遥感组合指数与不同分类技术结合提取农业用地方法

A methodology of agricultural land extraction using remote sensing combined index and different classification technologies

多光谱遥感影像因具有丰富的波谱信息,提高了地表覆盖的辨识能力,利用遥感数据高精度自动提取专题信息是目前研究的热点和难点。本文以北京市ASTER影像为例,通过对城市生态环境中土地类型及其光谱特征规律分析,组合归一化差异植被指数、修正归一化差异水体指数和归一化差异建筑指数三种指数,制作组合指数新影像。对组合指数影像采用基于支持向量机的面向对象分类方法进行农业用地信息提取,同时将该方法分别与基于原始影像、组合指数影像的最大似然及支持向量机的分类方法进行对比分析。实验结果表明:组合归一化差异指数影像压缩了数据维数,降低了覆盖地物相关性,易于农业用地信息提取。对组合指数影像采用基于支持向量机的面向对象分类方法精度达95.701%。

Because of multi-spectral remote sensing image has a wealth of spectral information, the ability of land cover identified is enhanced. Automatically extracting high-precise thematic information by remote sensing data is the current hotspot and difficuhy of research. In the paper, taking the ASTER image of Beijing city and surrounding area for example, with the detailed analysis of spectral characteristics law and land use types in urban ecological environment, the new combined index image was made of three kinds of indexes that are Normalized Difference Vegetation Index, Modified Normalized Difference Water Index and Normalized Difference Buih-up Index. The combined index image was used to classify with the object-oriented classification method based on support vector machine to extract agrieuhural land information. At the same time, the maximum-likelihood and support vector machine classification method based on the original image and combined index image were contrasted and analyzed with this method. Results indicated that the combined normalized difference index image could compress data dimension and reduce the correlation of the coverage surface features. It woald be easy to extract agricultural land information. For the combined index image, the object-oriented classification method based on support vector machine had higher precision than other methods. Accuracy reached at 95. 701%.