【作者】 柯维；

【导师】 张长胜；

【作者基本信息】 苏州大学 ， 纺织工程， 2011， 硕士

【摘要】 目前织物组织结构的分析还是借助于放大镜、分析针等简单工具,完全依赖人工目测完成。由于人工目测不可避免会受到个人的视力、情绪、疲劳、光线等因素的影响,带有很大的个人主观性,往往难以保证识别质量,造成测试结果具有一定的主观性和不可靠性,而且耗费操作人员大量的时间重复做相同的工作。加上效率低下已无法适应目前纺织生产小批量、多品种、高效率的需要。这成为了提高纺织企业信息化、生产自动化程度的一个瓶颈。如果能设计出一套织物组织结构参数的自动识别系统,将会大大推动纺织生产效率,实现全面自动化。图像处理技术和人工智能近年来取得了很大的发展,这使得利用图像处理技术实现织物组织结构的自动分析成为可能。本文运用灰度共生矩阵对织物图像进行特征提取,然后设计了一个三层BP神经网络对提取出来的特征值进行识别分类,经试验验证对平纹、斜纹、缎纹三种组织织物图像的正确识别率可达93.45%。在织物分类成功的基础上,运用小波分解理论,提取织物图像经纬向亮度信息,对织物组织点进行定位,利用组织点图像表面纹理信息,提取相关性特征,对其经纬属性进行判断,从而分析出织物具体组织结构。更多还原

【Abstract】 At present, the work to analyze fabric structure still depends on artificial visual measurement, which is easily influenced by personal sight, mood, mental state as well as optical line. Therefore this method of testing is of great individual subjectivity and unreliability. The recognition quality is difficult to guarantee as a result of that. Meanwhile, it takes much more time for operators to do the same testing work repeatedly with poor efficiency, which cannot meet the needs of current textile production for small quantities, multi-item and high efficiency. With so many drawbacks, artificial visual measurement has become a great bottleneck of textile industry to improve information technology and production automation. An automatic identification system on fabric structural parameters will realize automation fully and enhance textiles’production efficiency powerfully. With the development of image processing technology and artificial intelligence, automatic analysis on fabric structure as a replacement of manual labor is of great possibility.In this thesis, features of fabric-image have been extracted by GLCM (Gray Level Co-occurrence Matrix). These features were employed to a three layer BP neural network for analyzing. It was verified by experiments on three kinds of fabric structures, such as plain weave, cross grain and satin, that all the structures can be correctly identified up to the rate of 93.45%. Furthermore, wavelet decomposition theory was applied to extract fabric image brightness information to locate point area. Related features could be drawn from the surface texture information of interlacing images, hence the properties of warp and weft can be judged to analyze the specific textual structure of fabric.更多还原