A maximum likelihood detection of signals using feature mapping framework

In Xu and Reed a matched-filter based detector was developed for the problem of detecting a 2-D target signal where prior information about the target pattern or template as well as the statistical properties of the clutter is limited. This was accomplished by an ad hoc substitution of the maximum likelihood estimate (MLE) of unknown clutter covariance matrix and the MLE´s of the complex amplitudes of the significant features components of target into the matched filter test. The present paper provides a new approach for the problem based on the generalized likelihood ratio (GLR) principle which maximizes the GLR function over unknown clutter covariance matrix and the unknown significant feature components of target signal to be detected. This new GLR test is compared with the matched-filter based test in Xu and Reed for performance. The feature mapping and representation which can be incorporated into the test to characterize the unknown target pattern are various, including the short time Fourier transform, the discrete cosine transform, and the discrete wavelet transform