Radar signature database validation for automatic target recognition
AssigneeThe United States of America as respresented by the Secretary of the Air Force
InventorsJohn A Malas, Krishna Pasala
Life of this patent
Estimating the term…
Abstract
An information theoretic method for testing and/or validating the suitability of a multi-radar signature database to be used on radar systems having automatic target recognition. The database may include measured data and/or modeled synthetic data. The technique allows measured data to be compared to the synthetic data using modal mutual information. The present invention further includes an information theoretic method for real time calculation of automatic target recognition using modal mutual information calculation.
Claims16
A method for assessing radar signature databases for use in training a radar target recognition decision algorithm, the method comprising the steps of:
- generating radar signature data from at least three targets using a first process and a second process, wherein the radar signature data includes at least one of measured radar data and synthetic radar data, and wherein the targets comprise at least two different target classes, at least two targets being in a same target class and at least one target being in a different target class;
- producing a first radar signature database using the first process and a second radar signature database using the second process;
- comparing the first radar signature database and the second radar signature database using an information theoretic method;
- computing an amount of cumulative mutual information between the first radar signature database and the second radar signature database for targets comprising the same target class and for targets comprising different target classes, wherein the cumulative mutual information comprises a sum of modal mutual information, MMI, derived from comparison of radar signature data from at least two targets, whereby numerical instabilities often incurred in computation of cumulative mutual information are eliminated through incremental computation of MMI;
- and
- based on the amount of cumulative mutual information between the first radar signature database and the second radar signature database, determining whether the first process and second process are suitable for use together to generate a hybrid radar signature database for training the radar target recognition decision algorithm.
A method of using cumulative modal mutual information to identify real-time target measurements using automatic target recognition, the method comprising the steps of:
- generating within an onboard radar system a first radar signature process measurement derived from a real-time target measurement, wherein the first radar signature process measurement is generated through a mapping to a fixed measurement aspect angle to a field target and a range of frequency sub-bands centered at a center reference transmit frequency;
- producing a second radar signature process derived from a hybrid signature database, wherein the hybrid signature database comprises a combination of radar signature data derived from measured radar data and synthetic radar data;
- comparing the first radar signature process measurement and the second radar signature process using an information theoretic method;
- computing in real-time an amount of cumulative mutual information between the first radar signature process measurement and the second radar signature process, wherein the cumulative mutual information comprises a sum of modal mutual information, MMI, derived from comparison of the first radar signature process measurement and the second radar signature process, whereby numerical instabilities associated with computing cumulative mutual information are eliminated through incremental computation and combination of MMI;
- and
- based on the amount of cumulative mutual information between the first radar signature process measurement and the second radar signature process, determining an identity of the field target.
Dependencies are inferred from the claim text.