Authors:
Binbin Chen
;
Chuanrong Li
and
Zengguang Zhou
Affiliation:
Key Laboratory of Quantitative Remote Sensing Information Technology, Chinese Academy of Sciences and Beijing, China
Keyword(s):
Principal component analysis (PCA), multi-scale texture, classify, high resolution, SVM
Abstract:
Land cover classification of high spatial resolution data integrating textural information and spectral features remains limited, and the traditional extraction methods of high spatial resolution image have shortcomings of low accuracy and classification efficiency. In order to explore the practical application methods and effects of high-resolution remote sensing images in vegetation classification, this paper presents a support vector machine classification method for high-resolution image classification, combined using the spectral, principal component, HSV color space and texture features of the study object, which is based on the image data of Wuwei, one city in Gansu Province, China. The threshold values of NDVI are determined to separate vegetation area and non vegetation area. Surface objects in vegetation area mainly include special medicinal herbs, wheat, sorghum, sunflower and fruit tree. The overall classification accuracy is measured as high as 96.01%, and the Kappa coef
ficient is 92.49%. The results of ground truth check show that the method has high precision and good effect, which can be used to distinguish the vegetation of the same species. Meanwhile, the method could be used to extract the vegetation coverage information accurately and quickly, which can provide a reference for high resolution image classification. This method would have an extensive application prospect in crop information extraction from mass satellite data.
(More)