An adaptive Bayesian approach towards a robust classifier for UAVs and birds

In this thesis the problem of classification of birds and Unmanned Aerial Vehicles (UAVs) using a model-based Bayesian approach is considered. The conventional way of discriminating between birds and UAVs is based on the micro-Doppler signature which is induced by the micro motions of the target, such as the motion of wings and rotor blades for birds and UAVs respectively. The model-based Bayesian approach is able to automatically classify targets and learn from experience. Hidden Markov models are developed based on the radar return model for the target and the associated class likelihood functions are derived. Maximum likelihood estimation is performed to estimate unknown parameters, which are subsequently used for classification. Unsupervised data are used to learn class dependent parameters by applying the learning technique called Maximum Likelihood Adaptive Neural System (MLANS). This approach does not require any preprocessing of the radar return signals and can simultaneously learn and classify. Moreover, the approach is robust with respect to uncertainties on parameter values, such as the initial position of the blades. The classification algorithm is tested on synthetic data and is shown to be capable to classify birds and UAVs with a $95 % probability.