A Research on 3-D point clouds reconstruction of SAR images driven by electromagnetic scattering mechanisms

3-D reconstruction of the ground target has great significance in automatic target monitoring, urban safety and other fields. Besides, it is a hot research topic in computer vision, photogrammetry and remote sensing. Unlike optical imaging, SAR imaging mechanisms are of great complexity, which brings many challenges for information extraction and interpretation of SAR images. Thus, the current 3-D reconstruction methods are not precious enough to recover the detailed outlines of the ground target and new precious methods need to be developed. In recent decades, the research on scattering modeling of targets has developed from attribute scattering center models to geometric scatterer parametric models, which greatly improves the target scattering feature extraction and provides a new idea for the 3-D target reconstruction. Combining the geometric scatterer parametric models with convolutional neural networks, this paper proposes a 3-D target reconstruction network of SAR images based on 3-D point clouds. Through a reconstruction experiment, the effectiveness of the algorithm is validated. The method to construct this network shows a new research exploration for target 3-D target reconstruction of SAR images, and provides a reference for 3-D target reconstruction.