Design of USV System for Autonomous Return

Unmanned surface vessel (USV) in complex water environments is affected by many different forces, and the low friction and inertial drift of the water surface make it difficult for USV to accomplish complex autonomous homing tasks. We propose an integrated solution that incorporates multi-task sensing and dynamic path planning. Through the improved YOLOv8 algorithm combined with deformable convolutional network (DCNv4), and the semantic segmentation of the harbor area using the UNet++ algorithm, we complete the water surface object detection work. Meanwhile, a sparse topological graph model is constructed based on spatio-temporal dual constraints, and a Dijkstra algorithm with improved dynamic relaxation mechanism is proposed to accomplish the autonomous return task. Experiments show that the proposed method is better than the traditional method in terms of object detection accuracy, harbor segmentation accuracy and path planning efficiency, which provides reliable theoretical support and technical realization for autonomous navigation of USV.