A New Artificial Intelligence-based Method for Detecting Pavement Distresses

To improve the efficiency and accuracy of pavement distress detection, this paper utilizes artificial intelligence algorithms to analyze road condition videos. Although traditional detection algorithms (such as YOLO) excel in fast and accurate detection of large objects, their capability to detect small objects is limited. To address the issue of detecting small objects, this study introduces the concept of a segmented dataset and incorporates segmented image inference technology in the model inference module, specifically enhancing the detail detection algorithm in the Fine-tuning dataset module of the image detection system architecture. Experimental results demonstrate a significant enhancement in the accuracy of pavement distress detection when using this enhanced dataset and optimized inference model. Specifically, with the adoption of YOLOv7, the new algorithm increases the detection accuracy of “Crack” by approximately 53.34% and enhances the detection accuracy of “Corner Breaks” by 50.46%. When employing YOLOv7x, the detection accuracy for “Crack” improves by around 45.42%, and the detection accuracy for “Corner Breaks” rises by 46.2%.