An efficient detection and quantification of pavement defects is a significant challenge for intelligent road maintenance; nevertheless, current methodologies exhibit poor accuracy, inadequate lightweight characteristics, and substantial 2D quantification errors. This paper proposes a collection of efficient and lightweight methods. Initially, the lightweight detection network YOLO‐CGR is established, using the convolutional block attention module (CBAM) into the backbone network, alongside the integration of self‐developed modules C2f with GSConv (C2f‐GS) and ResBlock with efficient multiscale attention (Res‐EMA). The values of mAP@0.5 and mAP@0.5:0.95 on the RDD2022 dataset reach 72% and 44.9%, respectively, outperforming YOLOv8n by 4.8% in both metrics, with only a 0.6M increase in the number of parameters. The DeepLabv3+ network architecture is improved for the crack segmentation task: the encoding phase uses the lightweight backbone MobileNetV2 and cascade atrous spatial pyramid pooling (CASPP); the decoder incorporates GhostConv and the proprietary upsampling feature fusion (UFF) module, resulting in enhanced Intersection over union (IoU) and class pixel accuracy (CPA) values of the algorithm. The algorithm’s IoU and CPA values are increased to 73.49% and 82.55%, respectively, while the computational load is diminished by 31%. Ultimately, the crack segmentation results are utilized to implement the double‐layer edge protection (DLEP)–Zhang‐Suen refinement technique, which enhances the cracks, delineates the branching cracks, and optimizes the trajectory of each crack, achieving a precise measurement with an average relative error of 3.11%. This study presents a comprehensive solution encompassing defect detection, segmentation, and quantification, significantly outperforming existing methods in detection efficiency, segmentation accuracy, and quantification reliability, while providing improved technological support for intelligent pavement maintenance.
Zhou et al. (2026) studied this question.