Obstacle Identification Based on Deep Learning for Inspection Robot of FAST Wire Rope

In this paper, for the requirements of image recognition and localization of the obstacles existing on the moving path in the inspection task of FAST cable inspection robot, the deep learning method is used to realize the recognition of multi-kind and multi-scale obstacles. The obstacles on FAST wire rope have not been identified before. By enhancing the data set and improving the network model, the recognition accuracy and speed are both improved, which meets the efficiency and accuracy requirements of obstacle detection on the wire rope, and provides help for the subsequent robot pace planning and overhaul tasks.