Citation: Wang, L.; Sole, A.; Hardeberg, J.Y. Densely Residual Network with Dual Attention for Hyperspectral Reconstruction from RGB Images. Remote Sens. 2022, 14, 3128. https://doi.org/ 10.3390/rs14133128 Academic Editors: Antonio J. Plaza and Pedro Latorre-Carmona Received: 5 March 2022 Accepted: 10 June 2022 Published: 29 June 2022 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affil- iations. Copyright: © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). remote sensing Article Densely Residual Network with Dual Attention for Hyperspectral Reconstruction from RGB Images Lixia Wang 1,2 , Aditya Sole 2 and Jon Yngve Hardeberg 2, * 1 Xiaomi, Nanjing 210019, China; lixiawang@whu.edu.cn 2 Department of Computer Science, Faculty of Information Technology and Electrical Engineering, Norwegian University of Science and Technology, N-2815 Gjøvik, Norway; aditya.sole@ntnu.no * Correspondence: jon.hardeberg@ntnu.no Abstract: In the last several years, deep learning has been introduced to recover a hyperspectral image (HSI) from a single RGB image and demonstrated good performance. In particular, attention mechanisms have further strengthened discriminative features, but most of them are learned by convolutions with limited receptive fields or require much computational cost, which hinders the function of attention modules. Furthermore, the performance of these deep learning methods is hampered by tackling multi-level features equally. To this end, in this paper, based on multiple lightweight densely residual modules, we propose a densely residual network with dual attention (DRN-DA), which utilizes advanced attention and adaptive fusion strategy for more efficient feature correlation learning and more powerful feature extraction. Specifically, an SE layer is applied to learn channel-wise dependencies, and dual downsampling spatial attention (DDSA) is developed to capture long-range spatial contextual information. All the intermediate-layer feature maps are adaptively fused. Experimental results on four data sets from the NTIRE 2018 and NTIRE 2020 Spectral Reconstruction Challenges demonstrate the superiority of the proposed DRN-DA over state-of-the-art methods (at least −6.19% and −1.43% on NTIRE 2018 “Clean” and “Real World” track, −6.85% and −5.30% on NTIRE 2020 “Clean” and “Real World” track) in terms of mean relative absolute error. Keywords: attention mechanism; densely residual network with dual attention (DRN-DA); receptive field; channel attention; spatial attention 1. Introduction With the increasing applications of computer vision technology in various engineering fields [1–11], hyperspectral images (HSIs) have proved to obtain more helpful information than RGB images. Hyperspectral images contain the reflectance of objects or scenes in different spectral bands, usually ranging from several dozens to hundreds, even outside the visible spectrum (e.g., in the ultraviolet or infrared spectrum). Compared with traditional RGB images with increased spectral range and resolution, HSIs provide much richer infor- mation, which has been widely used in cultural heritage [1,2], medical diagnosis [3], remote sensing [4], food quality inspection [5], color quality control [6], and various computer vision tasks, such as face recognition, object tracking, and image classification [7–9]. Due to the growing need for HSIs, various hyperspectral imaging systems (HISs) have been developed in the last several decades. The first HISs, such as NASA’s airborne visible/infrared imaging spectrometer (AVIRIS) [12], employed a prism to disperse the reflected light and a linear array detector to record the reflected light. This kind of HISs can acquire images with high spatial/spectral resolution in “whisk broom” imaging mode, but image acquisition is time-consuming since they adopt the point-scanning method. Afterwards, “push broom” HISs, such as NASA’s advanced land imager (ALI) [13], and “staring” HISs, often used in microscopy or other lab applications, have been developed. Remote Sens. 2022, 14, 3128. https://doi.org/10.3390/rs14133128 https://www.mdpi.com/journal/remotesensing