You are here

AI and PONS for Object Identification and Annotation

Award Information
Agency: Department of Defense
Branch: National Geospatial-Intelligence Agency
Contract: HM047622C0062
Agency Tracking Number: M2D-0020
Amount: $999,964.82
Phase: Phase II
Program: SBIR
Solicitation Topic Code: NGA203-003
Solicitation Number: 20.3
Timeline
Solicitation Year: 2020
Award Year: 2022
Award Start Date (Proposal Award Date): 2022-02-02
Award End Date (Contract End Date): 2024-02-13
Small Business Information
103 Mansfield Street
Sharon, MA 02067-1111
United States
DUNS: 101306678
HUBZone Owned: No
Woman Owned: Yes
Socially and Economically Disadvantaged: No
Principal Investigator
 Richard Tolimieri
 (401) 849-5389
 richie@prometheus-us.com
Business Contact
 Jim Byrnes
Phone: (781) 784-2355
Email: jim@prometheus-us.com
Research Institution
N/A
Abstract

To attack the challenge of object identification and annotation across diverse families of image data, Prometheus and Raytheon will implement a software toolkit based on our new mathematically–based AI tool. The AI input will be matrix coefficients resulting from the unique Prometheus-developed energy-spreading transform, PONS, the Prometheus Orthonormal Set. PONS is currently in use by both the US military, in the form of novel radar waveforms, and commercially by Cisco in their Intelligent Proximity technology. PONS applicability here and elsewhere results because the transform expresses high-resolution signals as sums of special low-resolution signals, where each low-resolution component is both a highly compressed version of the original signal and contains roughly the same amount of information as any other such component. This enables employment of these components as image snippets, thereby greatly reducing the raw data required by the AI engine. Existing real-world applications of PONS transforms, to wireless communications, optical communications, robust transmission of digital data, watermarking, and radar, all rely upon one-dimensional PONS transforms. We have extended the mathematical foundation by creating multidimensional PONS bases, deriving a new tensor decomposition theorem, and showing that the crucial energy spreading property carries over to multidimensional PONS. In the initial portion of this project we will extend the suite of one-dimensional PONS algorithms and FFT-like codes in order to utilize the mathematically proven and published two-dimensional PONS expansions. We will then apply our proven Artificial Intelligence and Machine Learning technologies to identify the same or similar objects in massive image data sets by employing one, or at most two, two-dimensional PONS matrix coefficients for each image.

* Information listed above is at the time of submission. *

US Flag An Official Website of the United States Government