Remote sensing data from real-world scenarios manifests a long-tailed distribution, with the continuous emergence of new classes over time. Nevertheless, the existing class-incremental remote sensing classification models neglect the above long-tailed distribution phenomenon, which seriously damages their overall superior performance. Meanwhile, long-tail class-incremental learning developed in other areas focuses only on the classifier decision boundary optimization of the tail-class, while neglecting the robustness of the feature backbone. The feature backbone trained on the base classes causes a serious significant distribution shift for the incremental classes owing to the distributional differences between base and incremental classes. To solve these issues, we propose a continually evolved feature and classifiers learning (CEF-CL) framework for long-tail class-incremental remote sensing scene classification. Specifically, tail-class data are scaled and grafted onto head-class data to diversify the semantic information of the tail-class leveraging the rich context of the head classes, which can improve the generalization of the feature backbone. And then, an adaptive multi-scale feature fusion (AMFF) module is proposed to couple feature maps of head and tail classes scale by scale for generating virtual tail-class features that deeply perceive head-class information, which can further enhance the reliability of classifier decision boundary optimization. Furthermore, examples from old classes are regarded as pseudo-tail classes to participate in incremental learning, which greatly alleviates catastrophic forgetting of old classes. Extensive experiments on two remote sensing benchmarks demonstrate the superiority of the proposed CEF-CL in comparison with existing class-incremental learning.

The joint classification of hyperspectral image (HSI) and light detection and ranging (LiDAR) data has attracted considerable attention in the field of remote sensing. Integrating the advantages of the two data sources can provide precise data support and analytical decision-making for remote-sensing applications. However, due to the inherent differences in properties and semantic information from heterogeneous data, most existing deep-learning methods suboptimally extract the characteristic features of both data sources while utilizing their interactive information. In this letter, we propose a dual-branch feature fusion network-based cross-modal enhanced CNN and Transformer (DF2NCECT) to make full use of the respective features and interactive information of multisource data. DF2NCECT consists of two main stages. One is the basic feature extraction stage, which builds a hybrid convolution module based on 3DCNN and inception structure to fully extract the joint features of HSI from multiple spatial perspectives. The other is the deep feature fusion stage, where the CNN and Transformer are designed in parallel to fully explore and fuse deep features between HSI and LiDAR. More importantly, to achieve efficacious interactive information between HSI and LiDAR, a cross-modal enhanced CNN and Transformer module (CECT) is designed to deeply enhance the fused interactive features from global/local perspectives. Experiments show that the proposed method is superior and outperforms the comparison methods by an average of 3.06% in OA on Houston2013 and 1.79% on Summer, respectively.

Convolutional neural network (CNN)-based hyperspectral image (HSI) classification models often exhibit high volume and complexity. This not only poses challenges in deploying them on mobile and embedded devices due to storage and power constraints but also introduces a dilemma between the growing demand for labeled samples and the high cost associated with manual labeling. To address these challenges, we propose an ultra-lightweight hybrid CNN based on redundancy removal (ULite-R2HCN), specifically designed for HSI classification in scenarios with limited samples. To reduce computational costs and enhance feature extraction effectiveness, we focus on optimizing the widely used depthwise convolution (DW-Conv) and pointwise convolution (PW-Conv) in the lightweight HSI classification model. For DW-Conv, we design a spatial convolution with redundancy removal (R2Spatial-Conv). This involves the design of multi-scale 3D convolution kernels with specific structures instead of 2D convolution kernels, aiming to reduce redundant convolution kernels and extract multi-scale spatial features. Simultaneously, for PW-Conv, we design a spectral convolution with redundancy removal (R2Spectral-Conv). This utilizes a “copy-splicing-grouping” structure to extract spectral features within arbitrary range intervals, effectively reducing redundant spectral extractions and capturing long-range spectral relationships. Numerous experiments have shown that the proposed ULite-R2HCN achieves higher classification accuracy with an ultra-light volume for a few training samples. In addition, sufficient ablation experiments also verified the advanced performance of the designed R2Spatial-Conv and R2Spectral-Conv.

Few-shot class-incremental learning has recently received significant research focus in remote sensing scene classification (FSCIL-RSSC). The success of FSCIL-RSSC relies on the robustness of the feature backbone and classifiers. Existing works focus on improving classifier adaptation, but little attention is paid to the importance of backbone robustness on the recognition ability of new class samples’ embeddings. Due to the large distribution shift between old and new classes, FSCIL-RSSC using high-layer (single-scale) features may not adapt flawlessly to new categories. To solve the issue, we put forward a gradient guided multiscale feature collaboration network (G-MFCN) for FSCIL-RSSC. Specifically, we introduce a parallel hierarchy strategy to simultaneously capture the multifeature discriminative information of the same sample. Then, a gradient guide block is designed to automatically pick out the optimal values of different convolution blocks for multifeature fusion. Finally, the classical feature pyramid network is introduced for multiscale fusion to obtain more obvious discriminative features of RSSC. More importantly, our proposed G-MFCN is a simple and adaptable module, which can combine any existing FSCIL frameworks to further improve the optimized classifiers’ effectiveness for the FSCIL-RSSC scenario. Extensive experiments on four benchmarks demonstrate that the proposed G-MFCN achieves significant improvements in comparison to existing FSCIL-RSSC methods.

Class-incremental (CI) learning has recently received extensive research interest in remote sensing scene classification (CI-RSSC). The existing CI-RSSC methods’ superior performance seriously relies on old (base classes) and new classes (incremental classes) sampled independently from an identical distribution (dataset). In real-world RSSC scenarios, there exist significant distribution shifts between old and new classes, leading to the existing CI-RSSC methods being unable to adjust flawlessly to these new classes. In this article, we propose a novel cross-domain (CD) CI-RSSC framework to solve the above-mentioned problems, termed CDCI-RSSC. Specifically, a modular sharing-based dynamic extension module is first designed, which only updates specialized modules to extract new class feature embeddings for reducing memory footprint. Then, an effective dynamic alignment guided domain adaptive module (DAM) is further proposed to calculate the dynamic weights of each sample in various fields, which can minimize distribution shifts between source and target domains. Finally, a foreground enhancement module (FEM) is introduced to alleviate the issue of complex background interference in RSSC by increasing the weight of critical regions. Compared with the existing CI-RSSC and CD-RSSC, our proposed CDCI-RSSC framework surmounts the challenge of handling the distribution shifts between source (base session) and target domains (incremental session) while alleviating the limitations of continuous learning of new classes. Extensive experiments on three CDCI scenarios show that the CDCI-RSSC model achieves significant performance improvements in comparison to existing CI-RSSC and CD-RSSC methods.

Recent studies have demonstrated the potential of hybrid convolutional models that combine 3D and 2D convolutional neural networks (CNNs) for hyperspectral image (HSI) classification. However, these models do not fully utilize the benefits of hybrid convolution due to inefficient connections between the two types of CNNs. Moreover, most CNNs, including hybrid models, require a significant number of parameters and computational resources for accurate classification, which increases the need for labeled samples and computational cost. Although the common lightweight strategies like depthwise separable convolution (DSC) can reduce parameters and computation compared to normal convolution (NC), they often compromise accuracy. To address these challenges, we propose a lightweight hybrid convolutional neural network (Lite-HCNet) for HSI classification with minimal model parameters and computational effort. Firstly, we design a novel channel attention module (NCAM) and combine it with a convolutional kernel decomposition (CKD) strategy to propose a lightweight and efficient DSC (LE-DSC) deployed in Lite-HCNet. The LE-DSC not only reduces the DSC volume further but also enhances its performance. Secondly, a lightweight and efficient hybrid convolutional layer (LE-HCL) is designed in Lite-HCNet to explore the efficient connection structure between 3D CNNs and 2D CNNs. Experiments show that the Lite-HCNet reduces the required computational cost and practical deployment difficulty while offering advanced performance with a small number of training samples. Furthermore, abundant ablation experiments confirm the superior performance of the designed LE-DSC.

In recent years, convolutional neural networks (CNNs) have been widely used in hyperspectral images (HSIs) classification. However, 2-D CNN, 3-D CNN, and even the newly emerged hybrid CNN (HCNN) all require multiple or deep CNN layers to obtain excellent classification performance, which inevitably results in the high complexity and the need for a large number of training samples. Moreover, as a local operator, convolution is challenging to fully use global information. To solve the above two issues, we design a HCNN based on global reasoning (GloRe-HCNN) for HSI classification. On the one hand, the GloRe-HCNN uses only one layer of 3-D CNN and one layer of 2-D CNN to jointly extract the spatial–spectral features of HSI. On the other hand, we contrive a spatial–spectral global reasoning unit (SS-GloRe-Unit) to take the place of stacked multilayer 3-D CNN for extracting global features fully. We select small training samples in three standard datasets and compare them with state-of-the-art CNN methods. Numerous experiments show that our GloRe-HCNN performs advanced performance.