPSR) module is designed to adaptively learn different updating rules for different category prototypes using rich labeled information in the query set. Furthermore, a local discriminative domain adaptation (LDDA) method is constructed to align the global distribution between source and target domains while preserving domain-specific discriminative information. Experimental results on four public HSI datasets demonstrate that CPSRCL outperforms existing FSL and deep learning methods for HSI classification.

Recently, prototypical network based few-shot learning (FSL) has been introduced for small-sample hyperspectral image (HSI) classification and shown good performance. However, existing prototypical-based FSL methods have two problems: prototype instability and domain shift between training and testing datasets. To solve these problems, we propose a refined prototypical contrastive learning network for few-shot learning (RPCL-FSL) in this paper, which incorporates supervised contrastive learning and FSL into an end-to-end network to perform small-sample HSI classification. To stabilize and refine the prototypes, RPCL-FSL imposes triple constraints on prototypes of the support set, i.e., contrastive learning (CL), self-calibration (SC) and cross-calibration (CC) based constraints. The CL module imposes internal constraint on the prototypes aiming to directly improve the prototypes using support set samples in the CL framework, and the SC and CC modules impose external constraints on the prototypes by using the prediction loss of support set samples and the query set prototypes, respectively. To alleviate domain shift in the FSL, a fusion training strategy is designed to reduce the feature differences between training and testing datasets. Experimental results on three HSI datasets demonstrate that the proposed RPCL-FSL outperforms existing state-of-the-art deep learning and FSL methods.

Cross-scene hyperspectral image classification (HSIC) is a challenging topic in remote sensing, especially when there are no labels in target domain. Domain adaptation (DA) techniques for cross-scene HSIC aim to label a target domain by associating it with a labeled source domain. Most existing DA methods learn domain-invariant features by reducing feature distance across domains. Recently, contrastive learning has shown excellent performance in computer vision tasks, but there is little or no research on the performance of cross-scene HSIC. Considering that its idea is similar to reducing feature distance, this paper attempts to explore whether contrastive learning can achieve cross-scene HSIC. In this work, an instance-to-instance contrastive learning framework based on category matching (CLCM) is designed. The main idea is to take the category information as the premise in the feature space, regard the source sample as an anchor, and find its positive and negative matching samples across domains. The instance-level discriminative feature embeddings are learned through positive matching pairs attracting each other and negative matching pairs repelling each other. Among them, the target label is a pseudo-label. To further improve the quality of contrastive learning, it is considered to focus on extracting the spectral-spatial features of HSI to more accurately represent semantic information. Simultaneously, high-confidence target samples are screened to update the network. Three DA tasks confirm the effectiveness and feature discriminativeness of CLCM, while also providing new ideas for cross-scene image classification.

Convolutional neural network (CNN) has achieved great success in hyperspectral image (HSI) classification. However, the local receptive field of CNN leads to the drawback in extracting long-distance features. Transformer has excellent global modeling ability and shows good performance for HSI classification. The existing Transformer-based methods usually ignore a problem that the spatial information varies under different channels. To well describe the cross-channel dependencies, a cross-channel dynamic spatial-spectral fusion transformer (CDSFT) is proposed in this article. In the proposed CDSFT, the multi-scale and multi-channel features are extracted and then cross-channel global features are extracted through transpose multi-head self-attention (TMHSA). Next, a dynamic feature enhancement module and a spectral spatial position attention module are designed to extract and enhance spectral-spatial joint features for classification. Experimental results on three well-known HSI datasets demonstrate the effectiveness of the proposed CDSFT method.

Image registration aims to eliminate the geometric deviation between multi-source data with the same range, and to promote the collaborative application of data. In recent years, spaceborne hyperspectral (HS) and multispectral (MS) data have been widely used in Earth observation. However, the difference in the number of bands, spatial resolution, and spectral resolution puts forward higher requirements on the registration algorithm. The key to HS and MS image registration is to extract more common key points, weaken and eliminate the difference of radiation and spatial texture information to build superior descriptors, and achieve high-precision matching of key points. This paper introduces a new robust HS and MS registration method based on common deep feature subspaces. We first construct the common deep feature subspaces extraction network to extract consistent edge features and common subspace images of the image pair. Then, Harris algorithm is used to extract key points from consistent edge features between images, which reduces the impact of spatial resolution differences between images. Besides, the SIFT descriptor and subspace images are used to describe key points, which reduces the impact of radiation differences between images. Finally, Euclidean distance is used for the initial matching of key points, and the affine matrix is calculated after the outliers are eliminated, and image registration is performed. We perform experiments on spaceborne HS and MS datasets of different spatial resolutions and comparisons with state-of-the-art methods. Experimental results show that our method can obtain satisfactory registration results and is robust.

Due to sensor design limitations and the influence of weather factors, it is currently challenging to obtain remote sensing images with high temporal, spatial, and spectral resolution. Spatial-temporal-spectral fusion aims to integrate the temporal, spatial, and spectral information from multiple sources of remote sensing images to reconstruct a remote sensing image with high temporal, spatial, and spectral resolution. Existing methods typically require at least three types of data to achieve spatial-temporal-spectral fusion. However, acquiring remote sensing data observed at the same time poses significant difficulties. The major challenge lies in effectively utilizing hyperspectral images with low spatial and temporal resolution and multispectral images with high temporal and spatial resolution to reconstruct remote sensing images with high temporal, spatial, and spectral resolution. To address the aforementioned issues, we propose a novel unsupervised 3D tensor subspace decomposition network. Our method incorporates the theory of 3D tensor subspace decomposition, utilizing a 3D hyperspectral/multispectral tensor subspace extraction network to predict the hyperspectral tensor subspace features with low spatial resolution missing at other times (To better understand, the missing moment is defined as time 2). Subsequently, the 3D hyperspectral tensor subspace reconstruction network is employed along with the time 2 hyperspectral tensor subspace features with low spatial resolution and the time 2 multispectral image to reconstruct the time 2 hyperspectral image with high spatial resolution. In the experiment, we utilize three simulated datasets and two real datasets to evaluate the fusion performance of our proposed method. The results demonstrate that our method achieves high-quality fusion results and exhibits comparable performance, and has robustness and practicality.

Spatial-temporal-spectral fusion (STSF) of remote sensing imagery can produce data with the highest spatial and spectral resolution, only as well as fine temporal resolution, by integrating images with complementary information in both the temporal and spectral domains. Accuracy of temporal variation is an important guarantee for achieving fidelity fusion in STSF. However, current STSF methods estimate the temporal variation only by utilizing the temporal variation between observed multispectral image (MSI) and the relationship between MSI and hyperspectral image (HSI), which is difficult to obtain accurate temporal variation. To address this problem, this paper proposes a coupled temporal variation information estimation and resolution enhancement for remote sensing image spatial-temporal-spectral fusion (CTVRE-STSF). The temporal variation information estimation model estimates the temporal variation of the target image, while the resolution enhancement model provides additional constraints for estimating the temporal variation. For the temporal variation information reconstruction model, we build a temporal variation information estimation based on a generalized linear mixed model and use the temporal variation between MSIs. In addition, a resolution enhancement model is constructed to estimate the temporal variation of the target image by incorporating relevant prior knowledge. The introduction of the resolution enhancement model in the prior provides additional constraints on the estimation of the temporal variation high-dimensional information, thus facilitating the resolution improvement. Experimental results on two real datasets demonstrate the effectiveness and superiority of our proposed method over current state-of-the-art methods, especially in terms of spectral fidelity.

Graph convolutional network (GCN) has recently attracted great attention in hyperspectral image (HSI) classification due to its strong ability to aggregate information of neighborhood nodes. However, a GCN model usually suffers from the over-smoothing problem (i.e., all nodes’ representations converge to a stationary point) when the number of GCN layers is increased. In addition, GCNs always work on superpixel-level nodes to reduce computational cost, so pixel-level features cannot be well captured. To deal with these problems, a novel two-branch deeper GCN (TBDGCN) is proposed to combine the advantages of superpixel-based GCN and pixel-based CNN, which can simultaneously extract superpixel-level and pixel-level features of HSIs. In the GCN branch, a GCN module with the DropEdge technique and residual connection is designed to alleviate over-smoothing and over-fitting problem, which results in a deeper network structure with more than ten layers. In the CNN branch, to capture spatial positional information and channel information, a mixed attention mechanism is constructed to extract attention-based spectral-spatial features. The features of the GCN and CNN branches are then fused for classification. Experimental results on three benchmark HSI data sets show that the classification performance of our TBDGCN is better than existing GCN models especially in the case of small sample size.

The improvement of temporal resolution of hyperspectral (HS) data is a fundamental and challenging problem. In this paper, we propose a Temporal-Spectral fusion method based on Generative Adversarial Network (TSF-GAN). First, the generator is used to train the nonlinear relationship between multispectral (MS) and HS data pairs at time T1 and T3, and we map the relationship to the MS data at T2 to obtain the HS data. Second, the discriminator is used to identify whether the differential image of HS data at different times is consistent with that of MS data, and whether the HS data at time T2 after spectral down-sampling is consistent with that of MS data at time T2. Preliminary experimental results demonstrate that the proposed TSF-GAN achieves comparative fidelity and has strong practicability.

Domain adaptation (DA) aims to use a well-labeled source domain to predict the labels of the unlabeled or poor-labeled target domain. Most of the existing DA methods focus on the use of feature-level or sample-level information. Recent studies have shown that domain discriminative information is also important for classification. To jointly exploit feature-level information and discriminative information, a new DA method called distribution alignment and discriminative feature learning (DADFL) is proposed for hyperspectral image (HSI) classification in this letter. DADFL incorporates category-discriminative information preservation and structured prediction (SP)-based pseudolabeling into a unified framework to simultaneously reduce distribution and subspace differences between domains. Experimental results on three hyperspectral DA tasks show that the classification performance of the proposed DADFL is better than that of existing DA methods.

Nonnegative matrix factorization (NMF) is a widely used hyperspectral unmixing model which decomposes a known hyperspectral data matrix into two unknown matrices, i.e., endmember matrix and abundance matrix. Due to the use of least-squares loss, the NMF model is usually sensitive to noise or outliers. To improve its robustness, we introduce a general robust loss function to replace the traditional least-squares loss and propose a general loss-based NMF (GLNMF) model for hyperspectral unmixing in this letter. The general loss function is a superset of many common robust loss functions and is suitable for handling different types of noise. Experimental results on simulated and real hyperspectral data sets demonstrate that our GLNMF model is more accurate and robust than existing NMF methods.

This letter presents a multiscale low-rank decomposition (MSLRD) method to extract multiscale spatial structures from hyperspectral images. The MSLRD assumes that ground objects have divergent characteristics in changing spatial scales. It decomposes each band image into a series of block-wise matrices, where these low-rank blocks take detailed spatial structures at multiple scales. It formulates the low-rank matrix decomposition problem into minimizing the ranks of all block matrices and adopts the alternative direction of the multiplier method to optimize it. Experiments on Indian Pines and Pavia University data sets show that the MSLRD can greatly improve the classification performance of regular classification on spectral features (i.e., all bands) and perform better than five state-of-the-art spatial feature extraction methods.

The maintenance of spectral variability between subclass objects and the relationship between hyperspectral (HS) bands have been a fundamental but challenging problem for fusing low spatial resolution (LR) HS and high spatial resolution (HR) multispectral (MS) images. This article presents a locally optimized image segmentation fusion (LOISF) framework for HS super-resolution reconstruction. First, LR HS and HR MS are clustered and segmented, and the label attributes of the segmented objects are identified by the prior information. Then, a novel joint fusion model for different typical ground objects is constructed based on spectral unmixing. The fusion problem is formulated mathematically as a convex optimization of a Frobenius norm, which includes spatial, spectral, and index constraints, with an alternating-directions’ optimization featuring linearization providing the solution. Experimental results demonstrate that the proposed LOISF preserves both spatial details and texture, achieving high spectral fidelity, and yielding significantly improved image quality compared to other state-of-the-art fusion methods.

Most hyperspectral image (HSI) data exist in the form of tensor; the tensor representation preserves the potential spatial–spectral structure information compared with the vector representation, which can help improve the classification performance of HSI. In this article, a deep high-order tensor convolutional sparse coding (CSC) model is proposed, which can be used to train deep high-order filters. Based on the deep high-order tensor CSC model, a deep feature extraction network (DHTCSCNet) is constructed, which is used for feature extraction of HSIs. By combining the spectral–spatial feature and the features extracted by the proposed DHTCSCNet at each layer, a combined feature that incorporates shallow, deep, spectral, and spatial features can be obtained. Then, the graph-based learning (GSL) methods are used to classify the combined feature. Experimental results show that the DHTCSCNet can obtain better classification performance compared with other HSI classification methods.

Deep learning methods have shown considerable potential for hyperspectral image (HSI) classification, which can achieve high accuracy compared with traditional methods. However, they often need a large number of training samples and have a lot of parameters and high computational overhead. To solve these problems, this article proposes new network architecture, LiteDepthwiseNet, for HSI classification. Based on 3-D depthwise convolution, LiteDepthwiseNet can decompose standard convolution into depthwise convolution and pointwise convolution, which can achieve high classification performance with minimal parameters. Moreover, we remove the ReLU layer and batch normalization layer in the original 3-D depthwise convolution, which is likely to improve the overfitting phenomenon of the model on small-sized data sets. In addition, focal loss is used as the loss function to improve the model’s attention on difficult samples and unbalanced data, and its training performance is significantly better than that of cross-entropy loss or balanced cross-entropy loss. Experiment results on five benchmark hyperspectral data sets show that LiteDepthwiseNet achieves state-of-the-art performance with a very small number of parameters and low computational cost.

This article proposes a dual global–local attention network (DGLAnet), which is an end-to-end unsupervised band selection (UBS) method that fully utilizes spatial and spectral information in both global and local aspects. The DGLAnet assumes that BS can be realized using the hyperspectral image (HSI) reconstruction process. First, the DGLAnet implements a dual attention module to obtain spatial–spectral and global–local features to reweight the HSI data. It adopts bi-directional relations to grasp spatial and spectral features from a global perspective. Meanwhile, the DGLAnet extracts local features through max-pooling and mean-pooling and then merges them via the convolution operation. Global–local features are utilized to learn attention to recalibrate the original data, and the reconstruction module is adopted to restore the original image from the reweighted HSI data. Finally, a proper band subset is selected by the constructed band evaluation index. Experiments on three hyperspectral data show that the DGLAnet outperforms other state-of-the-art methods and uses all bands with a lower computational cost.

Recent studies have shown that deep domain adaptation (DA) techniques have good performance on cross-domain hyperspectral image (HSI) classification problems. However, most existing deep HSI DA approaches directly use deep networks to extract features from the data, which ignores the detailed information of HSI in spectral and spatial dimensions. To effectively exploit the spectral–spatial joint information for DA of HSIs, we propose a two-branch attention adversarial DA (TAADA) network in this article. In the TAADA network, a two-branch feature extraction (TBFE) subnetwork is first designed as a generator to extract the attention-based spectral–spatial features. Then, a discriminator based on two classifiers with the multilayer FC-BN-ReLU-Dropout structure is constructed. Based on adversarial learning between the generator and the discriminator, the ability of discriminative feature extraction and cross-domain classification is improved simultaneously. Finally, the TAADA network can adjust the distribution between the source and target domains and extract domain-invariant features. Experimental results on three cross-scene HSI classification tasks show that our proposed TAADA outperforms some existing DA methods.

Nowadays, convolutional neural networks (CNNs) are widely used in the field of hyperspectral image (HSI) classification. However, a major feature of HSIs is their rich spectral–spatial information with hundreds of continuous bands. This inevitably incurs the problems of high computational cost for network optimization and high interredundancy in the convolution kernels. To solve these problems, in this article, we rethink HSIs from the spectral perspective and introduce a lightweight operator called involution, which can effectively solve the above limitations. Different from traditional convolution kernels, the involution kernels pay more attention to the features of the channels but usually ignore the spatial features in the receptive field. To incorporate both spatial and spectral information, we construct a dual-pooling layer and design a novel involution-2D operator and its more lightweight version, involution-1D operator. Finally, an efficient spatial–spectral interaction network (ESSINet) for HSI classification is proposed based on these two new operators, which can make the spatial–spectral information in HSIs interact more closely. Extensive experimental results on four public datasets demonstrate the effectiveness and efficiency of the proposed ESSINet over some state-of-the-art CNN-based networks.

Hyperspectral unmixing is a significant task in remote sensing image analysis. Existing learning-based methods for hyperspectral unmixing generally are in the form of an autoencoder and take geometric distances, such as spectral angle distance (SAD) as loss functions. These methods ignored the distribution similarity between the observation and the reconstruction, which might help improve the unmixing performance. Besides, the autoencoder is trained by directly comparing the difference between the observation and the reconstruction, and the difference between their features has been neglected. Based on the above considerations, we propose a joint metric neural network for hyperspectral unmixing, by introducing theWasserstein distanceandfeature matchingas regularization terms and SAD as the underlying loss. The proposed neural network consists of two parts: an autoencoder is used for endmember extraction and abundance estimation, while a discriminator is used to compute the Wasserstein distance. The Wasserstein distance can stably provide useful gradient information that promotes the autoencoder to reach a solution with better unmixing performance. The feature matching is adapted to an intermediate layer of the discriminator for enforcing the features of the observation and the reconstruction to be equal, which can lead to further improvement of the unmixing performance. The model analysis and the regularization parameter analysis are conducted to demonstrate the effectiveness of our method. Experimental results on four real-world hyperspectral data sets show that our method outperforms the state-of-the-art methods, especially in terms of abundance estimation.

The nonoverlapped spectrum range between low spatial resolution (LR) hyperspectral (HS) and high spatial resolution (HR) multispectral (MS) images has been a fundamental but challenging problem for MS/HS fusion. The spectrum of HS data is generally 400–2500 nm, and the spectrum of MS data is generally 400–900 nm; how to obtain the high-fidelity HR HS fused image within the whole spectrum of 400–2500 nm? In this article, we proposed a band divide-and-conquer framework (BDCF) to solve the problem, by comprehensively considering spectral fidelity, spatial enhancement, and computational efficiency. First, the spectral bands of HS were divided into overlapped and nonoverlapped bands according to the spectral response between HS and MS. Then, a novel improved component substitution (CS)-based method by combing neural network was proposed to fuse the overlapped bands of LR HS. Then, a mapping-based method with the neural network was presented to construct the complicated nonlinear relationship between overlapped and nonoverlapped bands of the original LR HS data. The trained network was mapped to the fused overlapped HR HS bands to estimate the nonoverlapped HR HS bands. Experimental results on two simulated data sets and two realistic data sets of Gaofen (GF)-5 LR HS, GF-1 MS, and Sentinel-2A MS show that the proposed BDCF has superior performance in both high spectral fidelity and sharp spatial details, and it obtained competitive fusion behaviors compared with other state-of-the-art methods. Moreover, BDCF has relatively higher computational efficiency than optimal solution-based methods and deep learning-based fusion methods.

Fusing low spatial resolution (LR) hyperspectral (HS) data and high spatial resolution (HR) multispectral (MS) data aims to obtain HR HS data. However, due to bad weather and the aging of sensor equipment, HS images usually contain a lot of noise, e.g., Gaussian noise, strip noise, and mixed noise, which would make the fused image have low quality. To solve this problem, we propose the multiscale low-rank deep back projection fusion network (MLR-DBPFN). First, HS and MS are superimposed, and multiscale spectral features of the stacked image are extracted through multiscale low-rank decomposition and convolution operation, which effectively removes noisy spectral features. Second, the upsampling and downsampling network mechanisms are used to extract the multiscale spatial features from each layer of spectral features. Finally, the multiscale spectral features and multiscale spatial features are combined for network training, and the weight of the noisy spectrum features is reduced through the network feedback mechanism, which suppresses the noisy spectrum and improves the noisy HS fusion performance. Experimental results on datasets of different noise demonstrate that MLR-DBPFN has superior spatial and spectral fidelity, comparative fusion quality, and robust antinoise performance compared with state-of-the-art methods.

This article proposes a multiscale spectral features graph fusion (MSFGF) method for selecting proper hyperspectral bands. The MSFGF regards that the selected bands should reflect diagnostic spectral information of ground objects at different scales, and it explores band selection from the aspect of multiple spatial scales. First, it adopts the multiscale low-rank decomposition (MSLRD) model to find multiscale spectral features of different ground objects. The model considers divergent spatial structures or spatial correlations of ground objects at different scales, and factorizes the hyperspectral data cube into a series of low-rank block-wise data cubes, where the blocks take spatial structures of different ground objects at increasing scales. Second, the MSFGF presents the multiscale sparse spectral clustering (MSSC) model to fuse the separate connected graphs of multiscale spectral features into a consensus graph. The consensus graph combines the complementary information of multiscale spectral features and helps to reveal the intrinsic clustering structure of all spectral bands. Finally, the MSFGF utilizes spectral clustering to find clusters from the consensus graph and selects representative bands. Experimental results on three widely used hyperspectral data prove the superiority of MSFGF in selecting bands, where it outperforms other seven state-of-the-art methods in classification with an acceptable computational cost.

The hyperspectral images (HSIs) classification technique has received widespread attention in the field of remote sensing. However, how to achieve satisfactory classification performance in the presence of a large amount of noise is still a problem worthy of consideration. In this article, a local correntropy matrix (LCEM)-based spatial–spectral feature representation method is proposed for HSI classification. Motivated by the successful application of information-theoretic learning (ITL), we propose to adopt correntropy matrix to represent the spatial–spectral features of HSI. Specifically, the dimension reduction is first performed on the original hyperspectral data. Then, for each pixel, we select its local neighbors within a sliding window using cosine distance for the construction of the LCEM. In this way, each pixel can be characterized as an LCEM. Finally, all the correntropy matrices are fed into a support vector machine (SVM) for final classification. In addition, we also propose a novel way to determine the size of the local window based on standard deviation. Because the LCEM as the feature descriptor can characterize discriminative spatial–spectral features, the proposed method has shown great interclass separability and intraclass compactness. Compared with other advanced approaches, the proposed LCEM method has achieved competitive performance in both evaluation indexes and visual effects, especially when the training size is very small.

Image fusion effectively solves the trade-off between spatial resolution, temporal resolution, and spectral resolution of remote sensing sensors. However, most of existing methods focus on the fusion of two of the spatial, temporal, and spectral metrics of remote sensing images. The few spatial-temporal-spectral fusion (STSF) methods available are mainly for fusing MODIS and Landsat images, which are not suitable for the characteristics of the spaceborne hyperspectral images with low temporal resolution, such as Hyperion, ZY-1 02D, and PRISMA. For this purpose, we proposed a novel generalized linear spectral mixing model for spatial-temporal-spectral fusion (GLMM-STSF). In the method, the GLMM is introduced into the STSF problem, and the temporal variations of images at different times are transferred to the endmember and abundance matrix variations of images for estimation. To the best of our knowledge, for the first time, the STSF task of remote sensing images is handled from the perspective of spectral unmixing. Compared with existing STSF fusion methods, our method targets the task of fusing spaceborne HSI with low temporal and spatial resolutions with multispectral image featured by high temporal and spatial resolutions. Taking the STSF of ZY-1 02D hyperspectral and Sentinel-2 multispectral real datasets as an example, comparisons with related state-of-the-art methods demonstrate that our proposed method achieves superior fusion performance.

Considering that multiple features can improve the classification performance as they contain diversity information of images, a multiple-feature latent space learning-based method is proposed for hyperspectral image (HSI) classification in this letter. In the proposed method, a latent space that contains diversity information of multiple features and transformation matrices between the latent space and features are both learned. Moreover, spatial information is used for labeling unlabeled samples in the classification. Experimental results on the Indian Pines and University of Pavia data sets demonstrate the effectiveness of the proposed method.

Due to the limitation of the satellite attitude measurement accuracy and the system servo control error of the payload scanning mechanism, an optimal use of Micro-Wave Radiation Imager (MWRI) observations requires high geolocation accuracy. In the operational system, the MWRI geolocation accuracy reaches 1 pixel, and there still exists room for improvement. In this article, we improve upon the coastline inflection point method (CIM) and propose to assign the accurate correspondence by employing a nonrigid point set registration method. First, the method identifies a set of latent variables to recognize outliers and then applies nonparametric geometric constraints to the correspondence asa prioridistribution. Second, the maximuma posteriori(MAP) estimation is applied by the expectation–maximization (EM) algorithm to obtain correct inliers. The comparison with other methods demonstrates that the proposed method can provide more accurate estimation of geolocation bias. In addition, the pixel error and changes in spacecraft attitude with the long-term geolocation data in FY-3C MWRI before and after correction were analyzed during the period from April 1 to August 30, 2018. The results have shown that the geolocation errors are reduced from [0.50, 0.60] pixels to [0.20, 0.33] pixels in the along- and cross-track directions after the attitude correction. In addition, the reduction of the standard deviation shows that the geolocation quality of MWRI is improved.

The presence of mixed pixels in the hyperspectral data makes unmixing to be a key step for many applications. Unsupervised unmixing needs to estimate the number of endmembers, their spectral signatures, and their abundances at each pixel. Since both endmember and abundance matrices are unknown, unsupervised unmixing can be considered as a blind source separation problem and can be solved by nonnegative matrix factorization (NMF). However, most of the existing NMF unmixing methods use a least-squares objective function that is sensitive to the noise and outliers. To deal with different types of noises in hyperspectral data, such as the noise in different bands (band noise), the noise in different pixels (pixel noise), and the noise in different elements of hyperspectral data matrix (element noise), we propose three self-paced learning based NMF (SpNMF) unmixing models in this article. The SpNMF models replace the least-squares loss in the standard NMF model with weighted least-squares losses and adopt a self-paced learning (SPL) strategy to learn the weights adaptively. In each iteration of SPL, atoms (bands or pixels or elements) with weight zero are considered as complex atoms and are excluded, while atoms with nonzero weights are considered as easy atoms and are included in the current unmixing model. By gradually enlarging the size of the current model set, SpNMF can select atoms from easy to complex. Usually, noisy or outlying atoms are complex atoms that are excluded from the unmixing model. Thus, SpNMF models are robust to noise and outliers. Experimental results on the simulated and two real hyperspectral data sets demonstrate that our proposed SpNMF methods are more accurate and robust than the existing NMF methods, especially in the case of heavy noise.

Cross-scene classification is one of the major challenges for hyperspectral image (HSI) classification, especially for target scenes without label samples. Most traditional domain adaptive methods learn a domain invariant subspace to reduce statistical shift while ignoring the fact that there may not exist a shared subspace when marginal distributions of source and target domains are very different. In addition, it is important for HSI classification to preserve discriminant information in the original space. To solve this issue, discriminative cooperative alignment (DCA) of subspace and distribution is proposed to cooperatively reduce the geometric and statistical shift. In the proposed framework, both geometrical and statistical alignments are considered to learn subspaces of the two domains with preserving discrimination information. Furthermore, a reconstruction constraint is imposed to enhance the robustness of subspace projection. Experimental results on three cross-scene HSI data sets demonstrate that the proposed DCA is significantly better than some state-of-the-art domain-adaptive approaches.

Due to the high dimensionality of hyperspectral images (HSIs), more training samples are needed in general for better classification performance. However, surface materials cannot always provide sufficient training samples in practice. HSI classification with small size training samples is still a challenging problem. Multiview learning is a feasible way to improve the classification accuracy in the case of small training samples by combining information from different views. This article proposes a new spatial window-based multiview intact feature learning method (SWMIFL) for HSI classification. In the proposed SWMIFL, multiple features that reflect different information of the original image are extracted and spatial windows are imposed on training samples to select unlabeled samples. Then, multiview intact feature learning is performed to learn the intact feature of the training and unlabeled samples. Considering that neighboring samples are likely to belong to the same class, labels of spatial neighboring samples are determined by two factors including the labels of training samples that locate in the spatial window and the labels learned from the intact feature. Finally, unlabeled samples that have same labels under these two factors are treated as new training samples. Experimental results demonstrate that the proposed SWMIFL-based classification method outperforms several well-known HSI classification methods on three real-world data sets.

Non-negative matrix factorization (NMF) is a classical hyperspectral unmixing model which minimizes the Euclidean distance between the hyperspectral data matrix and its low rank approximation (i.e., the product of endmember matrix and abundance matrix), and it fails when applied to noisy data because the loss function is sensitive to outliers. In this paper, we propose a Cauchy NMF (CauchyNMF) model for hyperspectral unmixing which uses a Cauchy loss function (CLF) to replace the traditional least-squares loss. Compared with the least-squares loss, CLF can penalize the noise term for suppressing the large noise mixed in the real data and thus is much more robust. Experimental results on simulated and real hyperspectral data sets demonstrate that our proposed CauchyNMF method is more accurate and robust than existing NMF methods, especially in the case of heavy noise.

Known as input in the numerical weather prediction (NWP) models, microwave radiation imager (MWRI) data have been widely distributed to the user community. Nevertheless, the current operational geolocation accuracy is still on the pixel scale due to the presence of geolocation uncertainty. In this letter, we propose a new method to estimate the geolocation errors in MWRI data. Compared to the traditional coastline inflection method (CIM), the proposed method has two innovations. First, we establish a surface fitting interpolation model by involving more observations to detect the coastline. Second, we employ the iterative closest point (ICP) algorithm to determine the correspondences between the detected coastline and the actual coastline. Simulated experimental results demonstrate that the proposed method can provide a more accurate geolocation error estimation than the CIM. By applying our method, we have processed an MWRI data set from January 1 to February 28 in 2016. The experimental results have shown that the operational FY-3C MWRI geolocation errors are 0.4813 and 0.4909 pixels in the along-track and cross-track directions, respectively, which can be significantly reduced to 0.1299 and 0.1497 pixels after the attitude correction. It means that the geolocation accuracy has an average improvement up to 70%.

This letter presents a correntropy-based sparse spectral clustering (CSSC) method to select proper bands of a hyperspectral image. The CSSC first constructs an affinity matrix with the correntropy measure which considers the nonlinear characteristics of hyperspectral bands and can suppress effects from noise or outliers in measuring band similarity. The CSSC imposes the sparsity and block diagonal constraint on spectral clustering, which can further improve band clustering performance. Bands are finally selected from each cluster on the connected graph. Experimental results on two widely used hyperspectral images show that the CSSC behaves better than spectral clustering and other several state-of-the-art methods in band selection.

This letter proposes a lateral-slice sparse tensor robust principal component analysis (LSSTRPCA) method to remove gross errors or outliers from hyperspectral images so as to promote the performance of subsequent classification. The LSSTRPCA assumes that a three-order hyperspectral tensor has a low-rank structure, and gross errors or outliers are sparsely scattered in a 2-D space (i.e., lateral-slice) of the tensor. It formulates a low-rank and sparse tensor decomposition problem into a convex problem and then implements the inexact augmented Lagrange multiplier method to solve it. The experiments on two hyperspectral data sets show that the LSSTRPCA can successfully remove outliers or gross errors and achieve higher accuracies than both the original robust principal component analysis (RPCA) and tensor robust principal component analysis (TRPCA).

This article presents a fast and latent low-rank subspace clustering (FLLRSC) method to select hyperspectral bands. The FLLRSC assumes that all the bands are sampled from a union of latent low-rank independent subspaces and formulates the self-representation property of all bands into a latent low-rank representation (LLRR) model. The assumption ensures sufficient sampling bands in representing low-rank subspaces of all bands and improves robustness to noise. The FLLRSC first implements the Hadamard random projections to reduce spatial dimensionality and lower the computational cost. It then adopts the inexact augmented Lagrange multiplier algorithm to optimize the LLRR program and estimates sparse coefficients of all the projected bands. After that, it employs a correntropy metric to measure the similarity between pairwise bands and constructs an affinity matrix based on sparse representation. The correntropy metric could better describe the nonlinear characteristics of hyperspectral bands and enhance the block-diagonal structure of the similarity matrix for correctly clustering all subspaces. The FLLRSC conducts spectral clustering on the connected graph denoted by the affinity matrix. The bands that are closest to their separate cluster centroids form the final band subset. Experimental results on three widely used hyperspectral data sets show that the FLLRSC performs better than the classical low-rank representation methods with higher classification accuracy at a low computational cost.

This paper proposes multiple-feature ideal regularized kernel for hyperspectral image classification, which offers the advantage of combining the complementary discriminative information among multiple features with the ideal regularized kernel. Four different types of features including spectral feature, shape feature (i.e., extended multiattribute profiles), local feature (i.e., local binary pattern), and global feature (i.e., Gabor feature) are investigated in this paper. Furthermore, a majority votingbased ensemble method combining different features is adopted to further increase classification performance. Experimental results demonstrate that the proposed method can provide superior performance than the state-of-the-art classifiers.

This paper proposes an unsupervised domain adaptation (DA) method called correlation alignment based on sparse matrix transform (CORAL-SMT) for hyperspectral image (HSI) classification. In CORAL-SMT, the covariance of source and target domain are constrained to have an eigen-decomposition that can be represented as a sparse matrix transform. Under maximum likelihood framework, based on greedy minimization strategy, the covariances can be efficiently estimated and are always positive definite. The proposed method is compared with some classical unsupervised domain adaptation methods. Experimental results on the City of Pavia hyperspectral data set demonstrate the effectiveness of CORAL-SMT.

Kernel joint sparse representation (KJSR) performs joint sparse representation in the feature space and has shown good performance for the hyperspectral image (HSI) classification. In order to distinguish spatial neighbouring pixels in the feature space, we propose two weighted KJSR (WKJSR) methods in this paper. The first one computes the weight directly based on the kernel similarity between neighbouring pixels. The second weighted scheme uses a nearest regularisation strategy to simultaneously optimise the weights of projected neighbouring pixels and joint sparse representation coefficients. The proposed WKJSR methods can exploit the similarities and differences among neighbouring pixels to obtain accurate weights for the joint sparse representation and classification. Experimental results on two benchmark HSI data sets demonstrate the effectiveness of the proposed methods.

The original manifold alignment (MA) approach is for semisupervised domain adaptation. Since the target prior information is difficult to obtain, we conduct it in an unsupervised manner, resulting in an unsupervised MA (UMA) method. This approach utilizes the probabilistic prediction results of target data to construct the cross-domain similarity matrix, which characterizes the relationships between domains and is used for alignment. Due to the spectral drift, the prediction results may not be accurate, and thus affect the alignment. We employed spatial filtering and overall centroid alignment method as two preprocessing strategies to improve the prediction results. Furthermore, per-class maximum mean discrepancy (MMD) constraint is introduced to the UMA to further improve the alignment performance. The proposed UMA_MMD algorithm is applied for the classification of remote sensing images, and the experimental results using hyperion multitemporal remote sensing images demonstrated the effectiveness of the proposed approach.

Domain adaptation, which aims at learning an accurate classifier for a new domain (target domain) using labeled information from an old domain (source domain), has shown promising value in remote sensing fields yet still been a challenging problem. In this letter, we focus on knowledge transfer between hyperspectral remotely sensed images in the context of land-cover classification under unsupervised setting where labeled samples are available only for the source image. Specifically, a discriminative transfer joint matching (DTJM) method is proposed, which matches source and target features in the kernel principal component analysis space by minimizing the empirical maximum mean discrepancy, performs instance reweighting by imposing an ℓ2,1-norm on the embedding matrix, and preserves the local manifold structure of data from different domains and meanwhile maximizes the dependence between the embedding and labels. The proposed approach is compared with some state-of-the-art feature extraction techniques with and without using label information of source data. Experimental results on two benchmark hypersepctral data sets show the effectiveness of the proposed DTJM.

In this paper, a self-paced joint sparse representation (SPJSR) model is proposed for the classification of hyperspectral images (HSIs). It replaces the least-squares (LS) loss in the standard joint sparse representation (JSR) model with a weighted LS loss and adopts a self-paced learning (SPL) strategy to learn the weights for neighboring pixels. Rather than predefining a weight vector in the existing weighted JSR methods, both the weight and sparse representation (SR) coefficient associated with neighboring pixels are optimized by an alternating iterative strategy. According to the nature of SPL, in each iteration, neighboring pixels with nonzero weights (i.e., easy pixels) are included for the joint SR of a testing pixel. With the increase of iterations, the model size (i.e., the number of selected neighboring pixels) is enlarged and more neighboring pixels from easy to complex are gradually added into the JSR learning process. After several iterations, the algorithm can be terminated to produce a desirable model that includes easy homogeneous pixels and excludes complex inhomogeneous pixels. Experimental results on two benchmark hyperspectral data sets demonstrate that our proposed SPJSR is more accurate and robust than existing JSR methods, especially in the case of heavy noise.

A joint sparse representation (JSR) method has shown superior performance for the classification of hyperspectral images (HSIs). However, it is prone to be affected by outliers in the HSI spatial neighborhood. In order to improve the robustness of JSR, we propose a maximum likelihood estimation (MLE)-based JSR (MLEJSR) model, which replaces the traditional quadratic loss function with an MLE-like estimator for measuring the joint approximation error. The MLE-like estimator is actually a function of coding residuals. Given some priors on the coding residuals, the MLEJSR model can be easily converted to an iteratively reweighted JSR problem. Choosing a reasonable weight function, the effect of inhomogeneous neighboring pixels or outliers can be dramatically reduced. We provide a theoretical analysis of MLEJSR from the viewpoint of recovery error and evaluate its empirical performance on three public hyperspectral data sets. Both the theoretical and experimental results demonstrate the effectiveness of our proposed MLEJSR method, especially in the case of large noise.

Incremental learning is one of the most effective methods of learning accumulated data and large-scale data. The newly increased samples of the previously known works on incremental learning are usually independent and identically distributed. To study how dependent sampling methods influence the learning ability of incremental support vector machines (ISVM) algorithm, in this paper we introduce an ISVM based on Markov resampling (MR-ISVM), and give the experimental research on the learning ability of the MR-ISVM algorithm. The experimental results indicate that the MR-ISVM algorithm has not only smaller misclassification rates and sparser of the obtained classifiers, but also less total time of sampling and training compared to ISVM based on randomly independent sampling. We also compare it with other ISVM algorithms.

Recent advances have shown a great potential to explore collaborative representations in hyperspectral imagery (HSI) classification, including sparse representations and joint collaborative representations. In this letter, we propose a weighted regularized collaborative representation optimized classifier (WRCROC) that makes use of multiple collaborative representations. It strikes a balance between an optimized weighted joint collaborative representation classifier, which essentially classifies a test sample to the class that minimizes the distance between the sample and its representation in the selected class, and a weighted regularized collaborative representation classifier, which actually assigns a test sample to the class that minimizes the distance between the sample and its collaborative components. The proposed WRCROC algorithm is tested on two benchmark HSI data sets. Experimental results demonstrate that the proposed algorithm performs better than existing representation-based classifiers.

Joint sparse representation (JSR) has been a popular technique for hyperspectral image classification, where a testing pixel and its spatial neighbors are simultaneously approximated by a sparse linear combination of all training samples, and the testing pixel is classified based on the joint reconstruction residual of each class. Due to the least-squares representation of the approximation error, the JSR model is usually sensitive to outliers, such as background, noisy pixels, and outlying bands. In order to eliminate such effects, we propose three correntropy-based robust JSR (RJSR) models, i.e., RJSR for handling pixel noise, RJSR for handling band noise, and RJSR for handling both pixel and band noise. The proposed RJSR models replace the traditional square of the Euclidean distance with the correntropy-based metric in measuring the joint approximation error. To solve the correntropy-based joint sparsity model, a half-quadratic optimization technique is developed to convert the original nonconvex and nonlinear optimization problem into an iteratively reweighted JSR problem. As a result, the optimization of our models can handle the noise in neighboring pixels and the noise in spectral bands. It can adaptively assign small weights to noisy pixels or bands and put more emphasis on noise-free pixels or bands. The experimental results using real and simulated data demonstrate the effectiveness of our models in comparison with the related state-of-the-art JSR models.

This paper proposes an ideal regularized composite kernel (IRCK) framework for hyperspectral images (HSI) classification. In learning a composite kernel, IRCK exploits spectral information, spatial information, and label information simultaneously. It incorporates the labels into standard spectral and spatial kernels by means of ideal kernel according to a regularization kernel learning framework, which captures both the sample similarity and label similarity and makes the resulting kernel more appropriate for HSI classification tasks. With the ideal regularization, the kernel learning problem has a simple analytical solution and is very easy to implement. The ideal regularization can be used to improve and refine state-of-the-art kernels, including spectral kernels, spatial kernels and spectral-spatial composite kernels. The effectiveness of the proposed IRCK is validated on the benchmark hyperspectral data set: Indian Pines. Experimental results show the superiority of our ideal regularized composite kernel method over the classical kernel methods.

By means of a sparse collaborative representation mechanism, sparse-representation-based classifiers show a superior performance in hyperspectral image (HSI) classification. Exploiting the similarity and distinctiveness of HSI neighboring pixels, we propose a new nearest regularized joint sparse representation (NRJSR) classification method in this letter. In the classification process of the central test pixel, the weights of different neighboring pixels and the sparse representation coefficients of different training samples are optimized simultaneously within a regularized sparsity model, which can obtain adaptive weights with good joint sparse representation ability. An alternative iteration strategy is used to solve the regularized joint sparsity model. The proposed NRJSR algorithm is tested on two benchmark HSI data sets. Experimental results demonstrate that the proposed algorithm performs better than other sparsity-based algorithms and spectral and spectral-spatial support vector machine classifiers.

In this paper, we present a sparse matrix transform (SMT) based linear discriminant analysis (LDA) algorithm for high dimensional data. The within-class scatter matrix in LDA is constrained to have an eigen-decomposition that can be represented as an SMT. Then, under maximum likelihood framework, based on greedy minimization strategy, the within-class scatter matrix can be efficiently estimated. Moreover, the estimated within-class scatter matrix is always positive definite and well-conditioned even with limited sample size, which overcomes the singularity problem in traditional LDA algorithm. The proposed method is compared, in terms of recognition rate, to other commonly used LDA methods on ORL and UMIST face databases. Results indicate that the performance of the proposed method is overall superior to those of traditional LDA approaches, such as the Fisherfaces, D-LDA, S-LDA and newLDA methods.

This paper proposes a region kernel to measure the region-to-region distance similarity for hyperspectral image (HSI) classification. The region kernel is designed to be a linear combination of multiscale box kernels, which can handle the HSI regions with arbitrary shape and size. Integrating labeled pixels and labeled regions, we further propose a region-kernel-based support vector machine (RKSVM) classification framework. In RKSVM, three different composite kernels are constructed to describe the joint spatial-spectral similarity. Particularly, we design a desirable stack composite kernel that consists of the point-based kernel, the region-based kernel, and the cross point-to-region kernel. The effectiveness of the proposed RKSVM is validated on three benchmark hyperspectral data sets. Experimental results show the superiority of our region kernel method over the classical point kernel methods.

Dimension reduction (DR) is a necessary and helpful preprocessing for hyperspectral image (HSI) classification. In this paper, we propose a spatial and spectral regularized local discriminant embedding (SSRLDE) method for DR of hyperspectral data. In SSRLDE, hyperspectral pixels are first smoothed by the multiscale spatial weighted mean filtering. Then, the local similarity information is described by integrating a spectral-domain regularized local preserving scatter matrix and a spatial-domain local pixel neighborhood preserving scatter matrix. Finally, the optimal discriminative projection is learned by minimizing a local spatial-spectral scatter and maximizing a modified total data scatter. Experimental results on benchmark hyperspectral data sets show that the proposed SSRLDE significantly outperforms the state-of-the-art DR methods for HSI classification.

Integrating labeled pixels with prior knowledge of hyperspectral spatial homogeneous regions, we propose a region-based hyperspectral image classification method, called the support vector machine with the multi-scale patch based box kernel (SVM-MPBK). It models the local homogeneous region of each pixel as a box, and measures the similarity between different box regions using box kernel. The box is represented as multidimensional intervals computed band by band in a neighborhood pixel patch. Using multi-scale patches to calculate box, SVM-MPBK fuses the complementary classification results in different scales by a majority voting. Experimental results on benchmark hyperspectral data sets demonstrate the effectiveness of SVM-MPBK.