基于YOLOv8的路面病害检测体系研究

摘 要：路面裂缝、坑槽等病害需要及时识别，检测结果才能为道路养护提供依据。人工巡检受作业效率和人员经验限制，将目标检测模型部署到检测车上，是提高巡检效率的一条可行路径。本文以YOLOv8为基础，讨论路面病害检测从模型改进到车载应用的实现过程。针对细小病害特征不明显、路面背景复杂以及边缘设备算力有限的问题，分别从数据增强、特征提取、训练策略和模型压缩入手，梳理精度提升与轻量化之间的关系。在介绍已有研究及其局限的基础上，结合市政道路检测案例，分析模型的检测性能、运算开销和端侧运行效果。研究表明，该体系能够满足案例中的实时检测要求，但夜间反光和多病害重叠仍会影响识别结果，后续需要补充相应样本并完善部署接口。 

关键词：YOLOv8；路面病害检测；特征提取 ；轻量化优化

正文 

可下载并阅读全文PDF，请按照本文版权许可使用。
Download the full text PDF for viewing and using it according to the license of this paper.

Research on Pavement Disease Detection System Based on Yolov8

Abstract: A detector used on a road survey vehicle has to recognize small defects while processing images fast enough for the survey. This study examines how a YOLOv8-based pavement distress detection system can meet these requirements, with attention to both model changes and deployment. Timely identification of cracks and potholes supports maintenance planning; manual surveys, however, depend on inspector experience and are limited in their working efficiency. The difficulties considered here are faint features of small defects, interference from pavement backgrounds and restricted computing capacity on edge devices. The discussion relates data augmentation, changes to feature extraction and training, and model compression to the balance between detection accuracy and computational demand. A review of earlier work establishes the remaining problems. A municipal road detection case then provides an example for assessing detection performance, computational cost and operation on the vehicle. Under the conditions of that case, the system meets the required processing speed. Accuracy still falls in images with strong reflections at night or overlapping defects. Training data for these conditions and the deployment interfaces therefore remain areas for improvement.

Keywords: YOLOV8; pavement disease detection; feature extraction; lightweight optimization

参考文献

1. 邓天民，李亚楠，李庆营，等。改进 YOLOv8 的路面病害检测模型 [J]. 重庆理工大学学报 (自然科学), 2024, 38 (4): 138-145.
2. 秦海福。基于 YOLOv8 的公路路面病害检测方法研究 [D]. 广州：广东工业大学，2024.
3. 侯勇严，梁扩旺，郭文强，等。基于改进YOLOv8的路面病害检测模型 [J]. 陕西科技大学学报，2024, 43 (3): 166-173.
4. 陈旭丹，姜男，董丽娜，等。基于深度学习YOLO模型的高原山区沥青路面病害自动识别 [J]. 道路工程，2019 (11): 67-70.
5. Ultralytics. YOLOv8: Ultralytics Real-Time Object Detection[EB/OL]. 2023.
6. 交通运输部. 2024 年交通运输行业发展统计公报 [EB/OL]. 2025-03-20.
7. 李建忠，王强，刘敏。基于YOLOv8与知识蒸馏的路面病害轻量化检测模型 [J]. 交通运输工程学报，2024, 24 (2): 187-198.