Research on high-precision recognition model for multi-scene asphalt pavement distresses based on deep learning

Zhang, Sheng; Bei, Zhenghao; Ling, Tonghua; Chen, Qianqian; Zhang, Liang · 2024 · DOAJ

DOI: 10.1038/s41598-024-77173-4

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of accurately detecting asphalt pavement distresses, such as cracks and potholes, in complex, multi-scene environments. Traditional manual inspection is inefficient and subjective, while existing automated methods often struggle with varied distress patterns, complex backgrounds, and the high computational costs of two-stage detection models. To overcome these limitations, the authors propose SMG-YOLOv8, an improved object detection model based on the YOLOv8s framework. The research aims to enhance detection accuracy and generalization across diverse scenarios while reducing model parameters for practical deployment. The SMG-YOLOv8 model integrates three specific architectural optimizations. First, it incorporates a Space-to-Depth (SPD) Convolution module in the backbone to preserve fine-grained information and mitigate feature loss in blurry images or small targets. Second, it introduces a Multi-Scale Convolutional Attention (MSCA) mechanism to adaptively focus on distress targets of varying scales and distinguish them from complex backgrounds. Third, it replaces the standard C2f structure with a lightweight G-GhostC2f structure, which reduces computational burden and memory usage while maintaining feature extraction capabilities. The model was trained using PyTorch on a high-performance computing environment featuring an RTX 4090 GPU. Experimental validation included ablation studies for each module and comprehensive comparisons against baseline models, including YOLOv5n, YOLOv5s, YOLOv6s, YOLOv8n, YOLOv8s, and SSD. Experimental results demonstrate that SMG-YOLOv8 significantly outperforms the YOLOv8s baseline. The proposed model achieved a macro-precision (Pmacro) of 81.1% and a mean average precision at 50% IoU (mAP50) of 79.4%, representing increases of 8.2% and 12.5%, respectively, over the baseline. The model showed superior performance in identifying longitudinal cracks, transverse cracks, mesh cracks, and potholes compared to other state-of-the-art models. Furthermore, validation on real-world data collected from road inspection projects yielded a Pmacro of 80.5% and a macro-recall (Rmacro) of 86.2%. These results confirm the model’s strong generalization capability and robustness in practical, multi-scene applications. The significance of this work lies in providing a lightweight, high-precision solution for intelligent pavement distress detection. By optimizing the network structure to reduce parameters while enhancing accuracy, SMG-YOLOv8 facilitates deployment on mobile devices such as drones and inspection vehicles. The study offers substantial technical support for automated road maintenance, improving traffic safety and reducing the economic losses associated with undetected pavement defects.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• dms validation