The problem of classifying radar clutter as found on air traffic control radar systems is studied, and an algorithm is developed to carry out this classification automatically. The basis for the algorithm is Bayes decision theory and the parametric maximum a posteriori probability (MAP) classifier. This classifier employs a quadratic discriminant function and is optimum for feature vectors that are distributed according to the multivariate normal density. Separable clutter classes are most likely to arise from the analysis of the Doppler spectrum. Specifically, a feature set based on the complex reflection coefficients of the lattice prediction error filter (PEP) is proposed. These coefficients are also used in the maximum entropy method (MEM) of spectral estimation, and this link establishes many of their characteristics. A number of transformations are necessary, however, before they can be used as features. The classifier is thoroughly tested using data recorded from two L-band air traffic control radars at different sites. The collected data base contains extensive bird, rain, and ground clutter, as well as thunderstorms, aircraft and ground-based moving vehicle echoes. Their Doppler spectra are examined; and the properties of the feature set, computed using these data, are studied in terms of both the marginal and multivariate statistics. Several strategies involving different numbers of features, class assignments, and data set pretesting according to Doppler frequency and signal-to-noise ratio, were evaluated before settling on a workable algorithm. Final results are presented in terms of experimental misclassification rates and simulated and classified PPI displays.

Doctor of Philosophy (PhD)