Synthetic aperture radar (SAR) has played a crucial role in ship monitoring, maritime rescue, and other fields due to its ability to operate around the clock and in all weather conditions. The interpretation and analysis of ship targets in SAR images have always been an important research direction in the field of remote sensing. In recent years, with the application of deep learning technology to the field of SAR image interpretation, an increasing number of deep-learning-based ship interpretation methods have been proposed and have achieved many new results in various tasks such as ship detection, identification, parameter estimation, and others. Against this backdrop, it is necessary to review the recently proposed deep-learning-based SAR image ship interpretation technologies. This paper reviews English journal and conference articles published from January 2022 to May 2025. First, we provide a detailed organization and statistical analysis of the commonly used and newly released public datasets in this field. Then, focusing on the ship detection and identification task, which has the highest research interest and the largest number of research papers, we innovatively categorize the methods into two classes based on whether they consider the characteristics of SAR image: deep-learning-based methods directly transferred from optical and the methods specifically designed for the characteristics of SAR images. Subsequently, we sorts out and summarize the new tasks and challenges proposed in the field of SAR image ship interpretation, such as cross-domain SAR image ship detection and recognition methods, ship detection with incomplete SAR imaging data, ship wake detection, ship parameter estimation, and three-dimensional ship refocusing. Finally, we summarize the current research status of this field and speculates on possible future research directions.

Relative pose estimation of a satellite is an essential task for aerospace missions, such as on-orbit servicing and close proximity formation flying. However, the changeable situation makes precise relative pose estimation difficult. This paper introduces a deep-learning-based satellite relative pose estimation method for monocular optical images. The method is geared towards uncooperative target satellites with known 3D models. This paper proposes a novel convolutional neural network combined with 3D prior knowledge expressed by the 3D model in the form of the point cloud. The method utilizes point cloud convolution to extract features from the point cloud. To make the result more precise, a loss function that is more suitable for satellite pose estimation tasks is designed. For training and testing the proposed method, large amounts of data are required. This paper constructs a satellite pose estimation dataset BUAA-SID-POSE 1.0 by simulation. The proposed method is applied to the dataset and shows desirable performance on the pose estimation task. The proposed technique can be used to accomplish monocular vision-based relative pose estimation tasks in space-borne applications.

Attitude estimation for uncooperative known space objects plays a critical role in intelligent space technology. In recent years, deep-learning-based methods for estimating the attitude of space objects have made significant progress, particularly through indirect methods based on key point detection. While highly accurate, the indirect method requires high texture detail and is time-consuming, which limits its application in CNN-based methods. Therefore, we investigate a direct attitude estimation method that is faster, more generalized, and better suited for space applications. We propose a fast and accurate network called DSOAE-Net for estimating the attitude of space objects using monocular images. We introduce a pose representation method that is particularly suitable for space object attitude estimation, which enhances the performance of attitude estimation. We also propose a general pipeline for directly estimating attitude and explore and experiment with each stage of the pipeline. Through detailed experiments, we validate that our proposed method achieves optimal results within this framework. The proposed method demonstrates a balance between speed and accuracy, with the best accuracy results achieved among compared methods on a weakly textured dataset.