Abstract: Abstract： The instance segmentation of ships can serve tasks such as ship monitoring, identification, and tracking, supporting intelligent navigation. However, due to the variability in ship shapes and scales, as well as interference from environmental factors, existing instance segmentation methods perform poorly in ship contour extraction. Therefore, this paper proposes a Ship Contour method based on curve recursion. By improving the CenterNet method to extract hierarchical features and integrating the Deep Layer Aggregation-60 backbone network, the method balances accuracy and speed; the Block structure is improved, and the ECA channel attention mechanism is introduced to enhance the ability of feature extraction, using the Mish activation function instead of ReLU to adapt to deep learning; a translation-invariant contour deformation method and a Dynamic matching loss function are introduced to accelerate the extraction of the final contour. On the 2023Ship-seg dedicated dataset with 2300 samples, the accuracy AP0.5:0.95 of the proposed method reached 64.0%, and the recall rate AR0.5:0.95 reached 67.9%, outperforming all mainstream instance segmentation algorithms. The method will effectively enhance the visual processing effects in monitoring and intelligent navigation scenarios.