期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2007
卷号：XXXVI-7/C50
出版社：Copernicus Publications
摘要：In recent years the launch of (very) high-resolution (VHR) spaceborne multi-spectral (MS) scanners has made purely supervised remote sensing (RS) image analysis over extended target areas no longer feasible as reference data are difficult, tedious, or expensive to gather. To overcome limitations of a purely supervised classification approach to RS image classification problems, a two-stage hybrid learning classification scheme can be recommended. In this context, a novel spectral prior knowledge-based preliminary classifier (suitable as a hybrid learning first-stage classifier) is presented hereafter. The main operational and architectural properties of the proposed rule-based classifier are summarized below. o It is input with Landsat 5 TM and Landsat 7 ETM+ images calibrated into planetary reflectance (albedo) and at-satellite temperature. o It is fully unsupervised, i.e, it requires: . no free parameter to be user-defined and . no reference (supervised) data set of examples. o It is pixel-based. As a consequence, it is computationally efficient, requiring approximately 15 minutes per Landsat scene (from data calibration to output map generation). o It is robust to changes in the input data set, i.e., it is capable of dealing with the inherent fuzziness (variability) of class-specific spectral signatures.
