MED-YOLOv8s: a new real-time road crack, pothole, and patch detection model

Real-time road damage detection and assessment is crucial to ensure road safety. Traditional road damage detection methods mostly rely on manual labor, which is not only inefficient, but it is also difficult to guarantee its reliability. In this study, a road damage detection model, MED-YOLOv8s, based on YOLOv8s is proposed. MobileNetv3 is adopted as the backbone of the detection algorithm, which reduces the number of parameters and the number of computations in the process of feature extraction, enabling the model to achieve a good balance between the detection speed and the detection accuracy. The introduction of the ultralightweight attention mechanism, ECA, adapts the optimization of the correlation of channels to improve the model generalization performance. In addition, replacing the standard convolution with the DW convolution in the 21st layer of the network not only eliminates part of the redundant feature maps but also better extracts the correlation information between the feature maps. In this study, we also discuss the influence of the mix-up data augmentation weight parameter on the detection effect of the model. The experimental results show that the mAP@0.5 of the MED-YOLOv8s model proposed in this study is 95.2%, which is 1.1% higher than that of the original model, and at the same time, the calculation amount of the model is reduced by 46.2%. This method not only improves the detection accuracy but also greatly reduces the model complexity, providing a reference for subsequent model migration.