基于注意力机制与YOLOv5融合的树脂拉链缺陷检测算法研究

Research on Resin Zipper Defect Detection Algorithm Based on Attention Mechanismand YOLOv5 Fusion

针对传统的树脂拉链缺陷人工检测存在的效率低和劳动强度大等问题,本文将YOLOv5算法与注意力机制(convolutional block attention module,CBAM)相结合,对树脂拉链缺陷检测算法进行研究,给出了算法的结构原理,并对树脂拉链缺陷进行检测试验。采集带有坏齿、边缘、内部、挤出、开裂和污染的树脂拉链图像,建立数据集并据此标注。同时,利用数据集对YOLOv5网络模型进行训练,并选择900张树脂拉链缺陷图像进行测试。测试结果表明,不同树脂拉链缺陷的检测准确率不同,模型对坏牙、边缘、内部、挤压、开裂和污染6种树脂拉链缺陷的识别率分别达到99%,100%,100%,100%,100%和99%,检测目标的置信度范围为0.82~0.99,检测准确率较高,效果较好,证明模型测试的精确率达到100%,召回率达到100%,平均准确率达到98%,证明了本文方法的可行性和有效性。本文算法可实现对常见树脂拉链缺陷的检测、分类及定位。该研究对提升树脂拉链制造行业的生产效率具有一定的成效。

Aiming at the problems of low efficiency and high labor intensity in traditional manual detection of resin zipper defects,this paper combines YOLOv5 algorithm with convolutional block attention module(CBAM)to study the detection algorithm of resin zipper defects,the structure principle of the algorithm is given,and the detection experiment of resin zipper defect is carried out.Images of resin zippers with bad teeth,edges,interiors,extrusions,cracks,and contaminations are collected,and data sets are created and labeled accordingly.At the same time,the YOLOv5 network model is trained using the data set,and 900 resin zipper defect images are selected for testing.The test results show that the detection accuracy of different resin zipper defect targets is different.The recognition rates of the model for six resin zipper defect targets of bad teeth,edge,interior,extrusion,cracking and pollution are 99%,100%,100%,100%,100%and 99%,respectively.The confidence range of the detection target is 0.82~0.99.The detection accuracy is high and the effect is good.It is proved that the accuracy of the model test reaches 100%,the recall rate reaches 100%,and the average accuracy rate reaches 98%,which verifies the feasibility and effectiveness of the method.The algorithm in this paper can realize the detection,classification and positioning of common resin zipper defect targets.This research has a certain effect on improving the production efficiency of the resin zipper manufacturing industry.