Detection of asphalt roads degradation using Deep Learning applied to Unmanned Aerial Vehicle imagery

Abstract

Asphalt roads deteriorate over time due to wear and tear, weather conditions and the effect of traffic loads. These degradations cause enormous damage to road users and economic losses to countries. In Senegal, the inspection of roads for maintenance purposes is done by field surveys and measurements, which is tedious, slow and expensive. This paper proposes a solution for automatic detection of the degraded state of paved roads using Deep Learning applied to drone imagery. The methodology includes three phases: collection of pavement images by drone, processing (annotation, training and test) of the images by YOLOv8 and localization of degraded areas on GeoTiff results of reconstructed pavements. The model was trained and tested on a dataset with a wide range of pavement images and the results show a precision rate of 86.7%, a recall rate of 78.8% and an F1 score of 82.5%.