Performance results of recognizing various class types using classifier decision fusion

Classification performance is compared using data based on waveforms that are transmitted at both low and high frequencies. The waveforms are made up of one low, and one high, bandwidth type. Various features have been extracted from each waveform for training different classifiers. Specifically, the classifier types consist of various neural networks, Weighted voting (Linear), Fisher's Linear discriminant, the Expectation Maximization algorithm, and the Bayesian Data Reduction Algorithm. The contribution of this paper is to show that overall classification performance improves if the decision outputs of the individually trained classifiers are fused using majority voting. Also, it is shown that classification performance depends on the transmitting carrier frequency of the waveforms and the specific configuration of the classes used to train each classifier.