Construction of hyperspectral images from RGB images via CNN

Hyperspectral imaging (HSI) is an approach that captures information spanning a broad spectrum of spectral bands or wavelengths. Unlike traditional RGB, HSI data contains numerous contiguous or narrow spectral bands, allowing for detailed spectral analysis and identification of materials in an image. HSI finds application in various fields, including medical, document forgery, remote sensing, agriculture, and environmental monitoring. The high cost of hyperspectral systems restricts its usage. Hence, we propose directly converting RGB images to hyperspectral images to bridge the gap between high-cost systems. In this study, we trained a 3D Convolutional Neural Network (CNN) to generate hyperspectral images from RGB images. We employed the band-by-band approach to process each band image individually. We used the ICVL and ARAD1K datasets for training and evaluation. Conducted performance evaluation using the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) metrics. Compared to the proposed model against various state-of-the-art models, it demonstrated a successful generation of meaningful results.