Drone-based road disease detection

Aiming at the problems of low efficiency, poor accuracy and high cost of manual detection of traditional road diseases, this paper proposes an intelligent detection method based on the fusion of UAV images and deep learning. The study takes S227 provincial highway in Anhui province as the experimental area, and utilizes UAVs carrying a variety of sensors to collect images, constructs a dataset containing six types of diseases and 3121 images and carries out preprocessing. The model is optimized based on YOLOv7: the backbone network introduces ODConv to enhance feature extraction; the fusion module adopts an aggregation-distribution structure to facilitate multi-scale information interaction; and the loss function uses a size-adaptive PIoU to improve the accuracy and robustness of small target recognition. Comparison experiments show that the optimized model outperforms the original YOLOv7 and other mainstream models in terms of accuracy, recall, mAP, etc., especially in the detection of potholes and other small targets. Robustness tests and real-world validation confirm the stability and practicality of the model in complex backgrounds. The study shows that the method is efficient, accurate and reliable, which provides strong technical support and practical basis for building an intelligent road inspection system.