Pyramid Hierarchical Spatial-Spectral Transformer for Hyperspectral Image Classification

Abstract

The transformer model encounters challenges with variable-length input sequences, leading to efficiency and scalability concerns. To overcome this, we propose a pyramid-based hierarchical spatial-spectral transformer (PyFormer). This innovative approach organizes input data hierarchically into pyramid segments, each representing distinct abstraction levels, thereby enhancing processing efficiency. At each level, a dedicated transformer encoder is applied, effectively capturing both local and global context. Integration of outputs from different levels culminates in the final input representation. In short, the pyramid excels at capturing spatial features and local patterns, while the transformer effectively models spatial-spectral correlations and long-range dependencies. Experimental results underscore the superiority of the proposed method over state-of-the-art approaches, achieving overall accuracies of 96.28% for the Pavia University dataset and 97.36% for the University of Houston dataset. In addition, the incorporation of disjoint samples augments robustness and reliability, thereby highlighting the potential of PyFormer in advancing hyperspectral image classification (HSIC).