高低频增强与跨层图卷积聚合的高光谱图像分类OA

To address the problems of local texture and edge-detail loss,limited receptive fields in the convolution branch,and insufficient use of cross-layer structural information in the graph branch for hyperspectral image classification,this paper proposes a CNN-GCN joint classification model with high-low frequency enhancement and cross-layer graph convolution aggregation.The model improves input representation through high-low frequency residual enhancement,extracts multi-scale spectral-spatial features using a multi-stage dynamic convolution encoder,enhances regional structural modeling by cross-layer weighted aggregation in the graph branch,and performs collaborative modeling through cross-branch attention fusion.Experiments on three public datasets,Indian Pines,Pavia University,and Salinas,achieve overall accuracies of 92.94%,95.11%,and 97.50%,with corresponding Kappa coefficients of 91.94%,93.50%,and 97.22%,respectively.The results show that the proposed method can effectively integrate local details,spatial context,and regional topological structure information,achieving competitive classification performance across different types of hyperspectral classification scenes.