Fusion Convolution Hyperspectral Image Classification Based on Parallel Attention Strategy

Abstract:
For hyperspectral image classification, traditional 3D convolution can obtain spatial information without losing spectral dimension information. Therefore, hyperspectral classification model based on 3D convolution usually has high classification accuracy. However, 3D convolution models are usually complex and consume a lot of computing resources. In order to reduce the complexity of network structure of traditional convolution based on 3D convolution, this paper proposes a fusion convolution hyperspectral image classification model based on parallel attention strategy. The model uses 3D and 2D convolution to learn spectral-spatial information at the same time, and effectively improves the classification accuracy of the model through parallel attention strategy. Experiments on two real data sets demonstrate the effectiveness of our proposed model.