Research on Real-time Road Crack Detection Algorithm Based on Improved Knowledge Distillation

An intelligent real-time crack detection algorithm based on lightweight and knowledge distillation is proposed for existing crack detection systems that can hardly meet the road development needs. In this paper, YOLOv4 is used as the main framework, and MobileNetV3 is used as the backbone network, based on which the model is compressed and knowledge distilled. The experimental results show that the final model has 95.58% lower parameters than the YOLOv4 model, 66.22% higher inference speed, and 0.914 detection accuracy. The improved, streamlined model has high precision, good detection effect, and fast operation speed, and is suitable for deployment on miniaturized embedded systems with in-vehicle as the platform.