Geometric Feature Extraction of Satellites Based on 3D Geometry Reconstruction from ISAR Image Sequences

Extracting the geometric features of satellites from inverse synthetic aperture radar (ISAR) images is a meaningful and challenging task. Previous methods have focused on estimating three-dimensional (3D) geometric features based on their projection mapping relationships with planar features, such as key points, lines, and parallelograms. However, these planar features are unable to fully describe the satellite cabin structure, resulting in limited estimation of only the orientation and length of the cabin main axis. To extract more diverse features, we propose a framework for geometric feature extraction based on 3D geometry reconstruction. By reconstructing the complete target structure using target silhouettes extracted from images, our framework can extract a wider range of geometric features of the satellite cabin, including its 3D shape, 3D size, and absolute attitude. To ensure precise reconstruction, we design a component completion (CC) module to maintain the integrity of the extracted silhouettes and a deviation parameter estimation (DPE) module to provide accurate projection parameters. To extract features from the reconstruction results, we propose a 3D geometric fitting algorithm applicable to different cabin shapes, where the appropriate fitting objective function is selected based on automatic 3D shape discrimination. The effectiveness of the proposed methods is verified through experiments conducted on both simulated and measured data.