In order to improve the perception accuracy of unmanned ships while sailing, a cascade multi-class ship detection model based on detail enhancement is proposed.First,a pan-mixed data augmentation (PMDA) algorithm is proposed to reduce the network′s dependence on the overall outline of the ship.Secondly,a deformable convolution-balanced feature pyramid (DC-BFP) is designed to improve the model's ability to extract detail features of the hull.Thirdly,combine the fully connected layer and the convolutional layer to form a cascaded cross detector (CCD) to improve the model's ability to analyze the detail features of the hull.Finally, the label smoothing regularization (LSR) is used to improve the overfitting problem of multi-class detection.Ablation and comparison experiments are carried out on the self-built 11-category photoelectric ship detection dataset MCSD11.The feature extraction results and the data visualization of the experimental results show that the improved parts of the model can improve the ship detection effect,with an average accuracy of 91.53%,which is higher than that of the mainstream ones.The detection model algorithm has been greatly improved.