On the training patterns of a neural network for target localization in the spatial domain

In this letter: rye show that some spatial sampling requirements for the sc attered electric field are necessary bt order to obtain good performances w hen a neural-network approach is applied to the solution of target localiza tion problems in the spatial domain. By means of some examples, concerning dielectric cylinders either in free space or buried in a lossy half space, we point out the dependence of the object localization an rite spatial samp ling rate of the scattered field and its relation to the field spatial band width. (C) 2001 John Wiley & Sons, Inc.