基于多尺度特征的轻量级交通标志检测模型

打开文本图片集

中图分类号：TP399 文献标识码：A DOI： 10.7535/hbgykj.2026yx01002

Abstract：Toaddressthechallenges balancing accuracy speed，aswellas the high misseddetection rate small targets iexisting trafic sign detection methods，alightweightreal-time traffcsign detection model（LRTD）was proposed. BasedonYOLOl1nasthebaseline，themodel introducedafeatureenhancement block（FEB）aglobal-localcolaborative atention module（GLCAM）intothebackbone network；Intheneck network，a multi-scalereceptivefieldcolaborativemodule （MSRFC）wasdesigned the feature fusion strategy was optimized to construct a high-resolution detection head.On the public datasets CCTSDB GTSDB，performancecomparisons betwen theLRTD model state--the-artdetectionmodels wereconducted，ablationexperiments werecariedouttoverifythefunctionalityeach module.Theresultsshowthaton the CCTSDB GTSDB datasets，the LRTD model achieves mAP@50 83.1% 95.6% respectively， mAP@50-95 55.6% 81.5% respectively. Compared with the YOLO1ln model，it increases mAP@50 by 6.7 percentage points 2.1 percentage points respectively， mAP @ 50-95 by 6.O percentage points 4.5 percentage points respectively. Additionaly，themodel maintainsareal-time inferencespeed155.Ofpsonthe CCTSDBdataset，with itsparameter count computational complexity reduced by1.9 percentage points1.6percentage pointsrespectively.The proposed model canefectivelyimprovetherecognitionperformancetraffcsigns incomplexscenariosprovideafeasibletechicalsolution for real-time object detection tasks in intelligent transportation systems.

Keywords：computer image processing；trafic sign detection；feature enhancement；atention mechanism；multi-scale feature fusion

交通标志检测是智能交通系统和自动驾驶的关键技术，其可靠性直接影响道路安全与车辆决策效率。（剩余19520字）