Multiview Subspace Clustering of Hyperspectral Images based on Graph Convolutional Networks
This work addresses the problem of clustering high-dimensional hyperspectral images for remote sensing applications, representing an incremental improvement by combining existing techniques.
The paper tackled hyperspectral image clustering by proposing a multiview subspace clustering method using graph convolutional networks, achieving overall accuracies of 92.38%, 93.43%, and 83.82% on three datasets and outperforming state-of-the-art methods.
High-dimensional and complex spectral structures make clustering of hy-perspectral images (HSI) a challenging task. Subspace clustering has been shown to be an effective approach for addressing this problem. However, current subspace clustering algorithms are mainly designed for a single view and do not fully exploit spatial or texture feature information in HSI. This study proposed a multiview subspace clustering of HSI based on graph convolutional networks. (1) This paper uses the powerful classification ability of graph convolutional network and the learning ability of topologi-cal relationships between nodes to analyze and express the spatial relation-ship of HSI. (2) Pixel texture and pixel neighbor spatial-spectral infor-mation were sent to construct two graph convolutional subspaces. (3) An attention-based fusion module was used to adaptively construct a more discriminative feature map. The model was evaluated on three popular HSI datasets, including Indian Pines, Pavia University, and Houston. It achieved overall accuracies of 92.38%, 93.43%, and 83.82%, respectively and significantly outperformed the state-of-the-art clustering methods. In conclusion, the proposed model can effectively improve the clustering ac-curacy of HSI.