Urban Area Man-Made Target Detection for PolSAR Data Based on a Nonzero-Mean Statistical Model

Numerous opportunities for advancement exist in the field of man-made target detection using synthetic aperture radar (SAR). With the development of SAR sensors, the high spatial resolution violates the validity of the zero-mean assumption, particularly for urban applications. In 2012, we refined the conventional azimuth stationarity extraction method by adopting a nonzero-mean model-the Rician distribution-to better adapt to the high-resolution SAR imagery of urban areas and improved the man-made target detection result significantly. However, the model cannot make full use of the multipolarization information because it is calculated with the amplitude data. This study extends the method by employing another nonzero-mean model-the multivariate normal distribution-to consider the scattering vector of polarimetric SAR (PolSAR) images. The proposed method achieves an excellent performance with overall accuracy of 88.76%. Two previous methods, including the PolSAR method using the zero-mean model and the single polarization method using the nonzero-mean model, are also involved in the comparisons. The effectiveness of applying nonzero-mean models to urban area SAR images is well demonstrated by experimental results.