A study of PDF based approach to classify full polarimetric radar data

Accurate and precise statistical modeling of the radar data is an important problem. Probability density function (PDF) is a versatile statistical tool to provide functional description of the data. In this paper, a new PDF based approach is proposed to classify full polarimetric radar data into various land cover classes. Samples from each class are fitted with continuous domain PDFs to obtain statistical signatures. Pearson´s Chi-squared goodness of fit (GOF) test is used to identify the best PDF. Decision rules are formulated based on selected PDFs in a novel and intuitive manner. The proposed approach is applied on ALOS PALSAR data and is verified by computing various figures-of-merit.