Global Random Graph Convolution Network for Hyperspectral Image Classification

Machine learning and deep methods have been employed in the hyperspectral image (HSI) classification field. Of methods, convolution neural network (CNN) has widely used achieved promising results. However, CNN its limitations modeling sample relations. Graph (GCN) introduced to HSI due demonstrated ability processing Introducing GCN into classification, key issue is how transform HSI, a typical euclidean data, non-euclidean data. To address this problem, we propose supervised framework called Global Random Convolution Network (GR-GCN). A novel method of constructing graph adopted for network, where built by randomly sampling from labeled data each class. Using technique, size constructed small, which can save computing resources, obtain an enormous quantity graphs, also solves problem insufficient samples. Besides, random combination samples make generated more diverse robust. We use with trainable parameters, instead artificial rules, determine adjacency matrix. An matrix obtained flexible stable, it better represent relationship between nodes graph. perform experiments on three benchmark datasets, results demonstrate that GR-GCN performance competitive current state-of-the-art methods.