MVCH: Multiview Collaborative Hybrid Network for Hyperspectral Image Classification

Abstract

Although deep learning methods have achieved significant progress in hyperspectral image (HSI) classification, existing approaches typically focus on modeling a single dependency view, such as local spatial–spectral, structural relational, or global contextual view, making it difficult to coordinate complementary views within a unified representation space. To address this issue, a multiview collaborative hybrid network (MVCH) is proposed for HSI classification. MVCH integrates multiple dependency views within a unified representation framework. Specifically, the graph-depthwise cooperative (GDC) module collaboratively models structural relational and local spatial–spectral views to enhance feature discriminability for fine-grained variations and discontinuous structural regions. Furthermore, the context-interaction Transformer (CIT) collaboratively models global contextual, local perceptual, and dynamic dependency views, enabling effective contextual interaction and adaptive dependency learning in complex HSI scenes. Extensive experiments conducted on three representative urban and agricultural HSI datasets demonstrate that MVCH effectively improves feature representation capability in complex scenes through unified multiview collaborative feature learning, achieving competitive classification performance compared with existing representative methods. The code is available at https://github.com/dsj0507/MVCH