Pavement Distress Detection Based on Improved Yolov8

A road pavement distress identification and detection algorithm SG-YOLO based on improved yolov8 is proposed to address the issues of low accuracy and large detection models in complex road surface conditions. By introducing a space-to-depth layer into the backbone feature extraction network structure of the algorithm to construct SPD-ResNet, the pixel loss of the image is reduced to adapt to the detection of low-resolution and small-target lesions on the road surface, and meanwhile, to further reduce the model parameters and computational volume so that the model has a better computing speed, the GSConv is incorporated into the neck part and the bottleneck structure is improved by the GSConv. The experimental results show that the improved detection algorithm SG-YOLO not only improves the accuracy of disease detection on the two widely differing RDD2022-China and RDD2022-US datasets, but also quickly detects the location and type of pavement damage in the actual road pavement disease detection and identification.