基于机器视觉的泡罩药品缺陷检测系统研究

发布时间：2018-04-29 09:35

本文选题：检测技术与自动化装置 + 机器视觉　；参考：《石家庄铁道大学》2014年硕士论文

【摘要】：药品的包装是流水线上的一道重要工序，但是过程中常常会出现药品漏装、破碎、缺损、有污染物等包装缺陷问题。自动视觉缺陷检测AVI (Automated VisualInspection)已经成为现代制药行业质量控制的一个重要组成部分。我国的AVI发展起步比较晚，目前国内企业应用的AVI系统仍以国外产品为主。发展具有自主知识产权的、适合我国国情的AVI系统，具有重要的意义。以软件构架为主、基于PC的机器视觉检测系统以其经济、灵活的特点，已经逐渐成为国内厂家为新设备配套和旧设备改造的认可方案。基于PC的机器视觉的药品包装检测方法还具有非接触、智能化、高精度和高速度的特点。如何在泡罩药品包装过程中，应用机器视觉技术实现对药品缺粒、破损等缺陷进行在线检测，并做出相应的处置，是这篇论文主要讨论的问题。本文采用非接触式的机器视觉检测技术作为基本思路。根据药厂包装药品所用实际设备的情况，首先通过对包装机械的结构、运行情况和检测系统的应用的分析，对检测系统的整体方案及硬件设备进行介绍。论文的重点是对药品泡罩包装机器视觉的关键技术进行研究：首先对采集到的药品包装图像进行滤波预处理便于后续缺陷分割；然后采用一种改进的基于形态学重构和控制标记符的分水岭算法对图像进行分割，截取只包含单个药品目标的图像，便于后续缺陷特征的提取。该方法有效地克服了传统分水岭方法的过分割问题，具有分割效果好、抗干扰能力强、稳定的特点，且本方法不需要先验知识，实用性较强；接着对特征提取方法进行分析研究，对单个目标的几何形状特征进行提取，作为后续神经网络的输入，用来进行缺陷分类；对药品包装缺陷的分类方法的原理进行分析对比，最终比较了BP神经网络和RBF神经网络的优缺点，，采用RBF网络完成了对药品表面图像的分类，提高了识别效率；最后在Matlab平台上实现了算法的仿真，在VC++6.0实验平台上利用Opencv语言实现了整个软件的程序编写。试验结果表明本系统能够对泡罩药品包装缺陷进行正确的分类检测，取得了较好的检测效果。
[Abstract]:Drug packaging is an important process on the assembly line, but the packaging defects such as drug leakage, breakage, defect, contaminant and so on often occur in the process. Automated Visual Inspection (AVI) has become an important part of quality control in modern pharmaceutical industry. The development of AVI in China started relatively late. At present, the AVI system used by domestic enterprises is still dominated by foreign products. It is of great significance to develop a AVI system with independent intellectual property rights that is suitable for China's national conditions. Based on the software architecture, the PC-based machine vision inspection system has gradually become the approved scheme of the domestic manufacturers for the new equipment matching and the old equipment transformation due to its economic and flexible characteristics. PC-based machine vision based drug packaging testing method also has the characteristics of non-contact, intelligent, high precision and high speed. This paper mainly discusses how to use machine vision technology to detect and deal with defects such as lack of grain and breakage in the packaging process of drug bubble mask. In this paper, the contactless machine vision detection technology is used as the basic idea. According to the actual equipment used in drug packaging in pharmaceutical factory, the whole scheme and hardware equipment of the testing system are introduced through the analysis of the structure, operation and application of the testing system. The focus of this paper is to study the key technology of drug bubble packaging machine vision: firstly, filter and preprocess the collected drug packaging image to facilitate the subsequent defect segmentation; Then an improved watershed algorithm based on morphological reconstruction and control markers is used to segment the image and intercept the image which contains only a single drug target so as to facilitate the subsequent extraction of defect features. This method overcomes the over-segmentation problem of the traditional watershed method, and has the characteristics of good segmentation effect, strong anti-interference ability and stability. Moreover, the method does not need prior knowledge and has strong practicability. Then the feature extraction method is analyzed and studied, and the geometric shape feature of a single target is extracted as the input of the subsequent neural network, which is used for defect classification; the principle of the drug packaging defect classification method is analyzed and compared. Finally, the advantages and disadvantages of BP neural network and RBF neural network are compared, the classification of drug surface images is completed by RBF neural network, and the recognition efficiency is improved. Finally, the algorithm is simulated on Matlab platform. The program of the whole software is realized by using Opencv language on VC 6.0 experimental platform. The test results show that the system can correctly classify and detect the packaging defects of the bubble mask and obtain good results.
【学位授予单位】：石家庄铁道大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;TP274

【参考文献】