Investigation of using linear and quadratic time-frequency representations for electromagnetic feature extraction with reduced aspect sensitivity

The extraction of target features is the most essential part of a classifier design problem as the performance of the classifier is primarily determined by the descriptive power of target-specific features. The feature extraction problem becomes especially challenging in electromagnetic target recognition due to the strongly aspect dependent nature of the scattered data. In this paper, the feasibility of using some other well-known time-frequency representation (TFR) techniques, besides the Wigner distribution (WD), are investigated, following the same fundamental feature extraction procedure introduced in (G. Turhan-Sayan, Pattern Recognition, forthcoming in 2003). Four different TFRs, the Gabor expansion (GE), the short time Fourier transform (STFT), the Page distribution (PD) and the WD, are used for feature extraction and target recognition, based on the same set of target classes and the same reference database for fair comparison of the results.