Hyperspectral image classification is an important but challenging task. Conventional convolutional neural networks (CNNs) are able to extract local spectral spatial features but ignore long-range dependencies and global features. To address this problem, we propose a new model combining 3D-CNN and Convolutional Vision Transformer, aiming to improve the performance of the image recognition task by utilizing the advantages of CNN in local feature extraction while retaining the advantages of Transformer in long-range dependency processing. Our model is tested on three publicly available hyperspectral image datasets, and the results show that our model outperforms other state-of-the-art models in terms of classification accuracy and robustness. The source code for our work is available at [https://github.com/Dreamvai/ViT-Convolution]. The model proposed in this letter provides a new idea for hyperspectral image classification and expands a new field for the application of convolutional transformers. In the future, we intend to further explore the performance of the convolutional transformer and the possibility of combining it with other types of data.

Traditional training models often experience significant performance drops on test sets when the training and testing data distributions differ. To address this domain shift problem, we propose D3Net, a domain generalization classification network for hyperspectral images based on Generative Adversarial Networks (GANs). Specifically, D3Net consists mainly of a generator and a discriminator. The generator extracts domain-invariant information from the source domain to generate data with core classification features, while the discriminator employs a dual-confidence model to enhance the capture of domain-invariant features. Through adversarial iterations, the model is able to adapt to the domain shift effects of unknown data. Unlike existing domain generalization methods that rely on random perturbations for data augmentation, D3Net utilizes learnable convolutional neural networks (CNNs) rather than randomization to enhance the model’s learning capability. We conducted cross-scene classification experiments on datasets from Houston, Pavia, and Indiana, and the results demonstrate the effectiveness of our approach. Code for D3Net is available at: https://github.com/gmsjzyq123/D3Net.

In Hyperspectral Image (HSI) classification tasks, extracting surface features is crucial but the complexity of HSI brings considerable challenges. Traditional Convolutional Neural Networks (CNNs) and (Vision Transformer) ViT perform well but have large parameters and are time consuming. In contrast, Graph Convolutional Neural Networks (GCNs) are fast to train with few parameters. The SLIC segmentation algorithm generates pixel-level graph data and super-pixel-level graph data. Single-branch GCNs learn only one type of graph data resulting in insufficient feature richness. In two-branch GCNs, one branch extracts local spatial-spectral features from pixel-level data, another branch extracts global spatial-spectral features from super-pixel-level data, and finally the two branches are combined. But the existing two-branch GCNs do not focus on the close fusion of local and global features. Therefore, in this paper, we propose a new two-branch GCNs with local and global feature fusion (LG-GCN). The LG-GCN model proposed in this study not only has a short training time, but also has a high accuracy. The LG-GCN can capture more comprehensively the local and global spatial-spectral semantic information of HSI. Experimental results on three publicly available HSI datasets show that LG-GCN outperforms other state-of-the-art models in terms of classification accuracy and robustness. In addition, it has a shorter training time and is more resistant to noise. The source code of the LG-GCN model is available at https://github.com/Dreamvai/LG-GCN.

Since convolutional neural networks (CNN) can extract deeper features from hyperspectral images, they show good classification performance in the hyperspectral image (HSI) classification task. However, the performance of many CNN models is constrained by the complexity of hyperspectral pictures. The typical 2D-CNN method is unable to capture the discriminant spectral-spatial features due to the massive dimensionality of HSI data as well as potential noise and redundancy, while the 3D-CNN method significantly increases the computational burden of the model. Moreover, the gradient problem of the deeper CNN architecture impedes the convergence performance of the neural network, leading to poor classification accuracy. To mitigate these problems, this paper proposes a joint network of Hybrid 3D-2D CNN and pyramid residual (J-NHPR) to realize the classification of HSI. The combined deep pyramid residual network can significantly enhance the performance of the suggested network with the HSI data by progressively improving the diversity of high-level spectral-spatial properties across layers. Our tests, which included eight distinct approaches for classification and three well-known HSI data sets, revealed that our recently built model, J-NHPR, can provide comparative advantage (in terms of classification accuracy) over other HSI classification methods. The source code of the J-NHPR model is available at https://github.com/Dreamvai/J-NHPR.