Improved classification algorithm for hyperspectral remote sensing images based on the hybrid spectral network model

In order to make better use of the spatial information and spectral information to improve the classification accuracy of hyperspectral remote sensing images, this article proposes an improved classification algorithm, a-HybridSN, based on the HybridSN (hybrid spectral network) model. First, the hyperspectral remote sensing image is carried out by principal component analysis to reduce the dimensionality for eliminate spectral redundancy. Second, the channel attention mechanism and spatial attention mechanism are integrated between the flatten layer and the dropout layer of the HybridSN model, which helps alleviate the interference of inconsistent classification results caused by random sampling in the dropout layer. Finally, experiments are conducted on the images from the GF-5 and Sentinel-2. The experimental results indicate that the proposed method can effectively improve the classification accuracy and alleviate the phenomenon of salt and pepper in the classification result image. The proposed method achieved state-of-the-art performance with an overall accuracy of 89.55% and a Kappa coefficient of 0.8712. Comparative analyses demonstrate a-HybridSN’s superior performance against established methods such as HybridSN, random forest, SoftMax, decision tree, and support vector machine. These results underscore the robustness and adaptability of a-HybridSN.