Spectral-spatial graph convolutional networks for semei-supervised hyperspectral image classification

Collecting label samples is quite costly and time consuming for hyperspectral image (HSI) classification tasks. Semi-supervised learning framework, which combines the intrinsic information of labeled and unlabeled samples can alleviate the deficient labeled samples and increase the accuracy of HSI classification. In this paper, we propose a novel framework for semi-supervised learning on multiple spectral-spatial graphs that is based on graph convolutional networks (SGCN). In the filtering operation on graphs we consider the spatial information and spectral signatures of HSI simultaneously. The experimental results on three real-life HSI data sets, i.e. Botswana Hyperion, Kennedy Space Center, and Indian Pines, show that the proposed SGCN can significantly improve the classification accuracy. For instance, the over accuracy on Indian Pine data is increased from 78% to 93%.