Performance model for synthetic aperture radar automatic target recognition and method thereof

• US 8,681,037 B2
• Filed: 04/28/2011
• Issued: 03/25/2014
• Est. Priority Date: 04/28/2011
• Status: Active Grant

• Current Assignee: Raytheon Company (Rtx Corporation)
• Original Assignee: Raytheon Company (Rtx Corporation)
• Inventors: Doria, David M.
• Primary Examiner(s): Brainard, Timothy A
• Application Number:
  US13/096,913
• Publication Number:
  US 20120274506A1
• Time in Patent Office: 1,062 Days
• Field of Search: 342/25, 342/189, 342/190
• US Class Current: 342/25.R
• CPC Class Codes: G01S 13/9027 Pattern recognition for feature extraction
• CPC Class Codes: G01S 7/412 based on a comparison between measured values and known or stored values
• CPC Class Codes: G06F 18/2415 based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate

Abstract

A target correlation matrix is generated for multiple two-class combinations of target types each having a target correlation and a synthetic aperture radar observation space. A target probability density of a target radar cross-section signature and a background probability density of a background radar cross-section signature are utilized. The observation space of each of the two-class combinations is partitioned into a target partition and at least one background partition in accordance with the target correlation. A conditional log likelihood is calculated using at least one random number for each of the partitions in accordance with the target probability density and the background probability density, and summed according to the two-class combinations. A maximum log likelihood is calculated from the summed conditional log likelihoods given that one target type of the multiple two-class combinations is assumed to be true. An automatic target recognition performance prediction based on the maximum log likelihood is generated.

Claims

1. A computer-implemented method for predicting automatic target recognition performance, the method comprising:

generating a target correlation matrix each having a target correlation and a synthetic aperture radar observation space for multiple two-class combinations of target types;

generating a target probability density of a target radar cross-section signature and a background probability density of a background radar cross-section signature;

partitioning the observation space of each of the two-class combinations of target types into a target partition and at least one background partition in accordance with the target correlation;

calculating a conditional log likelihood for each of the partitions using at least one random number in accordance with the target probability density and the background probability density;

summing the conditional log likelihoods of the partitions according to each of the two-class combinations of target types associated with the correlation matrix and observation space;

calculating a maximum log likelihood from the summed conditional log likelihoods given that one target type of the multiple two-class combinations of target types is assumed to be true;

generating an automatic target recognition performance prediction based on the maximum log likelihood; and

outputting the performance prediction via an output device.

2. The method of claim 1, further comprising:

assigning the target probability density to the target partition and the background probability density to the at least one background partition.

3. The method of claim 1, wherein the target correlation matrix is based on a geometric similarity between a first target type of one of the two-class combinations of target types and a second target type of the one of the two-class combinations of target types.

4. The method of claim 1, wherein the observation space is partitioned by projecting a geometric representation of a first target type of one of the two-class combinations of target types over a geometric representation of a second target type of the one of the two-class combinations of target types, shifting the first target type in a first direction in accordance with the target correlation and shifting the first target type in a second direction opposite the first direction in accordance with the target correlation.

5. The method of claim 4, wherein the target partition comprises an overlapping portion between the first target type and the second target type, a first background partition comprises a first non-overlapping portion between the first target type and the second target type after the first target type is shifted in the first direction, and a second background partition comprises a second non-overlapping portion between the first target type and the second target type after the first target type is shifted in the second direction.

6. The method of claim 5, wherein the observation space is a union of the target partition, the first background partition and the second background partition.

7. The method of claim 6, wherein the partitions comprise resolution elements having a size determined by a quotient of the observation space and a synthetic aperture radar resolution.

8. The method of claim 7, wherein the at least one random number is generated for each of the resolution elements.

9. The method of claim 1, wherein the simulated target radar cross-section signature and the simulated background radar cross-section signatures are generated from a probabilistic model of the radar cross-section measurement and the radar background cross-section measurement, respectively.

10. The method of claim 1, wherein the target probability density is modeled by diffuse and specular components of the target radar cross-section signature.

11. The method of claim 10, wherein the diffuse component comprises Rayleigh parameters and the specular component comprises Rician parameters.

12. The method of claim 1, wherein the background probability density is generated from clutter and shadow components of the background radar cross-section signature.

13. The method of claim 12, wherein the clutter and shadow components comprise Rayleigh parameters.

14. The method of claim 1, wherein the target partition and the at least one background partition for each of the two-class combinations of target types are compiled into a set.

15. The method of claim 14, wherein the at least one random number is generated by Monte-Carlo simulation for each of the members of the set.

16. The method of claim 1, wherein the observation space is partitioned, for each pairwise combination of types of the correlation matrix, in accordance with the vector outer product of the partitions in each generalized representation of the union of targets and background, for each combination of target types.

17. A system for predicting automatic target recognition performance, comprising:

a target correlation matrix generator configured to generate a target correlation matrix for multiple two-class combinations of target types each having a target correlation and a synthetic aperture radar observation space;

a probability density generator for generating a target probability density of a target radar cross-section signature and a background probability density of a background radar cross-section signature;

a partition generator for partitioning the observation space of each of the two-class combinations of target types into a target partition and at least one background partition in accordance with the target correlation;

a random number generator for generating at least one random number for each of the partitions in accordance with the target probability density and the background probability density, wherein the target partition is associated with the target probability density and at least one background partition is associated with the background probability density;

an automatic target recognition performance calculator for calculating a conditional log likelihood using the at least one random number for each of the partitions in accordance with the target probability density and the background probability density, summing the conditional log likelihoods of the partitions according to the two-class combinations of target types, calculating a maximum log likelihood from the summed conditional log likelihoods given that one target type of the two-class combinations of target types is assumed to be true, and generating an automatic target recognition performance prediction based on the maximum log likelihood; and

an output device for outputting the performance prediction.

18. A system for predicting automatic target recognition performance, the system comprising:

an output device;

a processor coupled to the output device; and

a memory coupled to the processor and storing computer instructions therein, the processor being operable to execute the program instructions, the program instructions including;

generating a target correlation matrix each having a target correlation and a synthetic aperture radar observation space for multiple two-class combinations of target types;

generating a target probability density of a target radar cross-section signature and a background probability density of a background radar cross-section signature;

partitioning the observation space of each of the two-class combinations of target types into a target partition and at least one background partition in accordance with the target correlation;

calculating a conditional log likelihood for each of the partitions using at least one random number in accordance with the target probability density and the background probability density;

summing the conditional log likelihoods of the partitions according to the two-class combinations of target types;

calculating a maximum log likelihood from the summed conditional log likelihoods given that one target type of the multiple two-class combinations of target types is assumed to be true;

generating an automatic target recognition performance prediction based on the maximum log likelihood; and

outputting the performance prediction via the output device.