Performance Analysis of Atr from Sar Imagery

Synthetic aperture radar (SAR) images are used in a ten-class automatic target recognition system using data from the public release of the MSTAR program. A likelihood ratio-based approach for target recognition is adopted using the conditionally Gaussian model. The target orientation is assumed to be unknown and uniformly distributed. The target orientation is estimated, along with the target class. The data are separated into two parts, the training data and the testing data. Parameters in the conditionally Gaussian models are estimated from the training data. The resulting models are then used in likelihood ratio tests on the testing data. The likelihood functions are highly peaked so that results using maximum a posteriori estimation are nearly identical to those using minimum mean Hilbert-Schmidt squared distance estima-tors (MMSE). For optimally chosen thresholds, the probability of correct recognition on the testing set is 0.984. Most orientation estimates are within ve degrees of the truth.