Radar image reconstruction based on neural net models

A novel method for microwave diversity radar imaging based on neural net models was developed. It is based on reconstructing a function f( theta , x, y), called the range-profile, which is a one-dimensional (1-D) Fourier transform of a single p-space line (i.e. of a single frequency-response measurement made at a given aspect angle theta ) weighted by mod p mod , where p is the frequency variable. The issue is how to reconstruct f( theta , x, y) as accurately as possible from the incomplete frequency-domain data. The method utilizes the available frequency-response information and makes no assumption about unavailable frequency components, rather than assuming that they are zero as in conventional techniques. The algorithm developed has been successfully tested by simulation and experiments. Results are presented for an experiment involving two metallic cylinders.<>