【作者】 陈慧敏；

【导师】 张渭源； 顾洪波；

【作者基本信息】 东华大学 ， 服装设计与工程， 2007， 博士

【摘要】 服装面料的平整度等级是一个评价服装或面料外观性能的重要指标。目前,对服装面料平整度等级的评估方式普遍采用标样对照法,是对面料定性化的总体主观视觉印象,评级结果受试样、环境、评级者的状态等影响较大,具有不确定性和不唯一性。本文首先概述了服装面料平整度等级客观评级方法的研究现状,列举了计算机图像处理技术和激光扫描技术在面料表面信息的获取和面料平整度等级客观评定中的应用,在比较了三维数据的不同数学表达形式后指出,点模型表达法用密集采样的离散点隐式地表示模型的表面,非常适于具有复杂空间形态的服装面料的三维表达,将有助于再现面料的褶皱形态,为面料平整度等级的客观评定提供依据。因此,为了获取高精度的面料点模型,本文自主研制了一套服装面料三维非接触式坐标测量装置。所建系统既符合人眼的观测习惯,又能最大程度上保持面料原有的褶皱形态,且由于在主动式双目结构光栅投影的基础上,结合了相移法和光学三角测量原理的特点,使用两幅图像相互补偿的方法解决了反光和阴影的问题,减少了相噪声的影响,加快了图像匹配的速度,提高了测量准确性。然后,本文以AATCC-124标准模板为研究对象,在对测量数据进行去噪、配准和截取等预处理操作后,利用统计原理建立了模板点模型各数据点沿Z坐标方向上坐标值的四分位差R_d、平均差R_a、标准差R_q、峰度R_k等四个与平整度等级有关的几何评价指标,从数据的离散趋势及其分布形态方面探讨了点模型所代表的实体对象的整体弯曲性能。接着,本文又利用离散微分几何原理对点模型特征点邻域的MLS重构曲面进行分析,得到模板点模型的特征点密度ρ_c、特征点高度Z、特征点曲率H等三个与平整度等级有关的特征点评价指标,进一步探究了点模型所代表的实体对象的局部屈曲特性。结果表明,各项评价指标均与平整度等级之间存在良好的相关性。但是,模板表面的褶皱形态迥异,仅利用其中一个评价指标、或联合考虑多个几何评价指标、或联合考虑多个特征点评价指标,都不能有效预测点模型所代表的实体对象的平整度等级。粗糙集理论是一种研究不完整数据、不精确知识的表达、学习和归纳等的一种数学工具,是在保持分类能力不变的前提下,通过知识约简,导出问题的决策或分类规则,是一种分类规则挖掘的主流方法之一。基于粗糙集理论的特点,本文首次将粗糙集理论应用到服装面料外观性能的客观评价中,以120个模板点模型的所有评价指标值为输入样本,建立了一个基于粗糙集理论的平整度等级评级模型。基于粗糙集理论的平整度等级评级模型的分类规则具有简单、直观等优点。人工神经网络方法通过大量非线性并行处理器简单地模拟人脑中的神经元之间的突触行为,实现分布式记忆和自学习自组织的功能,从而具备很强的识别与分类能力。基于神经网络的特点,本文以120个模板点模型的所有评价指标值为输入样本,建立了基于BP网络的平整度等级评级模型。基于BP网络的平整度等级评级模型输入样本的平整度等级值与BP网络训练集的输出值之间的相关系数为94.84%。因此,基于BP网络的平整度等级评级模型具有容错能力强、可靠性高等优点。本文比较了基于粗糙集理论和基于BP网络的两个平整度等级评级模型的优缺点后,探讨了粗糙集理论与神经网络相集成的智能计算方法,并首次将粗糙集和BP网络相结合的技术应用到服装面料外观性能的客观评价中,建立了基于粗糙集－BP网络的平整度等级评级模型。基于粗糙集－BP网络的平整度等级评级模型首先基于粗糙集原理,对于样本的输入信息,通过挖掘数据间的关系,去掉冗余信息,并将已经降低了空间维数的样本信息作为神经网络的输入内容,使用神经网络作为后置的信息识别系统,从而简化了神经网络的结构,缩短了训练时间,避免了神经网络不能区分知识的重要性与冗余程度的缺陷。后续的基于神经网络的数据处理方法又很好地抑制了噪声的干扰,实现了信息的大规模并行处理,并最终从数据中产生平整度等级分类信息。结果表明,基于粗糙集－BP网络的平整度等级评级模型输入样本的平整度等级值与网络训练集的输出值之间的相关系数为97.19%。与基于粗糙集理论和基于BP神经网络的两个平整度等级评级模型相比,基于粗糙集－BP网络的平整度等级评级模型具有容错和抗干扰能力更强、网络训练时间更短、预测精度更高等优点。为了验证基于点模型的粗糙集－BP网络平整度等级评级模型的可靠性,本文选用30个颜色不同、纱线支数和组织结构与密度均不相同的常见纯棉服装面料进行平整度等级主客观评价,试样平整度等级主客观评级值的相关系数为93.45%。随后,本文从数据采集设备、试样的颜色和花型、试样的厚度、试样的组织结构、试样表面的褶皱形态和褶皱分布、评价指标的选择和评级模型等九个方面对平整度等级主客观评级值的影响展开讨论,最后得出结论:本文所设计的数据采集装置能准确获取试样的三维坐标数据,本文所建立的平整等级客观评价指标能区分试样表面的褶皱形态,本文所建立的评级模型可靠性强、预测精度高。最后,本文以前几章的研究工作为基础,利用虚拟仪器技术,建立了一个基于LabVIEW的服装面料平整度等级虚拟测评系统,在虚拟平台上实现了服装面料三维坐标数据的在线测量和平整度等级的客观评价。该测评系统具有操作简单、数据直观、运行速度快、界面可视化程度高等优点。更多还原

【Abstract】 For garment or fabric appearance, the cloth smoothness grade is one of the most important performance factors.Traditionally, the grade process is manually performed. After repeating home laundering, the fabric sample and appropriate reference standards are put side by side under a standard lighting and viewing area. The judges rate the fabric appearance by comparing the sample with the references. The results are usually affected by the rating surroundings, and the statuses of judges are highly subjective and non-repeatable.Development of vision systems that can be used to evaluate fabric smoothness has also preoccupied many researchers. In this thesis, practicably image processing techniques and laser scanning methods to perform smoothness grading that had been researched and developed in the past decade are described first. Compared with different kinds of mathematical expressions, point-sampled model which contains a large number of three-dimensional coordinate points and indirectly represents model surface is quite suitable for the expression of the irregular fabric surface. The point-sampled model is convenient to re-form the fabric wrinkle modality and can provide reliable items for fabric smoothness objective evaluation. A 3D non-contact measurement system is constructed to obtain these coordinate points. This system utilizes two charge coupled devices (CCD) and a grating projecting unit to sense the 3D topography of the fabric surface. The technique bases of the system are structured lighting, trigonometry and phase-shifting. Two images captured by different CCD compensate each other and reduce the influence of noises. The design of the system assures the fabric original and natural state and is insensitive to fabric colors and patterns.Subsequently, taken the AATCC-124 replicas’ point-sampled models as study objects, four statistical parameters based on the Z ordinates of the scatter points were established to characterize smoothness appearance. They were inter-quartile range R_d, arithmetic average deviation R_a, root mean square deviation R_q and Kurtosis value R_k. The discussions on the discrete tendency and distributing modality of scatter points well revealed the entity bending performance which was in existence as point-sampled model. Afterward, the principle of discrete differential geometry was applied to the replicas point-sampled models. Each vertex and its neighborhood were grabbled. MLS surface at vertex was constructed. These calculations made the following three geometry parameters decided: vertices densityρ_c, vertices height Z and vertices mean curvature H. The discussions on the significantvertices and their reconstructed surfaces well explained the surface local bending performance. All of the statistical and geometry characterizations are closely correlative to smoothness grades. Actually, no wrinkle modality is similar to each other even though they are on the same standard smoothness replica. So, no single characterization can be used alone to forecast the entity’s smoothness grade.Rough Set (RS) is one of the soft-computing methods dealing with indefiniteness and incompleteness. It can find the relationship between the data, pick up the useful characters and reduce the information process. In recent years, it has been successfully applied in data mining, knowledge acquisition, algorithm research, decision support systems and pattern recognition. In this thesis, for the first time, RS method was employed in the objective evaluation of fabric smoothness grade. The objective smoothness grading model based on RS theory took all the seven characterizations of 120 replicas’ point-sampled models as the inputs. The grading model was expressed as simple and intuitional classification rules.Artificial neural network (ANN) is also one of the popular computing methods dealing with indefiniteness and incompleteness. It has the essential nonlinear character, parallel processing ability and the ability of self organization and self-learning. The back-propagation (BP) algorithm was implemented in this thesis to train a feed-forward ANN which also took all the seven smoothness characterizations of 120 replicas’ point-sampled models as the inputs. The correlative coefficient between input smoothness grades and training grades was 94.84%.RS and ANN are both played important roles in intelligent computing methods. They are complementary, so the integration of RS and ANN is feasible. In this thesis, for the first time, an approach of data mining integrated RS and ANN was presented. It fully developed two methods’ advantages. RS efficiently processed the reduction of the seven smoothness characterizations, simplified the network’s structure, reduced the network’s training epochs and improved the judgment accuracy. The correlative coefficient between input smoothness grades and training grades based on RS-BP ANN was high up to 97.19%. Compared with the smoothness grading models which were based on RS or ANN alone, the RS-BP ANN smoothness grading model had the highest tolerance fault, disturbance resistibility, forecast precession and shortest training time.Simulation experiments were executed to verify the validity of RS-BP ANN smoothness grading model based on point-sampled model. 30 pieces of 100% cotton garment cloths with different color, printed pattern or structure were chosen. The experimental grades provided by the RS-BP ANN were highly consistent with the subjective results, and the correlative coefficient between objective and subjective evaluation results was 93.45%. Discussions were made about the influences to the objective grading results which include data acquirement, swatches color, pattern, thickness and structure. We came to the conclusion that the RS-BP ANN grading results were more believable and closer to real grades especially in the case of fabrics with color or pattern.Finally, based on virtual instrument technique, a fabric smoothness grade system was established. The system performs the functions of data acquirement, transmission, analyzing and fabric smoothness grade assessing. It makes full use of the advantages of LabVIEW and has a friendly interface. It was testified that the virtual system built in this thesis had good performance in running and smoothness grade assessing.更多还原