基于細頸結構的密集小目標車輛檢測研究

——以LSGC-YOLOv8為例

打開文本圖片集

中圖分類號：TP391.4；TP181 文獻標識碼：A 文章編號：2096-4706（2025）11-0114-08

Research on Dense Small Target Vehicle Detection Based on Thin Neck Structure -Taking LSGC-YOLOv8 as an Example

ZHOU Zijian， LIU Haizhong （School ofMathematicsand Physics，Lanzhou Jiaotong University，Lanzhou 73oo70，China）

Abstract：The traditional monitoring systemrelieson manual identification of targets，which is inefficient and easy to misjudge.This paper improves and researches the series of YOLOalgorithms.Although theYOLO algorithm has been widely usedinmanyfelds，itsillfacestheechallengesinactualtraffcscenarios.Weatherchangesandhardwarelimitationscaeasily leadtoadeclieindataqualityThedetectionabilityofnarrowareatarget，occlusiontargetandincompletetargetisinsuient. Inadition，inomplexesdetaileatureetractionaslitatios.Tothisd，tispperproposanimproedodetat integrates lightweight designandstructuraloptimization.The model contains threeinnovations.Firstly，the backbone network introducesalightweightarchitecture，whichsignificantlyreduces thecomputationalcomplexityunderthepremiseofcontolable accuracylossSecondlythedetectionheadisreconstructedbyVoVGSCSPmodule tonhancetheabilityoffe-grainedfeature capture.ThirdlytheintegratedAtentionMechanismimproves thequalityoffeaturefusion.Theexperimentalresultsshowthat theimprovedmodelachievesthebalanceoptimizationofaccuracyandefciencyincomplextraffcscenarios，ndhasapication value.

Keywords：TargetDetection;intelligenttransportation; image recognition;DeepLearning;vehicledetection

0 引言

隨著機動車的增加，實時反饋交通流量和監(jiān)測追蹤流動車輛在城市智慧交通系統(tǒng)中發(fā)揮著愈發(fā)重要的作用[1。（剩余10830字）