In this paper, we proposed a method and testing system for identifying unmanned aerial vehicle (UAV) electromagnetic signals in complex environments. We utilized signals collected in authentic environments alongside publicly available data to construct a dataset and optimized images using contrast enhancement techniques. We improved the YOLOv5 model to enhance detection accuracy and built a complete system based on this improved model. In real-world tests, our model achieved an accuracy of 90.7% and a recall rate of 87.4%. The system can identify UAVs within 4 milliseconds (ms) and output results at a speed of 100 frames per second (fps). The results indicate that the performance of the improved algorithm surpasses that of traditional methods, and the system demonstrates excellent real-time capability.