Spectral Data Filtering to Increase its Discriminative Power

The methodology of two-criteria optimization of comparative analysis of nonstationary spectral data observed against the background of normally distributed and uncorrelated additive interference is considered. The data is subjected to both linear and nonlinear filtering. The dependence of the similarity measures of the observed and reference spectral data on the linear filtration coefficients is found by maximizing the criterion of the likelihood ratio of the presence and absence of reference data, and the coefficients themselves by maximizing the criterion of discriminating ability: the ratio of similarity measures between the same and different data. The proposed methods for linear filtering provide the same reliability of their distinction as the well-known correlation methods of consistent filtering, but with exceeding the threshold level of interference. With nonlinear filtering, the effect of "over-discrimination" is obtained, when the reliability of distinguishing data is ensured by exceeding the threshold level of interference several times compared with linear filtering