Noise reduction in Hyperion high dynamic range hyperspectral data using machine learning and statistical techniques

Abstract

Numerous remote sensing applications rely heavily on hyperspectral imagery, but it is frequently plagued by noise, which degrades the data quality and hinders subsequent analysis. In this research paper, we present an in-depth analysis of noise removal techniques for hyperspectral imagery, specifically for data acquired from the Hyperion EO-1 sensor. Setting off with obtaining Hyperion data and the pre-processing stages, the paper discusses the acquisition and denoising of Hyperion data. The hyperspectral data considered is in the high dynamic range (HDR) format, which maintains the original imagery's complete dynamic range. The study explores various noise reduction methods, such as minimum noise fraction (MNF), principal component analysis (PCA), wavelet denoising, non-local means (NLM), and denoising autoencoders, aimed at enhancing the signal-to-noise ratio. The effectiveness of these techniques is evaluated through visual quality, mean square error (MSE), and peak signal-to-noise ratio (PSNR), alongside their impact on mineral exploration. Furthermore, the paper investigates the application of machine learning algorithms on denoised data for mineral identification, highlighting the potential of integrating denoising techniques with machine learning for improved mineral exploration. This comparative analysis aims to identify the most efficient noise removal methods for hyperspectral imagery, facilitating higher quality data for subsequent analysis.