TY - GEN
T1 - Fast Reconstruction of Hyperspectral Image from its RGB Counterpart Using ADMM-Adam Theory
AU - Lin, Chia Hsiang
AU - Lin, Tzu Hsuan
AU - Lin, Ting Hsuan
AU - Lin, Tang Huang
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - This paper aims at recovering the hyperspectral image from its RGB counterpart. This highly challenging inverse problem has profoundly impactful applications, including hyperspectral imaging for metamaterial-driven miniaturized satellite. Popular inverse imaging theories include convex optimization (CO, wherein ADMM is a key optimizer) and deep learning (DL, wherein Adam plays a fundamental role); the former usually involves math-heavy optimization procedure, while the latter often requires time-consuming big data collection. We adopt the ADMM-Adam theory, recently investigated in the remote sensing literature for blending the advantages of CO and DL, in order to achieve outstanding hyperspectral signature reconstruction (HSR) without support from heavy math or big data. Simply speaking, a deep regularizer is devised to extract useful information embedded in the rough solution learned from small data. Then, such information is used to design a simple convex regularizer via Q-quadratic function for designing an effective HSR algorithm, whose effectiveness is experimentally illustrated.
AB - This paper aims at recovering the hyperspectral image from its RGB counterpart. This highly challenging inverse problem has profoundly impactful applications, including hyperspectral imaging for metamaterial-driven miniaturized satellite. Popular inverse imaging theories include convex optimization (CO, wherein ADMM is a key optimizer) and deep learning (DL, wherein Adam plays a fundamental role); the former usually involves math-heavy optimization procedure, while the latter often requires time-consuming big data collection. We adopt the ADMM-Adam theory, recently investigated in the remote sensing literature for blending the advantages of CO and DL, in order to achieve outstanding hyperspectral signature reconstruction (HSR) without support from heavy math or big data. Simply speaking, a deep regularizer is devised to extract useful information embedded in the rough solution learned from small data. Then, such information is used to design a simple convex regularizer via Q-quadratic function for designing an effective HSR algorithm, whose effectiveness is experimentally illustrated.
KW - adaptive moment estimation (Adam)
KW - Alternating direction method of multipliers (ADMM)
KW - hyperspectral signature reconstruction
KW - small data learning
UR - http://www.scopus.com/inward/record.url?scp=85143161672&partnerID=8YFLogxK
U2 - 10.1109/WHISPERS56178.2022.9955079
DO - 10.1109/WHISPERS56178.2022.9955079
M3 - 會議論文篇章
AN - SCOPUS:85143161672
T3 - Workshop on Hyperspectral Image and Signal Processing, Evolution in Remote Sensing
BT - 2022 12th Workshop on Hyperspectral Imaging and Signal Processing
PB - IEEE Computer Society
T2 - 12th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing, WHISPERS 2022
Y2 - 13 September 2022 through 16 September 2022
ER -