Unsupervised learning for hyperspectral recovery based on a single RGB image

Hyperspectral imagery often suffers from the degradation of spatial, spectral, or temporal resolution due to the limitations of hyperspectral imaging devices. To address this problem, hyperspectral recovery from a single red-green-blue (RGB) image has recently achieved significant progress via deep learning. However, current deep learning-based methods are all learned in a supervised way under the availability of RGB and correspondingly hyperspectral images, which is unrealistic for practical applications. Hence, we propose to recover hyperspectral images from a single RGB image in an unsupervised way. Moreover, based on the statistical property of hyperspectral images, a customized loss function is proposed to boost the performance. Extensive experiments on the BGU iCVL Hyperspectral Image Dataset demonstrate the effectiveness of the proposed method.