机器视觉检测技术机器应用研究/Machine Vision Inspection and Its Engineering Application

2018-08-12 14:54:46

PCB project processing tracking 焊点



机器视觉技术是一个发展十分迅速的新研究领域,是计算机科学的重要研究领域之一。机器视觉在国外已经得到蓬勃发展,而国内却是刚起步,随着我国各行业的快速发展,机器视觉技术已显得日趋重要,机器视觉技术的应用研究正成为国内业界研究的热点。
本文对机器视觉的理论知识进行了研究,并将这些技术应用在空间交会模拟目标检测跟踪和PCB焊点检测与识别两个项目中。理论与实际相结合,将算法有效的组织,解决了实际问题。
在空间交会模拟目标检测跟踪项目中,针对光照环境不稳定的问题,设计对二极管和对纸片两套跟踪方案,并在HSV颜色空间进行图像处理,提高了系统的适应能力;针对动态目标跟踪运算量大、易于受管道边缘噪声影响等问题,采用自适应背景相减法检测目标,运用kNN分类方法识别两个相似目标,用加权的移动式管道滤波跟踪算法进行目标跟踪处理,提高了空间交会模拟目标检测跟踪系统的实时处理性能和抗干扰能力。软件用Labwindows/CVI和VC++混合编程实现。
在PCB焊点检测与识别分类项目中,通过对PCB焊点图像的分析,提取其灰度图像特征和与二值图像特征,并采用画图分析方法对特征进行降维处理,进而针对PCB焊点图像样本少、各类特征交叉严重的特点,采用基于小样本、非线性的SVM(支持向量机)算法中的One-vs-Res方法对PCB焊点进行分类。算法在Matlab下仿真,软件用C#.net实现。
机器视觉技术中的图像处理和目标跟踪技术在空间交会模拟目标检测跟踪项目中得到应用,使系统的算法处理时间达到0.05s,可以跟踪运动速度为0.2m/s的目标,达到了项目要求的算法处理时间0.2s和目标运动速度0.01m/s;图像处理和模式识别技术在PCB焊点检测与识别项目中得到应用,使系统的算法处理时间小于0.5s,分类正确率达到96.57%,满足了项目要求的算法处理时间0.5s和分类正确率90%。


Machine vision technology is a new research field and developing fast. It is one of the most important fields of computer science and is attaching importance to many countries. Machine vision has been well developing in abroad, but it is still in its infancy in China. With the fast development of industry, machine vision technology has become a hot point research field of industry in our country. The scope of graphics display technology is also increasing. It applies in many areas of the national economy.

The thesis summarized the basic algorithms of machine vision and described their principle, method and technology by two projects: “Target Detection and Tracking Technology of Simulating Spacecraft Rendezvous System” and “Fault Detection and Fault Grouping of Visual PCB Soldering Inspection”. These algorithms were effectively organized to solve practical problems.

In the project “Target Detection and Tracking Technology of Simulating Spacecraft Rendezvous and Docking System”, As for the problem that the light of the environment is unstable, the thesis designed two tracking program – tracking diode and tracking white scrip and HSV color instead of RGB color was used for image processing to enhance the system’s adaptive ability. As for the problem that the computation of dynamic target tracking is so large and the tracking result is often affected by pipeline border noise, the thesis adopted adaptive background difference method to detect the target, used kNN(k-Nearest Neighbor) classification method to distinguish two similar objects, and used a mobile pipeline wave filtering tracking algorithm to solve target tracking. The methods above effectively resolved the problems and improved the real-time processing function and anti-jamming ability of the system.

In the project “Fault Detection and Fault Grouping of Visual PCB Soldering Inspection”, as for the given image sample is small and various features are serious cross, the paper analyzed the PCB soldering image, extracted the features of the image, reduced feature dimension and classified the soldering with the SVM method basing on small samples and nonlinear character.

Image processing and target tracking technology of machine vision was applied to the project “Target Detection and Tracking Technology of Simulating Spacecraft Rendezvous and Docking System”, so that the algorithm processing just spent 0.05s and the target moving at a speed of 0.2m/s could be tracked, which met the project demand that the algorithm processing time less than 0.2s and the speed of target greater than 0.01m/s. Image processing and pattern recognition technology was applied to the project “Fault Detection and Fault Grouping of Visual PCB Soldering Inspection”, so that the algorithm processing spent less than 0.05s and the correct classification rate reached 96.57%, which met the project demand that the algorithm 0.5s and the correct classification rate 90%.