Semisupervised Hyperspectral Image Classification Based on Hybrid Spatial-Spectral Tensor Network

Deep learning-based methods have shown great promise in hyperspectral image (HSI) classification due to their ability to learn discriminative feature representations from abundant labeled data. However, these methods often struggle with the curse of dimensionality and computational complexity when faced with the high dimensionality of HSI data and the limited number of annotated samples. To overcome these challenges, we propose a novel Hybrid Spatial-Spectral Tensor Network (HSSTN), which integrates novel low-rank tensor analysis techniques and Convolutional Neural Networks(CNNs) in a synergistic manner to effectively capture the latent relationships among sub-tensors. Additionally, we introduce a semisupervised framework S-HSSTN, which builds upon HSSTN by incorporating diverse pathways and functional modules. S-HSSTN effectively exploit the spectral-spatial features while minimizing computational demands and requiring fewer labeled samples. Experiments conducted on the PaviaU dataset verify the robustness and efficiency of the proposed S-HSSTN model.