【作者】 刘成霞；

【导师】 傅雅琴；

【作者基本信息】 浙江理工大学 ， 纺织工程， 2015， 博士

【摘要】 织物在穿着和洗涤护理的过程中,由于揉搓、挤压、拧绞、扭曲等外力而产生折痕或皱纹的现象称为起皱,抵抗起皱变形的能力称为织物的抗皱性。抗皱性不仅是织物的基本服用性能,也是影响服装外观的重要性能。因此对织物抗皱性进行客观、准确地测试和评价就显得异常重要。然而现有的织物抗皱性测试方法存在诸多问题,测试过程中织物所处的状态、受力方向、发生的变形等与实际着装时的情况差异甚大,致使现有的测试方法不能真实评价织物做成服装后实际穿着过程中的抗皱能力。针对这一现状,本文提出一种模拟实际着装的织物抗皱性测试新方法,构建了模拟装置,并利用图像处理技术对模拟装置产生的折皱进行了分析,提取了折皱密度、灰度共生矩阵以及小波特征参数,进而基于信息融合技术,将这些指标融合成一个综合折皱指标,最后利用这些特征对织物折皱等级进行了聚类研究,并利用神经网络技术对折皱等级进行了分类识别,研究内容及研究结果如下:1)构建的织物折皱模拟装置产生的折皱,无论在外观形态上,还是在专家主观评价结果上(包括起皱程度的排序、评分以及折皱等级)都与实际着装产生的折皱具有良好的一致性;且模拟装置的测试稳定性比现有的折皱回复角法更好。2)针对折皱回复角测试时只考虑织物经纬向所带来的片面性,提出应该增加45°折皱回复角的测试,以提高检测结果与实际着装的吻合度,并建立了折皱密度与不同方向折皱回复角的模型,根据此模型可由折皱回复角预测实际着装时服装的起皱程度,无需经过工作量巨大的服装制作及实际着装实验。3)利用图像处理技术,对模拟装置产生的折皱进行了特征提取与纹理分析,并将这些纹理特征与折皱回复角和专家主观评分结果进行了对比,研究表明:织物折皱灰度图像经Sobel边缘检测提取的折皱密度与专家视觉评分之间存在较好的一致性,且斜向折皱回复性与折皱密度之间相关性显著;越平整的织物,灰度共生矩阵的能量和相关性越大,熵和惯性矩越小,与织物折皱程度相关性由大到小依次是:熵、能量、相关性和惯性矩;折皱越严重的织物经小波分解后的三个方向(水平、垂直、斜向)的细节系数标准差越大,且水平方向的细节系数标准差明显大于垂直和斜向标准差。折皱密度、灰度共生矩阵和小波分析都表明:织物经向折皱回复角对实际着装时抗皱能力的贡献率大于纬向。4)利用信息融合技术,将折皱密度(WD),灰度共生矩阵中4个方向的熵值之和(Entropy),小波分解1层时的水平细节系数标准差(SH1)这三项指标进行数据融合,得到了织物综合折皱指标(CWI),研究结果表明这一综合指标比三项单一指标能更客观、更全面地评价织物折皱。5)分别以WD,Entropy,SH1三项指标和以CWI这一综合指标为特征,对织物折皱等级进行了三种方法的聚类分析,研究表明:从指标选取来看,进行k-均值聚类以及自组织特征映射(SOM)网络聚类时,用CWI为特征的聚类结果好于以三项单独指标为特征,说明CWI比这三项单一指标对织物折皱的表征更符合视觉观察的结果;从聚类方法来看,SOM神经网络的聚类结果与主观评价结果的一致性好于系统聚类和k-均值聚类。6)以WD,Entropy,SH1以及CWI为特征,分别利用LVQ和PNN神经网络对织物折皱等级进行了分类识别,结果表明:特征向量中增加CWI可以使LVQ和PNN神经网络的预测准确率有不同程度的提高;且将LVQ与PNN串联后组成的LVQ-PNN混合神经网络比单一的LVQ和PNN神经网络的识别率有较显著的提高。本文提出的模拟实际着装的织物抗皱性测试方法在一定程度上弥补了现有测试方法的不足,可有效提高测试结果与实际着装时折皱情况的吻合度,同时也为纺织品检验领域提供了新手段,还可以根据测试结果指导面料设计,减少服装加工中面料的错误使用带来的浪费;同时,本文基于图像处理和模式识别技术,实现织物抗皱能力的客观评定和自动识别,符合当前信息化发展的趋势和方向,有助于促进计算机在线检测技术的发展和实现。更多还原

【Abstract】 Wrinkling is caused by twisting, shearing, compressing and bending, etc in the process of washing, drying, or wearing, which is difficult to recover even when external force is removed. The ability for the fabric to resist wrinkling is called wrinkle resistance. The behavior of wrinkle resistance is not only the basic performances of fabric, but also is one of the most important characteristics in determining the visual aesthetic of clothes. Therefore, it is of vital importance to precisely and objectively measure and evaluate the wrinkle resistance of fabrics. However, the commonly used wrinkle resistance measuring methods have some drawbacks in which the deformation and force direction of fabric are very different from that in actual wear, resulting in that the testing methods can’t be used to measure the ability of fabric to resist wrinkle during actual wear.Aiming at this, a novel fabric wrinkling resistance measuring method simulating actual wear is put forward and a new wrinkle producing instrument is set up in this paper. Furthermore, image process technology is used to analyze the wrinkles produced by the simulating method and wrinkle density, gray level co-occurrence and wavelet analysis parameters are extracted. Then based on information fusion technology, these parameters are combined into a comprehensive wrinkling parameter. Finally, these specifications are used to cluster fabric wrinkling grade with different clustering analysis method and neural networks. The following results have been drawn.1) Fabric wrinkle resistance measuring instrument simulating actual wear is set up, which can produce wrinkle very similar with wrinkles during actual wear in appearance. And there is good agreement between experts’ subjective evaluation results of wrinkles produced by the new method and that in actual wear, including rank, score and grade of wrinkling degree. Besides, variation coefficient of wrinkle density in the new method is smaller in value than that of wrinkle recovery angle. Therefore, the measurement stability of the new method is better than most commonly used the wrinkle recovery angle method.2) In the method of wrinkle recovery angle, only wrinkle resistance in the warp and weft direction of fabric is considered which is unilateral. Aiming at this, it is put forward that wrinkle recovery angle at 45° should be included to improve the consistency of the testing result with actual wear. Moreover, equation between wrinkle recovery angle at different direction with wrinkle density which can be used to predict wrinkling degree during actual wear according to wrinkle recovery angle, avoiding the time and energy consuming fashion making and actual wear experiment.3) Feature and texture of fabric wrinkled image are extracted and analyzed with image processing technology. Results show that there is good agreement between wrinkle density extracted(WD) by Sobel operator detected from gray level image and score of the experts’ subjective evaluation. Wrinkle recovery angle in bias direction also plays import part in fabric wrinkle resistance. Fabric with smoother surface tends to have larger energy and correlation, but smaller entropy and contrast of the gray level co-occurrence matrix(GLCM). The correlation between GLCM parameters and wrinkling degree from large to small is entropy, energy, correlation and contrast in turn. The more severely wrinkled fabric tends to have larger detailed coefficient standard deviation in three horizontal, vertical and diagonal directions through wavelet analysis. The detailed coefficient standard deviation in horizontal direction is larger than that of vertical and diagonal directions in value. Analysis of wrinkle density, GLCM and wavelet transform show that contribution of wrinkle recovery angle in warp direction with in actual wear is larger than that of weft direction.4) Fabric comprehensive wrinkling index CWI is obtained by data fusing from WD,Entropy, SH1 using Information fusion technology. The test of rank correlation coefficient shows that there is good agreement between the rank according to CWI value of 48 fabrics and that of experts’ subjective evaluation. CWI can characterize fabric wrinkle more objectively and comprehensively than the three specifications.5) Three clustering methods are employed to analyze wrinkled image with specifications of WD, Entropy, SH1 and CWI, respectively. When using k-means cluster method and Self-Organizing Feature Map(SOM) neural network method, the result of choosing CWI as specification is better than that of WD,Entropy, SH1, which proves that CWI can characterize what eyes observe on fabric wrinkle. With respect to clustering method, the clustering result of SOM agrees with subjective evaluation results better than that of system and k-means method.6) LVQ and PNN neural network are employed as supervised learning method to analyze fabric wrinkle, with specifications of WD,Entropy, SH1 and CWI, respectively. It proves that the prediction accuracy can be improved when CWI is added into the specifications and the prediction accuracy of VQ-PNN combined neural network is improved significant with the single LVQ and PNN neural network.The measurement for fabric wrinkle simulating actual wear put forward in this paper makes up for the defect in the measurements now commonly used and the agreement between testing results and wrinkling in actual wear can be increased significantly. Besides, a new approach has been provided for the textile measuring field, the testing result of which can be used to instruct fabric design and help reduce the wasted caused by misuse of fabrics. On the other hand, the objective evaluation and automatic recognition of fabric wrinkling is based on image processing and pattern recognition technology, which meets the trend and direction towards information and can promote the development of on-line inspection technology of computer.更多还原