In this paper, we propose an accurate ship detection algorithm based on center segmentation and angle prediction. Center segmentation module is constructed to accurately predict the ship location. Specifically, this module predicts the center region of each ship via semantic segmentation. With center region, the generated region proposals are more accurate. In order to detect the rotated ships, we design the angle prediction module. This module predicts the angle range of each ship, which is beneficial to more precise rotated region proposal generation. Then, we integrate the center segmentation module and the angle prediction module into a semantic segmentation network to generate more accurate rotated region proposals, which is fast and effective. Then, we classify and regress the proposals to output the final ship detection results. Experiments demonstrate the effectiveness of the proposed ship detection algorithm.