Hyperspectral image classification based on spatial pyramid attention mechanism combined with ResNet

In order to extract spatial-spectral joint features of hyperspectral images, this paper proposes a hyperspectral image classification model based on an improved spatial pyramid attention mechanism residual network. Firstly, principal component analysis is used to remove spectral redundancy, and combined with spatial pyramid attention mechanism, a residual network based hyperspectral image classification model is improved to obtain refined features. Then, the spatial pyramid attention model is used to achieve multi-scale joint feature attention, improve sensitivity to joint features, and effectively emphasize and focus on spatial and spectral information for information exchange. Finally, the classification label is obtained through the Softmax classifier. The proposed method in this paper is tested on MUUFL and Trento datasets, and the experimental results show that the overall classification accuracy of the proposed algorithm reaches 94.08% and 98.32%, respectively. Compared to other hyperspectral classification models, the convergence speed of this model is faster, and it achieves significant improvement in classification performance with higher ground object classification accuracy.