TY - GEN
T1 - Reconstructing Hyperspectral Images from RGB Using a Spectral-Informed Neural Network
AU - Kang, Ryuna
AU - Ku, Zahyun
AU - Kwak, Yunsang
N1 - Publisher Copyright:
© 2025 IEEE.
PY - 2025
Y1 - 2025
N2 - Hyperspectral imaging is a technology that captures rich spectral information across a wide range of wavelengths. However, its application in extreme environments is limited due to bulky hardware and a complex acquisition process. To address these limitations, this study proposes a network based on a spectral-informed neural network (SINN) that reconstructs high-dimensional hyperspectral images from standard RGB inputs. The proposed model predicts over 250 spectral bands within the visible range by applying spectral basis embedding and a neural network-based mapping structure. It also demonstrates strong generalization performance through training on real hyperspectral datasets. Experimental results show that the proposed SINN outperforms a baseline model in both quantitative metrics and spectral reconstruction accuracy. These findings validate the potential of SINN as a lightweight and interpretable solution for hyperspectral image recovery.
AB - Hyperspectral imaging is a technology that captures rich spectral information across a wide range of wavelengths. However, its application in extreme environments is limited due to bulky hardware and a complex acquisition process. To address these limitations, this study proposes a network based on a spectral-informed neural network (SINN) that reconstructs high-dimensional hyperspectral images from standard RGB inputs. The proposed model predicts over 250 spectral bands within the visible range by applying spectral basis embedding and a neural network-based mapping structure. It also demonstrates strong generalization performance through training on real hyperspectral datasets. Experimental results show that the proposed SINN outperforms a baseline model in both quantitative metrics and spectral reconstruction accuracy. These findings validate the potential of SINN as a lightweight and interpretable solution for hyperspectral image recovery.
KW - heavily Illposed inverse problem
KW - Hyperspectral image reconstruction
KW - RGB sensitivity attention
KW - Spectral informed neural network
UR - https://www.scopus.com/pages/publications/105031891358
U2 - 10.1109/ICMIC66299.2025.11257785
DO - 10.1109/ICMIC66299.2025.11257785
M3 - Conference contribution
AN - SCOPUS:105031891358
T3 - ICMIC 2025 - 4th International Conference on Mobile, Military, Maritime IT Convergence: Promoting Ultimate Convergence of Wireless, Military and Maritime Communications in the 6G Era
SP - 278
EP - 280
BT - ICMIC 2025 - 4th International Conference on Mobile, Military, Maritime IT Convergence
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 4th International Conference on Mobile, Military, Maritime IT Convergence: Promoting Ultimate Convergence of Wireless, Military and Maritime Communications in the 6G Era, ICMIC 2025
Y2 - 27 August 2025 through 30 August 2025
ER -