基于YOLOv8n的輕量化道路裂縫檢測(cè)算法

打開文本圖片集

Lightweight road crack detection algorithm based on YOLOv8n

TURSUNMamat2，QIUJianzhuo12，ZHUXinglin1，XULi1，2 （1.Collegeof TransportationandLogisticsEngineering，XinjiangAgricultural University，Urumqi83oo52，China; 2.EngieeringResearchCenterforIntelligentTransportation，XinjiangAgriculturalUnversityUrumqi83o52，China）

Abstract：In view of the wideobjectdistribution scale，complexanddiverse featuresand thedemandof dealing witha large numberofdatasetsinautomaticroadcrack detection，alightweightroadcrack detectionalgorithmGCW-YOLO basedon YOLOv8isproposed.Firstly，theglobalatentionmechanismisintroducedintothebackbonenetworktoenhancetheabilityto extractandfuseroadcrackfeaturesfirst，andthentheoriginallossfunctionisreplacdwithWise-IoUlossfunctiontogetbeter featurefocusandreducethelossoffeaturesandclasification inprediction.Finall，thelightweightnetworkstructureGhostNet isintroduced intotheC2fmodule toimprovethefeatureextractioneficiencyofthemodelandreducethecomputational complexity.Experimentswereconductedonaself-madeexpresswaycrack diseasedatasetwithatotalof15116images，andthe generalizationperformanceofthealgorithmwasverifiedonpublicdatasets.Experimentalresultsshow that themeanaverage precision （mAP）oftheproposedalgorithmreaches63.5%，whichisimprovedby6.0%incomparison with thatoftheoriginal model， itsspatial and temporal efficiencyisimproved by 3.0% and 8.5% ，respectively，and itsdetection speed reaches 250 f/s. Thecomparativeexperimentalresults show thattheGCW-YOLOalgorithmcombines the advantagesof lightweightand high detectionaccuracyandshowsgood generalization，soithasgoodpracticalvalueandpromotionprospectinroadmaintenance.

Keywords：road crack detection; deep learning; YOLOv8n;atention mechanism; lightweighting; feature focus

0 引言

公路作為重要的交通基礎(chǔ)設(shè)施，對(duì)促進(jìn)社會(huì)經(jīng)濟(jì)發(fā)展具有重大意義。（剩余10690字）