一种轻量化网络的道路病害检测方法研究  

Study on Road Defect Detection Method for Lightweight Networks

作　　者：刘鹏杰 姚金杰[1] 高晶[1] 郭钰荣 LIU Pengjie;YAO Jinjie;GAO Jing;GUO Yurong(Shanxi Key Laboratory of Information Detection and Processing,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学信息探测与处理山西省重点实验室,太原030051

出　　处：《计算机测量与控制》2025年第4期40-47,共8页Computer Measurement &Control

基　　金：山西省重点研发计划项目(202102010101002);内蒙古自治区科技计划项目(2022YFSJ0031)。

摘　　要：针对道路病害检测中道路情况复杂、实时检测较为困难,缺检漏检等问题,采集并制作了多类型道路病害数据集R-CRACK,在YOLOv5s模型基础上,在Neck模块上使用轻量化卷积GSConv模块替换部分标准卷积构建轻量化网络颈部;在Head模块上利用SimSPPF对空间金字塔池化方式进行改进并应用轻量级上采样算子CARAFE,而后得到GSC-YOLO模型;将GSC-YOLO模型利用矩形推理、图像加权,标签平滑处理方式对数据集R-CRACK中训练集部分进行训练;模型训练后的结果表示,与YOLOv5s基础模型相比,GSC-YOLO参数量减少6.8%、计算量减少4.8%、mAP(@.5)上涨了9.2%;利用改进前后的网络模型分别对单一及复杂环境下的道路病害进行检测,通过对比不同模型的检测效果,证明了GSC-YOLO模型针对YOLOv5s缺检漏检等问题有所改进,此类轻量化检测网络对解决道路病害检测有着重要意义。In response to the challenges of complex road conditions,difficult real-time detection,and missing detection and inspection in road distress detection,a multi-type road distress dataset R-CRACK is collected and presented.Based on the YOLOv5s model,lightweight GSConv modules are used to replace some standard convolutions in the Neck module to construct a lightweight network neck.In the Head module,the SimSPPF is applied to improve the spatial pyramid pooling method,utilize the lightweight upsampling operator CARAFE,and obtain the GSC-YOLO model.The rectangle inference,image weighting,and label smoothing techniques are used to train the GSC-YOLO model on the training part in the dataset R-CRACK.The results of the trained model show that compared with the base YOLOv5s model,the GSC-YOLO model reduces the parameter by 6.8%,the computation by 4.8%,and improves the mAP(@.5)by 9.2%.The improved network models are used to detect road defects in both single and complex environments.Comparison results of different models show that the GSC-YOLO model improves the shortcomings of YOLOv5s in terms of defect missing detection.It is of significance for this lightweight detection network to solve road defect detection.

关 键 词：道路病害检测 YOLOv5 缺陷漏检 CARAFE GSConv 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]