Synthetic Aperture Radar (SAR) is a high-resolution space-borne (or airborne) radar systemand Automatic Target Recognition (ATR) in SAR image is one of the key technologies of intelligent image interpretation.Traditional SAR ship target detection algorithms are mostly limited by the scene and have poor generalization abilityso a detection model based on YOLOv3 network is designed.Multi-scale prior box is used to detect the targetand the optimal weight of the model is obtained through training to realize end-to-end target detection.The test results show thatcompared with Faster R-CNN algorithmYOLOv3 has better performance in accuracy and running speed.