Article
Automated Road Damage Detection via UAV Imaging and Deep Neural Networks
This study introduces an AI-driven system for detecting and localizing road surface damage using deep learning methods. The approach employs three variants of the YOLO (You Only Look Once) architecture—YOLOv5, YOLOv7, and YOLOv8—to analyze and classify road images. A custom dataset containing multiple categories of road defects, including potholes, cracks, and surface irregularities, was used for training and evaluation. Comparative results indicate that YOLOv8 delivers the best performance, achieving an accuracy of 85%, followed by YOLOv7 at 82% and YOLOv5 at 65%. Additional evaluation metrics such as precision, recall, and confusion matrices further confirm the superior detection capability of YOLOv8 with reduced misclassification rates. The system also supports real-time detection, visually marking damaged areas with bounding boxes. Overall, the proposed framework demonstrates the effectiveness of deep learning in automating road condition assessment, with practical applications in smart infrastructure management and autonomous transportation systems.
Full Text Attachment
