Comparison of Compact Polarimetric with Full Polarimetric Radar Data for Land Use Discrimination Based on Svm Classification

This study comes within the framework of the global cartography and inventory of the Polynesian landscape. An AIRSAR airborne acquired fully polarimetric data in L band, in August 2000, over the main Polynesian Islands. This study focuses on Tubuai Island, where several ground surveys allow the validation of the different results. While they preserve some of the polarimetric information as those that would be recorded by a full polarimetric (FP) radar sensor, compact polarimetry architectures are relevant for systems constraints reduction. Focus is put on the “!/4” mode that is simulated from FP data. It has been shown that this mode is particularly efficient for applications dealing with distributed targets like land use classification. In this study, the SVM (Support Vector Machine) algorithm is used as classification method due to its ability to handle linearly non separable cases by using the kernel method. In particular, it is well suited for combining numerous heterogeneous indicators such as intensity channels, polarimetric descriptors, or textural parameters. The results show that for full polarimetric data, the SVM classification performance when only the elements of the polarimetric coherence matrix are involved is comparable to the Wishart classification one. The addition of polarimetric indices improves significantly the classification. On the other hand when “!/4” mode is simulated, the overall classification performance is similar (" lower of 3%) than those observed with full polarimetric data, with a higher confusion for the Pinus class. Moreover, the “!/4” mode shows much better performance for the land use discrimination of the studied scene than ENVISAT Alternate Polarisation modes involving intensities acquired in co or cross polarization.