Analysis of High Resolution Polarimetry Data of Static Targets in Automatic Target Recognition Context
Abstract
Reduction of false alarm with acceptable accuracy of classification rate is a challenge in Automatic Target Recognition (ATR) using Synthetic Aperture Radar (SAR) images. This report addresses the evaluation of polarimetric techniques features, benefits of applying Two-Dimensional Fourier Transform on the polarimetric features, and the technique of training data selection to improve the classification accuracy and to reduce the false alarm in small stationary targets of high-resolution full polarimetric SAR images. The Pauli and Cameron decompositions, Self Organizing Maps, and Span techniques are applied on the polarimetric data and then Two-Dimensional Fourier Transform is applied to improve the performance. Two types of training data (one with samples of target only and the other with samples of target and Not-a-target) are used to train the Holographic Neural Technology (a neural network) classifier. The results show the Self Organizing Map feature extraction technique with Fourier Transform algorithm has a better classification rate and low false alarm. The ATR system trained with samples of target and not-a-target, produced low false alarm compared to the one trained with samples of target alone.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2007
- Accession Number
- ADA479331
Entities
People
- Chen Liu
- Nicholas Sandirasegaram
Organizations
- Defence Research and Development Canada