DiffDet4SAR: Diffusion-based Aircraft Target Detection Network for SAR Images

Aircraft target detection in SAR images is a challenging task due to thediscrete scattering points and severe background clutter interference.Currently, methods with convolution-based or transformer-based paradigms cannotadequately address these issues. In this letter, we explore diffusion modelsfor SAR image aircraft target detection for the first time and propose a novel\underline{Diff}usion-based aircraft target \underline{Det}ection network\underline{for} \underline{SAR} images (DiffDet4SAR). Specifically, theproposed DiffDet4SAR yields two main advantages for SAR aircraft targetdetection: 1) DiffDet4SAR maps the SAR aircraft target detection task to adenoising diffusion process of bounding boxes without heuristic anchor sizeselection, effectively enabling large variations in aircraft sizes to beaccommodated; and 2) the dedicatedly designed Scattering Feature Enhancement(SFE) module further reduces the clutter intensity and enhances the targetsaliency during inference. Extensive experimental results on theSAR-AIRcraft-1.0 dataset show that the proposed DiffDet4SAR achieves 88.4\%mAP$_{50}$, outperforming the state-of-the-art methods by 6\%. Code isavailabel at \href{https://github.com/JoyeZLearning/DiffDet4SAR}.