Pavement Defect Detection Model Based on Fusion of Multiscale Shared Features

Aiming at the problems of low detection performance and large memory consumption in the current road defect detection, a fusion of MSDhead (Multi-scale shared decoupled detection header) detection header and CSPP (Cross-stage spatial pyramid pooling) of YOLOv8 road defect detection model. First, the model uses the proposed CSPP to replace the SPPF in the original YOLOv8 model, which improves the ability of the model to extract key information from input features of different sizes; secondly, the resnet18 backbone is used to replace the original backbone of YOLOv8, which improves the feature extraction ability of the model; and finally, the proposed MSDhead detection head is used to replace the detection head of YOLOv8, which reduces the number of model parameters, reduces memory consumption, and effectively decouples feature information. On the VOC2007 dataset, the model improved by 0.5%,2.9%,3.1 %,2.5%,4.8%, and 3.1 % on AR, AP50, AP75, AP50-95, APM, and APL, respectively.