A Triple-Path Spectral–Spatial Network With Interleave-Attention for Hyperspectral Image Classification

To exploit hyperspectral image's (HSI) spectral-spatial information and reduce network complexity, a triple path convolution neural (CNN) with interleave-attention mechanism is constructed for high-precision classification. A hybrid branch proposed to capture joint features, which are later utilized as complementary purely spectral spatial features. Furthermore, the elaborately designed increase interaction of data from spectral, spatial, branches they propagate through feature integration. Meanwhile, two attention modules adopted in corresponding optimize extracted features better representation. We utilize several real HSI datasets evaluate performance, demonstrates that can obtain very satisfactory performance fewer parameters low computational complexity.