Skip to content
Open access

Axial Reorganization and Group Offset Mamba for Hyperspectral Image Classification

2026 · IEEE Transactions on Geoscience and Remote Sensing · Vol 64, pp. 5533015-5533015 · 0 citations · 54 references

Abstract

Hyperspectral image classification (HSIC) is a core task in the field of remote sensing. Its inherent characteristics of high spectral redundancy, complex spatial structure, and strong heterogeneity of land cover make it difficult to capture spatial global correlations and inefficiently model spectral long-range dependencies. Traditional methods are difficult to balance feature discriminability and modeling efficiency, and the accuracy of fine classification in complex scenes is limited. In this article, an axial reorganization and group offset Mamba (ARGO-Mamba) network is proposed. First, a local–global attention guidance (LGAG) module is proposed, which preserves key fine information through local feature aggregation and then dynamically enhances the representation of important features during the discriminative feature calibration process. Second, an axial reorganization spatial Mamba (ARSM) module is introduced, which abandons the optimization of scanning strategies in spatial modeling and directly encodes the directional structural information of land features through axial reorganization and captures global spatial contextual information more efficiently. Finally, a group offset spectral Mamba (GOSM) module is constructed, which creates diverse spectral grouping views through spectral grouping and group shift and simultaneously models its long-range spectral dependencies using Mamba. Experiments on the Indian Pines, Pavia University (PU), Houston 2013, and WHU-Hi-Longkou datasets show that the OA values of the proposed ARGO-Mamba increase by 0.56%–2.5% compared to the most advanced methods in recent years, with an average increase of 1.3%. It performs more outstandingly in small sample situations and highly heterogeneous scenes, providing a new solution for complex remote sensing classification tasks. The source code will be released on https://github.com/liang-zhenlong/ARGO-Mamba

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.