A Semi-Supervised Arbitrary-Oriented SAR Ship Detection Network based on Interference Consistency Learning and Pseudo Label Calibration

The rapid development of deep learning cannot be achieved without the support abundant labeled data. However, obtaining such a large amount annotated data needs professionals in field synthetic aperture radar (SAR) image understanding, which leads to scarcity SAR datasets with annotations. annotations poses bottleneck performance ship detectors based on learning. Recently, semi-supervised has become hot paradigm, can mine effective information from unlabeled further improve detectors. existing detection studies all adopted multi-stage frameworks, are complex and inefficient. In this article, we first design an end-to-end framework for detection. To overcome strong interferences resulting imaging or quantization processes SAR, Introduce interference consistency mechanism enhance model's robustness. solve background inshore scenario, pseudo-label calibration network is designed calibrate according context knowledge around ships. Based HRSID other four datasets, superiority proposed approach over several state-of-the-art frameworks been evaluated under various labeling ratios, i.e., 1%, 5%, 10%, 100%.