Crack Detection and Quantification of Rural Cement Concrete Pavements Based on Deep-Learning Fusion Model

In order to meet the needs of engineering specifications for crack detection and quantitative assessment of rural cement concrete pavements, a method for accurate detection and quantification of cracks is proposed. A deep learning fusion model, YOLO algorithm for rural cement concrete pavement (YOLO-RCCP), is proposed for rural cement concrete pavement crack detection. The YOLO-RCCP model uses the object detection network as a baseline model and adds a grid classification network through a shared backbone network. The box detection head of the object detection network provides crack identification and localization results. The grid classification head of the classification network provides crack classification results for crack quantification. YOLO-RCCP is equipped with SimCSPSPPF pyramid pool structure, adaptive spatial feature fusion (ASFF) adaptive spatial feature fusion, and C3Ghost lightweight structure, which enhances the feature representation ability of the model, effectively solves the problem of missing detection of multiscale cracks, alleviates the overfitting problem, and realizes the synchronous improvement of detection accuracy of box and grid. A continuous highway is used to validate the effectiveness of the proposed method. Combined with the pavement cracking ratio calculated from the manual inspection results, the accuracy of the proposed method is 92%, which meets the specification requirements.