基于改进YOLOv3算法的堆叠工件检测  

Stack Workpiece Detection Based on Improved YOLOv3 Algorithm

作　　者：于微波[1] 胡刘东 刘克平[1] 李岩[1] YU Wei-bo;HU Liu-dong;LIU Ke-ping;LI Yan(School of Electrical and Electronic Engineering,Changchun University of Technology,Changchun 130000,China)

机构地区：[1]长春工业大学电气与电子工程学院,长春130000

出　　处：《组合机床与自动化加工技术》2023年第4期87-90,共4页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金(61773075);吉林省发改委产业技术研究与开发(2020C018-1)。

摘　　要：针对传统物体检测算法识别堆叠工件存在准确率低以及漏检的问题,提出一种基于改进YOLOv3算法的堆叠工件检测方法。首先,引入Inception结构增强特征检测网络的特征提取能力,提高堆叠工件检测的准确率;其次,引用增强型特征金字塔结构(enhanced feature pyramid network,EFPN),提高模型多尺度特征融合能力,改善算法漏检率高的问题;最后,利用K-means聚类融合交并比损失函数(intersection over union,IOU)重新确定工件锚框,解决YOLOv3网络预设锚框尺寸不适合现有工件的问题。实验结果表明,改进算法均值平均精确度(mean average precision,mAP)达到92.89%,相较于原始YOLOv3算法提高了5.32%,F1值为0.95,召回率为93.33%,精确率为97.65%,满足堆叠工件检测的指标要求。Aiming at the problems of low accuracy and missing detection in traditional object detection algorithm,a stack workpiece detection method based on improved YOLOv3 algorithm is proposed.Firstly,the Inception structure is introduced to enhance the feature extraction ability of the feature detection network and improve the accuracy of stack workpiece detection.Secondly,the enhanced feature pyramid network(EFPN)structure is used to improve the multi-scale feature fusion ability of the model and improve the problem of high missed detection rate of the algorithm.Finally,the K-means clustering algorithm combined with the intersection union ratio loss function is used to re-determine the workpiece anchor frame,so as to solve the problem that the original preset anchor frame of YOLOv3 network is not suitable for the existing workpieces.The experimental results show that the mean average precision(mAP)of the improved algorithm is 92.89%,which is 5.32%higher than the original YOLOv3 algorithm,F1 value is 0.95,recall rate is 93.33%,and accuracy rate is 97.65%,which can meet the index requirements of stack workpiece detection.

关 键 词：堆叠工件检测 YOLOv3算法 Inception结构 增强型FPN结构 

分 类 号：TH16[机械工程—机械制造及自动化] TG506[金属学及工艺—金属切削加工及机床]