AUTHOR=Hu Shaoxiang , Hou Rong , Ming Luo , Meifang Su , Chen Peng 

TITLE=A hyperspectral image reconstruction algorithm based on RGB image using multi-scale atrous residual convolution network

JOURNAL=Frontiers in Marine Science

VOLUME=Volume 9 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2022.1006452

DOI=10.3389/fmars.2022.1006452

ISSN=2296-7745

ABSTRACT=Reconstruction of hyperspectral images from RGB images has important research value as a low-cost and fast method to obtain ocean hyperspectral images. We designed an algorithm to reconstruct hyperspectral images from RGB images using multi-scale Atrous convolution residual network (MACRN). The algorithm includes three parts: Low-level extraction, High-level feature extraction and feature transform. The High-level feature extraction module is composed of multi-scale Atrous convolution residual blocks (ACRB) cascaded. It uses the stacking of multiple modules to form a depth network for extracting high-level features in the input RGB image. The algorithm uses jump connection for residual learning, and the final High-level feature adopts the method of adding the output of the Low-level feature extraction module and the output of the cascaded Atrous convolution residual block element by element, so as to prevent the phenomenon of gradient dispersion and gradient explosion in the deep network. Without adding too many parameters, the model can extract multi-scale features under different receptive fields, make better use of the spatial information in RGB images, and enrich the context information. The experiment of hyperspectral reconstruction of Sentinel-2 satellite data on the northern coast of Australia shows that the reconstruction result of this algorithm is stable, and it can reconstruct high-precision hyperspectral images from RGB images with less computational cost.