TY - GEN
T1 - A Variational Approach for Robust Online Fusion of Multiresolution Multispectral Images
AU - Li, Haoqing
AU - Borsoi, Ricardo
AU - Imbiriba, Tales
AU - Closas, Pau
N1 - Publisher Copyright:
© 2024 European Signal Processing Conference, EUSIPCO. All rights reserved.
PY - 2024
Y1 - 2024
N2 - Multi-resolution image fusion is a key problem for real-time satellite imaging, which has a central role in detecting and monitoring the intensity of key natural phenomena such as floods. It aims to solve the trade-off between high temporal and high spatial resolution in remote sensing instruments. Although several algorithms have been proposed to solve this problem, the presence of outliers caused by, e.g., cloud cover downgrades their performance. In this paper, an online image fusion method based on a robust Kalman filter with a weakly supervised approach for temporal dynamics estimation is proposed. Outliers are modelled as a discrete variables, where the probability of contamination for each pixel and spectral band is modelled by a latent variable whose distribution is approximated under variational inference. Experiments fusing images from the MODIS and Landsat 8 sensors show that the proposed approach is significantly more robust against cloud cover, without losing its efficiency when no clouds are present.
AB - Multi-resolution image fusion is a key problem for real-time satellite imaging, which has a central role in detecting and monitoring the intensity of key natural phenomena such as floods. It aims to solve the trade-off between high temporal and high spatial resolution in remote sensing instruments. Although several algorithms have been proposed to solve this problem, the presence of outliers caused by, e.g., cloud cover downgrades their performance. In this paper, an online image fusion method based on a robust Kalman filter with a weakly supervised approach for temporal dynamics estimation is proposed. Outliers are modelled as a discrete variables, where the probability of contamination for each pixel and spectral band is modelled by a latent variable whose distribution is approximated under variational inference. Experiments fusing images from the MODIS and Landsat 8 sensors show that the proposed approach is significantly more robust against cloud cover, without losing its efficiency when no clouds are present.
KW - Bayesian Filtering
KW - Image fusion
KW - Multispectral imaging
KW - Super-resolution
KW - Variational Inference
UR - https://www.scopus.com/pages/publications/85208416063
UR - https://www.scopus.com/pages/publications/85208416063#tab=citedBy
U2 - 10.23919/eusipco63174.2024.10715320
DO - 10.23919/eusipco63174.2024.10715320
M3 - Conference contribution
AN - SCOPUS:85208416063
T3 - European Signal Processing Conference
SP - 666
EP - 670
BT - 32nd European Signal Processing Conference, EUSIPCO 2024 - Proceedings
PB - European Signal Processing Conference, EUSIPCO
T2 - 32nd European Signal Processing Conference, EUSIPCO 2024
Y2 - 26 August 2024 through 30 August 2024
ER -