基于自组织特征映射网络的纹理分类研究

提出了基于自组织特征映射网络（ＳＯＭ）的纹理分类方法。采用了适合纹理分析的纹理谱（ＴＳ）的概念，并在分类过程中引入了纹理谱特征向量，纹理谱向量是ＴＳ经过降维处理得到的．该特征向量反映了空间模式的纹理特征．在学习（训练）与分类识别中，采用了神经元网络模型．与ＴＳ相对应的特征向量重复地送入ＳＯＭ网络的输入端，网络的权向量则逐渐地将样本值聚类到各自的样本中心．计算机模拟实验表明，作者提出的纹理分类方案十分有效而且实用．本方案计算量小，学习周斯短，识别率高．本文最后给出了实验结果及分析．

A Study of Texture Classification Based on Self-Organizing Feature Maps

This paper presents a method of texture classification based on Self --Organizing Feature Maps (SOM). The concept of Texture Spectrum (TS) is adopted. In classification,TS Vector which is obtained from TS through decreasing dimension is introduced. TS vector indicates texture character of space pattern. Neural Network (NN) pattern is used in learning (training) and classification,TS vector which is correponding to TS is sent into the inpUt of SOM repeatly and then the sample values are gradually clustering into the center of every class of samples by SOM weight vector. Computer experiments indicate that method presented in this paper is effective and practical With its less computation,less learning time and high classification ratio' Finally the experimental results and analysis are given.