In order to realize the fast recognition of charging target in the process of long-distance charging of flying unmanned aerial vehicle (UAV) by laser, an improved Yolov3 algorithm is proposed. The lightweight network model is used as the feature extraction network to realize the accurate and fast recognition of charging UAV target by laser launching system. Compared with the original Yolov3 network, the average detection speed is increased from 17 frame/s to 33 frame/s, and the weight of network model is reduced from 236.0 MB to 29.7 MB, which greatly reduces the dependence of Yolov3 model on hardware. The research results show that the improved algorithm has high accuracy and real-time performance, which provides a valuable technical means for real-time remote charging of UAV by laser.