The potential of polarimetric radar data in the coarse classification of semi-urban landcovers

Radar sensor developments have shown a transition trend from single to multi polarisation. A literature review revealed that few studies have assessed and quantified the potential of using Quad-Pol over Single-Pol data in supervised land-cover classifications. Two semi-urban Convair C-Band polarimetric datasets were obtained to quantify the potential improvements. The HH/HV/VVintensity layers of the datasets were generated using the European Space Agency's POLSARPRO and classified in various combinations with a Minimum Distance, a Maximum Likelihood and a Neural Network (NN) algorithm. Results demonstrate that Quad-Pol data increases the overall classification accuracy rates of all algorithms by 2 to 12% and confirms the NN algorithm as the preferred method to use for fully polarimetric data. It achieved accuracies of about 75% for both datasets. The results from other methods were less acceptable since they are in the low 60%. The research presented in this report provides a neophyte user some avenues to exploit polarimetric data in the most efficient way possible.;Keywords. SAR, polarimetry, land-cover, classification, neural network.