Evaluation of L-band Polarimetric SAR Images for Urban Land Cover Classification

Several polarimetric characteristics are extracted from ALOS PALSAR data,containing coherent and non-coherental characteristics.Then,through qualitative and quantitative indicators,their abilities to distinguish five typical urban land covers including artificial features,bare land,farmland,woodland and water are analyzed and compared.The experimental results show that circular polarization correlation coefficient,linear polarization correlation coefficient,total power and XPI are the best set of polarization characteristics to classify these land covers.Classification using such a group of characteristics can achieve an overall accuracy of 75.5% with Kappa being 0.651 1.In addition,we find that artificial features are easier to be distinguished from other land cover types,while bare land and farmland are prone to be mixed up.