A Lightweight Position-Enhanced Anchor-Free Algorithm for SAR Ship Detection

As an active microwave device, synthetic aperture radar (SAR) uses the backscatter of objects for imaging. SAR image ship targets are characterized by unclear contour information, a complex background and strong scattering. Existing deep learning detection algorithms derived from anchor-based methods mostly rely on expert experience to set series hyperparameters, it is difficult characterize unique characteristics targets, which greatly limits accuracy speed. Therefore, this paper proposes new lightweight position-enhanced anchor-free algorithm called LPEDet. First, resolve target contours multiscale performance problems, we used YOLOX as benchmark framework redesigned backbone, NLCNet, balances speed accuracy. Second, scattering target, designed attention strategy, suppresses clutter adding position information channel that highlights more accurately identify locate target. The experimental results two large-scale datasets, SSDD HRSID, show our method achieves higher faster than state-of-the-art methods.