Analysis of polarimetric techniques using high-resolution polarimetry data in an automatic target recognition context

In recent years, large numbers of radar images are collected but there is neither time nor enough manpower to go through each collected image. Researchers in the automatic target recognition (ATR) field have developed automated algorithms and tools to analyse each image and obtain higher recognition rate and fewer false alarms but there is still a need for improvement in these aspects. In this study, we have investigated various polarimetric and non-polarimetric techniques and recommended the best ATR approach among those analysed for higher recognition rate and least false alarm rate. The experimental results show that self-organising map (SOM) feature extraction technique with a two-dimensional Fourier transform (2DFFT) algorithm has a better classification rate and a lower false alarm rate. The classifier used here was AND Corporation´s holographic neural technology (UNeT) classifier. The SOM technique using |HH|, |HV| and |W| achieved 98.9% correct classification over the detected targets and reduced the false alarm rate to 8.2%. An ATR system trained with both target and not-a-target class data produced a lower false alarm rate compared with ATR systems trained with target samples alone. This study will help in selection of appropriate methods for future ATR system implementations. In addition, it will assist image analysts (IAs) in choosing appropriate techniques and training datasets to perform their operational tasks.