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：2004
卷号：XXXV Part B7
页码：1354-1358
出版社：Copernicus Publications
摘要：This paper concerns the evaluation of a feature selection and classification techniques for land cover classification and potential mon- itoring of temporal changes. For the purposes of our study, an ASTER satellite image was used, acquired on 5 October, 2001, with 14 spectral bands resampled at the spatial resolution of 15 m. The study area concerns the geographic area of the water basin of Lake Vegoritis, that is located in the northern part of Greece. A variety of industrial and agricultural activities take place in the Vegoritis lake basin, which result to constant lowering of the lake water table and the change of land use, and subsequently, lead to degrading of the lake environment. Besides the intensity level of the original bands, several features were created including spectral indices (e.g. NDVI), band ratios and products among selected bands, and Haralick texture features together with their second order combinations. The total of the features was used as input to a novel feature selection process, the Greedy Non-Redundant (GreeNRed) feature selection algorithm. This algorithm is based on information theory, and greedily selects features with no redundant information. The algorithm succeeded in keeping the complexity at low levels by restraining evaluations in one-dimensional feature spaces, whereas non-redundancy is achieved by a boosting-like sample weighting. After the feature selection process different categories of classification methods were applied, namely, K-nearest neighbors and support vector machine. Classification accuracy assessment followed in order to derive the best classification method and consequently, to give further feedback as far as the performance of the feature selection algorithm is concerned and the usefulness of machine learning algorithms for land cover classification as a prerequisite for the assessment of the extended lake environment
关键词：Remote Sensing; ASTER; Feature Selection; Classification; Algorithms; Land cover
