Hyperspectral image (HSI) classification faces significant challenges due to the spatial–spectral heterogeneity of land covers and the geometric rigidity of standard convolutions. Although Transformers offer powerful global modeling capabilities, their quadratic computational complexity limits practical efficiency. To address these limitations, this paper proposes a novel hierarchical framework named MDS3-Net (Multiscale Deformable Spectral–Spatial Sequence Network). Specifically, we design a Multiscale Spectral-Deformable Convolution (MSDC) module that adopts a cascaded strategy to sequentially extract discriminative spectral features and adaptively align spatial receptive fields with irregular object boundaries. To capture long-range dependencies efficiently, a Spectral–Spatial Sequence (S3) Encoder is introduced based on a gated large-kernel convolution mechanism, achieving global context modeling with linear complexity. Furthermore, a Dual-Path Feature Extraction (DPFE) module is proposed to perform semantics-preserving dimension reduction via spectral reorganization and spatial attention. Experimental results on four public datasets demonstrate that the proposed MDS3-Net achieves state-of-the-art classification performance and exhibits superior robustness under limited training samples compared to existing methods.
30 March 2026 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Hyperspectral image (HSI) classification has emerged as a critical task in remote sensing. However, the inherent high spectral dimensionality, limited labeled data, and complex spatial structures in HSIs present substantial challenges to traditional and even deep learning-based approaches. Traditional convolutional neural networks (CNNs) often struggle to capture the irregular, non-Euclidean spatial relationships among pixels, whereas standard Transformer architectures are limited in modeling localized spectral-spatial dependencies and are prone to overfitting in scenarios with limited training data. In this paper, we propose a novel Dual-Scale Graph Transformer with Spectral Spatial Convolution (DGTSS) framework that integrates multi scale structural modeling and deep spectral feature extraction in a unified architecture. Specifically, we construct two complementary graph representations: a pixel-level graph that captures fine grained spatial relationships and a super-pixel level graph that encodes broader semantic structures. These dual-scale graphs are processed by a Graph Transformer module that incorporates node features and enables effective modeling of long-range dependencies and context-aware feature aggregation. Furthermore, we introduce a dedicated Spectral-Spatial Convolution (SSC) module to enhance spectral feature discrimination by maintaining local spatial continuity. Extensive experiments on four benchmark hyperspectral datasets demonstrate that our method achieves state-of-the-art classification performance. Moreover, compared with existing Transformer-based models, our framework significantly reduces model complexity and computational cost, while maintaining superior performance.