基于改进DiffusionDet的印刷电路板缺陷检测研究

打开文本图片集

Research on PCB Defect Detection Based on Improved DiffusionDet

WangXiuyou，HuangGuangren，LiuHuaming，unHaoyu （Fuyang Normal University，Fuyang236037，China）

【Abstract]Defects inprintedcircuitboards （PCBs）candirectlyimpairthe performanceof electronicdevices，yeteffective lydetectingpotentialPCBdefectsremainsapersistentchallngeintheindustry.Tosolvethisproblem，animprovedalgo rithmbasedonDifusionDetisproposedtodetectPCBdefects directlywithoutrelyingonthequeryof experienceobjectsor learnableobjects.Thekeyimprovements includethat，firstlyObjectDetectionConvolution（ODConv）isusedtoeplacepart ofthestaticconvolution kernelofthefeatureextractionnetwork，expandingthelearningdimensionofconvolutionaloperation，soastoobtainfeaturemapsthataremoreconducivetodefectdetection，andsecondly，theGUlossfunctionisusedto replacetheoriginalIoUlossfunctiontoimprovethemodel'sabilitytooptimize thepositionofpreselectedframes.Theexperimental results show that the improved DiffusionDet achieves a mean Average Precision （mAP） of 98.83% ，up from the original 98.12% ，showing better performance in PCB defect detection tasks.

【Key words] printed circuit board; defect detection; DiffusionDet; ODConv; GIoU

[中图分类号］TN41 [文献标识码]A

[文章编号]1674-3229（2025）02-0057-07

0 引言

印刷电路板承载着电子元件间信号传输、电源供应和数据处理等重要功能。（剩余9985字）