基于深度学习的汽车保险片识别插接研究  

Research on vehicle insurance tablets recognition and insertion based on deep learning

作　　者：曹宇[1,2] 张庆鹏[1,2] CAO Yu;ZHANG Qingpeng(School of Automation,Harbin University of Science and Technology,Harbin 150080,CHN;Key Laboratory of Advanced Manufacturing and Intelligent Technology Ministry of Education,Harbin 150080,CHN)

机构地区：[1]哈尔滨理工大学自动化学院,黑龙江哈尔滨150080 [2]先进制造智能化技术教育部重点实验室,黑龙江哈尔滨150080

出　　处：《制造技术与机床》2020年第12期138-141,共4页Manufacturing Technology & Machine Tool

基　　金：黑龙江省普通本科高校青年创新人才培养计划项目(UNPYSCT-2015045)。

摘　　要：针对传统汽车保险片采用人工插接无法满足工厂批量化插接的问题,提出一种基于深度学习的汽车保险片自动插接算法,该方法使用CCD工业摄像机结合远心镜头采集保险片的图像信息。采用基于KNN算法匹配保险盒的保险片插件槽位置,使用Faster R-CNN网络对保险片识别定位,算法对9种颜色的保险片准确识别,最后由SCARA四轴机器人自动完成插接操作。经过实验验证,对常见的9种颜色的保险片识别准确率能达到99.8%,对保险片插接平均周期为1 s和1.5 s时,SCARA机器人对6个汽车保险盒同时插接保险片的准确率达96.87%以上。Aiming at the problem that the traditional automobile insurance tablets can not meet the batch insertion of the factory by manual plugging,an automatic interpolation algorithm based on deep learning is proposed.The method uses the CCD industrial camera combined with the telecentric lens to collect the image information of the insurance film.The acquired image is preprocessed by Gauss-median filtering.Using the KNN algorithm to match the position of the fuse insert slot of the fuse box,use the Faster R-CNN network to identify the location of the fuse,use algorithm to accurately identify the 9 colors of the fuse,and finally the SCARA four-axis robot automatically completes the docking.operating.After the experimental design of this paper,the accuracy of the identification of the common 9 colors of the insurance film can reach 99.8%,and the average period of the insertion of the insurance piece is 1s and 1.5s,the SCARA robot simultaneously inserts insurance for 6 car insurance boxes.The accuracy of the film is over 96.87%.

关 键 词：机器视觉 深度学习 KNN Faster R-CNN 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]