Harmonic Analysis With Adaptive Water and Land Separation for High Performance Hyperspectral Image Quality Enhancement

The stripe noise exists in the images acquired by the imaging spectrometer in orbit, which seriously restricts the subsequent high-precision quantitative application of hyperspectral images (HSIs). This article proposes the stripe removal method for HSIs, which can be applied to various scenes. It can remove the stripe noise of the images with large water area coverage, different stripe widths, and different stripe brightness to the greatest extent, without affecting the details of the image. The method uses the maximum between-class variance method adaptive threshold to extract the boundary of water and land areas in the image, and uses harmonic analysis to eliminate the cumulative stripe noise of image mean and variance in the frequency domain to obtain the theoretical true value of the image. Extensive experiments are carried out on GF5, ZY-1-02D, and Huanjing-2A (HJ-2A) satellite HSIs to compare the visualization effects of eight different algorithms for stripe removal, and the performance of the eight different algorithms is quantitatively evaluated by information entropy (IE) and noise estimation. The results show that the proposed algorithm has the most superior overall performance in terms of quantitative evaluation, processing efficiency, adaptability, and robustness, and is the best solution for engineering applications.