SIA OpenIR  > 光电信息技术研究室
Alternative TitleThe Method Research of Material Classification based on Machine Vision
Thesis Advisor郝颖明
Keyword物料识别 余弦变换 小波变换 Gmm模型 Em算法 智能相机
Call NumberTP391.4/M15/2010
Degree Discipline模式识别与智能系统
Degree Name硕士
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract随着生产自动化水平的日益提高,企业对产品质量的要求越加严格,基于机器视觉的检测技术以其非接触、精度高、速度快、成本低等优点逐渐被企业所了解和接受。基于视觉的物料识别作为机器视觉的一个应用领域,用于对生产过程中的多种物料进行分类、分拣和分级等,是后续的计量、剔除另类、检测缺失(陷)、实时监控和数据管理等处理工作的基础,正在成为当今检测技术研究的热点之一。本文针对广东韶山钢铁厂亟需解决的对同一传送带上分时传送的几种典型炼钢物料(球矿、焦炭、铁矿)和空载传送带状态的自动识别,开展基于视觉的物料识别方法研究。完成的工作包括: (1) 为减少空载图像对三类物料图像识别的干扰,首先提出将空载状态识别出来的研究思路。鉴于空载图像的有效边缘大大少于三类物料图像的有效边缘,提出了利用图像边缘量来识别空载状态的方法。边缘量用5级提升小波变换的高频子带系数绝对值的平均值表示,比较实验得出DB4小波识别空载图像最优。 (2) 针对球矿、焦炭、铁矿三种物料的识别,提出了基于颜色和变换域的物料识别方法。鉴于焦炭图像整体近于黑色,球矿和铁矿偏红色的特点,提出利用图像红色分量(RGBred)与对应灰度值(RGBgrey)距离的平均值识别焦炭;针对球矿形状相似、纹理规则,铁矿形状不一、纹理杂乱的特点,提出利用两类图像频域差异分类的方法。 (3) 针对多数物料图像小波域高频子带系数稀疏的特点,提出了基于混合高斯模型(GMM)的物料识别方法,该方法不仅适用于前述三种典型物料,也可用于其他物料识别。 (4) 建立了基于VC4465智能相机的物料识别工程实现模拟实验平台,分别在PC机和相机内实现了两种物料识别算法。基于该平台,进行了炼钢物料和三种农产品的识别实验,实验得出两种方法识别率高于96%,最大识别时间在380ms之内,能够满足实际要求。
Other AbstractWith the automation level of production increasing, product quality requirement concerned by business become more stringent increasingly, detection technology based on machine vision is gradually being understood and accepted due to its non-contact, precision, high speed and low cost. As a machine vision application, material classification used to a variety of manufacture materials for classification, sorting and grading, etc., is becoming one of the research focus of today's detection technology, which is the foundation of measurement, excluding heterogeneous, missing (depression) detection, real-time monitoring and data management in follow-up work. In a steel factory of guang dong, some typical steel materials (ball ore, coke, iron ore) and the empty conveyor belt state are urgent needed automatic identification which are in the same conveyor belt and time-sharing transmission. In light of this demand, this paper carried out the material classification method research based on machine vision. The main research content and complete tasks are following: (1) In order to reduce the affect of the three materials classification, we proposed the research ideas of classifying conveyor belt in first. Because of much less effective edge in conveyor belt images than the three types of materials images’. This paper proposed using of edge content to identify the empty state. Edge content is expressed by the absolute average value of high frequency sub-band coefficients after 5-times wavelet transform, and experiment results DB4 wavelet identification is optimal. (2) For the ball ore, coke and iron ore three types of materials classification, this paper proposed a color and transform domain based methods. For the coke’s overall picture is nearly black, while the ball mine and iron ore are biased red, therefore this paper classifies coke images using the absolute average value of the red image component (RGBred) and the corresponding Grey value (RGBgrey); for the following characters: ball mine’ shapes are similar, texture being rule, while iron ore has different shapes, texture being clutter, they are classified by making use of the differences between two types of image in frequency domain. (3) Against the sparse feature of many material images' high frequency sub-band coefficients of wavelet domain, this paper proposed a Gaussian Mixture Model (GMM) based method, pointing out method 2 can be applicable not only to the first three typical materials nut also classifying other materials. (3) Experimental platform was built based on VC4465 smart camera, and two kinds of material classification algorithm are respectively implemented within PC and VC4465. Based on this platform we carry out classification experiments of steel materials and 3 agricultural products, experimental classification accuracy of two methods are both higher than 96%, maximum classification time within 380ms,meeting practical needs.
Contribution Rank1
Document Type学位论文
Recommended Citation
GB/T 7714
马怀志. 基于视觉的物料识别方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2010.
Files in This Item:
File Name/Size DocType Version Access License
基于视觉的物料识别方法研究.pdf(2501KB) 开放获取LicenseApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[马怀志]'s Articles
Baidu academic
Similar articles in Baidu academic
[马怀志]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[马怀志]'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.