Research on Asphalt Pavement Disease Detection Based on Object Detection Algorithms

The present manuscript endeavors to furnish a comprehensive scrutiny of the model framework implemented by YOLOv5s for object detection, with a particular emphasis on its enhancements for detecting irregularities on asphalt road surfaces. The evaluation of the model encompasses precision and recall metrics, with mAP serving as the principal performance indicator. In order to augment the precision of the model, the backbone network has been fortified by integrating the ResNet18 residual network and the ECA attention mechanism. This augmentation has significantly enhanced the model's capacity to detect cracks and potholes, resulting in a notable 6.0% increase in mAP compared to the original YOLOv5s model. The integration of the ResNet18 network and ECA attention mechanism has considerably amplified the model's ability to discern intricate features. Simultaneously, this integration has also facilitated in curtailing computational costs, thus rendering the model more efficient. The experimental analysis conducted in this context provides explicit evidence of the superiority of the YOLOv5s-RD model over the original YOLOv5s model in identifying and localizing road imperfections. The findings underscore the potential of this enhanced model framework to make a significant contribution to the domain of road maintenance and safety.