Applied YOLOv5 for Freeway Foreign Object Recognition in UAVs Imagery: Classification Criteria and Algorithm Comparison

Freeways play an important role in economic development. To avoid the security issues caused by foreign objects on freeways, freeway foreign object recognition algorithms are applied. This paper used the YOLOv5 algorithm in UAV imagery to recognize foreign objects based on a freeway foreign object data set. Firstly, the data set was established by unmanned aerial vehicle (UAV) acquisition, open data set, and nearby road section photography. Then, the foreign objects were classified into 1, 3, 6, and 12 classifications according to their different characteristics. Then, this paper constructed the YOLOv5-based object detection algorithm. Finally, 12 sets of comparison algorithm experiments were designed based on the YOLOv5 with 3 algorithm versions and 4 foreign object classifications. The results show that the YOLOv5m has the best performance, especially in the case of 3 classifications, where the total loss is 0.023 and the medium foreign mAP is 0.804, which can effectively recognize freeway foreign objects.