Ship detection by satellite-carried synthetic aperture radar (SAR) imagery is currently a hot topic in the development of space remote sensing technology. Due to the limitation of hardware resources and the requirement of high-speed processing, the current deep learning model has the problems of low detection rate and high false alarm density in ship detection in complex scenes. To address the issues, we propose a lightweight method to detect ships in SAR images for embedded platforms based on the upgraded YOLOv7-tiny, called Lightweight Ship Detection-YOLO (LSD-YOLO). Firstly, the MobileNetV3Block module is introduced into the backbone network, and incorporating a ULSAM attention mechanism to enhance the capacity for the method to capture important features of ships in complex scenarios. Secondly, we utilize SPPFCSPC module and replace certain standard convolutions with GsConv. The design not just reduces computational complexity but also significantly enhances the method’s detection performance. Notably, the average detection accuracy on the SSDD dataset is 97.7%, which is a 1.9% improvement over the yolov7-tiny model, with the number of parameters and computation decreasing by 21.7% and 24.4%, respectively, and significantly outperforming the other compared models. Finally, on the Cambricon Siyuan220, the proposed method has a detection speed of 153 frame/s, and achieves the accurate and real-time detection in SAR image ship detection task.