基于改进Y0L0v8n模型的隧道裂缝检测算法研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2026）03-0057-06

Research on Tunnel Crack Detection Algorithm Based on Improved YOLOv8n Model

WANG Wei， LU Yang （College ofMathematicsand Computer，JilinNormal University，Siping 36ooo，China）

Abstract： With the continuous aging of tunnel structures，tunnelcrack detection isof greatsignificance for ensuringthe safeoperationoftuels.Aimingat teenvironmentalcharacteristicsofextremelylowillumination，lowcotrast，andhighoise intunels，thispaper proposesatunnelcrack detectionmethodbasedontheimprovedYOLOv8nmodel.Firstlytheattention mechanism moduleisintroducedintothe backbone network toenhancethefeature extractioncapabilityof the modeland the atentiontokeyinformatio，educetheifuenceofirelevantbackground，andimprovetebustnsof thenuraletwork. Secondly，aimingatthecomplex shape changes ofcracks，thedeformableconvolution module DCNv2 isadded to the neck structuretofexiblydealwithtargetsofdiffrentscalesandimprovedetectionaccuracy.FalltheShape-IoUlossfuctionis introduced topromotemoreaccurate targetpositioningand improve theoveralldetectioneficiencyandaccuracy.Experments showthattheimproved modelreaches0.919and0.860 inmean Average Precision （mAP）andF-scorerespectively，whichare improved by 2.57% and 3.61% compared with the original YOLOv8n model，and can effectively meet the actual needs of tunnel crack detection.

Keywords：YOLOv8; tunnel crack identification; Attention Mechanism; deformable convolution; Shape-IoU

0 引言

随着城市发展对空间需求的不断增长和地上空间限制的加剧，地下空间的合理开发与利用已经成为一个重要的科技课题。（剩余8798字）