CASIA OpenIR  > 毕业生  > 博士学位论文
智能检测系统关键技术研究
其他题名Research on Key Techniques of Intelligent Inspecting Systems
路香菊
学位类型工学博士
导师王云宽
2008-06-06
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业控制理论与控制工程
关键词视觉检测 异纤分拣 对称性 二维条码 Visual Inspection Impurities Detection Symmetry Measure 2d Barcode
摘要智能检测是计算机通过智能传感器或敏感元件从客观事物中获取信息,借以认识客观事物并加以判断或分类的一种综合性的科学方法。 信息获取是智能检测系统最关键的环节之一,而机器视觉的发展使计算机可以快速获取大量信息,且易于自动处理,并易于同设计信息以及加工控制信息集成。因此,视觉检测技术在现代自动化大生产中被大力的发展和应用,成为智能检测的主要研究方向和研究热点。然而,视觉检测系统中很多具有挑战性的关键问题一直没有得到很好的解决,如目标的准确定位及表达、算法的可靠性与自适应性等。特别是在国内,自主视觉应用系统的研发才刚刚开始,其发展很不完善。 本文针对几种典型视觉检测系统中的若干关键技术问题进行了深入研究,主要研究内容包括不变性特征提取、鲁棒性的识别与检测算法等。 本论文主要的工作和贡献有: (1)针对现有棉花异纤检测系统在高速情况下的可靠性差、误检率高等问题,给出了一种有效的纯色纤维中异类物质的检测方法。 与其他同类算法相比较,本文算法优点包括:一是计算量更低,降至为4n,这里n为要处理的图像的尺寸;二是可靠性更高,能在一定程度上实现对光线变化以及阴影的不变性检测,因此,其通用性更强。 (2)针对现有棉花异纤检测系统中的精度低的问题,提出了一种基于灰度图像分析的高精度棉花异纤检测方法,并定义了一种上下文特征空间(CFS),此CFS包含:图像的灰度特征、梯度特征、局部熵特征以及尺度形态滤波特征。基于CFS中的不同特征,分别应用模糊推理规则,实现了对棉花中混杂的各类杂质的分类检测,特别是对一类细微的不明显杂质的高精度检测。 (3)基于图像中两对称点的距离、相位及密度权重的关系分析,提出了一种计算图像对称性测度的方法。并针对复杂背景下目标的难以定位与识别等问题,给出了一种基于图像的对称性测度来定位复杂背景下圆形目标的方法,实现了任意圆形物体的基于内容的不变性检测。 (4)针对目前基于机器视觉读取二维条码技术中所存在的自适应性差、受背景干扰严重等问题,给出了一种二维条码PDF417的鲁棒性定位与识读方法。对PDF417的定位,提出了基于边界起止符号的灰度投影曲线匹配的定位方法;对PDF417的译码,提出了基于特征值、投影以及其差分向量的分析方法。本文算法最终实现了复杂背景中一类信息密度较大、条与空对比不明显的PDF417条码的识读。 总的来说,本文在针对以上几种典型的视觉检测应用系统中的若干关键技术问题,作了深入的分析和有益的探索,在一定程度上提高了系统的性能。
其他摘要The technique of intelligent inspection is a comprehensive scientific method which is capable of describing or classifying objective things, relying on a computer's decision-making deduced from dealing with the acquired information of smart sensors. Vision-based inspecting systems are widely and increasingly developed in today's diverse automatic fields. However in China, vision-based applications are still fresh and not well developed. In this thesis, we study on the key techniques of several typical visual inspecting systems. The main contributions of this thesis include the following issues: (1)The main shortcomings of the current vision-based systems for inspecting cotton include the low efficiency and adaptability, etc. And we present an efficient algorithm for detecting impurities in a kind of unicolor fibres. The superiority of our method include: firstly, the amount of its computational work reduces to 4n, where n is the total size of the image, much smaller than that of the previous work. Secondly, it is invariant to shadows and illuminations, hence it is more adaptively to process images captured under different situations. (2)Based on analysis of gray-scale images, we propose a method for detecting different impurities in cotton with a high-precision, in order to improve the precision of the current systems. Our approach aims at constructing an adaptive fuzzy rule based on a feature space of gradients, gray values, local entropies and local features from scalable morphological filters, to color different types of impurities. Due to the fact that the method is based on a full contextual information, it can handle small changes in images more adaptively. And it is perfectly suitable to distinguish tiny and indiscernible impurities mixed in fibre masses. (3)It's still a challenging work to locate and recognize objects in a strong backgrounds. And we propose a computational method based on distance, phase and intensity weight functions, to quantitatively measure the symmetry of an object. Then based on the symmetry measure, a systemic approach is given to robustly locate and invariantly inspect arbitrary circular objects. (4)The previous work to read out 2D barcode based on image analysis failed when the barcode is with high stripes' densities in a cluttered environment. We develop a robust approach to read 2D barcode PDF417. Firstly, we propose a stop/start pattern projecting curve matching method to locate PDF417. And secondly, we implement a method to analyze the eigenvalues, projections, the difference vector of eigenvalues and projections, to decode PDF417. The main advantage of the method is its high performance to decode PDF417 with high-density,weak contrast between bars and spaces, and under uncertain complex background. To sum up, we have made a lot of fruitful attempts and significant progresses on some key techniques of several typical vision-based inspecting systems.
馆藏号XWLW1283
其他标识符200518014628023
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6120
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
路香菊. 智能检测系统关键技术研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2008.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20051801462802(7402KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[路香菊]的文章
百度学术
百度学术中相似的文章
[路香菊]的文章
必应学术
必应学术中相似的文章
[路香菊]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。