棉、莫代尔纤维的图像识别

【作者】 林涛；

【导师】 杨庆斌；

【作者基本信息】 青岛大学 ， 纺织材料与纺织品设计， 2010， 硕士

【摘要】 中国作为纺织品出口大国,纺织品出口需要经过严格检验,本论文就是利用图像处理方法测定棉/Modal混纺纱混纺比。在前人研究的基础上,针对纤维纵向制取切片,对切片进行拍照,对图像进行二值化处理,提取纤维特征值,在MATLAB神经网络中判断纤维的种类,并设计系统界面使操作简单化。纤维纵向切片制取的好坏对以后的处理影响很大,采用哈氏切片器切片,通过蒸馏法来获取切片,研究统计制取过程中纤维存在的状态。采用东华大学的棉/苎麻自动拍照系统的部分功能,人工手控来获取图片,分析环境因素对照片的影响。按照图像处理的流程,以数学形态学为基础,设计了每一步的原理和MATLAB程序,处理后通过图片效果进行分析,实现了纤维水平状态的图像二值化,为纤维特征提取打下了基础。对纤维提取特征,取棉、Modal纤维差异大的直径CV值,矩形度,平均灰度值做为区分棉与Modal纤维的特征,用origin对提取的特征值数据进行分析,确定直径CV值和平均灰度值两者相结合来区分两种纤维。设计了感知器和BP两种神经网络,用样本数据进行训练,优化神经网络的参数,确定较佳的结构参数,用检测样本进行检测,最后采用BP网络来判断棉、Modal纤维。用VB设计系统界面,把复杂的算法放到后台,使操作人员只面对简单的界而。解决了VB与MATLAB之间的接口问题,使两种程序能互相访问。更多还原

【Abstract】 China is a major exporter of textiles; textile must be tested strictly when exported. This paper is aim to measure cotton/Modal yarn blend ratio by image processing according the order of fiber longitudinal section processing. It covers preparation of the fiber longitudinal section, photographing, fiber binarization, feature extraction, MATLAB neural network Modal determination and designing the system interface. Preparation of fiber longitudinal slice is a key process in image processing. First, slice is made through Y172 equipment, then distilled. The paper analyses state of the existing fiberPhoto is acquired with cotton/ramie features automatic camera system of Donghua University. The part function of system is applied, we get the fiber picture by manual, analyze the impact of environmental factors.In accordance with the process of image processing, the paper designs process step and MATLAB program based on mathematical morphology, analyses effects of every step process and achieve fiber horizontal state and make a preparation for feature extraction.Diameter CV, rectangular, the average gray value is extracted as difference feature of cotton/modal. The characteristic values of the data are analyzed with the origin to prove the distinction. In the end, the paper determining to combing diameter CV and the average gray value to distinguish fiber.The paper compares perception and BP neural network, trains network using the sample data, optimizes neural network parameters, makes sure the type of network. Finally BP network is proved to better suit to the experimentationSystem interface is designed by combing VB and MATLAB, solving the problem of different program interface, making operators face a simple interface.更多还原