Pavement damage image classification using deep learning with inspection system: a case study in Morocco

Abstract

Road and highway authorities rely on pavement management systems(PMS), in particular regular pavement condition inspections, to manage and preserve this infrastructural heritage. To this end, visual surveys are regularly conducted to detect and classify pavement damage, assess pavement condition, and derive performance indicators. However, manual pavement inspection can be a subjective and time-consuming process that requires a high level of skill from those responsible for inspection and monitoring. This study proposes a machine learning (ML) technique to automatically classifying digital images of national road surfaces captured by a camera mounted on a smart vehicle equipped with a multifunctional road inspection system (SMAC). The image dataset, captured on different roads in Morocco, includes five classes of pavement damage and one class of no damage. The experimental results indicate that the ResNet50 model achieves superior classification accuracy of approximately 94%. This research contributes to the automation of road monitoring processes and provides road managers with an effective tool for planning and executing maintenance operations with enhanced reliability and efficiency.