节点文献
基于神经计算的服装缝纫性能模糊评价研究
Fuzzy Evaluation Research of Dress Seam Pucker Based on Neural Computation
【作者】 潘永惠;
【导师】 王士同;
【作者基本信息】 江南大学 , 轻工信息技术与工程, 2008, 博士
【摘要】 服装缝纫平整度是影响服装外观质量的重要因素,成衣加工过程中的缝纫平整度控制及评价是一个困扰了服装行业多年的问题。目前在大多数商业行为中仍然以主观评定法AATCC-88B(American Association of Textile Chemists and Colorists-Method 88B)作为服装缝纫平整性能的主要评价标准,在参考样品的基础上目视评级,即1、2级是不可接受的,3级是临界级,4、5级是可接受的。虽然这种方法具有简单,易操作等特点,但由于受主观个人因素的影响较大。因此,客观评价服装缝纫平整度的研究一直在持续,而设计有效的客观评价算法是当前服装行业急需解决的热点和难点问题。本文针对这些问题,在对面料的FAST力学性能数据进行相关分析和核主成分分析的基础上,提出基于监督模糊聚类的径向基神经网络(SFCM-RBFNN)来客观评价服装缝纫性能,取得了较好的预测效果。首先用面料性能测试仪FAST仪器对面料的各项力性能指标进行了测试,然后对面料各项力学性能指标与缝纫性能进行相关性分析,分别分析了面料结构力学性能、面料拉伸伸长性能、面料弯曲性能、面料成形性能和面料尺寸稳定性能与面料经、纬向缝纫平整度等级之间的关系,分别作图考察面料缝纫性能随FAST各项力学指标的变化分布情况,并剔除部分与缝纫平整度相关性低的指标。同时从面料斜向力学性能理论和实验研究出发,研究了面料各向异性与缝纫外观平整性能的关系,分析了面料的拉伸性能、弯曲性能及成型性随面料斜裁角度变化的改变情况,及它们对缝纫外观平整性能产生影响。其次针对面料的高维非线性FAST力学性能数据,运用核主成分分析法(KPCA)将这些FAST指标数据(输入空间)影射到一个新的线性特征空间去,然后在这个空间用线性分类方法进行有效的分析。用高斯核函数对面料FAST力学性能数据做了核主成分分析试验,并提取了相应的主成分,达到了非线性数据降维的目的。然后在模糊C均值(FCM)算法基础上进一步提高模糊聚类的精度,提出了SFCM算法,即有监督的模糊C均值算法,将输入空间样本x(面料FAST力学性能数据主成分)和输出空间的期望值y(缝纫平整度)同时引入到模糊聚类中来,并通过相应的局部线性回归模型来表征相应的模糊聚类的每一个类别,用这些局部模型的线性组合求和来计算输出y的实际值,从而达到该算法的实现。在前面KPCA分析的基础上,对面料FAST力学性能数据进行SFCM模糊聚类,聚类效果较好。该算法除了反映输入空间的聚类特性外,还反映了输出空间的逼近特性。最后对在径向基神经网络(RBFNN)基础上对它的隐层节点基函数进行了调整,将SFCM聚类算法中的模糊分割矩阵U和聚类中心v引入基函数,从而使RBF神经网络隐层节点的宽度(感受野)和中心得到了有效的优化和控制,提高了系统的泛化能力。实验证明,本文提出的基于监督模糊聚类的径向基神经网络(SFCM-RBFNN)算法对不同结构和材料面料的缝纫平整度预测具有较好的精度和鲁棒性。
【Abstract】 Fabric sewing ability is one of important factors that influence the quality of clothing appearance. Controlling and evaluation of seam pucker is a lasting problem of fashion industry during the course of clothing manufacture.AATCC (American Association of Textile Chemists and Colorists) Method 88B has been commonly used for the subjective evaluation of seam pucker. According to this method, the appearance of seams are compared with photographic standards and the severity of seam pucker is graded into five classes, Class 5 being little or no pucker, and Class 1 severe pucker. Normally, Class 5 and 4 are acceptable, Class 3 is critical or borderline, and Class 2 and 1 are unacceptable. The merit of this method is directness, simplicity, low investment and easy to master, but it is influenced by uncertain factors from the evaluator and loses its objectivity easily. So, the research for objective evaluation of seam pucker is on going, and designing an effective evaluation algorithm continues to be hotspot and difficult problem of the fashion design and manufacture industry.In this paper, a supervised fuzzy clustering RBF neural network (SFCM-RBFNN) based on kernel principal component analysis is introduced for constructing the garment seam evaluation system. Experimental results demonstrate that the proposed system could efficiently evaluate the fabric sewing ability.Firstly, this paper use FAST(Fabric Assurance by Simple Testing) apparatuses to test the mechanic indexes of fabric, and analyze the correlation between the indexes and fabric sewing ability of warp action and weft action, such as fabric structure mechanical properties, fabric extension mechanical properties, fabric formality ability, dimensional stability. Then figures are plotted to observe the trend of seam pucker grade with these FAST mechanical properties data. According to the correlation analysis, a few low related indexes with seam pucker are eliminated.With the fabric inclined mechanical properties theory, research was done on the relations between fabric anisotropy and sewing ability, such as extension properties, bendability properties, formality ability. This paper also plots figures to observe the trend of seam pucker grade as the fabric diagonal cutting angle.Secondly, according to these high dimension and nonlinear FAST mechanical properties data (in input space), The kernel principal component analysis (KPCA) method was applied to project these data in a whole new linear feature space, then linear clustering method was used to analyze it effectively in the new space. This paper extracts the principal component of fabric FAST mechanical properties by KPCA method with gaussion function and reduces the dimension of the data.Thirdly, to improve the accuracy of fuzzy clustering based on fuzzy C mean (FCM) algorithm, this paper proposes a novel supervised fuzzy C mean clustering algorithm including input space sample x (principal component of fabric FAST mechanical properties) and output space expectation y (seam pucker grade) at the same time. Using local linear regression model to represent each fuzzy cluster, the actual output y can be calculated by sum of these local linear model. And the clustering effect of the fabric FAST mechanical properties PCA data by SFCM algorithm is quite good. The SFCM algorithm proposed by this paper not only shows the clustering action in input space, but also shows the approximating action in output space.This paper modifies the radius base function of hidden nodes of RBF neural network by importing the fuzzy partition matrix U and clustering center v of SFCM algorithm to construct a SFCM-RBF neural network. The modification makes the width (receipt field) and center of hidden nodes of RBF neural network become more effectively optimizing and controlling. Experimental results demonstrate that the proposed system could efficiently be used as an objective garment seam evaluation system with high accuracy and is robust for various structures fabric.