When a ship is sailing on the sea, the attitude of the ship changes in real - time due to the influence of the ship's own movement, wind, and waves. And it is difficult to accurately measure the ship's attitude in real - time from an aircraft. In order to solve the above problems, a pose - estimation method integrating the traditional template - matching method and the deep - learning method was designed. The deep - learning method improves the accuracy, robustness, and environmental adaptability of pose estimation, while the real - time performance of pose estimation is enhanced by combining it with the template - matching method based on contour features. Firstly, the three - dimensional model of the target ship was used to render the multi - pose images of the ship, and the ship attitude template library was established through the instance segmentation algorithm. Then, the visible - light images of the target ship were collected. The ship - matching images were obtained through the target - detection and instance - segmentation algorithms. These ship - matching images were matched with the images in the ship - attitude template database, and the attitude corresponding to the successfully - matched ship - attitude template image was the attitude of the ship. Through simulation verification, the accuracy of 3D attitude estimation could reach 1°.