Classification of pothole distress severity in asphalt pavements using YOLOv8

Abstract

Potholes are a type of distress that occurs in pavement surfaces. According to the method adopted for distress surveys, potholes are classified into three levels of severity: low, medium, and high. The severity assessment is traditionally performed through slow and labor-intensive manual procedures. To automate this process, this study employed the YOLOv8s and YOLOv8m models to detect pothole distress and classify its severity. During the training phase, YOLOv8m achieved the best evaluation metrics, while YOLOv8s outperformed in the testing phase, particularly in recognizing high-severity potholes. However, both models failed to effectively detect low and medium severity levels, indicating the need for improvements before field application. One possible explanation for this limitation is the lack of depth information in the input images, a factor that will be addressed in future research.