Satellite chl-a image inpaintin

Speaker: 
Julien Brajard
Affiliation: 
LOCEAN, Sorbonne UniversiteĢ, NERSC
Seminar Date: 
25. September 2018 - 12:00 - 12:30
Location: 
Lecture room, Ground Floor, NERSC

In this work, we evaluate the efficiency of a deep-convolutional neural network (CNN) to reconstruct missing data in chlorophyll-a satellite images. The missing area are mainly due to the presence of cloud above the ocean. The CNN-based methodology is compared with a sate-of-art krigging algorithm.
The reconstruction algorithms were applied on a region from the West Mediterranean Sea. It shows good reconstruction abilities of both the krigging and the CNN approach (correlation between 0.81 and 0.86). The CNN seems particularly well suited to reconstruct the small spatial structures of the missing area.