A robust estimation of virtual dimensionality in hyperspectral imagery

Estimate the Intrinsic dimensionality which is normally characterized by virtual dimensionality is the first step of hyperspectral data processing. A robust estimation algorithm called the noise constrained virtual dimensionality analysis is proposed in this paper. It decomposes date matrix into the product of Q which is the unit orthogonal matrix and R which is the triangle matrix through the QR decomposing for decreasing the computational complexity, adopts glide noise detection window to filter the noise to improve the accuracy of the estimation of dimensionality and synthesizes the Least squares algorithm to modify threshold for the reasonable results. The result obtained by applying the classical algorithms and the proposed algorithm both on simulated hyperspectral date and real hyperspectral date are also presented and discussed.