Objective To address the poor perception of marine floating obstacles such as aquaculture areas, buoys, and small floating objects by unmanned surface vessel （USV） navigation radar, a unified technical solution is proposed for stable and accurate perception of various types of marine floating obstacles.

Methods  An environmental perception method based on constructing and updating an occupancy grid map from navigation radar echo video data is presented. First, a multi-level set approach is adopted to describe the inclusion relationship between radar tracks and echo points, laying the foundation for the construction of the grid map. During this process, neighboring tracks are aggregated based on group adjacency to mitigating trajectory deviation caused by target splitting. Then, a probability update algorithm for the occupancy grid map, based on the natural logarithm function, is proposed to effectively distinguish sea clutter from minor marine floating obstacles by leveraging historical data. Finally, a probability diffusion model for the grid map, grounded in track attributes, is established to ensure real-time updates for typical dynamic targets' occupied grids.

Results The results from actual ship trials show that the proposed method can accurately acquire the contour information of extensive marine floating obstacles like aquaculture areas and buoys and suppress target splitting phenomena. Compared with classical methods, the initial detection distance for small floating objects with a freeboard of 0.5m was improved by 78.34m, and the positioning accuracy was improved by 2.97m.

Conclusion The proposed method ensures accurate perception of marine floating obstacles and moving targets on the sea surface, safeguarding the safe navigation of USVs.