Scientists at HRL Laboratories have published their new framework for training computer deep neural networks to be able to classify synthetic aperture radar (SAR) images without a large labeled data set, solving the problem of SAR image identification when only a few labeled data were available.
The system uses a shared learning space between a large labeled electro-optical (EO) dataset and the SAR system. Both systems learn to transfer data within the shared space, which enables the SAR system to classify data from the electro-optical system without having a label on each data element.
“We train two coupled deep encoders to map data points from the EO and SAR domains to the shared embedding space,” said Mohammad Rostami, first author on the paper. “This enables an image classifier program trained from the embedding space to generalize image classification well on the SAR domain, using mostly the EO data. Thus the new system can learn from the large amount of EO data to identify images for the SAR system without the labels.”
Synthetic aperture radar is usually installed in moving vehicles such as aircraft. By scanning over objects, SAR creates two-dimensional images or three-dimensional reconstructions of objects, such as landscapes.
“In this paper we demonstrated the effectiveness and applicability of our approach for the problem of ship classification in the area of maritime domain awareness,” said Soheil Kalouri, an HR. coauthor on the paper.
The paper, Deep Transfer Learning for Few-Shot SAR Image Classification, was published in the journal Remote Sensing on June 8, 2019. Other HRL authors included Kyungnam Kim.
HRL Laboratories, LLC, Malibu, California (hrl.com) is a corporate research-and-development laboratory owned by The Boeing Company and General Motors specializing in research into sensors and materials, information and systems sciences, applied electromagnetics, and microelectronics. HRL provides custom research and development and performs additional R&D contract services for its LLC member companies, the U.S. government, and other commercial companies.
Media Inquiries: Michele Durant, (310) 317-5321