双注意力机制下的钢材缺陷检测方法

打开文本图片集

【Abstract】Toaddressthe problems of missed detection，false detectionand low detection accuracyin thedetectionof steel surfacedefects，adual-atentionmechanismbasedonYOLOv5 isproposed todetectsteeldefects.Firstly，toaress thedefects onthestelsurfacethatare similar tothe backgroundandprone to misseddetection，aBRAmodule isadded to the backbone network，theBRAC3 moduleisdesigned，andanovelandeficientchannelatention mechanism，SECAisintroduced.Thedualatentionmechanismenhances thesemantic informationof steeldefectsand increasesatentiontomportantpixels；secondly，thecross-scalebidirectionalfeature fusionnetwork isusedtofurther integratefeature information from shallowand dep networks tocapture more feature details;finally，anew lossfunction，SMPDIoU，isdesignedbycombining the ideasof SIoUand MPDIoU，which makesthealgorithm fasterand further improvingthe detectionaccuracy.Experimentalresults showthattheimproved model improvesthe mAPby5.1 percent，theregresionrate by3.5 percentand the FPS value by 21.4f/s ，compared with the classical YOLOv5 model. While improving detection accuracy， the detection speed is also enhanced， demonstrating its application value in industrial production.

Key words】YOLOv5; defect detection; attention mechanism; BiFPN; MPDIoU

［中图分类号〕TP391.4 [文献标识码］A [文章编号]1674-3229（2025）01-0110-10

0 引言

其在制造的过程中，容易产生表面缺陷，而表面缺陷会降低钢材的强度，缩短钢材的使用寿命，影响其质量。（剩余8384字）