Differentiable Voxel Reconstruction of Satellite Target from Multi-Station ISAR and Optical Image Sequence

The situation awareness of non-cooperative satellites is essential, but the unknown parameters of the target spin motion makes the 3D (three-dimensional) reconstruction of the target from ISAR (inverse synthetic aperture radar) images difficult. We propose to adopt a joint optical-ISAR system to observe the satellites, and use the complementary information brought by the two different imaging geometries of the two sensors to accomplish the cross-range scaling of ISAR and obtain the target attitude information. Then, a differentiable voxel reconstruction network with updatable view angle parameters is introduced for 3D reconstruction. The experiments show that the proposed method can improve the robustness in the 3D reconstruction of a non-cooperative target.