Hyperspectral Image Super-Resolution Using Convolutional Neural Network and Wavelet Transform

Abstract

Hyperspectral images have many purposes in the industry, the spectral information, which these images provide, allows to perform various sorting or object detection tasks. However, most of these images are obtained at a low-spatial resolution, thus reducing the effectiveness of the tasks, in which they can be used. In this study, a novel framework has been proposed to increase the resolution of hyperspectral images without affecting the spectral properties of the pixels. The designed system consists of two sections: the first section is the spatial section where wavelet transform is used for increasing spatial resolution; the second section represents the spectral procedures where a neural network is employed especially to correct the spectral distortions generated in the spatial section. Numerous experimental results have confirmed the better performance of the novel framework via objective and subjective criteria.