To address the issue of inadequate extraction of spectral and spatial local details in hyperspectral image(HSI)classification,an innovative spectral-spatial attention network named MSSAN is proposed.It includes spectral and spatial feature extraction modules with multi-scale dilated convolution blocks,residual and dense extraction blocks,and attention mechanism.The network effectively integrates shallow and deep features,with the multi-scale dilated convolution block enhancing the extraction of local details.The subsequent attention mechanism highlights key features and makes full use of spectral and spatial information.Comparative experiments demonstrate MSSAN's superior performance on IP,UP,and SV datasets compared to existing advanced algorithms.Ablation experiments confirm the efficacy of MSSAN's modular combination.
国家科技期刊平台
NSTL
万方数据
