项目名称: 基于图论的高速织机织布过程中细小疵点监测方法研究
项目编号: No.51205294
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 机械工程学科
项目作者: 胡峰
作者单位: 武汉纺织大学
项目金额: 25万元
中文摘要: 高速织机织布过程中监测和辨识萌芽状态细小疵点对织机控制与决策及提高产品质量具有重要意义。细小疵点图像信号具有对比度低、维数高、冗余信息多和随机性大等特点,使现有特征提取与辨识方法的精度和效率下降。本项目拟针对细小疵点给特征提取与辨识方法带来的挑战,提出基于图论的细小疵点特征提取与辨识方法。采用测地线、高斯相似准则和Procrustes分析方法等,发展邻域图建模和嵌入维数选择方法,建立细小疵点特征提取模型,提出基于图嵌入方法的细小疵点特征提取算法,克服细小疵点不利因素对特征提取精度的影响,突破特征灵敏度低的瓶颈。探讨谱聚类方法在疵点辨识中的不足,研究相似图建模方法,建立细小疵点的相似图模型,提出基于改进规范切割方法的疵点辨识算法,解决细小疵点类型繁多和学习样本不完备制约辨识精度提高的问题。为细小疵点监测提供核心理论和关键技术,为下一步织机控制研究提供技术支持,具有重要学术意义和工程应用价值。
中文关键词: 局部线性嵌入算法;图嵌入;细小疵点;特征提取;Gabor小波
英文摘要: The fabric defect monitoring in the production process of high speed loom is the most important to increase the fabric qualities.The crux in whole project is the fabric defect feature extration and identification.The methods of fabric feature extraction will be studied based on the locally linear embedding algorithm,which include the automatical computation of the number of nearest neighbors and the selection of embedding dimension of samples.We shall apply Neighborhood Variance,geodestic distance and Gaussion similarity criteria to automatically computing the number of nearest neighbors and employ Procrustes analysis method and mix scatter matrices in selecting embedding dimension of samples.The methods will be able to improve the sensitivity of fabric defect features and reduce computation time.In order to solve the problem of fabric defect identification, we shall improve the normalized cut algorithm,which will be utilized to enhance identification precision of fabric defect and shorten identification time of fabric defect.The improved normalized cut algorithm consists of two components that are processed sequentially: a greedy agglomerative hierarchical clustering procedure and a local refinement.Based on this condition, we will put forward a technical of fabric defect monitoring in the production process of
英文关键词: LLE;graph embedding;small fabric defect;feature extraction;Gabor wavelet