基于深度學(xué)習(xí)的車輛部件半自動標注研究

打開文本圖片集

中圖分類號：TP391.4 文獻標志碼：A 文章編號：2095-2945（2025）13-0014-07

Abstract：Intheraofartifialinteligene，intelligentlossassssmenttechnologyusingdeplearningisageneraltrendin thedevelopmentandtransformationofautomobileinsurance.Itcangreatlyreduce thecostofmanuallossassessmentand improvetheefciencyoflossassessment;deeplearningmodelsrequirealargenumberofannotatedsamplesfortraining，but currentlythereisnopublicdatasetonvehicleinjurytypesandcomponents.Thispaperproposesasemi-automaticlabeling methodforvehicleparts，first，amodelthatcanberoughlylabeledistrainedthroughaseddataset，andthenthemodelis usedforautomaticlabelingofvehicleparts;onthebasisofautomaticlabeling，asmallnumberofmanualcorectionsaremade， andthenexpandedtothedataset，andthemodelisre-trainedtoimprovetheaccuracyofre-labeling.Theexperimentalresults showthatthesemi-automaticlabelingmethodcanefectivelyimprovethelabelingefciencyofvehiclepartsandreducethe labor and time costs required for data labeling work.

Keywords： deep learning; image segmentation; semi-automatic annotation; vehicle parts; data

據(jù)公安部統(tǒng)計，截至2024年底，我國汽車保有量突破4.5億輛。（剩余9277字）