Extracting robust discriminative features is a critical challenge in person re-identification (ReID). While Transformer-based methods have successfully addressed some limitations of convolutional neural networks (CNNs), such as their local processing nature and information loss resulting from convolution and downsampling operations, they still face the scalability issue due to the quadratic increase in memory and computational requirements with the length of the input sequence. To overcome this, we propose a pure Mamba-based person ReID framework named ReIDMamba. Specifically, we have designed a Mamba-based strong baseline that effectively leverages fine-grained, discriminative global features by introducing multiple class tokens. To further enhance robust features learning within Mamba, we have carefully designed two novel techniques. First, the multi-granularity feature extractor (MGFE) module, designed with a multi-branch architecture and class token fusion, effectively forms multi-granularity features, enhancing both discrimination ability and fine-grained coverage. Second, the ranking-aware triplet regularization (RATR) is introduced to reduce redundancy in features from multiple branches, enhancing the diversity of multi-granularity features by incorporating both intra-class and inter-class diversity constraints, thus ensuring the robustness of person features. To our knowledge, this is the pioneering work that integrates a purely Mamba-driven approach into ReID research. Our proposed ReIDMamba model boasts only one-third the parameters of TransReID, along with lower GPU memory usage and faster inference throughput. Experimental results demonstrate ReIDMamba's superior and promising performance, achieving state-of-the-art performance on five person ReID benchmarks. Code is available athttps://github.com/GuHY777/ReIDMamba.

Hyperspectral image (HSI) clustering is a fundamental yet challenging task that typically operates without training labels. Recent advancements in deep graph clustering methods have shown promise for HSI due to their ability to effectively encode spatial structural information. However, limitations such as inadequate utilization of structural information, poor feature representation, and weak graph update capabilities hinder their performance. In this article, we propose an adaptive homophily structure graph clustering (AHSGC) method for HSI. Our approach begins with the generation of homogeneous regions to process HSI and construct the initial graph. Next, we design an adaptive filter graph encoder that captures both high and low-frequency features for subsequent processing. We then develop a graph embedding clustering self-training decoder using KL Divergence to generate pseudo-labels for network training. To enhance graph learning, we introduce homophily-enhanced structure learning, which updates the graph based on the clustering task. This involves estimating node connections through orient correlation estimation and dynamically adjusting graph edges via graph edge sparsification. Finally, we implement joint network optimization to facilitate self-training and graph updates, with K-means used to express latent features. The clustering accuracy on three datasets is 83.60%, 63.65%, and 86.03%, the FLOPs are 3.57G, 30.62G, and 2.95G. The source code will be available athttps://github.com/DY-HYX.

Graph Neural Networks (GNN) are capable of modeling complex non-Euclidean structures through information transfer, and thus have been party widely used in the field of Hyperspectral Image (HSI) classification. However, conventional GNNs often have difficulty in handling regular grid data, which in turn loses positional information or spatial coherence, as well as in capturing long-range dependencies, which affects their performance in heterogeneous and limited-sample condition. To address these limitations, this paper proposes a novel Graph-based Transformer and Convolution Fusion Network (GTCFN) that integrates the local representation power of Convolutional Neural Networks (CNNs) with the global reasoning capability of graph-based Transformers. GTCFN consists of two synergistic branches: a Graph Transformer sub-network (GTsN) that models high-level semantic structures among superpixels via attention-based topology learning, and a Spectral–Spatial Convolutional sub-network (S2CsN) that extracts multi-scale fine-grained features using 5×5, 7×7, and 9×9 convolutional kernels. To enhance efficiency and generalization, GTCFN incorporates kernelized attention with random feature mapping, reducing the complexity fromO(M2) toO(M). At the same time, attention oversmoothing is avoided by introducing a Gumbel-based multi-head random aggregation mechanism. Experiments conducted on four benchmark datasets, namely Indian Pines, Pavia University, Salinas and WHU-Hi-HongHu, show that GTCFN achieves state-of-the-art performance with OA of 95.62%, 98.34%, 97.88% and 96.69%, which is significantly better than 12 other algorithms, such as CNNs, graph-based models and hybrid network models. The core code for GTCFN is posted on https://github.com/ Majunyi310321/GTCFN.

Self-supervised hyperspectral image (HSI) clustering remains a fundamental yet challenging task due to the absence of labeled data and the inherent complexity of spatial-spectral interactions. While recent advancements have explored innovative approaches, existing methods face critical limitations in clustering accuracy, feature discriminability, computational efficiency, and robustness to noise, hindering their practical deployment. In this paper, a self-supervised efficient low-pass contrastive graph clustering (SLCGC) is introduced for HSIs. Our approach begins with homogeneous region generation, which aggregates pixels into spectrally consistent regions to preserve local spatial-spectral coherence while drastically reducing graph complexity. We then construct a structural graph using an adjacency matrix A and introduce a low-pass graph denoising mechanism to suppress high-frequency noise in the graph topology, ensuring stable feature propagation. A dual-branch graph contrastive learning module is developed, where Gaussian noise perturbations generate augmented views through two multilayer perceptrons (MLPs), and a cross-view contrastive loss enforces structural consistency between views to learn noise-invariant representations. Finally, latent embeddings optimized by this process are clustered via K-means. Extensive experiments and repeated comparative analysis have verified that our SLCGC contains high clustering accuracy, low computational complexity, and strong robustness. The code source will be available athttps://github.com/DY-HYX.

The joint clustering of multimodal remote sensing (RS) data poses a critical and challenging task in Earth observation. Although recent advances in multiview subspace clustering have shown remarkable success, existing methods become computationally prohibitive when dealing with large-scale RS datasets. Moreover, they neglect intrinsic nonlinear and spatial interdependencies among heterogeneous RS data and lack generalization ability for out-of-sample data, thereby restricting their applicability. This article introduces a novel unified framework called anchor-based multiview kernel subspace clustering with spatial regularization (AMKSC). It learns a scalable anchor graph in the kernel space, leveraging contributions from each modality instead of seeking a consensus full graph in the feature space. To ensure spatial consistency, we incorporate a spatial smoothing operation into the formulation. The method is efficiently solved using an alternating optimization strategy, and we provide theoretical evidence of its scalability with linear computational complexity. Furthermore, an out-of-sample extension of AMKSC based on multiview collaborative representation-based classification is introduced, enabling the handling of larger datasets and unseen instances. Extensive experiments on three real heterogeneous RS datasets confirm the superiority of our proposed approach over state-of-the-art methods in terms of clustering performance and time efficiency. The source code is available at https://github.com/AngryCai/AMKSC.

Incorrect labeling (i.e., noisy label learning) in HSI classification has attracted so much attention in recent years, which holds the assumption that the given pixels of an HSI may be incorrectly labeled and only one candidate label is required to provide for a typical pixel. However, instead of offering only one candidate label that may be incorrect, partial label learning often provides a candidate label set that contains the ground-truth label for each pixel in an HSI, which is also an essential problem of great practical value and has recently started to attract attention. This paper proposes a novel framework for partial label learning in HSI classification, namely unlabeled data guided partial label learning (UPLL). The proposed framework is an iterative process that can fully exploit the benefits of unlabeled data. Specifically, during each iteration, we conduct the semi-supervised label propagation; the resulting labeling confidence matrices of the original training samples and the unlabeled testing samples are further enhanced by exploiting the spatial information. Then, we select qualified original training samples and unlabeled testing samples with high confident predictions to disambiguate and expand the original training set, leading to a more robust representation of training data. Such phases are repeated until convergence. The comprehensive experiments show the superiority of the proposed UPLL method over the existing state-of-the-art methods. Especially, the classification accuracy improves more than 5% with very few training samples than the second best comparing method.

With the aid of transformers, significant progress has been achieved in hyperspectral image change detection (HSI-CD) in recent times. Nonetheless, most contemporary detection methods fail to incorporate diverse diagnostic features extracted from hyperspectral (HS) images. In addition, relying solely on algebraic-based techniques to extract information of difference is insufficient for achieving satisfactory detection performance. In this regard, we propose an innovative triple-path local-to-global network (TL2GN), complemented by a hybrid head transformer (HybridHT), called TL2GH2T, tailored for HSI-CD tasks. To be specific, TL2GH2T first investigates spatial, spectral, and spatial–spectral features from a local-to-global perspective. Then, a novel spatial and spectral token fusion (SSTF) module is developed to integrate the above three tokenized features, producing discriminative features from two HS images separately. Moreover, drawing inspiration from chromosomal crossover mechanisms, we propose a HybridHT. Its goal is to simultaneously learn cross correlation and self-correlation information of bitemporal features from a global perspective, producing highly discriminative distinctions. Our approach, validated through extensive experimentation on four varied HS benchmarks, exhibits exceptional performance in HSI-CD, outperforming contemporary methods in both visual and quantitative evaluations.

Researchers have been paying more attention to hyperspectral image (HSI) classification based on semi-supervised superpixel-level graph convolutional networks (SGCNs) due to their aggregation ability of rich contextual information. Although these SGCNs achieve good classification performance, the influence of the positional distributions among labeled superpixels has been overlooked. The locations of labeled superpixels, such as located at class boundaries or centers, exert a substantial influence on the final performance. To address this issue, this article proposed a novel graph neural network (GCN) method with the guidance of positional distributions of labeled superpixels, abbreviated as LPDGCN. Specifically, we first propose to utilize the sparse, low-rank as well as feature smoothness restrictions to optimize the initial superpixel graph structure because the connectivity relationships of labeled superpixels located at class boundaries or centers are easily influenced by spectral variation. Second, in order to effectively determine the positional distributions of labeled superpixels and make full use of the position relationships, we propose to utilize the information conflict from the above topology connectivity to determine the positional distributions of labeled superpixels and develop the reweighted strategy to weaken the influence of labeled superpixels located at class boundaries and strengthen the influence of that located at class centers. Finally, we evaluate the LPDGCN method on four public HSI datasets, demonstrating its superiority over other advanced classification methods in terms of three metrics, i.e., overall accuracy (OA), average accuracy (AA), and kappa coefficient (KC).

Transformer-based methods have a great ability to model nonlocal interactions between spectral and spatial information, while the local features are easily ignored. Graph convolutional neural networks (GCNs) tend to do well in exploiting neighborhood vertex interactions based on their unique aggregation mechanism, while the ability to extract global information is limited. In this article, we study to comprehensively utilize the advantages of transformer and graph convolution by combining the two structures into a unified Transformer (Graphormer) to construct both local and global interactions for hyperspectral image (HSI) classification, and spatial–spectral features enhanced Graphormer framework (S2GFormer) is proposed. Specifically, a follow patch mechanism is first proposed to transform the pixel in HSI to patches while preserving the local spatial features and reducing the computational cost. Moreover, a patchwise spectral embedding block is designed to extract the spectral features of the patch, in which a neighborhood convolution is inserted for comprehensive spectral information extraction. Finally, a multilayer Graphormer Encoder module is proposed to extract the representative spatial–spectral features from the patch for HSI classification. In our network, we jointly integrate the three aforementioned parts into a unified network, and each component benefits the other. The experimental results demonstrate its suitability for HSI classification when compared with other state-of-the-art (SOTA) classifiers, particularly in scenarios with very limited labeled samples. The code of S2GFormer will be made publicly available at:https://github.com/DY-HYX.

Efficient extraction of spectral sequences and geospatial information is crucial in hyperspectral image (HSI) classification. Recurrent neural networks (RNNs) and Transformers excel in capturing long-range spectral features, while convolutional neural networks (CNNs) excel in aggregating spatial information through convolutional kernels. However, RNNs and Transformers suffer from low-computational efficiency, and CNNs have limitations in perceiving global contextual information. To address these issues, this article proposes GraphMamba—an efficient graph structure learning vision Mamba for HSI classification. Specifically, GraphMamba is a novel hyperspectral information processing paradigm that preserves spatial-spectral features by constructing spatial-spectral cubes and employs a linear spectral encoder to enhance the operability of subsequent tasks. The core components of GraphMamba include the HyperMamba module, which enhances computational efficiency, and the SpatialGCN module, designed for adaptive spatial context awareness. The HyperMamba mitigates clutter interference by employing a global mask (GM) and introduces a parallel training and inference architecture to alleviate computational bottlenecks. Meanwhile, the SpatialGCN utilizes weighted multihop aggregation (WMA) for spatial encoding, emphasizing highly correlated spatial structural features. This approach enables flexible aggregation of contextual information while minimizing spatial noise interference. Notably, the encoding modules of the proposed GraphMamba architecture are both flexible and scalable, providing a novel approach for the joint mining of spatial-spectral information in hyperspectral images. Extensive experiments were conducted on three different scales of real HSI datasets. When compared with state-of-the-art classification methods, GraphMamba demonstrated superior performance. The core code will be released athttps://github.com/ahappyyang/GraphMamba.

Due to the scarcity of prior information and the high complexity of spectral data, hyperspectral image (HSI) clustering presents a significant challenge. Although recent deep clustering methods have demonstrated remarkable performance, their intricate network structures and poor robustness hinder their practical application. To address this issue, we propose an efficient and lightweight spectral-spatial feature graph contrastive learning (S2GCL) framework for robust HSI clustering. Specifically, we have designed a novel spectral-spatial feature encoder that fully leverages the information in HSI by incorporating both spatial structure and spectral similarity matrices. To establish a lightweight model, we implement several effective designs: First, S2GCL eliminates the commonly used data augmentation and discriminator in GCL during the generation of positive embeddings. Second, we use a multilayer perceptron (MLP) to produce low-dimensional embeddings instead of relying on graph convolutional networks (GCNs). Third, negative embeddings are generated through row-shuffling, avoiding the use of neural networks. Finally, we propose a multiple boundary loss function to extract complementary information from spatial structures and neighboring nodes, while also constraining the interclass differences between positive and negative examples. We conducted extensive experiments on four publicly available datasets and compared S2GCL with state-of-the-art clustering methods. The results indicate that S2GCL achieves satisfactory performance. The code for S2GCL will be released athttps://github.com/ahappyyang/S2GCL.

Insufficient prior knowledge of a captured hyperspectral image (HSI) scene may lead the experts or the automatic labeling systems to offer incorrect labels or ambiguous labels (i.e., assigning each training sample to a group of candidate labels, among which only one of them is valid; this is also known as partial label learning) during the labeling process. Accordingly, how to learn from such data with ambiguous labels is a problem of great practical importance. In this letter, we propose a novel superpixelwise low-rank approximation (LRA)-based partial label learning method, namely SLAP, which is the first to take into account partial label learning in HSI classification. SLAP is mainly composed of two phases: disambiguating the training labels and acquiring the predictive model. Specifically, in the first phase, we propose a superpixelwise LRA-based model, preparing the affinity graph for the subsequent label propagation process while extracting the discriminative representation to enhance the following classification task of the second phase. Then to disambiguate the training labels, label propagation propagates the labeling information via the affinity graph of training pixels. In the second phase, we take advantage of the resulting disambiguated training labels and the discriminative representations to enhance the classification performance. The extensive experiments validate the advantage of the proposed SLAP method over state-of-the-art methods.

Remote sensing image (RSIs) segmentation is widely used in urban planning, natural disaster detection and many other fields. Compared with natural scene images, RSIs have higher resolution, complex imaging, and diverse object shapes and sizes, while semantic segmentation methods based on deep learning often require many data labels. In this paper, we propose a semi-supervised RSIs segmentation network with multi-scale deformable threshold feature extraction module and mixed attention (MDMANet). First, a pyramid ensemble structure is used, which incorporates deformable convolution and bole convolution, to extract features of objects with different shapes and sizes and reduce the influence of redundant features. Meanwhile, a mixed attention (MA) is proposed to aggregate long-range contextual relationships and fuse low-level features with high-level features. Second, an FCN-based full convolution discriminator task network is designed to help evaluate the feasibility of unlabeled image prediction results. We performed experimental validation on three datasets, and the results show that MDMANet segmentation provides more significant improvement in accuracy and better generalization than existing segmentation networks.

Federated learning is a promising solution in several industries for co-training models among distributed clients via centralized servers without leaving private user data on the devices. Thus, federated learning can be seen as a stimulus for the edge computing paradigm as it supports collaborative learning and model optimization. In view of the strict requirements for data security and system reliability of hyperspectral classification techniques for surveillance, aerospace, and military missions, this paper proposes a novel stereo attention cross-decoupling fusion-guided federated neural learning algorithm for hyperspectral image classification, which first trains client devices using a scalable federated learning approach consisting of master server, secure aggregator and edge client devices of a certain size.The distributed devices train local models of the neural network for classifying hyperspectral images and send them to the secure aggregator, which aggregates the local models using a weighted averaging strategy and sends them to the master server for iteration. In addition, the stereo attention cross-decoupling fusion module is used to mine the multidimensional spatial details of the hyperspectral images, specifically by first extracting the most discriminative features from different directions (horizontal, vertical, and spatial) using the attention mechanism, and then using the decoupling fusion strategy to classify the original feature map into three levels: significant, minor, and redundant, and use them to model the multidimensional spatial relationships, thus strengthening the capability to represent features. Extensive experiments on several public datasets have shown that the proposed method provides competitive performance and, more importantly, is effective in enhancing privacy and reliability for hyperspectral image classification.

Although convolutional neural networks (CNNs) have shown superior performance to traditional machine learning algorithms for hyperspectral image classification tasks, the ability of traditional CNNs to model remote dependencies in the spatial orientation of HSIs is still limited, and they always extract similar low-level features, leading to feature redundancy. To cope with this limitation, this paper proposes a novel multi-order statistical representation-guided graph convolution and continuous context threshold-aware network for the classification of hyperspectral images with limited training samples. Initially, the spectral spatial information is separately modeled using first-order features and second-order pooling operators. Secondly, we propose graph-structuring the patch’s features. By employing a random walk transition probability matrix, graph-structured convolution can mine more discriminative direction features. In addition, we design a continuous context threshold-aware network to model multidimensional spatial relationships, thereby enhancing the representation of graph features. Specifically, the cross-attention mechanism is used to calculate the attention weights in the vertical and horizontal directions, and the features are divided into two levels—important and secondary—by solving the cosine distance between feature vectors, and the former is retained and the latter is punished. Extensive experiments on multiple HSIs datasets demonstrated that the proposed method delivers competitive performance. The code will be available at: https://github.com/vivitsai/GSC-CCTA.

Transformer has been widely used in classification tasks for hyperspectral images (HSI) in recent years. Because it can mine spectral sequence information to establish long-range dependence, its classification performance can be comparable with the convolutional neural network (CNN). However, both CNN and Transformer focus excessively on spatial or spectral domain features, resulting in an insufficient combination of spatial-spectral domain information from HSI for modeling. To solve this problem, we propose a new end-to-end graph convolutional network (GCN) and Transformer fusion network with the spatial-spectral feature extraction (GTFN) in this paper, which combines the strengths of GCN and Transformer in both spatial and spectral domain feature extraction, taking full advantage of the contextual information of classified pixels while establishing remote dependencies in the spectral domain compared with previous approaches. In addition, GTFN uses Follow Patch as an input to the GCN and effectively solves the problem of high model complexity while mining the relationship between pixels. It is worth noting that the spectral attention module is introduced in the process of GCN feature extraction, focusing on the contribution of different spectral bands to the classification. More importantly, to overcome the problem that Transformer is too scattered in the frequency domain feature extraction, a neighborhood convolution module is designed to fuse the local spectral domain features. On Indian Pines, Salinas, and Pavia University datasets, the overall accuracies (OAs) of our GTFN are 94.00%, 96.81%, and 95.14%, respectively. The core code of GTFN is released at https://github.com/1useryang/GTFN.

This article presents FLGC, a simple yet effective fully linear graph convolutional network for semi-supervised and unsupervised learning. Instead of using gradient descent, we train FLGC based on computing a global optimal closed-form solution with a decoupled procedure, resulting in a generalized linear framework and making it easier to implement, train, and apply. We show that (1) FLGC is powerful to deal with both graph-structured data and regular data, (2) training graph convolutional models with closed-form solutions improve computational efficiency without degrading performance, and (3) FLGC acts as a natural generalization of classic linear models in the non-Euclidean domain (e.g., ridge regression and subspace clustering). Furthermore, we implement a semi-supervised FLGC and an unsupervised FLGC by introducing an initial residual strategy, enabling FLGC to aggregate long-range neighborhoods and alleviate over-smoothing. We compare our semi-supervised and unsupervised FLGCs against many state-of-the-art methods on a variety of classification and clustering benchmarks, demonstrating that the proposed FLGC models consistently outperform previous methods in terms of accuracy, robustness, and learning efficiency. The core code of our FLGC is released at https://github.com/AngryCai/FLGC .

Graph convolutional network (GCN) has shown potential in hyperspectral image (HSI) classification. However, GCN is a transductive learning method, which is difficult to aggregate the new node. The available GCN-based methods fail to understand the global and contextual information of the graph. To address this deficiency, a novel semisupervised network based on graph sample and aggregate-attention (SAGE-A) for HSIs’ classification is proposed. Different from the GCN-based method, SAGE-A adopts a multilevel graph sample and aggregate (graphSAGE) network, as it can flexibly aggregate the new neighbor node among arbitrarily structured non-Euclidean data and capture long-range contextual relations. Inspired by the convolution neural network (CNN) self-attention mechanism, the proposed network uses the graph attention mechanism to characterize the importance among spatially neighboring regions, so the deep contextual and global information of the graph can be learned automatically by focusing on important spatial targets. Extensive experimental results on different real hyperspectral data sets demonstrate the performances of our proposed method compared with the state-of-the-art methods.

Recently, graph convolutional network (GCN) has received more and more interest in the field of hyperspectral image classification (HSIC). The existing GCN-based models for HSIC propagate and aggregate information through the GCN network based on the graph, which is constructed according to spatial location or spectral similarity. However, the constructed graph may not be ideal for the downstream classification task due to the variety of spectral characteristics. In this paper, a fully connected graph is adaptively constructed to make full use of local spatial information and global spectral information. Besides, we apply a neural sparsification technique to remove potentially task-irrelevant edges in case of misleading message propagation. Furthermore, label propagation (LP) serves as regularization to assist the graph network in learning proper edge weights that lead to improved classification performance. The resulting network is end-to-end trainable. The experimental results on three popular benchmarks, including Indian Pines, Pavia University, and Kennedy Space Center, demonstrate the superiority of our algorithm.

Graph convolutional networks (GCN) have begun to show their potential in hyperspectral image classification in recent years. However, most of the current GCN methods are designed to learn node features on fixed and homogeneous graphs, and it is difficult for them to learn effective node features on heterogeneous graphs. The limitation is particularly evident in hyperspectral classification because of the different types of nodes and edges. The Graph Transformer Network with the graph attention mechanism (GTN-A) is proposed to address this shortcoming in this paper. It can generate a new graph structure, which is represented by a more useful meta-path, so that node features can be better aggregated. The experiments conducted on two benchmark datasets illustrate the effectiveness of our method.

Deep subspace clustering has achieved remarkable performances in the unsupervised classification of hyperspectral images. However, previous models based on pixel-level self-expressiveness of data suffer from the exponential growth of computational complexity and access memory requirements with increasing number of samples, thus leading to poor applicability to large hyperspectral images. This paper presents a Neighborhood Contrastive Subspace Clustering network (NCSC), a scalable and robust deep subspace clustering approach, for unsupervised classification of large hyperspectral images. Instead of using a conventional autoencoder, we devise a novel superpixel pooling autoencoder to learn the superpixel-level latent representation and subspace, allowing a contracted self-expressive layer. To encourage a robust subspace representation, we propose a novel neighborhood contrastive regularization to maximize the agreement between positive samples in subspace. We jointly train the resulting model in an end-to-end fashion by optimizing an adaptively weighted multi-task loss. Extensive experiments on three hyperspectral benchmarks demonstrate the effectiveness of the proposed approach and its substantial advancement of state-of-the-art approaches.

Due to prior knowledge deficiency, large spectral variability, and high dimension of hyperspectral image (HSI), HSI clustering is extremally a fundamental but challenging task. Deep clustering methods have achieved remarkable success and have attracted increasing attention in unsupervised HSI classification (HSIC). However, the poor robustness, adaptability, and feature presentation limit their practical applications to complex large-scale HSI datasets. Thus, this article introduces a novel self-supervised locality preserving low-pass graph convolutional embedding method (L2GCC) for large-scale hyperspectral image clustering. Specifically, a spectral–spatial transformation HSI preprocessing mechanism is introduced to learn superpixel-level spectral–spatial features from HSI and reduce the number of graph nodes for subsequent network processing. In addition, locality preserving low-pass graph convolutional embedding autoencoder is proposed, in which the low-pass graph convolution and layerwise graph attention are designed to extract the smoother features and preserve layerwise locality features, respectively. Finally, we develop a self-training strategy, in which a self-training clustering objective employs soft labels to supervise the clustering process and obtain appropriate hidden representations for node clustering. L2GCC is an end-to-end training network, which is jointly optimized by graph reconstruction loss and self-training clustering loss. On Indian Pines, Salinas, and University of Houston 2013 datasets, the clustering accuracy overall accuracies (OAs) of the proposed L2GCC are 73.51%, 83.15%, and 64.12%, respectively.

Hyperspectral image (HSI) clustering is an extremely fundamental but challenging task with no labeled samples. Deep clustering methods have attracted increasing attention and have achieved remarkable success in HSI classification. However, most existing clustering methods are ineffective for large-scale HSI, due to their poor robustness, adaptability, and feature presentation. In this paper, to address these issues, we introduce unsupervised self-correlated learning smoothy enhanced locality preserving graph convolution embedding clustering (S2LGCC) for large-scale HSI. Specifically, the spectral-spatial transformation is introduced to transform the original HSI into a graph while preserving the local spectral features and spatial structures. After that, a locality preserving graph convolutional embedding encoder is designed to learn the hidden representation from the graph, in which the deep layer-wise graph convolutional network (LGAT) is proposed to preserve the adaptive layer-wise locality features. In addition, the self-correlated learning smoothy module is developed to learn the smoothy information and the non-local relationship in the hidden representation space for clustering. Finally, a self-training strategy is proposed to cluster the graph node, in which a self-training clustering objective employs soft labels to supervise the clustering process. The proposed S2LGCC is jointly optimized by the fusion graph reconstruction loss and self-training clustering loss, and the two benefit each other. On IP, Salinas, and UH2013 datasets, the OAs of our S2LGCC are 71.76%, 82.61%, and 63.82%, respectively.

The application of graph convolutional networks (GCNs) to hyperspectral image (HSI) classification is a heavily researched topic. However, GCNs are based on spectral filters, which are computationally costly and fail to suppress noise effectively. In addition, the current GCN-based methods are prone to oversmoothing (the representation of each node tends to be congruent) problems. To circumvent these problems, a novel semi-supervised locality-preserving dense graph neural network (GNN) with autoregressive moving average (ARMA) filters and context-aware learning (DARMA-CAL) is proposed for HSI classification. In this work, we introduce the ARMA filter instead of a spectral filter to apply to GNNs. The ARMA filter can better capture the global graph structure and is more robust to noise. More importantly, the ARMA filter can simplify calculations compared with the spectral filter. In addition, we show that the ARMA filter can be approximated by a recursive method. Furthermore, we propose a dense structure, which not only implements the ARMA filter in the structure, but is also locality-preserving. Finally, we design a layerwise context-aware learning mechanism to extract the useful local information generated by each layer of the dense ARMA network. The experimental results on three real HSI datasets show that DARMA-CAL outperforms the compared state-of-the-art methods.