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：2008
卷号：XXXVII Part B6b
页码：307-312
出版社：Copernicus Publications
摘要：Effective and understanding exploration of hyperspectral remote sensing data necessitates the development of sophisticated schemes that represent images. Such schemes ideally preserve and recognize significant features. However, uncertainty arises in classification problems when the input pattern is not perfect or measurement error is unavoidable. It would be need to obtain the estimation of uncertainty classification associated with a new observation and its membership within a particular class. Typically, existing methods supplying uncertainty information has monotonic neural network model, back propagation NN (BPNN) and the fuzzy membership model (FMM). The paper describes that an efficient combinational algorithm for uncertainty estimation on spectral dimension in hyperspectral remote sensing images is proposed for classification accuracy improvement and computing efficiency. The combination of fuzzy clustering and neural network adopted, which input patterns are divided into several small neural networks based on fuzzy clustering, is provides the classification boundaries based on the degree-of-dissimilarity measurement of the input pattern associated with each classification class. And we proposed the neural network with dynamical neuron created during learning NN algorithm. In the experiment, We tested three methods for misclassification using the same data and compared the performances with our method
