基于改進(jìn)YOLOv8s的交通標(biāo)志識(shí)別方法

打開文本圖片集

引用格式：，，，等.基于改進(jìn)YOLOv8s的交通標(biāo)志識(shí)別方法[J].現(xiàn)代電子技術(shù)，2025，48（17）：145-152.

中圖分類號(hào)：TN911.73-34；TP391.41 文獻(xiàn)標(biāo)識(shí)碼：A 文章編號(hào)：1004-373X（2025）17-0145-08

Trafficsignrecognitionmethodbasedon improvedYOLOv8s

QINLunming1，ZHANGYunqi1，CUI Haoyang1，BIANHouqin1，WANGXi2 （1.CollegeofElectronicsandering，ShanghaiUniversityofElectricPower，ShanghaiO3O6，China; 2. eering，BeijingJiaotongUniversity，Beijing1Oo44，China）

Abstract：Inviewofthe missed detectionand false detectioncaused bythelarge scalechangeof traficsign images capturedbythecamerasofautonomousvehiclesandthelargeproportionofsmallobjects，aCMF-YOLOv8straffcsigndetection algorithmisproposed.Firstlythelarge-objectfeatureextractionlayersanddetectionheadsoftheoriginalnetworkmodelare removed，andthesmall-objectdetectionheadsareaddedtoimprovethedetectionprecisionforsmallobjectsandreducethe parameterquantity.Secondly，animprovedcoordinateatentionmechanismisaddedtothebackbonenetwork，andtheglobal channelinformationisobtainedbyglobalaveragepoolingandconvolution，andthefeaturemapisprocessed incombinationwith positioninformation，soastoimprovetheobjectdetectionacuracyofthemodelincomplexenvironments.Additionalythe FasterNetmoduleisused torefietheC2fmoduleoftheoriginalmodelandreduce theparameterquantity.Finaly，the MPDIoU lossfunctionisintroducedtotakeaccountofthenon-overlappingareas，centerdistances，andwidth-heightdeviations comprehensivelybyvertexcoordinates，whichoptimizes thecomputationprocessandimproves thedetectionacuracy. Experimental resultsshow that theCMF-YOLOv8salgorithmachieves amean average precision （mAP）of91.2%，with amodel parameter quantity of 2.7×106 .In comparison with the original YOLOv8s model，the mAPof the proposed model is improved by 5.7% ，anditsparameterquantityisreducedby76%.Inaddition，itsrecognitionaccuracyofoverlappngandoccludedtraffic signs is higher，which is of practical significance to the safety of autonomous driving.

Keywords：traficsigndetection；MPDIoUlossfunction；Channel_CAmechanism；FasterNet module;smallobject detection;autonomous driving;CMF-YOLOv8s

0 引言

隨著人工智能的快速發(fā)展，帶動(dòng)了各行各業(yè)的一系收稿日期：2024-09-13 修回日期：2024-11-15列科技變革，在自動(dòng)駕駛領(lǐng)域掀起了開發(fā)熱潮。（剩余12694字）