Deep Contrastive Learning Network for Small-Sample Hyperspectral Image Classification

Recently, deep learning methods have been widely used in hyperspectral image (HSI) classification and achieved good performance. However, the performance of these methods may be limited because of the scarcity of labeled samples in HSI data. To solve the small-sample classification problem, a deep contrastive learning network (DCLN) method is proposed in this paper. The proposed DCLN method first constructs contrastive groups and trains the network through contrastive learning. Then, it uses the trained network to extract spectral-spatial features of HSI pixels and generates pseudo-label for each unlabeled sample based on the spatial-spectral mixing distance. Finally, the pseudo-labeled samples with higher confidence are selected and added to the original training set to retrain the network. By gradually increasing pseudo-labeled samples and refining the contrastive learning network, the model shows good feature learning ability and classification performance with the limited labeled samples. Experimental results on 4 public HSI datasets demonstrate that the proposed DCLN method can achieve better performance than existing state-of-the-art methods.