In order to realize the precise detection of obstacles in the process of autonomous navigation of unmanned ships, an obstacle detection method for unmanned ships based on quadtree sector layer value clustering was proposed. Firstly, the obstacle point cloud data was retrieved based on the quadtree sector division, and the untrustworthy data in the sector image limit was eliminated. Secondly, the obtained quadtree layer value was used to calculate the global density distance, and then the layer value threshold was obtained to detect irregular multi-linear obstacle features. Finally, the reference distance was obtained by establishing the spatial topological relationship between data points, and the obstacle point cloud data was clustered and judged based on the reference distance to improve cluster segmentation accuracy. The results of multi-linear obstacle feature recognition performance test and surface unmanned ship obstacle detection experiment show that compared with other density clustering algorithms, in terms of positive detection rate, false detection rate and missed detection performance index, the proposed algorithm decreases by 9.86%, 5.04% and 3.10% respectively during multi-linear obstacle feature recognition performance test, and the proposed algorithm decreases by 10.50%, 6.97% and 2.95% respectively during surface unmanned ship obstacle detection experiment. In the performance indicators of positive detection rate, false detection rate, and missed detection rate.