基于轻量化网络的分心驾驶检测方法研究

打开文本图片集

关键词：分心驾驶；YOLOv8n；Dysample；轻量化；共享卷积；HGNetV2；注意力机制中图分类号：TN911.73-34；TP391.41 文献标识码：A 文章编号：1004-373X（2026）09-0178-07

Research on distracted driving detection method based on lightweight networks

ZhongYalu1，²，KongYanqil，²，LinChen1，²，Zhang Hong1，²，Yang Fugui³，Xie Zhi¹，² （1.CollegeofMechanicalandElectricalEngineering，FujianAgricultureandForestryUiversity，Fuzhou350o，ina; 2.FujianProvincialKeyLaboratoryofAgricultural InformationPerceptionTechnology，Fuzhou35oO2，China; 3.FujianJiangxiaUniversity，Fuzhou35ooO2，China）

Abstract：A lightweight detection network named HDSL-YOLO isdesigned toavoid the hardware resource constraintsof in-vehiclesystemsandimprovethedeployabilityandaccuracyrateofdistracteddrivingdetectionmodels.Onthebasisof the YOLOv8n，themodelisimprovedbyincorporatingmultipleoptimizationstrategies，includingthefolowingfourkeyaspects： firstlytheHGNetV2lightweightbackbonenetworkisintroducedtoefectivelyreduceparametersizewhilesignificantly improvingdetectionspedandoperationaleffciency；secondly，thedynamicupsamplingmodule （Dysample）isintegratedto enhancefeaturerepresentation，particularlyinmulti-scaleojectextraction;additionaly，theSimAM（simpleatentionmodule）is incorporatedtofurterstrengthenthemodel'spereptionandrecognitioncapabilitiesforsmallobjects；andfinally，alightweight detection head is adopted to further streamline parameters.Experiments demonstrate that the mAP and mAP@0.5：0.95 of the HDSL-YOLOalgorithmisimprovedby92.4%and55.4%，respectively.Incomparisonwith theoriginalYOLOv8nalgorithm， theimprovedalgorithmrealizesnotonlyanimproveddetectionaccuracyrate，butalsoafurtherlightweighting，soitachieves dualoptimization.DeployingtheHDSL-YOLOontheembedded platformJetsonNanoconfirmsshorterresponse times，which validates the effectiveness of the proposed improvements.

Keywords：distracteddriving;YOLOv8n;Dysample;lightweighting;shared convolution; HGNetV2;atentionmechanism

0 引言

每年因机动车碰撞造成的死亡人数高达135万人，受伤人数超过7800万人，其中大量的事故与分心驾驶直接相关，分心驾驶已成为威胁道路安全的主要因素之交通安全是威胁公共安全的重要社会问题，全球一。（剩余11045字）