Title : Chaotic Prediction and Modeling of Sea Clutter Using Neural Networks

Radar clutter, the unwanted radar echoes, has a long history of being modeled a.s a stochastic process. The main reason for using this model is that radar clutter appears to be very random to our naked eyes. Due to this stochastic assumption, radar detection is based on statistical decision theory. More precisely, the probability distribution of noise or clutter is obtained to derive the likelihood function for making decision. In this thesis, we try to justify the stochastic assumption of radar clutter, in pa.rticular, sea clutter. We find that assuming sea clutter as a random process uses unnecessarily high degrees of freedom. In other words, sea clutter does not have to be modeled by a random process to handle its random behavior. Using different real-life sea clutter data, we show that the random nature of sea clutter is possibly a result of the chaotic phenomenon. After showing that sea clutter is not truely random, we then try to model sea clutter data by a deterministic dynamical system. To construct a useful model for sea clutter, we need to reconstruct the dynamics of sea clutter, and neural network is lIsed here as a tool to achieve this purpose. Two novel neural networks are developed to reconstruct the clutter dynamics. The first one is called ratio~al function neural network which has an unique local minimum and a rapid learning phase. The second network, which uses the idea that sea clutter can be embedded as a manifold, does not require any learning, and is very robust and accurate. Both networks have excellent performances in reconstructing the dynamics of the real-life sea clutter data. The model for sea clutter is then used for detection of small targets in ocean environment. Now detection is no longer a statistic decision problem, but rather a process of distinguishing two different dynamical systems. One dynamical system