SSFSL: Self-Supervised and Few-Shot Learning for Cross-Domain Hyperspectral Image Classification

Few-shot learning (FSL), a kind of meta-learning, has shown great advantages in hyperspectral image (HSI) classification with a small number of labeled samples. It allows the classification model trained with a limited number of labeled samples on a given HSI (in the source domain) to classify other HSIs (in the target domain). However, due to omitting the continuity of HSI spectral sequences, existing FSL based classification methods have limitations in feature extraction and exploiting a large number of unlabeled samples in the target domain. To address the aforementioned two issues, we propose a novel cross-domain HSI classification method based on self-supervised learning and FSL (SSFSL). Specifically, self-supervised learning and FSL are jointly constructed to extract transferable knowledge in the source domain. Then, a novel adaptive soft label generation method is proposed to make full use of the information of unlabeled samples in the target domain. Next, the designed residual shrinkage network is used as a feature extractor to train in both the source and target domains. The qualitative and quantitative experiments on four publicly available datasets demonstrate the superiority of our SSFSL over the state-of-the-art methods, in terms of both visual effect and quantitative metrics.