NASA SBIR 2019-I Solicitation

Proposal Summary

PROPOSAL NUMBER:
 19-1- S5.03-4464
SUBTOPIC TITLE:
 Bridging the Gap of Applying Machine Learning to Earth Science
PROPOSAL TITLE:
 Machine Learning for Earth Science
SMALL BUSINESS CONCERN (Firm Name, Mail Address, City/State/Zip, Phone)
Kraenion
17094 Lon Road
Los Gatos, CA 95033- 8510
(650) 575-7816

Principal Investigator (Name, E-mail, Mail Address, City/State/Zip, Phone)

Name:
Mr. Binu Mathew
E-mail:
binu@kraenion.com
Address:
17094 lon rd los gatos, CA 95033 - 8510
Phone:
(650) 283-9142

Business Official (Name, E-mail, Mail Address, City/State/Zip, Phone)

Name:
Mr. Binu Mathew
E-mail:
binu@kraenion.com
Address:
17094 lon rd los gatos, CA 95033 - 8510
Phone:
(650) 283-9142
Estimated Technology Readiness Level (TRL) :
Begin: 3
End: 5
Technical Abstract (Limit 2000 characters, approximately 200 words)

Kraenion's is an applied machine learning company developing a Computer Vision and Machine Learning (CVML) software platform named Vision Engine. It provides robust pre-trained visual intelligence solutions and keeps up with the latest in AI research, but can be customized for specific customer needs. It was engineered to produce good neural networks quickly on modest computers. This engineering investment has been worthwhile: we were able to get 100% accuracy on the DeepSat Earth Science dataset in under 1 hour on a laptop. Vision Engine includes an active learning based neural network training technology where the training software is aware of the cost of labeling data. Unlike traditional neural network training that assumes a large labeled dataset, our system carefully picks samples that maximize the learning opportunity and presents it for labeling to a human annotator. This provides much higher return on dollars invested for data annotation in areas like Earth Science where unlabeled data is abundant, but there is a scarcity of labeled data.

Web: http://kraenion.com

Potential NASA Applications (Limit 1500 characters, approximately 150 words)
  1. Provide active learning based neural network training technology where the training software is aware of the cost of labeling data. Unlike traditional neural network training, our system carefully picks samples that maximize the learning opportunity while data minimizing labeling cost.
  2. Deploy semantic segmentation for Earth Science data. 
  3. User friendly training and inference support for Earth Science Machine Learning applications.
  4. Visualization support for large multi-spectral datasets.
  5. Augment human annotators using semantic segmentation.
Potential Non-NASA Applications (Limit 1500 characters, approximately 150 words)
  1. Satellite data based claims processing for insurance companies.
  2. Threat detection in baggage via 3D CT scan at airports for DHS
  3. Smart wheelchairs
  4. Smart airport ground vehicles

 

Duration: 6
Form Generated on 06/16/2019 23:33:21