基于混合纹理的计算机自动分类方法  

Computer Automatic Classification Method Based on Mixed Texture

作　　者：崔红霞[1] 黄科涵 

机构地区：[1]渤海大学信息科学与技术学院,辽宁锦州121000

出　　处：《计算机技术与发展》2018年第2期158-162,共5页Computer Technology and Development

基　　金：国家自然科学基金(41371425)

摘　　要：针对高分辨率光学图像光谱信息单一,纹理信息丰富的特点,设计了一种基于粗、细纹理两种特征相结合的计算机自动分类方法。通过提出一种基于Tamura的局部纹理特征和灰度共生矩阵的细小纹理特征混合的7维特征向量,实现图像基于k-means聚类的7维特征空间的计算机自动分类。针对耕地、森林、裸露地、水域四类典型地物,通过对1 600张样本影像(每类400张)的分类探测,自动确定Tamura特征和灰度共生矩阵特征移动窗口的最佳尺寸。模拟地物合成影像自动分类和低空高分辨率光学影像的典型地物自动分类的实验结果表明,该方法的自动分类精度优于单种纹理特征的分类精度,采用混合纹理对遥感图像进行地物分类是计算机自动分类的研究方向之一。According to the characteristics of high resolution optical image with single spectral information and rich texture information,wedesign a computer automatic classification method based on the combination of two features of coarse and fine texture. By presenting a new7 dimensional feature vector based on Tamura texture feature and gray level co-occurrence matrix,automatic classification of 7 dimensionalfeature space based on k-means clustering is achieved. According to four kinds of typical objects like the cultivated land,forest and bareland,waters,through the image of 1600 samples （each 400） classification detection,the optimal size of Tamura features and gray level co-occurrence matrix features moving window is determined automatically. The experiments on automatic classification of simulation object syn鄄thetic image and that of typical objects for the low optical image with high resolution show that the classification accuracy of the proposedmethod is better than that of single texture features. Texture features using the mixture of remote sensing image classification is one of the re鄄search directions of computer automatic classification.

关 键 词：计算机分类 纹理特征 灰度共生矩阵 聚类 非监督分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]