Classification of radar non-homogenous clutter based on statistical features using neural network

This paper presents a robust clutter classifier based on the neural network to assist the radar receiver by choosing optimal constant false alarm rate where this classifier has been trained for 16 classes, four radar return distribution with different situations. The return radar signal distributions are Rayleigh, Weibull, lognormal and K-distribution, while the situations are, signal, multi-target, closed multi-target, and clutter edge. Multilayer perceptron with back-propagation as a neural network with seven features, mean, variance, mode, kurtosis, skewness, median, and entropy, have been used to classify the return signal. A least mean square error is used to evaluate the classifier performance. The simulation is evaluated for the signal to clutter ration from +35 dB to −35 dB, with 5-20 neurons of the hidden layer, and 60-360 samples. By performing, the optimisation has been gained by using 240 samples and 20 neurons then lead to 98.1% return signal classification.