In real sea scenes, the appearance of ship targets is similar and the edge information is blurred. The existing algorithms cannot meet the demands for fine-grained and real-time detection at sea. Therefore, a fine-grained detection method is proposed for ship objects based on multi-scale coordinate attention and multi-network self-supervised learning. First, a multi-scale coordinate attention and multi-network self-supervised learning module is designed. Feature enhancement is carried out on the basis of the original feature pyramid network and path aggregation network to improve fine-grained detection accuracy. Secondly, an unmanned surface vehicle (USV) sensing platform based on pods and electronic compass is constructed, and a dataset containing different ship objects such as fishing boats, speedboats, and merchant ships is prepared. Finally, the algorithm is tested and integrated into public and self-made datasets. The results show that the proposed algorithm has high detection accuracy for ship targets. The mAP50 reaches 94.6% in the real sea scene, which is 1.1% higher than that before the improvement. The operation speed is 27 fps, which verifies the robust and real-time fine-grained detection capability of USVs.