【作者】 韩芳芳；

【导师】 段发阶；

【作者基本信息】 天津大学 ， 测试计量技术及仪器， 2012， 博士

【摘要】 产品表面质量是产品质量的重要组成部分,也是产品商业价值的重要保障。机器视觉检测技术作为先进的产品质量监测手段,受到了生产企业越来越多的重视。本文对表面缺陷视觉在线检测的关键技术进行了较为系统的研究。以表面缺陷视觉检测的主要过程:图像获取、缺陷分割和缺陷判别为主线,对图像处理流程和关键算法进行了设计和实验分析;针对在线检测的需求,提出了算法效率分析方法和在线检测软件结构的多线程方案,建立了较为完整的表面缺陷视觉在线检测体系结构;以钢板表面缺陷检测为应用实例进行了实验验证。本文主要研究内容如下:1.以获取高质量图像为目标,提出以被测对象的特性为主导的照明方案设计原则;以表面凹坑缺陷检测为例,建立基于线阵CCD系统进行凹坑检测的数学模型,提出凹坑缺陷的图像特征;建立表面缺陷检测成像系统景深的数学模型。2.对不同背景模式的缺陷分割问题进行研究;设计完整的基于边缘的缺陷分割算法流程;提出基于小波系数层间相关性的容噪性边缘检测算法;讨论有理系数小波滤波器的设计步骤和关键问题,提出一种长度为8-4的有理系数对称紧支集双正交小波滤波器;对不同小波滤波器的多种应用效果进行比较分析。3.在缺陷图像的空间域、投影域、小波变换域进行了图像特征参数提取的研究,并利用主成分分析法进行特征空间降维;设计基于DAG SVM的缺陷分类决策树;提出采用谱系聚类优化决策树结构设计的方案。4.采用事前分析法和事后测试法对本文提出的关键算法进行时间效率分析;采用实时采集加准实时处理、多线程技术,提出适用于表面缺陷在线检测的软件系统结构方案;设计基于内存映射文件技术的存储文件系统。5.对钢板表面缺陷的在线检测进行应用研究。根据钢板表面缺陷检测指标要求,进行系统结构设计;对钢板表面缺陷的分割、特征提取、模式分类算法进行实验验证。建立了实验室环境下滚筒转动系统的实验样机,为高速表面缺陷的在线检测研究提供实验条件。更多还原

【Abstract】 Product surface quality is an important part of the product quality, and it is also an important guarantee for the product commercial value. As an advanced product quality monitoring method, machine vision inspection technology has been paid more and more attentions by manufacturing enterprises. This paper does a systematic study for key technologies of surface defects online detection based on machine vision.Based on the main procedures of surface defects visual detection, which is image acquisition, defects segmentation and defects discrimination, this paper does design and experimental analysis for image processing and key algorithms. Aiming at requirements for online detection, this paper also proposes methods for algorithm efficiency analysis and multithreading scheme for online detection software, establishes a more complete system structure for surface defects visual online detection. Some testing experiments have been made with the steel plate surface defects detection as the applying practice.1. Aiming at the acquisition of high quality images, proposes the lighting scheme design principle that based on the characteristics of measured object. Taking surface pit defect detection for instance, dose mathematical modeling for pit defect detection based on linear CCD system, and presents the image characteristic for pit defect, and does mathematical modeling for the depth field of surface defects detection imaging system.2. Studies for defects segmentation with different mode and different background; Designs completely algorithm flow chart for defects segmentation based on edge feature. Proposes a set of anti-noise edge detection algorithm based on the correlation feature of wavelet transform coefficients of different levels. Makes a discussion and research for rational coefficients wavelet filter design, gives a length 8-4 rational symmetric compactly-supported biorthogonal wavelet filter, and does experiments for the comparison of different wavelet filters application results.3. Studies for the extraction of image characteristic parameters in space domain, projection domain and wavelet transform domain, and does dimension reduction by the method of Principal Component Analysis. Designs the decision tree for defects classification based on DAGSVM algorithm, and adopts hierarchical cluster method to optimize the decision tree design. 4. Using prior analysis and afterwards testing methods, analyses the time efficiency for the key algorithms proposed by this paper. With the techniques of real-time acquisition, quasi real-time processing and multithread programming, gives the software structure for surface defects online detection, and designs the storing file system based on memory mapping file technique.5. Studies for the application of steel surface defects online detection. Makes system structure design according to measurement indicators, and does testing experiments to verify the algorithms for defect segmentation, feature extraction, and pattern classification. Establish the roller experiment prototype in the laboratory environment, which provides experimental conditions for high speed online surface defect detection.更多还原