NASA SBIR 2018-I Solicitation

Proposal Summary

 18-1- S5.02-9802
 Earth Science Applied Research and Decision Support
 Deep Learning Enhanced Fidelity InSAR Toolkit (DEFIT)
SMALL BUSINESS CONCERN (Firm Name, Mail Address, City/State/Zip, Phone)
Lynntech, Inc.
2501 Earl Rudder Freeway South
College Station , TX 778456023-6023
(979) 693-0017

Principal Investigator (Name, E-mail, Mail Address, City/State/Zip, Phone)
Dr. Jason Hill PhD
2501 Earl Rudder Freeway South College Station, TX 77845 - 6023
(979) 764-2200

Business Official (Name, E-mail, Mail Address, City/State/Zip, Phone)
Darla Hisaw
2501 Earl Rudder Freeway South College Station, TX 77845 - 6023
(979) 764-2219
Estimated Technology Readiness Level (TRL) :
Begin: 2
End: 4
Technical Abstract

Lynntech, in collaboration with Southern Methodist University (SMU) Earth Sciences, proposes to develop a new deep learning-based toolkit that is useful for enhancing the fidelity of results derived from Interferometric Synthetic Aperture Radar (InSAR) interferograms. The automated deep learning tool performs a spatial–temporal analysis of multiple InSAR images, to yield a high fidelity estimate of the deformation of the topography and estimate of atmospheric water vaper when a recent Digital Elevation Model is also known. There are existing methods used by earth science experts to detect and mitigate the atmospheric anomaly that effects the time of flight of backscattered radar, either from multiple InSAR images or when integrating other sources of elevation or meteorological observations or models.  The automated image reconstruction algorithm will minimize a loss function, an inferred empirical error based on a large sample set, rather than the heuristic or incompletely modeled statistical algorithms currently employed, through a three-step process: first detect the regions affected by the atmospheric anomaly, and then second without a-priori knowledge use a generative network to reconstruct the interferogram or deformation map without the atmospheric effect, and use another network to train the loss function to evaluate the generator’s result and adjust its internal parameters. This type of approach has not been implemented for InSAR imagery, but has been applied to similar image processing problems and generalized to other tasks. This tool is meant for big data analysis of very fast revisit InSAR that covers the entire globe. Lynntech and SMU-Earth Science propose to develop and validate this approach for developing a new image processing tool in Phase I , while developing the deep learning enhanced fidelity InSAR toolkit in Phase II and III, raising the TRL from 2 to 4 within the Phase I work plan and planning for testing on relevant datasets in Phase II.

Potential NASA Applications

This technology would be useful to many Earth science and meteorological applications involving changes in terrain, ecology and the weather.  Digital elevation models can be produced and updated in near real-time. Ground level deformations due to various processes could be monitored on almost a daily basis. Also highly fidelity spatio-temporal analysis of fast revisit InSAR data to track changes in the Earth’s surface and atmosphere would help in the zenith dry delay correction of GPS signals.

Potential Non-NASA Applications

Regularly monitoring changes in the ground would be useful in disaster prediction and recovery (e.g. mudslides, flooding, sinkholes), evaluating the settling of infrastructure, preventing property damage and is also vital to land management strategies. Estimating atmospheric water vapor would assist in nowcasting. InSAR imagery can be useful in decision making in a wide range of applications. SAR imagery enhancement algorithms, with a few changes, can be used in other terrestrial applications.

Form Generated on 05/25/2018 11:45:35