摘要
硅胶圈检测由于缺陷具有种类多、易变形、复杂性高等特点,通过提出多工位、多角度、不同打光方式结合的基于视觉的机电检测系统来解决这类产品的检测问题。整个检测系统的图像处理使用了基于OpenCV库按照实际项目需求进行改进,使用滤波算子、阈值分割和形态学方法。使用VGG16卷积神经网络训练模型在小样本条件对三种缺陷分类和传统视觉检测方法得到的准确率进行对照实验。在小样本下传统机器视觉的检测准确率>98%,优于深度学习的方法。通过Qt完成UI界面的设计。该软件系统使用多线程设计模式编写,并提出了一种多工位循环剔除新方法,具备在线检测计数和剔除功能。就一种特定的硅胶圈产品检测进行实际实验和测试,验证了本系统的可靠性和实用性。
Silicone ring detection has the characteristics of multiple types,easy deformation,and high complexity.By proposing a vision-based electromechanical detection system that combines multiple stations,multiple angles,and different lighting methods to solve the detection problems of such products.The image processing of the entire detection system is improved based on the OpenCV library in accordance with actual project requirements,using filter operators,threshold segmentation and morphological methods.The VGG16 convolutional neural network training model is used to control the accuracy of the three defect classifications and traditional visual inspection methods under small sample conditions.In a small sample,the detection accuracy of traditional machine vision is>98%,which is better than deep learning methods.Complete the UI interface design through Qt.The software system is written in a multi-threaded design model,and a new method of multi-station cyclic elimination is proposed,which has the functions of online detection,counting and elimination.The actual experiment and test on a specific silicone ring product test verified the reliability and practicability of the system.
作者
毕德学
程威盛
郝孝其
丁彦玉
BI De-xue;CHENG Wei-sheng;HAO Xiao-qi;DING Yan-yu(Tianjin Key Laboratory of Integrated Design and On-Line Monitoring for Light Industry and Food Machinery and Equipment,College of Mechanical Engineering,Tianjin University of Science&Technology,Tianjin 300222,China)
出处
《机械设计与制造》
北大核心
2024年第2期125-130,共6页
Machinery Design & Manufacture
基金
天津市科技支撑计划项目(18ZXRHGX00020)
天津市教委科研计划项目(2018KJ118)
天津市科技计划项目(20YDTPJC00130)
天津市科技计划项目(20YDTPJC00140)。
关键词
机器视觉
缺陷检测
硅胶圈
多线程
VGG
Machine Vision
Defect Detection
Silicone Ring
Multithreading
VGG
