Radar Image Modelling and Detection Using Neural Networks

Abstract

Apparently random behaviour of a deterministic nonlinear dynamical system is referred to as chaos. Chaotic systems arise naturally in many circumstances. If we assume that sea clutter is the result of a chaotic process, we can apply an alternative approach to clutter elimination in radar signals. A neural network can be used to model the underlying system dynamics; in this case a radial basis function (RBF) network used. The radar signal is input to the network, producing a single step prediction. An error signal is generated by comparing each network prediction with the next element of the actual radar signal. As a target will not conform to the same dynamics as the clutter, a large prediction error should be observed when a target is present in the signal. Classical detection schemes are applied to the error signal to implement target detection. This approach has been tested using data collected at Osborne Head, Nova Scotia, Canada, by an instrumental quality X-band coherent radar. Quantization error limits the prediction accuracy, but the RBF is capable of reaching the best prediction for both temporal data and spatial data produced by a radar sweep through a range of azimuth. The RBF predictive detector is shown to be efficient in detecting small targets in sea clutter.