Research on Key-Point Detection of Space Target Based on Lightweight-HRNet

To solve the computational problem of key-point detection of space target, a lightweight High-Resolution Net (HRNet) method is proposed. Firstly, the redundancy analysis of the four stages of HRNet is carried out, and it is found that the fourth stage is highly redundant. Then, the fourth stage with the highest redundancy is directly removed to form de-redundancy HRNet_S3. Then, to solve the performance degradation problem of HRNet_S3, the basic module of HRNet_S3 is reconstructed based on ShuffleNet, and SHRNet is obtained. Then, a comparative experiment is carried out based on the SPEED dataset. Finally, the keypoints are combined with PnP method to realize the pose estimation of spatial objects.