DCAPT-Net: A Multi-Scale Ship Detection Algorithm for SAR Images

Within the recent timeframe, the detection of ships in SAR images has become a critical task for many countries. To tackle the issue of multi-scale ship detection, this study presents DCAPT-Net, a new ship detection algorithm. This algorithm enhances multi-level feature fusion by introducing a novel dense connection attention pyramid structure. Furthermore, it incorporates a dedicated small object detection layer and a four-head Transformer structure detection head to improve the representation of small targets. Collectively, these innovations achieve effective multi-scale ship detection. Studies performed using the High Resolution SAR Images Dataset (HRSID) and the SAR Ship Detection Dataset (SSDD) highlight the DCAPT -Net's superior performance in identifying ship targets. Additionally, ablation studies confirm the efficacy of the network improvements introduced.