Abstract
Crop classification maps are required for the management of crops and for the estimation of agricultural disaster compensation. In this study, classification using TerraSAR-X data (including TanDEM-X) was performed. Applying the m-chi decomposition to the dual-polarized SAR data (HH and VV polarization), the three components, double (even) bounce, randomly polarized and single (odd) bounce, were derived. Then, besides gamma naught (HH and VV polarization) data, the three components were obtained and evaluated regarding their usefulness in crop classification. The comparisons between the Kernels based Extreme Learning Machine (KELM) and Random Forests (RF) algorithms were also performed. It was found that KELM performed better, achieving an overall accuracy of 93.4% based on the three components and gamma naught values for HH and VV polarizations.
