Performance model for synthetic aperture radar automatic target recognition and method thereof
First Claim
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.
1 Assignment
0 Petitions
Accused Products
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.
4 Citations
18 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. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-
-
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.
-
Specification