棉花异性纤维中麻绳与羽毛的分类特征

摘要：为准确识别棉花异性纤维中较难识别的羽毛和麻绳异性纤维，采用机器视觉技术，通过图像处理方法采集异性纤维目标，对羽毛和麻绳异性纤维的色彩和纹理特征进行有效的特征提取，形成异性纤维目标的特征向量。再通过一种自底向上的凝聚型层次聚类算法对提取的羽毛和麻绳的色彩与纹理特征进行层次聚类分析，选择最优特征向量。将8个特征向量进行降维分析并比较各维数下的层次聚类效果，试验结果表明，选取红色(R_ave)、绿色(G_ave)、蓝色(B_ave)、能量、熵、惯性矩等6个特征进行层次聚类效果最好，羽毛识别率达到94%，麻绳识别率达到95%， 说明选择的特征向量对这2种异性纤维具有理想的区分性。该研究可为棉花异性纤维的正确识别提供参考。

Classification features of feather and hemp in cotton foreign fibers

Feather and hemp are two kinds of foreign fibers frequently found in cotton, which are difficult to identify using existing image processing methods. A novel image processing method was proposed to classify the two impurities in lint. Three color and five texture features were extracted for these impurities from machine-acquired images of lint samples. An agglomerate hierarchical cluster analysis was conducted, and dimensionality reduction was performed to determine the optimal number of color and texture features. Such agglomerate hierarchical cluster analysis resulted in rates of correct identification of 94% for feather and 95% for hemp. The optimal combination was obtained with six features (color coordinates R, G, B and energy, entropy, and moment of inertia) in the hierarchical cluster analysis. The research can provide a reference for the correct recognition in cotton foreign fibers.