Low-Light Image Enhancement of Permanently Shadowed Lunar Regions with Physics-Based Machine Learning

Reference : Low-Light Image Enhancement of Permanently Shadowed Lunar Regions with Physics-Based ...


	Document type :	Scientific congresses, symposiums and conference proceedings : Poster
	Discipline(s) :	Engineering, computing & technology : Computer science
	To cite this reference:	http://hdl.handle.net/10993/49471

Title :	Low-Light Image Enhancement of Permanently Shadowed Lunar Regions with Physics-Based Machine Learning
Language :	English
Author, co-author :	moseley, ben [University of Oxford]
	Bickel, Valentin [ETH Zurich/MPS Goettingen]
	Rana, Loveneesh [University of Luxembourg > Interdisciplinary Centre for Security, Reliability and Trust (SNT) > Remote Sensing >]
	Lopez-Francos, Ignacio [NASA Ames Research Center]
Publication date :	2021
Peer reviewed :	Yes
Audience :	International
Event name :	GPU Technology Conference 2021 (GTC 2021)
Event date :	12–16 April 2021
Event organizer :	NVIDIA
Abstract :	[en] We show that it is possible to significantly enhance the quality of extremely low-light images of permanently-shaded regions (PSRs) on the moon by using two physics-based deep neural networks, which we called HORUS, to remove CCD sensor-related noise and photon noise. To inform our distribution of training data, we perform ray tracing over a digital elevation model of the moon and derive the distributions of secondary illumination angles in PSRs. Our network provides high-resolution, low-noise images that will help enable future ground missions to plan and execute safe and effective traverses into, around, and out of lunar PSRs — a critical step in our endeavor to explore the moon and beyond.
Permalink :	http://hdl.handle.net/10993/49471

There is no file associated with this reference.

All documents in ORBi^lu are protected by a user license.