Sea clutter neural network classifier: Feature selection and MLP design

Abstract

The design of radar detectors in sea clutter environments is really a complex task. A neural network based automatic sea clutter classifier has been designed, as part of an adaptive detector capable of exploiting all the capabilities of detectors designed for specific clutter environments. The most extended sea clutter models have been considered (Gaussian, Weibull and K-distributed). Results show that an MLP with 3 inputs (the variance, the entropy of the modulus of the samples and the correlation coefficient), 6 hidden neurons and 4 outputs, is able to provide a performance similar to the K - NN algorithm with K = 10 with a significant reduction in computational cost, a very important feature in real time applications. © 2013 Springer-Verlag Berlin Heidelberg.