基于大数据的胶料表面缺陷视觉识别检测研究  

Research on Visual Recognition and Detection of Rubber Surface Defects Based on Big Data

作　　者：黄景涛 杨波[2] 杨益军 田文涛 HUANG Jing-tao;YANG Bo;YANG Yi-jun;TIAN Wen-tao(Information Center,State Administration for Market Regulation,Beijing 100820,China;Library of Guangdong University ofTechnology,Guangzhou 510006,Guangdong,China;Guangzhou Yuexin Technology Co.,LTD.,Guangzhou 510006,Guangdong,China)

机构地区：[1]国家市场监督管理总局信息中心,北京100820 [2]广东工业大学(广东省)图书馆,广东广州510006 [3]广州粤信科技有限公司,广东广州510006

出　　处：《合成材料老化与应用》2023年第3期24-27,共4页Synthetic Materials Aging and Application

基　　金：国家质量监督检验检疫总局质检公益性行业科研专项(编号:201510041)。

摘　　要：胶料表面缺陷影响胶料使用,实际生产过程中要强化表面缺陷检测。对人工检测胶料表面缺陷存在的效率低、精度低、成本高等问题,提出了基于大数据的胶料表面缺陷检测方法。采用线扫描工业相机获取胶料表面缺陷图像,对获取的图像灰度化处理。将图像白色树脂转换为白雾,并去除白雾。在此基础上,采用灰度变换法对图像进行增强。采用Darknet-19卷积神经网络提取图像特征,并通过YOLOv4算法进行表面缺陷检测。胶料表面缺陷检测结果表明:对裂缝缺陷检出率最高,误检率最低;对毛团缺陷检出率最低,误检率最高;对三种表面缺陷的检测均能够满足实际胶料生产的精度和实时性要求。The surface defects of rubber compound aff ect the use of rubber compound,and the surface defect detection shall be strengthened in the actual production process.To solve the problems of low efficiency,low precision and high cost in manual detection of rubber surface defects,a rubber surface defect detection method based on big data is proposed.The line scanning industrial camera was used to obtain the image of rubber surface defects,and the obtained image was grayed.Convert the image white resin to white fog,and remove the white fog.On this basis,the gray transformation method is used to enhance the image.Darknet-19 convolution neural network was used to extract image features,and YOLOv4 algorithm was used to detect surface defects.The detection results of rubber surface defects show that the detection rate of crack defects is the highest,and the false detection rate is the lowest;The detection rate of wool defects is the lowest,and the false detection rate is the highest;The detection of three kinds of surface defects can meet the precision and real-time requirements of actual rubber production.

关 键 词：胶料表面缺陷 Darknet-19卷积神经网络 YOLOv4算法 图像处理 

分 类 号：TS57[轻工技术与工程—皮革化学与工程] TP391.5[自动化与计算机技术—计算机应用技术]