Crop Classification by Machine Learning Algorithm with Combination of X- and C-band SAR Data

Abstract

A crop classification method using satellite data is proposed as an alternative to the existing ground survey. In this study, crop types were classified using two kinds of SAR data (i.e., TerraSAR-X X-band dual-polarization data and Radarsat-2 C-band fully-polarization data) and Random Forests. Sigma naught polarimetric parameters were calculated from SAR data and classifications were conducted using the following four different datasets ; Case 1 : all parameters calculated from Radarsat-2, Case 2 : all parameters calculated from Radarsat-2 and sigma naught calculated from TerraSAR-X data, Case 3 : all parameters calculated from Radarsat-2 and polarimetric parameters calculated from TerraSAR-X data, and Case 4 : all parameters calculated from Radarsat-2 and both sigma naught and polarimetric parameters calculated from TerraSAR-X. The highest overall accuracy of 0.934 was achieved by Case 4, and there were significant differences with the other classification results (p>0.05, based on Z-test). These results reveal that combining two kinds of SAR data can be improved classification accuracy.