Electromagnetic Signal Processing for Feature Extraction and Classification of Lossy Dielectric Targets

Turhan-Sayan G., Barut B. I.

Progress In Electromagnetics Research Symposium, Stockholm, Sweden, 12 - 15 August 2013, pp.184-187 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • City: Stockholm
  • Country: Sweden
  • Page Numbers: pp.184-187
  • Middle East Technical University Affiliated: Yes


This paper presents the application of a real-time electromagnetic target classification technique to recognize dispersive dielectric objects with varying loss characteristics. The suggested classifier design technique is operative at the resonance region and is based on the use of Singularity Expansion Method (SEM) to represent a given electromagnetic scatterer by its natural response. A multi-aspect database of wide band scattered signals are processed to extract target features with the ultimate aim of target characterization in an aspect invariant manner. The Wigner distribution (WD), a quadratic time-frequency transformation, is used to extract pole-related feature vectors from non-stationary scattered signals. The Principal Components Analysis (PCA) is further used to fuse multi-aspect feature vectors to obtain a single unified feature vector for each library object. A suitable late-time design interval needs to be chosen to obtain fused features with minimized aspect sensitivity that is a vital requirement for improved classifier accuracy.