A Comparative Study of Artificial Neural Networks and Naïve Bayes Techniques for the Classification of Radar Targets

Abstract

The classification of radar targets is one of the most important study topics, especially in the defense and automotive industries. However, in most of the studies in the literature, raw radar signals are used. Raw radar signals may be subject to ambient noise and signal modulation effects. This may make it difficult to classify radar targets. In this study, instead of using raw data, Fourier-based features extracted from Radar Cross-sectional Area have been used. These extracted features are then input to two types of classifiers, ie, Naive Bayes (NB) and Artificial Neural Networks (ANN) for the classification of radar targets. In addition, both classifiers were trained with different algorithms and their performances were compared. In the ANN-based classifiers, the best accuracy was found that 96.69% with using Bayesian regularization and back propagation training function. On the other hand, the best accuracy with the NB classifier was achieved at 93.95% using Epanechnikov Kernel Distribution. The result presented here demonstrates that Fourier transform based feature extraction can be used effectively in radar target classification applications.