Ship Detection for Polarimetric SAR Images Based on Anchor-Free Network in High Sea State

Ship detection of polarimetric SAR (PolSAR) images has been developed for marine monitoring due to the abundant scattering features. However, PolSAR images in high sea state generally contain serious surface clutter interference and targets are blurry, which both hamper the performance of ship detection. To overcome the issue, a novel anchor-free ship detection network is proposed for PolSAR images. In the proposed method, polarimetric scattering features are extracted by polarimetric decomposition firstly, and multi-head anchor-free detector is developed to improve the detection performance of ships in high sea state. Moreover, Shannon entropy normalization and center-penalty are developed for optimization of model robustness, which aims to enhance the robustness of the detector. Experimental results on the GaoFen -3 dataset in high sea state demonstrate that the proposed method has superior detection performance than other advanced deep detection approaches.