SIA OpenIR  > 自动化系统研究室
基于机器视觉的火腿肠在线包装质量检测方法研究
Alternative TitleResearch on the Online Sausage Packaging Quality Inspection Method Based on Machine Vision
陈帅1,2
Department自动化系统研究室
Thesis Advisor史海波
ClassificationTP242.62
Keyword质量检测 机器视觉 图像分割 特征提取 分类识别
Call NumberTP242.62/C47/2011
Pages107页
Degree Discipline机械电子工程
Degree Name博士
2011-11-28
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract食品包装不仅要求外形美观、方便实用,更重要的是要保证包装的质量,确保食品安全。随着同类产品的竞争加剧,各火腿肠生产厂家在不断提升产品的内在品质的同时,越来越注重产品的包装质量检测,但是目前的检测手段还是传统的人工检测方式。机器视觉检测技术作为先进的检测手段,具有快速、准确和可靠等优点,因此,开发基于机器视觉的火腿肠在线包装质量检测系统对于减轻火腿肠生产企业员工的劳动强度,提高企业的生产效率和自动化程度,提高食品包装检测水平,保证食品安全都具有十分重要的意义。 本文在总结机器视觉检测系统应用现状和相关理论发展的基础上,分析了基于机器视觉的火腿肠在线包装质量检测系统所面临的难点,对系统相关的图像分割、特征提取和缺陷识别方法进行了研究。 目标主体区域和缺陷区域的图像分割方法是检测系统的基础。本文根据火腿肠的缺陷分布特点,对火腿肠进行了区域划分,提出了一系列分割方法。针对主体区域分割,由于火腿肠具有种类繁多、表面图案复杂、背景易发生变化的成像特点,本文深入研究了自适应阈值的分割方法,提出了基于直方图波谷自适应阈值和基于改进模板的二维最大熵自适应阈值的分割方法,通过对比分析,最大二维熵法能够准确地分割出不同种类的火腿肠主体区域,并具有较强的抗干扰性。针对破袋区域分割,提出了一种基于空间分布密度特征和颜色特征组合的缺陷区域分割方法。针对接缝区域分割,提出了一种基于颜色特征和投影特征组合的图像分割方法。 缺陷图像的特征提取和选择是缺陷分类识别的前提。本文对火腿肠图像的颜色特征、轮廓特征、边缘特征、几何形状特征和纹理特征的提取方法进行了研究。针对轮廓特征提取,提出了一种多尺度自适应阈值的轮廓角点提取方法,该方法能够准确提取出鼓泡区域轮廓存在的角点特征。针对边缘特征提取,提出了一种基于预测的边缘检测方法,该方法能够准确提取出划破肠衣的边缘特征,并具有较高的运行效率。 火腿肠包装缺陷识别是一个多类、多特征的模式识别问题,本文根据火腿肠外观缺陷特征描述,对专家规则方法和神经网络方法进行了对比研究和实验,设计了一种基于专家规则与神经网络的分层识别算法,实现了火腿肠各种包装缺陷图像的识别。 系统分析了火腿肠在线包装质量检测系统的功能和性能需求,设计了一种基于机器视觉的检测系统,实现了产品的有序摆放、全方位图像采集、实时包装质量检测和次品剔除等功能,并将本文提出的图像分割、特征提取和缺陷识别方法应用到系统中,经过实验验证,系统达到了预期目标,填补了火腿肠生产行业相关包装检测设备的空白。
Other AbstractFood packaging needs not only beautiful appearance, but also being convenient and practical, more importantly, must ensure the quality of packaging and food safety. With the increasing competition of similar products, sausage manufacturers put more and more emphasis on the product’s packaging quality inspection except keeping upgrade the product’s inner quality. However, the inspection method is still the traditional manpower. Machine vision inspection technology is a kind of advanced inspection method with fast, accurate and reliable characristic, therefore, the development of the online sausage packaging quality inspection system based on machine vision has great significance in reducing the sausage production employe’s labor intensity, and improving production efficiency and automation level, and promoting food quality inspection standards and ensuring the food security. The application status of the machine vision inspection system and the development of the related theory are summarized in this paper, and the difficulty of developing the sausage inspection system based on machine vision is analyzed, and the related technology such as image segmentation, feature extraction and defect recognition are investigated. Image segmentation methods for segmenting the sausage body region and defect region are the basis of the inspection system. According to the distributed characteristics of the sausage defects, the sausage is diveded into several regions, and a series of segmentation methods are proposed. For the segmentation of body region, because the sausage has a wide range of types and surface patterns, and the background image is easy to change, the adaptive threshold image segmentation method is studied in this paper, and two adaptive threshold segmentation methods based on histogram trough and the model improved two-dimensional maximum entropy are proposed,the two-dimensional maximum entropy method can segment different kinds of sausage body region precisely and has good noise immunity. For the burst defect region segmentation, a new segmentation method combination of spatial distribution of density and color feature is proposed. For the seam region segmentation, a new segmentation method combination of color feature and projection feature is proposed. Feature extraction and selection of the defect image is the precondition of the defect classification and recognition. The extraction methods of color feature, contour feature, edge feature, geometry feature and texture feature of the sausage image are investigated in this paper. For the extraction of contoure feature, a new corner extraction method based on multi-scale adaptive threshold is proposed, and the corner feature of the bulb region contour is extracted precisely. For the extraction of edge feature, a new edge detection method based on forecast theory is proposed, and the laceration defect edge feature is accurated extracted with high efficency. Sausage packaging defect recognition is a multi-class, multi-feature pattern recognition problem, according to the sausage’s appearance defect feature, the expert rules and neural network method are compared in this paper, and a series of experiments are carried out, and a layer recognition method based on expert rules and neural network is proposed. By this method, the all sausage defects’ recognition are realized. The function and performance requirement of the online sausage packaging quality inspection system are analyzed, and a kind of machine vision inspection system is designed, and some functions are realized such that products are arranged in order, and the image is captured all around, and the package quality is online inspected, and the defects are eliminated. And the proposed image segmentation method, feature extraction method and defect recognition method are applied in the system, after a series of experiments, the system is proved to achieve the predefined target, and the system fill the blank of sausage package inspection device in the sausage manufacture industry.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/9320
Collection自动化系统研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院研究生院
Recommended Citation
GB/T 7714
陈帅. 基于机器视觉的火腿肠在线包装质量检测方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2011.
Files in This Item:
File Name/Size DocType Version Access License
基于机器视觉的火腿肠在线包装质量检测方法(1767KB) 开放获取LicenseApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[陈帅]'s Articles
Baidu academic
Similar articles in Baidu academic
[陈帅]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[陈帅]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.