Semi-supervised Classification of Hyperspectral Image through Deep Encoder-Decoder and Graph Neural Networks

The hyperspectral image (HSI) classification is a challenging task due to the high dimensional spectral feature space, and a low number of labeled training samples. To overcome these issues, we propose a novel methodology for HSI classification, called DAE-GCN, which is based on deep neural networks. The main goal is to preserve both spectral and spatial features in the classification task by using only a few number of labeled training samples. Firstly, we propose a deep autoencoder (DAE) model, which learns to extract relevant features from the HSI. It seeks to find a better representation of the HSI in order to improve the classification rates. Secondly, we construct a spectral-spatial graph using the obtained latent representation space. The aim is to take into account the spectral and spatial features by considering distances between neighboring pixels. Finally, a semi-supervised graph convolutional network (GCN) is trained based on the latent representation space to perform the spectral-spatial classification of HSI. The main advantage of the proposed method is to allow the automatic extraction of relevant information while preserving the spatial and spectral features of data, and improve the classification of hyperspectral images even when the number of labeled samples is low. Experiments are conducted on two real HSIs, including Indian Pines, and Pavia University datasets. Experimental results show that the proposed model DAE-GCN is competitive in classification performances compared to various state-of-the-art methods.