Hyperspectral Image Classification Using Multi-feature Fusion Residual Hypergraph Convolution Network

Convolutional neural network (CNN) has been broadly adopted on hyperspectral image (HSI) processing due to its impressive feature extraction capabilities. Nevertheless, it is still a challenge for CNN-based hyperspectral image classification methods to extract more effective spectral-spatial features considering all spectral bands. The recent advances of graph convolutional neural networks (GCN), a combination of CNN and graph representation learning, have shown great potential in the HSI. However, the existing GCN-based methods have lower accuracy than the mature CNN methods due problems such as overdependence on the adjacency matrix, usage of a single modal feature. In this paper, we propose the Multi-Feature Fusion Residual Hypergraph Convolution Network (MF-RHCN) for HSI classification. We present a novel Hypergraph Convolution Network that fuses multimodal features effectively, and automatically updates the hyperedge weights, which solves the difficulties encountered by the previous GCN-based hyperspectral classification methods. MF-RHCN first generates hyperedges from features of different modalities to construct a hypergraph representing multimodal features in HSI. Then, the HSI and the extracted hypergraph are input into the hypergraph convolutional neural network for learning. We enhance the hypergraph convolutions with residual connection and input mapping, which helps reduce the over-smoothing issue associated with deep hypergraph models. Sufficient experiments on three datasets have proved the effectiveness of MF-RHCN.