MS3D-CoordAttention algorithm forHyperspectral Remote Sensing Image Classification

In recent years, the rapid advancement of remote sensing technology has furnished a robust data foundation for landcover classification through hyperspectral image. However, conventional classification methods often struggle with effective feature extraction and achieve limited accuracy when processing such data. Further challenges, including the scarcity of samples, spectral variability within classes, and spectral similarity between different classes, limit the efficacy of classification. To address these issues, this paper introduces a landcover classification approach based on the MS3D-CoordAttention framework. The method employs a 3D-CNN for deep feature extraction from hyperspectral images and uses weighted fusion of multi-scale features to generate enriched remote sensing data. Experimental results on the Indian Pines, Pavia University, and Botswana datasets demonstrate that the enhanced method substantially outperforms traditional approaches in both accuracy and F1-score. It also shows the improvements in training efficiency. This study provides valuable technical insights for automated landcover classification, laying a strong foundation for scalable ecological conservation, more effective resource management, and informed environmental policymaking.