In recent years, the visual detection method based on deep learning has been widely used in the field of ship target detection on the sea surface. In order to solve the problems of low detection accuracy and poor detection performance for small targets in traditional visual detection methods, a sea surface ship target detection method based on Attention-YOLOv3 is proposed, which effectively improves the detection performance of ship targets. Based on the research and analysis of the structure and characteristics of mainstream One-stage and Two-stage models, YOLOv3's feature is used to extract network Darknet-53 for obtaining image features, and the deep and shallow semantic information in the network is extracted through the fusion of features in the Feature Pyramid Network (FPN) network structure, and attention mechanism modules are added to further optimize network performance. The improved Attention-YOLOv3 model is applied to the verification of ship detection scenarios on the sea surface. It is trained based on the collected data set of ship targets in COCO format, and tested using images containing ship targets on the sea surface as the test set. The experimental results show that the improved Attention-YOLOv3 network solves the problem of insensitivity to small object detection and achieves higher detection performance compared to the original detection network model.