Review of Deep Semi-supervised Learning in SAR Image Interpretation

The rapid development of deep learning brings effective solutions and remarkable achievements for synthetic aperture radar (SAR) image interpretation. The significant improvement of deep model performance usually relies on huge amount of labeled sample data. However, it is a challenging work to manually collect and label sufficient samples for training a complex deep convolutional neural networks. To make full use of small amount of labeled samples and large amount of unlabeled samples, we resort to semi-supervised learning (SSL) to improve the generalization ability of deep models. In this article, we first briefly introduce the setting of SSL methods and analyze the existing work related on SSL algorithms. Then, the mainstream SSL methods for SAR image interpretation are listed according to specific applications, together with experimental dataset and evaluation results. Finally, the strengths and shortcomings of the stated algorithms are discussed, and the possible future directions are summarized. This article aims to provide an systematic and comprehensive overview for scholars working on semi-supervised learning research in the SAR image interpretation field.