Comparison of Various Polarimetric Decomposition Techniques for Crop Classification

Abstract

Differential response of crops to polarimetric radar signals provides the basis for crop discrimination, classification and monitoring. Polarimetric decomposition approaches provide a measure of the relative contributions of backscatter from different scattering mechanisms and hence, the selection of proper decomposition method plays a vital role in the classification of natural distributed targets especially in crop discrimination. The study focused towards assessing the performance of both coherent and incoherent decomposition methods in terms of discrimination of different cover types, mainly the crops like paddy and cotton using their classification accuracies. Decomposition techniques viz., Freeman, van Zyl, Yamaguchi-3 component and Krogager were applied on three Radarsat-2 fully polarimetric SAR data sets (Fine Quad mode) during the crop growth period. The decomposition parameters along with their temporal dynamics were studied over paddy and cotton crops. Temporal dynamics and the field observations revealed that the manifestation of paddy and cotton crops was best in second date of acquisition (26-Oct.-2010), coinciding with the peak vegetative stage of paddy crop and vegetative stage of cotton crop. It is observed that the scattering mechanisms occurring in paddy and cotton crops are distinct. Supervised parallelepiped minimum distance to mean classification was employed to classify the decomposed data. The analysis showed that Krogager and van Zyl gave best class accuracies as well as overall kappa accuracies (K^ of Krogager = 0.7265 and K^ of van Zyl = 0.6804). The crop type viz. paddy and cotton could be better discriminated and mapped from the classification of van Zyl and Krogager decompositions.