This paper extends general radial basis function networks (RBFN) with Gaussian kernel functions to generalized radial basis function networks (GRBFN) with Parzen window functions, and discusses applying the GRBFNs to recognition of radar targets. The equivalence between the RBFN classifiers (RBFNC) with outer-supervised signals of 0 or 1 and the estimate of Parzen windowed probabilistic density is proved. It is pointed out that the I/O functions of the hidden units in the RBFNC can be extended to general Parzen window functions (or called as potential functions). We present using recursive least square-backpropagation (RLS–BP) learning algorithm to train the GRBFNCs to classify five types of radar targets by means of their one-dimensional cross profiles. The concepts about the rate of recognition and confidence in the process of testing classification performance of the GRBFNCs are introduced. Six generalized kernel functions such as Gaussian, Double-Exponential, Triangle, Hyperbolic, Sinc and Cauchy, are used as the hidden I/O functions of the RBFNCs, and the classification performance of corresponding GRBFNCs for classifying one-dimensional cross profiles of radar targets is discussed.
A Cartesian grid technique based on one-dimensional integrated radial basis function networks for natural convection in concentric annuli
