摘要
高分一号卫星是我国2013年自主发射的第一颗高分辨率对地观测卫星(GF-1),其在湿地中的应用还比较少见。以洞庭湖湿地为研究区,GF-1影像为主要遥感数据源,在对研究区湿地主要植被柳树、杨树、芦苇和苔草的光谱特征分析基础上,建立决策树分类算法。同时结合GF-1影像特有的纹理信息,引入纹理均值和相异性指数对决策树算法进行改进,结果表明:通过采用纹理均值和相异性指数,总体精度从传统决策树的85.64%提高到了92.66%,Kappa系数从0.82提高到0.91,说明该方法对湿地植被识别的效果较好。这对于同等空间分辨率遥感数据的植被分类具有指导和借鉴作用。
GF-1 satellite was launched independently in 2013 in China and it is the first high resolution earth observation satellite, and its application in the wetland is relatively rare. This paper used Dongting Lake wetland as the research area based on GF-1 image. We adopted classification algorithm of decision tree, built classification regulation on the basis of vegetation spectral characteristic and classfied the vegetation into four categories including poplars, willow, reeds, and moss grass. Then the method was improved by combining the proper textural information of GF-1 image with using texture mean and dissimilarity index. The result showed that the overall precision of the vegetation classification method based on spectral characteristic was 85.64%, but that of the improved method was upgraded to 92.66%, and kappa coefficient was 0.82, but that of the improved method was upgraded to 0.91 by using the texture mean and dissimilarity index. The result showed the vegetation classification method can classify and discriminate vegetation effectively. As to vegetation classification with similar high-resolution data source, the method will provide scientific guide and reference value.
出处
《中南林业科技大学学报》
CAS
CSCD
北大核心
2015年第11期27-31,共5页
Journal of Central South University of Forestry & Technology
基金
国家重大专项(N21-Y30B05-9001-13/15-2)
国家自然科学基金项目(31370639)
湖南省高校产业化培育项目(13CY011)
关键词
高分辨率遥感
决策树
纹理
植被分类
洞庭湖
high-resolution remote sensing
decision tree
texture
vegetation classiifcation
Dongting lake
