Contribution of PolInSAR Parameters of the Coherence Region to Crop Classification with Bistatic TanDEM-X Data

This study investigates the contribution of PolInSAR features, which correspond to characteristic parameters of the coherence region, in crop-type classification based on single-pass TanDEM-X bistatic data. An agricultural area in Sevilla, Spain, in which 13 crop types are cultivated, is selected for the experiment. Crop classification is carried out by means of a random forest classifier. Accuracy metrics demonstrate the suitability of the PolInSAR observables for this purpose. A maximum accuracy of 90% and 94% at pixel and field level is obtained, respectively, when all the evaluated PolIn-SAR parameters are employed jointly as input features. Notably, some PolInSAR feature subsets, such as the coherence measured at the Pauli channels or the foci of the ellipse which represents the coherence region, yield accuracy levels very close to these maxima. This demonstrates their sensitivity to the crop structure, which is relevant for classification purposes.