Abstract: A ship target detection method was proposed based on improved Mask R-CNN (mask region-based convolution neural networks) to solve the problem that the classical rectangular detection boxes within the typical target detection algorithms based on the convolutional neural network will frame many irrelevant areas in ship target detecting, which can lead to problems such as missed detection and false detection. On the basis of Mask R-CNN, a discrimination module, category prediction branch, and semantic segmentation branch were added, thus, the target localization and category prediction of the ship targets in the optical images collected by the vision system were performed. At the same time, the edge contour of the ship target was obtained and semantic segmentation of warship targets was realized, providing more accurate information for the maritime unmanned combat system. Experimental results show that, keeping a high detection rate and operation efficiency, the proposed method can achieve lower false detection rate, higher average accuracy of ship target detection, better ship target detection performance.