Texture classification of vegetation cover in high altitude wetlands zone

35th International Symposium on Remote Sensing of Environment

地球科学;生态环境科学

Hua, Liu^2 ; Wentao, Zou^1,2 ; Bingfang, Wu^1 ; Hongbo, Ju^2

Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, China^1

Research Institute of Forest Resource Information Techniques, CAF Beijing, China^2

关键词: Classification accuracy;    Classification procedure;    Decision tree classifiers;    Gray level co-occurrence matrix;    Remote sensing images;    Texture classification;    Texture measures;    Vegetation cover;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/17/1/012083/pdf
DOI  :  10.1088/1755-1315/17/1/012083

学科分类：环境科学（综合）

来源: IOP

【 摘 要 】

The aim of this study was to investigate the utility of datasets composed of texture measures and other features for the classification of vegetation cover, specifically wetlands. QUEST decision tree classifier was applied to a SPOT-5 image sub-scene covering the typical wetlands area in Three River Sources region in Qinghai province, China. The dataset used for the classification comprised of: (1) spectral data and the components of principal component analysis; (2) texture measures derived from pixel basis; (3) DEM and other ancillary data covering the research area. Image textures is an important characteristic of remote sensing images; it can represent spatial variations with spectral brightness in digital numbers. When the spectral information is not enough to separate the different land covers, the texture information can be used to increase the classification accuracy. The texture measures used in this study were calculated from GLCM (Gray level Co-occurrence Matrix); eight frequently used measures were chosen to conduct the classification procedure. The results showed that variance, mean and entropy calculated by GLCM with a 9*9 size window were effective in distinguishing different vegetation types in wetlands zone. The overall accuracy of this method was 84.19% and the Kappa coefficient was 0.8261. The result indicated that the introduction of texture measures can improve the overall accuracy by 12.05% and the overall kappa coefficient by 0.1407 compared with the result using spectral and ancillary data.