【作者】 贺福强；

【导师】 陈子辰； 王文；

【作者基本信息】 浙江大学 ， 机械制造及其自动化， 2012， 博士

【摘要】 随着科技发展及社会生活水平的提高,家用沙发、汽车座椅、皮衣、皮鞋等皮革制品的应用越来越广泛,对皮革制品加工质量与效率的要求也越来越高。由于皮革原料取自动物毛皮(牛皮、羊皮、猪皮等),其表面不可避免地存在疤痕、斑点、皱褶、孔洞和凹陷等多种瑕疵,因此在皮革原料加工中必须检测出可用的有效区域(剔除瑕疵,获取可用区域范围)。长期以来皮革制品加工生产中,皮面检测、排样、切割主要依靠人工完成,存在劳动强度大、主观因素影响严重、一致性差、皮革原料浪费率高等问题。为了有效利用皮革原料、提高产品质量和生产效率、降低生产成本,皮革制品加工工艺与设备逐步向高效、可靠的自动化方向发展,特别是皮革表面的人工检测工艺已逐步被机器视觉检测所替代。本学位论文以浙江省科学技术厅科技攻关重大项目“皮革制品准柔性制造技术开发(项目编号：2003C11023)”为背景,针对大面积皮革原料(牛皮)表面可用区域的机器自动识别需求,采用机器视觉检测方法获取可用区域,研究皮革表面瑕疵检测和可用区域提取的理论方法,解决若干关键技术,研制相应的样机,并通过实验研究验证理论方法的可行性和系统的有效性。论文的主要研究内容包括：1.综述目前图像去噪方法的基本原理及特点,针对皮革图像采集过程中混合噪声对图像质量影响问题,在不破坏皮革纹理细节信息的条件下,研究基于小波自适应阈值和中值滤波相结合的联合去噪技术,准确保留皮革边界及线条、斑点、皱褶、孔洞、疤痕和凹陷等重要皮革表面信息,并通过实验验证其可行性。2.针对工业CCD视场小、皮革材料柔软特性和皮革裁剪机结构限制等问题,从理论和实用角度探讨工业现场中图像拼接的实时性和可靠性,研究基于Gabor-Zernike矩几何相似三角纹理特征块的图像拼接技术,结合硬件条件,解决图像拼接算法复杂、速度慢等问题,以实现大面积皮革视觉检测中序列图像的快速精确拼接,获得皮革的整体轮廓特征,展现大面积皮革的全局信息。3.针对皮革纹理和瑕疵全局随机性、特征提取难和计算量大等问题,研究基于粒子群优化模糊聚类方法的皮革表面瑕疵实时检测技术,利用粒子群快速全局寻优,结合模糊聚类的特点,采用类间最大距离和类内最小距离的方法快速提取瑕疵区和优质皮区的最佳特征,解决传统模糊聚类算法基于梯度下降的迭代耗时过程,提升全局搜索能力和提高聚类效率,实现瑕疵区域和皮革优质区域的快速聚类,通过图像分割处理得出可用区域。针对皮革瑕疵检测的实时性问题,通过对每一幅待检皮革图像进行多层正交小波分解,提取各层低频子图像的能量和局部同质性等共生矩阵特征进行分析,自动确定分辨率级数和选取分解子图像的小波频带进行重构,最后在重建图像中采用自适应的二值阈值法分割出皮革的瑕疵区域和优质区域,把纹理表面图像的瑕疵检测转化为非纹理图像的瑕疵检测。4.为实现皮革样件智能排样和数控裁剪等后续工序,对皮革可用区域边界和轮廓进行位图矢量化处理。针对在生成复杂不规则皮革可用区域边界与轮廓矢量图形时的断点、不连续和支线等问题,提出一种改进链码表示的轮廓边界位图矢量化技术,以实现皮革可用区域矢量化图形的任意编辑,满足智能排样要求,实现皮革样件的无图纸加工。5.采用基于机器视觉的检测技术研制大面积皮革原料表面的视觉检测系统,实现图像采集、拼接、瑕疵识别与可用区域提取等功能,满足工业生产的技术需求。6.结合理论研究,在皮革数控裁剪机上进行混合噪声去噪处理、纹理特征块图像拼接、瑕疵检测及皮革可用区域轮廓位图矢量化试验,验证各项技术方案的可行性与系统工作的有效性。更多还原

【Abstract】 With the development of science and technology and the improvement of standard of living, home sofa, car seat, leather garments, leather shoes and other leather products are used more and more widely, leather products processing quality and efficiency requirements are also getting higher and higher. As a result of leather raw materials from animal fur (leather, sheepskin, pigskin), its surface will inevitably exist various defects, such as spots, wrinkles, scars, holes and other defects, therefore in the raw leather processing, the available efficient area (excluding defective, acquiring available area) must be first detected.Long term since the leather products processing in detection, layout, cutting mainly rely on manual completed, the labor intensity is large, the subjective factors influencing severity, poor consistency, raw leather waste rate high. In order to make effective use of leather raw materials, improve product quality and production efficiency, reduce production costs, leather products processing technology and equipment gradually to the efficient, reliable automated direction, especially the leather surface artificial detection technology has gradually been replaced by machine vision detection.This dissertation combines Zhejiang province key science and technology project" leather products pseudo flexible manufacturing technology development (Grant NO.2003C11023)", for the large area of raw materials for leather (cowhide) surface using computer vision detection method to get the available area, study of leather surface defects detection and the available area extraction theory method, to solve some key technologies, the corresponding development the prototype, and through the experimental research to validate the feasibility of theoretical method and the effectiveness of system. The main contents of this paper include:First, This paper summarizes the current image denoising methods of basic principle and characteristics, In view of the problem of mixed noise influence on image quality in the process of leather image acquisition.In the conditions of do not damage the leather texture detail information, Study on the joint denoising technology of based on wavelet adaptive threshold and median filtering, Accurate retention important leather surface information of leather boundary, lines, spots, wrinkles, scars, depressions, holes and other defects, And through experimental verification of the denoising effect.Second, In view of the problems of small field of view in industrial CCD, leather soft material properties and leather cutting machine structure of restrictions and other issues, from theoretical and practical perspective to study on industrial scene image mosaic of the real-time and reliability, Research the image splicing technology based on the Gabor_Zernike moments of geometric similarity triangle texture block, combined with the hardware conditions, to solve the problems of image mosaicing algorithm complex, slow speed. In order to realize a large area of leather fast and accurate image sequence stitching in visual detection, to obtain the overall characteristics of leather, show large leather global information.Third. In view of the problems of leather texture and defects of global random, feature extraction is difficult and the large amount of calculation. Study on fuzzy clustering based on Particle Swarm Optimization of leather surface defect real-time detection technology, the use of particle swarm fast global optimization, combining fuzzy clustering features, using the maximum distance between and within class minimum distance method for rapid extraction the best feature of defect area and high grade skin area, instead of the traditional fuzzy clustering algorithm based on gradient descent iterative process, enhance the ability of global search and improve the clustering efficiency, realize the defect region and leather quality region rapid clustering, Through image segmentation is processed to derive usable area. Aiming at the problem of real-time leather flaw detection, by each of a detected leather image wavelet decomposition. Extraction of each layer of low frequency sub image energy and local homogeneity feature analysis. Automatically determine the resolution series and select decomposition sub image of wavelet frequency to reconstruct, finally in the reconstructed image by using the adaptive threshold method to divide into the defect area and the quality area, the texture image defect detection is converted into a non-texture image defect detection..Fourth, In order to realize the leather sample intelligent nesting and cutting process, for leather available area boundary and contour vectorization of bitmap processing. In view of the problems of breakpoint, discontinuous and branch lines in the generation of complex irregular leather available area boundary and contour vector graphics, a new contour bitmap vectorization technology of improved chain code representation is presented, realization of vector graphics arbitrary edit for leather available area. To meet the requirements of intelligent nesting, implementation of the leather sample without drawing processing.Fifth, Based on machine vision detection technology to build a large leather raw surface of the vision detection system, Implementation the functions of image acquisition, stitching, defect recognition and the available area extraction, to meet technical requirements of the industrial production.Sixth. The combination of theoretical research. In the numerical control leather cutting machine for test of mixed noise denoising effect, texture block image stitching, defect detection and leather available region contour vectorization of bitmap, validation of the technical feasibility and system working effectiveness.更多还原