The application of manifold learning in dimensionality analysis for hyperspectral imagery

While reducing the dimensionality of hyperspectral data, linear dimensionality analysis methods are usually adopted to acquire intrinsic dimensionality (ID) of high-dimensional hyperspectral data. This paper uses an unsupervised manifold learning method to conduct the dimensionality analysis of hyperspectral data, providing a manifold-learning-based algorithm for hyperspectral data dimensionality analysis. ISOMAPDA, LLEDA, LEDA and LTSADA algorithms are adopted to estimate the intrinsic dimensionality of simulated and real hyperspectral data, and obtain the two-dimension manifold figures of high-dimensional data. At last, this article discusses the relative advantages and disadvantages of those algorithms in the process of hyperspectral dimensionality analysis.