The marine environment is polluted and the marine ecological environment continues to be destroyed because of the over-exploitation and utilization of marine resources. Countries around the world have taken measures to solve the problem. In order to protect the coastal ecological environment, China has also begun to implement measures to monitor ships in the territorial sea and inshore waters. Ship detection is the focus of current research. Synthetic aperture radar has been widely used in the field of marine remote sensing monitoring because of its advantages such as all-time and all-weather monitoring, which provides strong data guarantee and technical support for multi-scale ship detection. With the improvement of the algorithm, researchers put more energy on the detection accuracy, while ignoring the detection speed and landing application of the algorithm. At present, the popular detection algorithms basically rely on the powerful graphics processing unit, so they can not be deployed in the front line of ocean monitoring. In order to solve the above problems, a lightweight SAR ship detection algorithm which can effectively enhance the receptive field is proposed in this paper. Firstly, ShuffleNetV2 is used as the backbone network. It is helpful to reduce the number of calculation parameters and memory consumption. Secondly, the improved space pyramid pool module is introduced. It not only expands the receptive field of the model effectively, but also makes the ship feature information be further mined. Simultaneously, the spatial attention module is added to enhance the model's attention to the spatial location information and improve the target positioning ability. Then, the improved path aggregation network is used to transfer more abundant ship positioning features from bottom to top. It is helpful to increase the shallow position information and multi-scale features. Finally, the experimental results on the SAR ship detection dataset show that the model size is 5.3 MB, the average detection accuracy is 94.7%, and the detection speed is 46 FPS. Compared with the current mainstream detection methods, it has fewer parameters, floating-point operations and smaller model size. It not only ensures high detection accuracy, but also realizes fast detection. The deployment verification is carried out based on the mobile phone. The results show that the ship can be accurately identified under the inshore complex background and offshore small target scene, the recognition accuracy is 85.7 %, and the test speed is 32 FPS. Compared with the computer equipment with excellent GPU performance, the detection accuracy and reasoning speed are all reduced, but still meet the real-time requirements. This method is helpful to transplant to FPGA or embedded mobile devices. It has great practical application value in real-time maritime safety monitoring and protection of marine ecological environment.