Triple-Attention Residual Networks for Hyperspectral Image Classification

The attention mechanism has been widely applied in hyperspectral image classification, yet most current methods rely on a relatively simple form of attention, which fails to extract features deeply. To address this issue, this paper proposes a triple-attention residual networks (TARN) for hyperspectral image (HSI) classification. This method can not only emphasize the importance of different bands in hyperspectral image classification, but also capture pairwise relationships between spatial feature pixels at the pixel level. By integrating these attention mechanisms, we achieve efficient extraction of spectral-spatial features. Extensive experiments demonstrate that the proposed artificial neural network significantly outperforms state-of-the-art methods, exhibiting its superior performance in hyperspectral image classification.