Hyperspectral Images Target Recognition Using Projection Pursuit

Abstract

In order to solve the recognition problem of the specific target, a hyperspectral target recognition method using minimum noise fraction (MNF) and projection pursuit was used. Firstly, MNF was used to calculate the intrinsic dimension as well as image denoising. Then, using spectral information divergence (SID) as the projection index, remove the background and extract the spectral curve via adaptive threshold of the value of the projection index. When calculating each projection index, we used a simplified approach to maximize the projection index, which does not require an optimization algorithm. It searches for a solution by obtaining a set of candidate projections from the data and choosing the one with the highest projection index. Finally, identify the target and its location using the spectral angle mapping method. Through a series of hyperspectral images experiments, the results show that the MNF image preprocessing can make the projections better and reduce the computational effectively. Remove the background using adaptive threshold of the projection index quickly and efficiently. The method can not only reduce the images noise more effectively but also extract endmember and identify the target quickly, reliably, and fastly.