Solar Panel Recognition of Non-cooperative Spacecraft Based on Deep Learnin

Abstract

Recognition and acquisition of non-cooperative objects is important but challenging for space on-orbit service missions. In this paper, the position of solar panels on space non-cooperative objects is detected based on the deep learning technique. The data set of pictures that consists of spacecraft is first established for training and testing. Then both Faster R-CNN and YOLOv3 algorithms are used to train the network to automatically detect the position of solar panels. Experiments show that the Faster R-CNN and YOLOv3 algorithms can achieve good results in the spacecraft solar panel detection. YOLOv3 has higher precision rate and lower recall rate than Faster R-CNN. Compared with the traditional object detection algorithm, the object detection algorithm based on deep learning does not need manual design features, and the detection performance is more robust. Though this paper focus on the recognition of solar panels, the proposed method can be easily extended to other scenarios. It can also be used to detect other parts of spacecraft.