Abstract

Road maintenance is crucial to facilitate safe and sustainable transportation systems. Conventional inspection methods, which are mainly based on human observations, are often slow and inefficient, however, particularly as expansion of urban environments continues. Contemporary advances in technology and smart systems and the expansion of deep learning capabilities in some networks has led to the automation of this process and the development of intelligent monitoring systems. Deep learning models enable the detections and classification of road defects issues in efficient ways, leading to the focus of this paper. The goal of this study was to develop a model to detect and classify various road issues using deep learning models. The datasets used to facilitate this included four classes representing four different road issues, namely cracks, potholes, water collection, and drain cover damage. Three models were evaluated across these datasets: YOLOv11X, Faster R-CNN, and RetinaNet, with similar Precision results emerging for all three models at 82.1%, 88.1%, and 82.4%, respectively. However, while Faster R-CNN achieved the highest precision value, taking overall performance and evaluation metrics into account showed YOLOv11X to be the most balanced model to offer satisfactory results. Overall, the findings of this study show this to be a promising approach that can be evolved and embedded into smart systems to generate real-time reporting for responsible authorities to ensure that the latter can develop safer and better managed and maintained urban infrastructures.

Statistics