M-YOLO based Detection and Recognition of Highway Surface Oil Filling with Unmanned aerial vehicle
YuChen Liu, Gang Shi, Yanxiang Li, Ziyu Zhao · 2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP) · 2022
Pavement oil repair is a kind of repair scheme for damaged pavement. Due to the problems of difficult and low detection efficiency in oil repair area, a light network architecture algorithm (M-YOLO) based on Mobilenet V3-YOLOv5S is proposed in this paper. MobileNetv3 structure was introduced to replace YOLOv5s backbone network to reduce model size and improve target detection speed. SPPNet network structure was introduced to remove repetitive features and improve target detection accuracy. After experiments, the accuracy of M-YOLO algorithm is up to 98.3%, the average accuracy is up to 95.5%, and the detection speed is up to 96.6fps. Compared with YOLOv3, the p-value, mAP value and FPS of the proposed algorithm are effectively improved by 10.4%, 5.1% and 30.8F*s-1, respectively. It can be seen that the algorithm proposed in this paper greatly reduces the number of model parameters and computation, and improves the detection accuracy and detection speed.