Effectiveness Analysis of Example-Based Machine Learning and Deep Learning Methods for Super-resolution Hyperspectral Images

Hyperspectral imagery has found widespread use in various applications due to its rich spectral information. However, the inherent limitation of low spatial resolution attributed to imaging hardware constraints has led to the development of methods aimed at increasing the spatial resolution of hyperspectral images. Although pan-sharpening and sparse representation techniques based on dictionary learning are effective approaches for this purpose, artificial intelligence-based deep learning methods have also been used recently. This study conducts a quantitative analysis of super-resolution methods applied to hyperspectral images, focusing on determining the optimal method in terms of both reconstruction quality and processing time. To address the performance of super-resolution algorithms, this study defines and employs three main categories: (1) multi-image and interpolation-based methods also known as pan-sharpening methods (IHS and PCA); (2) example-based machine learning methods, especially, dictionary learning-based sparse representation methods (K-SVD, ODL, and Bayesian); and (3) deep learning-based methods (CNN). This research contributes valuable information on the comparative effectiveness of super-resolution methods for hyperspectral images, providing a basis for informed selection and implementation in relevant applications.