PERFORMANCE OF RESONANT RADAR TARGET IDENTIFICATION ALGORITHMS USING INTRA-CLASS WEIGHTING FUNCTIONS (TARGET CLASSIFICATIONS, PATTERN, NOISE)

This dissertation is concerned with the use of calibrated resonant-region radar cross-section (RCS) measurements of targets for the classification of large aircraft.; Errors in the RCS estimate of full scale aircraft flying over an ocean, introduced by the ionospheric variability and the sea conditions were studied.; The "Weighted Target Representative" (WTR) classification algorithm was developed, implemented, tested and compared with the nearest neighbor (NN) algorithm. The WTR-algorithm has a low sensitivity to the uncertainty in the aspect angle of the unknown target returns. In addition, this algorithm was based on the development of a new catalog of representative data which reduces the storage requirements and increases the computational efficiency of the classification system compared to the NN-algorithm.; Experiments were designed to study and evaluate the characteristics of the WTR- and the NN-algorithms, investigate the classifiability of targets and study the relative behavior of the number of misclassifications as a function of the target backscatter features.; The classification results and statistics were shown in the form of performance curves, performance tables and confusion tables.; Experimental results showed that the use of vertical polarization, 8.0-25.0 MHz frequency band and 0(DEGREES)-30(DEGREES) aspect angle range produce the lowest number of misclassifications of all the target feature parameters studied. When the number of spectral components exceeded 10 no substantial decrease in the number of misclassifications was produced.; The use of confusion tables proved to be convenient in studying the separability of the targets in the database. Confusion between specific targets which caused a large number of misclassifications were easily seen.; Recommendations for future study on the target representative techniques, weighting function schemes and on composite classification algorithms were included.