基于Haar特征及Adaboost的焊点检测算法研究

发布时间：2018-02-26 03:00

本文关键词： 印制电路板焊点检测 Haar特征 Adaboost算法色彩特征　出处：《西安电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着现代化科技日新月异的进步,电子产品越来越朝着数字化和小型化发展,印制电路板也朝着元件密集化和微型化发展,传统的电路板检测技术在检测能力和速度上都不能满足新的表面贴装技术的要求,因此对印刷电路板表面贴片安装质量进行可靠、快速的自动检测成为提高电子制造业自动化水平和产品质量的重要途径。基于机器视觉的自动检测技术越来越受到人们的关注。这种检测技术采用了计算机技术、高速图像处理技术和识别技术、自动控制技术、精密机械技术和光学技术,是多种高技术的综合产物,具有自动化、高速化和高分辨率的检测能力,它大大减轻了人的劳动强度,提高了质量判别的客观性和准确性。对印制电路板进行检测的过程中,焊点检测是其中一个重要的检测环节。焊点的位置以及其有无缺陷,直接影响PCB印制电路板能否正常工作,所以针对焊点的检测方法研究,是目前一个研究热点。现有的焊点检测方法中,基于机器视觉的方法处于优势明显的地位,特别是在近几年很热门的机器学习领域的算法,是其中一个热门的研究方向。机器学习以其强大的泛化能力,在预测上能达到很高的准确率,且算法模型建立好后,学习过程可以自适应地达到优化,甚至在随着样本的增加,可以在检测的过程中得到二次学习。将机器学习运用到印制电路板的焊点检测中,不但能够解决人工检测速度慢,生产率低下的问题,而且还可以降低人工成本,提高检测准确率。本文围绕对印制电路板的焊点检测算法,做了以下工作:1、对印制电路板的检测技术的发展的过程做了总结,详述了目前的主要检测手段,对模板匹配法、支持向量机法和BP神经网络算法的使用做了理论研究以及性能比较。2、详细研究了Haar特征的算法理论和识别原理。对Adaboost分类器做了深入研究,同时也分析了其他相似特征即HOG特征与LBP特征的算法理论,使用OpenCV丰富的代码库,在Vusial Stdio2013的开发平台,对代码进行编写、修改和调试,通过软件实现,将Haar特征融合Adaboost分类器算法应用到了焊点检测上来。使用样本来训练生成分类器,并将生成的分类器应用到对PCB印制电路板的焊点检测过程中,最后将其检测结果与其他特征做了分析比较,确定基于Haar特征的Adaboost算法是一种效果良好的焊点检测算法。3、针对现有算法误检焊点的缺陷,提出了将色彩特征加入Haar特征的设想,首先对颜色空间模型进行了研究和分析,选取了最优的HSV模型作为颜色空间,同时通过降维处理,把三维向量的问题转换到一维空间,然后通过软件实现,成功地将颜色特征加入到了Haar特征里进行样本训练,最后采用本文所提算法对印制电路板的图像进行检测,并与之前算法的检测效果进行对比分析,验证了改进后新算法的正确性、有效性和可靠性。
[Abstract]:With the rapid development of modern science and technology, electronic products are becoming more and more digital and miniaturized, and printed circuit boards (PCB) are becoming more and more intensive and miniaturized. The traditional PCB testing technology can not meet the requirements of the new surface mounting technology in terms of testing ability and speed, so the quality of the PCB surface patch installation is reliable. Rapid automatic detection has become an important way to improve the level of automation and product quality in electronic manufacturing industry. The automatic detection technology based on machine vision has attracted more and more attention. High speed image processing and recognition technology, automatic control technology, precision mechanical technology and optical technology are the integrated products of many kinds of high technology, with automatic, high speed and high resolution detection ability, It greatly reduces the labor intensity of human beings and improves the objectivity and accuracy of quality discrimination. In the process of testing printed circuit boards, solder joint detection is one of the important detection links. It directly affects whether PCB printed circuit board can work properly, so the research of solder joint detection method is a hot spot at present. Among the existing solder joint detection methods, the method based on machine vision is in a dominant position. Especially in the field of machine learning, which is very popular in recent years, it is one of the hot research directions. Machine learning, with its powerful generalization ability, can achieve high accuracy in prediction, and after the algorithm model is established, The learning process can be optimized adaptively, even with the increase of samples, the second learning can be obtained in the process of detection. The application of machine learning to the solder joint detection of printed circuit boards can not only solve the problem of slow manual detection, The problem of low productivity, but also can reduce labor costs, improve the detection accuracy. This paper around the printed circuit board solder joint detection algorithm, did the following work: 1, to the printed circuit board testing technology development process has been summarized. In this paper, the main detection methods, the template matching method, are described in detail. The use of support vector machine and BP neural network algorithm is studied in theory and performance comparison. The algorithm theory and recognition principle of Haar feature are studied in detail. The Adaboost classifier is deeply studied. At the same time, it also analyzes the algorithm theory of other similar features, that is, HOG feature and LBP feature. Using the rich code base of OpenCV, the author writes, modifies and debugs the code on the platform of Vusial Stdio2013, and realizes it by software. The algorithm of Haar feature fusion Adaboost classifier is applied to solder joint detection, and the generated classifier is trained by samples, and the generated classifier is applied to the solder joint detection process of PCB printed circuit board. Finally, comparing the detection results with other features, it is determined that the Adaboost algorithm based on Haar features is a good solder joint detection algorithm. Aiming at the defects of existing algorithms to detect solder joints, the idea of adding color features to Haar features is put forward. Firstly, the color space model is studied and analyzed, and the optimal HSV model is selected as the color space. At the same time, the problem of 3D vector is transformed into one dimensional space by dimensionality reduction, and then realized by software. The color feature is successfully added to the Haar feature for sample training. Finally, the algorithm proposed in this paper is used to detect the image of the printed circuit board, and the detection effect is compared with that of the previous algorithm. The correctness, validity and reliability of the improved algorithm are verified.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41;TN41

【相似文献】