【作者】 谢莉青；

【导师】 于伟东；

【作者基本信息】 东华大学 ， 纺织材料与纺织品设计， 2010， 博士

【摘要】 织物构成要素涉及结构和品质两要素,是机织物本身特征的表述,也是其质量的基本依据。人们以往只是根据应用特征将其简单地划分为外观、物理(力学)品质,并多为定性的表述与评价。显然,建立这两类要素的表征及对应方法,进而明确结构要素与织物性能间的关系,对织物的生产和使用都具有十分重要的理论指导意义。机织物构成要素包括织物组织、结构、纱线构成、纤维的类别与混合比等。要从一块织物上获得如此大的信息,最佳检测方案是采用同试样同机测量的技术来实现,但目前尚无此类多特征、易交互纺织面料的组合测量理论和装置。本文针对覆盖织物构成要素的基本评价,将原本分立进行的检验集中到一台操作系统上同步完成,探索实现织物大容量同机同试样测量的新途径。研究在以下方面取得成果：(1)研制了一个集成光、机、电、算和控制技术一体化的织物构成要素的组合原位测量装置(FaCFAS系统)。FaCFAS系统由变焦、高分辨、动态摄像系统,多光路光源系统和力学测量系统组成。它可以有效地摄取织物的组织和结构特征及纱线的结构与形态、实现纤维的观察识别和图像获取,为利用数字图像处理分析技术对织物构成要素的原位测量提供了新的手段。(2)针对以往图像算法难以解决紧密织物,歪斜织物,扭曲变形织物,绒面织物及组织结构不规则织物的纱线定位及纱线排列密度的换算问题,建立了一种力学拉伸矫正技术。通过力学拉伸方式去矫正织物的结构,从而改变了织物中纱线间隙的遮盖、移位的情形,获得清晰化的纱线排列的图像要素特征。特征的归一化,可统一纱线的分割算法,实现了纱线的准确定位和纱线排列密度的测量。实践证明效果较好,因此开辟了一种力学拉伸矫正技术和图像处理技术组合的新思路和新方法。(3)在色纱排列分析方面：讨论了力学拉伸对织物颜色识别的影响,设计了若干针对色纱划分和颜色分析的算法。获取色纱信息的浮动窗口算法能够提取到纯组织点的像素；所设计的减色算法有效降低了数据的处理量,简化了运算过程；改进的模糊C均值算法实现了色纱分色；所设计的调色板算法,采用了点位取样和颜色统计值中位数做色彩,有效克服了随机的颜色变异点的失真影响,明显优于通常使用均值生成色彩的效果;具有自验功能的模板叠合算法,纠正色纱排列图的少数错误点,计算复杂度低,是对初始识别结果进行二次识别复验的把关工具。(4)在织物组织的识别方面：提出了力学拉伸与图像处理结合的识别方案和相应算法。基于遗传算法的特征筛选和可分性原则的特征整合,构造出可以克服因毛羽或不确定因素所致个别组织点特征数据变异的组合特征,使缩减特征维数和提高特征性能兼备。力学拉伸织物的作用,强化了图像纹理,使组织点的异类特征值更趋分离,简化为明暗特征和纹理特征的织物的两个判析模式。对多个平行试样的组织意匠图,利用最小规格化和经组织点配准算法及模板叠合算法进行互验,可纠正原意匠图中少数的误判点,重构意匠图的识别正确率达到100%。(5)在织物中纱线结构的测量方面：其一,从织物中原位测量纱线的直径,不同测量点的测量精度不同。通过实验得到织物组织点处纱线的投影宽度dc与纱线的真实直径的线性度最好,应选择该处作为测量点。每个样品按照本文的方法测量26次,取置信度为0.95,则测量值与真实值的误差不超过3.5%。提出了纱线的压扁系数η是一个具有很强实用性的纱线几何参数,通过对η的测算对于预测纱线和面料的一些性能具有参考价值。η小的织物,手感柔软,质地轻薄,覆盖接触面积大,防风保暖性好,适宜做秋冬的贴身面料。其二,利用图像全局特性检测纹理的方向,将空间域比较困难的全局检测问题转化为适合探测纹理形状和主体方向的变换域求解。设计了基于Fourier变换的纤维倾角算法,从角向功率谱中提取出纹理的主方向,由于受到样品和技术因素的影响,造成测量结果波动,代表性和稳定性尚不令人满意。设计了基于Hough变换的纤维倾角算法,即使当纤维粘结成块,代表条纹方向直线出现的几率仍然是最大的,用简单的数据结构表达条纹的外形,它的抗噪能力强,直观性较好。与人工测量结果的相关系数为0.959(?)误差基本控制在5%之内,具有应用潜力。其三,长期以来,纱线密度(δ)一直沿用简化假设基础上的经典理论和公式,被视为均值的表达,无法正确反映影响纱线中纤维实际的分布状态。本文以纱线直径与纱线线密度关系的经典模型为基础,利用δ随纱线结构松紧变化引起纱线外形轮廓的变化,而捻角能够体现纱线结构松紧变化的规律,本算法采用有监督的方式,引入了捻角对传统的δ值的修正系数kβ,由于质量沿长度的随机变化,将原本为恒定质量概念的纱线密度,难以通过称重获得其分布,而采用基于图像形状特征的间接测量,实现了一种纱线密度δT分布测量的新方法。更多还原

【Abstract】 Woven fabric structure, including structure and quality, not only refers to woven fabric characteristics, but also presents fabric fundamental quality. People used to assess fabric appearance, physical (mechanical) quality according to the application characteristics, most of assessments are qualitative, which are inadequate for identifying fabric properties and application features. It is highly desirable to develop a new simple and quantitative approach to find the relationship of fabric structural elements and fabric properties, which is important to give advice to production and use.The constituent elements of fabrics include fabric texture, structure, yarn composition, fiber type, and blending ratio and so on. The best effective method of obtaining so much information from fabric is using the same sample on the same mechanics, but there are no such textile fabric measurement theory and devices combining multi-feature and easy interaction.In this paper, we present a novel approach by which the same samples are tested on the same mechanics, carrying out synchronization tests in a single operating system according to the basic evaluation of the constituent elements of fabrics.Research results are as following areas:(1) Develop a measuring device (FaCFAS system) integrating light, machines, electricity, computing and control technology. FaCFAS system consists of a zoom, high-resolution, dynamic camera system, multi-source optical system and mechanical measurement system components. It can been effectively used to capture clear image and obtain materials’ mechanical properties and structural features, providing a new mean to measure the constituent elements of fabrics using digital image processing and analysis technology.(2) In order to overcome the difficulties and low precision for automatic recognition of yarn array and yarn density from woven fabrics such as tight structures, skew, twisted, suede or non-uniform structures, we invent a new mechanical Stretching & Correcting Technique. The sample is forced by biaxial stretching undestructively; it can not only improve the woven fabric structure, but also produce or enhance image features for yarn arrays. Normalized features can unify segmentation algorithm of yarns to achieve yarn location accuracy and yarn density measurements. The new method explored a new technology which combined stretching- correcting and image processing, and it is proved to be very effective and practical.(3) Analyze dyed yarn array:discuss mechanical stretching effect on the fabric color recognition, design a number of algorithms for yarn division and color analysis. Floating window arithme, which obtained dyed yarn information, can extract pure organizational point of the pixel; subtractive algorithm is designed to reduce the amount of data processing, simplify the computing process; improved Fuzzy C Means algorithm achieves dyed yarn separation; the designed palette algorithm, using Point-bit sampling and Color statistical value median as color, effectively overcomes effects of random color variation point of distortion, effects of which are better than mean-generated; a self-test function template composite algorithm corrects a few mistakes of dyed yarn arrangement diagram, the computational complexity is low, this algorithm are the initial recognition results of the second re-inspection and checking identification tools.(4) In the aspects of fabric recognition:propose the mechanical stretching and corresponding algorithm. According to the feature selection of genetic algorithm and the separable principle of feature integration, combination of characteristics of the data which can lead to variation of individual organizations is constructed to overcome the hairiness or uncertainties in order to reduce the characteristics of dimension and improve the performance characteristics at the same time. The role of mechanical stretching fabric reinforced the image texture features of the organizational point of the value of more heterogeneous separation, which reduces to light and dark features and texture features of the fabric analysis of two sub-modes. The smallest normalization and the organization point of the template matching algorithm and the algorithm in each composite examination are used for the organization of multiple parallel samples’pattern grid. In this way, a few of the original pattern grid misjudgment points can be corrected. The rate of reconstruction of pattern grid which is corrected can be up to 100%.(5) In the aspects of fabric yarn structure measurement:First, if the yarn diameter is measured from the basic point of the fabric, there is different measuring accuracy from different points. It is the projecting width of the yarn from the organizational point that is best in according with the actual diameter in the linearity. And the relative error of correcting is the minimum.So it should be selected as the measurement points. Each sample should be measured 26 times in accordance with the articles. The confidence level should be taken as 0.95. In this way, the error between the measured value and the true value is less than 3.5%.There is reference value for the prediction of some properties of yarns and fabrics by the calculation of pressing coefficient of the yarn which brought forward highly practical value among all the geometric parameters of yarns. If the fabric’s is smaller, then it is softer in handle, lighter in texture, the contact area covered is larger, wind proofness is better. This kind of fabric is suitable for autumn and winter.Secondly, the image’s Overall Specification is used in detecting texture’s direction to transform the problem of overall detection which is rather difficult in spatial domain to transform domain problem which is suitable for the shape and texture detection.The inclination angle algorithm based on the Fourier transform of fiber is designed. The main direction of texture form is extracted from the angular power spectrum. The sample and impact of technical factors result in fluctuations in measurement results. And the representativeness and stability has not been satisfied.The Hough transform algorithm based on the fiber angle is designed. Even when the fiber is bonded into blocks, the probability of the straight line representing the direction of texture remains the largest.Texture’s shape is expressed with simple data structure whose anti-noise capability and intuitive are better. The correlation coefficient with manual measurements is up to 0.959. The error under control keeps within 5%. It proves potential applications.Thirdly, in a long time, the yarn density (δ) which is seen as the expression of the mean has been in use on the basis of simplifying assumptions and formulas of the classical theory for a long time, which can not correctly reflect the fiber distribution in yarns. This paper is based on the classical model of relation between yarn diameter and yarn linear density. The variationδfollowing the variation of elastic yarn structure is used to lead to the contour changes in the yarn, while the twist angle can reflect the variation in the law of elastic yarn structure. This algorithm uses a supervised way. The twist angle’s correction factor kβtoδis introduced, because yarn density used to be a constant quality concept is difficult to obtain their distribution through weighing due to the random change of quality along the length. So, image-based shape feature of indirect measurement is a new method to achieve yarn density (δT)distribution.更多还原