Article’s

Timely detection of road-surface defects such as potholes, longitudinal cracks, alligator cracking, and patch repairs is essential for keeping transportation networks safe and for planning cost-effective maintenance. Manual road surveys remain slow, expensive, and place field personnel at risk, which motivates automated alternatives. This paper presents an AI-driven pipeline that pairs Unmanned Aerial Vehicle (UAV) image acquisition with deep-learning-based object detection to identify and localize road damage without requiring an inspector to walk or drive the surveyed stretch. Drone imagery is first passed through a preprocessing stage — resizing, contrast normalization, and Gaussian-filter-based noise suppression — before being fed to a Convolutional Neural Network (CNN) detection backbone. Successive members of the YOLO object-detection family, namely YOLOv4, YOLOv5, and a Transformer-Prediction-Head variant of YOLOv5, were trained on a merged corpus built from the public RDD2022 road-damage dataset and a regional Spanish road imagery collection, so that the resulting detector generalizes across differing pavement types, marking conventions, and camera geometries. Afourth configuration, YOLOv7, was additionally evaluated to gauge how far continued progression within the YOLO family alone could push detection accuracy. On a held-out test split, the detectors reached a mean Average Precision at an IoU threshold of 0.5 (mAP@0.5) of 26.8% for the YOLOv4 baseline, 59.9% for YOLOv5, 65.7% for the Transformer-augmented YOLOv5 variant, and 73.2% for YOLOv7, indicating that both architecture progression and attention-based prediction heads improve localization of irregularly shaped defects such as alligator cracks. The complete system is delivered as a Django web application with separate user and administrator roles, allowing non technical maintenance staff to upload imagery, trigger detection, and review flagged damage on a dashboard, while administrators retrain and monitor the underlying model. The results suggest that combining UAV-based data collection with modern anchor-based and transformer-augmented detectors is a practical route toward continuous, low-cost, low-risk road-condition monitoring for municipal and highway authorities.