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：2009
卷号：XXXVIII-4/W10
出版社：Copernicus Publications
摘要：Virtual globes support users with remote images from multiple sources, and support data analysis, information extraction and even knowledge discovery. But when extracting thematic information, those remote images are so complex that we should provide a large amount of label data, which is much expensive and difficult for manual collection, to get sufficient classification result. Semi-Supervised Classification, which utilizes few labeled data assigned with unlabeled data to determine classification borders, has great advantages in extracting classification information from mass data. We find Gauss Mixture can excellently fit the remote sensing image's spectral feature space, propose a novel thought in which each class's feature space is described by one Gauss Mixture Model, and then apply the thought in Semi-Supervised Classification. A large number of experiences shows by using a small amount of label samples, the method proposed in this paper can achieve as good classification accuracy as other supervised classification methods(such as Support Vector Machine Classification, Object Oriented Classification), which need large amount of label samples, and so has a strong application value
