Multiorder Graph Convolutional Network With Channel Attention for Hyperspectral Change Detection
Yuxiang Zhang, Rui Miao, Yanni Dong, Bo Du
Abstract
Hyperspectral change detection aims to obtain the change information of objects in the multitemporal hyperspectral images. Recently, with the advantages in fully extracting the image features of irregular areas, graph convolutional network (GCN) has attracted increasing attention for hyperspectral change detection. Existing GCN-based change detection methods usually use the graph structure constructed by superpixels to reduce computational cost, which ignores the multi-order difference information among graph nodes and the local difference information within superpixels. To addressing these problems, this paper proposes an efficient multi-order GCN with channel attention module for hyperspectral change detection. Specifically, the multi-order GCN module is designed by repeatedly mixing the feature representations of neighborhoods. The channel attention module is then proposed to enhance the difference features of bitemporal HSIs. After that, the pixel-wise change detection is accomplished by a lightweight feature fusion module and a fully connected layer. Experiments on three hyperspectral datasets illustrated the effectiveness of the proposed algorithm.