1-D feature extraction using a dispersive scattering center parametric model

Automatic object recognition, the ability for a machine to classify an object it has detected, is an area of continual interest. The classification of objects requires extraction of discriminatory information, and this process is called feature extraction. The extractable features are inextricably tied to the type of sensor used, and this work addresses one-dimensional, high resolution radar (HRR). Several approaches have been proposed for feature extraction in HRR. One approach is based on finding sinusoids in noise by a sum-of-damped exponentials (DE) model. The features of the DE-based scattering models, range and amplitude, have been used for feature extraction, but the DE model simplistically assumes that all scattering centers on the object have the frequency response of an ideal point scatterer. The high potential improvement to the damped exponential model lies in estimating the frequency response of each scattering center uniquely. This improvement is based on the geometric theory of diffraction of (GTD) and the uniform theory of diffraction (UTD) and includes electromagnetic scattering theory as a basis for scattering center modeling. Research in adapting the DE models to include GTD/UTD has been performed, but has had limited implementation in object recognition problems. This work provides an analysis and a proof-of-concept for implementing the GTD/UTD-based dispersive scattering center (DSC) model to perform object recognition. The DSC model presented is based directly upon Potter´s (see IEEE Trans. Antennas Progagat., vol.43, p.1058-67, 1995) work at the Ohio State University.