Naïve Bayesian radar micro-doppler recognition

A comprehensive evaluation of a naive Bayesian classifier used for micro-Doppler signature (mu-DS) radar automatic target recognition has been performed. An initial estimate of performance is made using the Bhattacharyya bound on the error probability that gives results of approximately 60% of the measured value. The classifier input data is pre-processed using the CLEAN algorithm, the Fourier transform and principal component analysis to provide feature vectors exhibiting only mu-DS information. The classifier includes ldquounknownrdquo input rejection that falsely declares known targets with a probability of just 0.07. The probability of correct classification is 0.94.