Research on enhancing road apparent crack detection based on the improved YOLOv8n model

Abstract

The improved YOLOv8n algorithm is proposed for the existing target detection algorithms to solve the issues of insufficient detection accuracy and leakage due to the target scale variability and complex background interference during road surface crack detection. This algorithm introduces the convolutional block attention module (CBAM) attention mechanism and integrates it with the cross-stage partial-feature fusion (C2f) module in the backbone network. The spatial pyramid pooling faster cross-stage partial channel (SPPFCSPC) module is introduced by integrating the spatial pyramid pooling (SPP) module with the Fully Cross-Stage Partial Convolution (FCSPC) module, which efficiently extracts multi-scale features. Then, the fine Slim-Neck paradigm is adopted to enhance the learning capability of the model while effectively reducing the number of model parameters. Ultimately, to mitigate the detrimental gradients produced by low-quality pictures, the weighted intersection over union (WIOU) loss function is employed instead of the complete intersection over union (CIOU), hence augmenting the bounding box regression efficacy of the network. After the aforementioned enhancements, the experimental outcomes on the road apparent crack dataset indicate that in comparison to the benchmark model YOLOv8n, the average precision (mAP@50), mean average precision (mAP@50–95), and recall of the enhanced algorithm have risen by 1.8%, 1.7%, and 1.8%, respectively. This indicates that the detection accuracy of road fractures is significantly enhanced by the enhanced YOLOv8n, which can more effectively accommodate the requirements of road maintenance.