Rural road surface distress detection algorithm based on mask R-CNN with data augmentation

Traditional manual detection of rural road surface distress is time-consuming and labor-intensive. In this paper, we propose a Mask R-CNN algorithm specifically designed for detecting rural road surface defects. To enhance precision and recall rates, data augmentation techniques—such as image translation, flipping, and noise perturbation—were applied to a dataset of 4,000 high-quality images of rural road pavement defects. This combination of Mask R-CNN with data augmentation is a novel approach that addresses the unique challenges of rural road distress detection. Experimental results demonstrate that data augmentation significantly improves recognition precision. The Mask R-CNN algorithm outperforms the ScNet algorithm in terms of precision for detecting and segmenting rural road defects. Among the various models and backbones tested within Mask R-CNN, the ResNeXt-101-FPN backbone achieved the highest precision and recall rates. Additionally, three field tests further validate the feasibility and reliability of the developed algorithm for rural road distress detection. The system, combining the Mask R-CNN algorithm with data augmentation, effectively distinguishes between varying levels of severity and classifies defects based on characteristics such as size, shape, and location. This enables maintenance crews to prioritize repairs more efficiently, resulting in significant improvements in road safety and durability.