Abstract—In recent years, Graph Convolutional Networks (GCN) have witnessed increasing applications in hyperspectral image classification tasks. In comparison to Convolutional Neural Networks, graph representations providing a more effective means to exploit the complex interplay of spatial and spectral features in hyperspectral images, emphasizing their potential to address the challenges associated with limited labeled data in hyperspectral image classification tasks. Although Graph Convolutional Networks are able to capture Hyperspectral Image (HSI) spatial context structure well, they lack the ability to capture pixel-level spectral spatial features compared to Convolutional Neural Networks (CNNs). In order to fully utilize the advantages of Convolutional Neural Networks and Graph Convolutional Networks, in this paper, we propose a model that combines superpixel-based Hypergraph Convolutional Networks features with patch-based Convolutional Neural Network features, engaging in feature learning on both small-scale regular regions and large-scale irregular regions. To test the model, we select 2% of the total number of dataset labels for training, 2% of the total number of dataset labels for validation and the 96% labels for testing. An overall accuracy of 92.37% and 95.86% was obtained in the Indian Pines and Pavia University dataset which is higher than other state-of-the-art methods and achieved a more accurate classification results on the landcover boundary areas.
 
Keywords—computer vision, hyperspectral image classification, graph learning, convolutional networks


Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.