基于遗传算法的遥感图像纹理特征选择  

Texture feature selection in remote sensing image based on genetic algorithms

作　　者：冯莉[1] 李满春[1] 李飞雪[1] 

机构地区：[1]南京大学地理信息科学系,南京210093

出　　处：《南京大学学报（自然科学版）》2008年第3期310-319,共10页Journal of Nanjing University（Natural Science）

摘　　要：本文基于Landsat 7 ETM+全色图像,以浙江省临安市大峡谷镇为研究区,进行基于遗传算法的遥感图像纹理特征选择的研究.首先利用灰度共生矩阵法和灰度-梯度共生矩阵法对研究区遥感图像进行纹理特征提取,共得到23个纹理特征,然后利用遗传算法对这23个纹理特征进行纹理特征选择,最后得到一组最优纹理特征集.实验结果表明,遗传算法因其自适应性、并行性、能较好地处理大规模复杂数据,且特别适合于解决多目标优化问题等诸多特性,所以是解决特征选择问题的理想方案.In this paper, an approach to texture feature selection in ETM4-panchromatic image based on genetic algorithms is described. Daxiagu town of Linan county, Zhejiang province is taken as the research area. Firstly, the texture features were extracted from the ETM^pan image by using gray co-occurrence matrix and gray-gradient co occurrence matrix, and all the 23 texture features were obtained. Then texture feature selection was carried out by genetic algorithms. The optimal texture sub-set could be got finally. The results show that the method of genetic algorithms can deal with the large and complicated data because of its characteristics of self-suitability and parallel, and it is very suitable for solving the problem of multi-object optimization, so the method of genetic algorithms is a better way to solve the problem of feature selection.

关 键 词：纹理特征提取 遗传算法 纹理特征选择 LANDSAT7 ETM+ 全色图像 

分 类 号：P208[天文地球—地图制图学与地理信息工程]