Spatial-Aware Network for Hyperspectral Image Classification

Deep learning is now receiving widespread attention in hyperspectral image (HSI) classification. However, due to the imbalance between a huge number of weights and limited training samples, many problems difficulties have arisen from use deep methods HSI To handle this issue, an efficient learning-based classification method, namely, spatial-aware network (SANet) has been proposed paper. The main idea SANet exploit discriminative spectral-spatial features by incorporating prior domain knowledge into architecture, where edge-preserving side window filters are used as convolution kernels. Thus, small parameters optimize. This makes it fit for sample sizes. Furthermore, able not only aware local spatial structures using filtering framework, but also learn making hierarchical architecture label information. experimental results on four widely data sets demonstrate that our significantly outperforms state-of-the-art approaches when samples available.