An On-Orbit Ship Detection and Classification Algorithm for Sar Satellite

Ship detection using synthetic aperture radar (SAR) plays a vital role in the wide area ocean surveillance. Especially the real-time ship detection and classification, significantly promotes the illegally operating ships monitoring performance. In this study, an on-orbit ship detection and classification method is developed for SAR satellite. An adaptive sliding window is developed to extract the water connected domain and propose the suspected ship target areas. The OpenSARShip dataset and manually selected non-ship slice images are adopted to train a deep learning model, which is applied to classify the proposed ship slice images into three types (cargo ship, other ship and false alarm). The results demonstrate the improvement performance of the proposed method over the constant false alarm rate (CFAR) method, where the detection accuracy improved from 88.5% to 98.4% and the false alarm rate mitigated from 11.5% to 0.7% compared with CFAR respectively. Meanwhile, the proposed method can achieved verification and testing accuracy of 97.2% and 93.4% respectively for ship type classification.