Vision Language Model (VLM) for Synthetic Aperture Radar (SAR) target search and classification from Sensor Independent Complex Data (SICD)

TECHNICAL POINT OF CONTACT (TPOC)
Tom Majumder
PROJECTED CMMC LEVEL REQUIREMENT
Level 2 (Self)
TECHNOLOGY AREAS
Air Platform
|
Ground Sea
|
Space Platforms
MODERNIZATION PRIORITIES
Integrated Network Systems-of-Systems
|
Integrated Sensing and Cyber
|
Trusted AI and Autonomy
KEYWORDS
Synthetic aperture radar (SAR); Sensor Independent Complex Data (SICD), Metadata; single look complex (SLC)
OBJECTIVE
Develop VLM based SAR imagery search and classification from sensor independent complex data (SICD)
DESCRIPTION
Vision language models (VLMs) combine multimodal (Video, SAR and Infrared Imagery) data/imagery and generative AI models to understand and process video, image, and text. VLM integrates large language models (LLM) with a vision encoder that allows LLM to explain contents of an imagery. The goal of this research is to analyze (detect and classify) synthetic aperture radar (SAR) imagery from sensor independent complex data (SICD) data.
PHASE I
This topic is accepting Direct to Phase II proposals ONLY. To qualify for a Phase II award, proposers must submit Feasibility Documentation demonstrating: successful classification (>95% accuracy) of single SAR image chips in SICD format; that the performer is very well qualified (e.g., PhD, research level) and has demonstrated experience classifying complex SAR imagery. Qualification requires a scientist's extensive knowledge of SAR signal and image processing, SAR operations from space, and deep learning (machine learning) understanding and implementation. The purpose of this documentation is to prove that the underlying technology is mature and ready to transition immediately into Phase II requirements.
PHASE II
Develop a VLM based SAR SICD Imagery classification system, where input contains SAR imagery with multiple targets and output should be location and classification of the targets. Use SAR SICD data provided by government or your own/commercial data.
PHASE III DUAL USE APPLICATIONS
Transition the VLM based SAR SICD classification tool for government utilities.
REFERENCES
https://www.nvidia.com/en-us/glossary/vision-language-models/
https://pypi.org/project/sarpy/1.1.7/
W. Schwartzkopf, T. Cox, F. Koehler and R. Fiedler, "Generic Processing Of SAR Complex Data Using the SICD Standard in Matlab," IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium, Yokohama, Japan, 2019, pp. 4438-4440, doi: 10.1109/IGARSS.2019.8898903