Evaluate Road Depressions with Deep Learning and Mobile Application

Road damage such as potholes and cracks potentially poses significant safety challenges, economic outlays, ecological impact, and governance concerns. This study presents an integrated approach using deep learning and mobile technology to automate the detection and severity classification of road depressions. The system compares three YOLO models, which are the YOLOv5s, YOLOv8s, and YOLOv11s, trained on the RDD2022 dataset to identify the optimal model for deployment. YOLOv8 is selected based on its balanced performance, achieving 65 percent precision, 59 percent recall, and 62 percent $\mathbf{F 1}$-score. A React Native mobile app enables users to upload images and receive feedback, while a Django backend handles prediction and status updates. This solution improves road maintenance efficiency and promotes public participation in infrastructure reporting.