基于SOM网络的织物起球客观评价

Objective Evaluation of Fabric Pilling Based on SOM Network

提出一种基于自组织映射算法的神经网络,用于织物起球的等级评定。起球织物都包含有重要的纹理信息,首先创建起球图像的灰度共生矩阵,从这些矩阵中提取特征向量,再以这些特征值作为网络的输入信息,建立SOM神经网络对图像的特征值进行训练、分类,也就是将不同等级的起球图像进行分类。本文详细介绍SOM网络的基本原理与学习算法,以及共生矩阵的计算,最后提取7种起球特征参数进行实验,结果表明该方法有效可行。

This paper proposes a new neural network based on self-organizing map algorithm method for the fabric pilling grading. Pilling fabric contain important texture information. Firstly, this paper creates the gray-level co-occurrence matrix of pilling image, then extracts its feature vector. Secondly, with the input features, the established SOM neural network is used for training and classification. That is classifying the pilling image as different levels. The paper details the basic principles of SOM networks and learning algorithms, as well as the calculated method of co-occurrence matrix. Finally, the seven kinds of fabric pilling feature parameters are imported to experiment. The results show that the method is efficient and feasible.