Feature aligned ship detection based on improved RPDet in SAR images

Abstract

Ship detection in Synthetic aperture radar (SAR) images is an important and challenging task. A number of deep learning-based methods have been proposed for the SAR ship detection task and achieved remarkable performance. However, existing methods still suffer from some problems, such as feature misalignment, multi-scale, and small target missing. To address these problems, we first introduce RPDet into the SAR ship detection task to address the feature misalignment problem, then propose an improved RPDet to address the multi-scale and small target missing problems. Specifically, the Path Aggregation Feature Pyramid Network (PAFPN) is utilized in our method to fuse multi-scale features, area distribution of ships in SAR images is analyzed for better feature-levels selection, and Complete-IoU (CIoU) loss is adopted for more accurate and stable bounding box regression. Extensive results on a publicly available SAR ship detection dataset SSDD show that the improved RPDet achieves 3.70% higher average precision (AP) than the original RPDet, and obtains SOTA results compared to other methods.