重参数化YOLOv8路面病害检测算法.pdfVIP

ComputerEngineeringandApplications计算机工程与应用2024，60（5）191

重参数化YOLOv8路面病害检测算法

王海群，王炳楠，葛超

华北理工大学电气工程学院，河北唐山063000

摘要：路面病害检测是保障人民交通安全的重要方式，为了提高路面病害检测的准确率，实现及时、精准的路面病

害检测，提出了一种重参数化YOLOv8路面病害检测算法。在主干网络引入CNX2f模块，提高网络对路面病害特征

的提取能力，有效解决路面病害特征与背景环境特征易混淆问题；引入RepConv和DBB重参数化模块，增强多尺度

特征融合能力，解决路面病害尺度差异较大问题；改进头部采用共享参数结构，并引入RBB重参数模块，解决头部参

数冗余问题，并提高特征提取能力；引入SPPF_Avg模块，解决路面病害特征丢失问题，丰富多尺度特征表达。实验

结果表明，改进后的路面病害检测网络精度为73.3%、召回率为62.3%、mAP为69.3%，较YOLOv8网络分别提高了

2.6、3.0、2.8个百分点，提高了模型的检测效果。

关键词：路面病害检测；特征提取；重参数化；多尺度特征；YOLOv8

文献标志码：A中图分类号：TP391.4doi：10.3778/j.issn.1002-8331.2309-0354

Re-ParameterizedYOLOv8PavementDiseaseDetectionAlgorithm

WANGHaiqun,WANGBingnan,GEChao

SchoolofElectricalEngineering,NorthChinaUniversityofScienceandTechnology,Tangshan,Hebei063000,China

Abstract：Roaddiseasedetectionisanimportantwaytoensurepeople’strafficsafety.Inordertoimprovetheaccuracy

ofroaddiseasedetectionandachievetimelyandaccurateroaddiseasedetection,apavementdiseasedetectionmodelof

re-parameterizedYOLOv8isproposed.Firstofall,CNX2fmoduleisintroducedintothebackbonenetworktoimprove

theabilityofthenetworktoextractpavementdiseasefeatures,andeffectivelysolvetheproblemthatthepavementdis-

easefeaturesareeasilyconfusedwiththebackgroundenvironmentalfeatures.Secondly,RepConvandDBBreparameter-

izationmodulesareintroducedtoenhancethecapabilityofmulti-scalefeaturefusionandsolvetheproblemoflargescale

differenceofpavementdiseases.Atthesametime,thesharedparameterstructureoftheheadisimproved,andRBBrepa-

rameterizationmoduleisintroducedtosolvetheproblemofheadparameterredundancyandimprovethefeatureextrac-

tioncapability.Finally,theSPPF_Avgmoduleisintroducedtosolvetheproblemofpavementfeaturelossandenrichthe

multi-scalefeatureexpression.Thee