Ship Detection based on M2Det for SAR images under Heavy Sea State

Abstract:
Deep learning has been applied to ship detection of SAR images, which is more advantageous than traditional methods. However, sea state affects the imaging characteristics of ships in SAR images, making ship detection of SAR images under heavy sea state difficult. So far, the impact of sea state on the results of deep learning has rarely been studied. In this paper, a ship detection framework using a single-shot object detector based on multi-level feature pyramid network is proposed. By extracting multi-level and multi-scale features, the detection rate in heavy sea state background is improved. The adjustable receptive field is added to the detection framework by multi-scale input to reduce the false alarm rate in the land. The training data is an established SAR image ship database. Experiments on the SAR images under heavy sea state verify that the proposed method can achieve an excellent performance.