基于多尺度Swin-transformer的汽车证件检测系统研究

打开文本图片集

中图分类号：TP311 文献标识码：A 文章编号：2096-4706（2026）05-0078-06

Research on Vehicle Certificate Detection System Based on Multi-scale Swin-transformer

WEI Xinquan1²， WANG Wenhong²， ZHANG Zhijia' （1.College ofArtificial Intelligence，Shenyang University of Technology， Shenyang 11o870，China; 2.Guangzhou Industrial Intelligence Research Institute，Guangzhou 511458，China）

Abstract：Toaddressthelow eficiencyand high misjudgmentand misseddetectionrates associated with manual detection inthecurrentcertificate printingquality，this paperdesignsandimplementsan integrated detectionsystem.Itintroducesa Swin-Transformeralgorithm，andincoporatesmulti-scalefeatureextraction，spatialresolutionductionndcomputatioal complexityoptimization.Ahybridquerystrategyandlow-dimensionalembedingrefinementareemployedtoguide theAtetion Mechanism towardkeyinstancesegmentationtasks，therebyimprovingtheacuracyofcriticalfieldrecognition.Experimental results show that the detection system achieves a comprehensive recognition rate of 99.35% . It outperforms other traditional OCR approaches and improves efficiency by 34.64% ， meeting the real-time processing requirements of production lines. The system is currently deployed in actual production lines and significantly improves vehicle certifcate detection efficiency.

Keywords： systemdesign; Deep Learning;Atention Mechanism; encoder;decoder;instance segmentation;real-time processing;industrial automation

0 引言

在汽车证件打印过程中，证件内容存在发生打印错误的风险。（剩余8186字）