卫星遥感影像中耕地信息的自动提取方法研究

Study on Automatic Abstraction Methods of Cultivated Land Information from Satellite Remote Sensing Images

以县 (市区 )为基本单位 ,采用遥感图像处理ENVI3.2软件和TM资料 ,进行了 7个县 (市区 ) 10个时相的耕地信息自动提取技术研究。结果证明 ,选择春季时相 ,TM 432波段合成 ,经几何校正和增强处理 ,通过交互的非监督分类提取耕地信息 ,可以获得满意的结果 ,平均分类精度达到 98.2 2 %

ion techniques on cultivated land information of seven counties and ten temporal phases were studied. Because of the spectral characteristic difference between cultivated land subtypes, the interactive unsupervised classification method can be used for obtaining a good result. Among supervised classification methods the ENVI Spectral Angle Mapper can reach a better result because of its data purification function to maintain the typicality of training samples. Compared with salty and mountainous regions, the classification precision can be higher in inner plain region where cultivated land has comparatively accordant spectral information. Based on comparison analysis the best temporal phase of cultivated land information abstraction is in spring which is from March to May. In addition the using of non remote sensed data such as land use map as well as expert experience is also important for promoting the precision of cultivated land acquiring. Selecting spring TM432 band composite images, processed by geometric calibration and enhancing processing, using interactive unsupervised classification method; the average precision of cultivated land classification is up to 98.22%.