Crop Type Classification Using TerraSAR-X in Tokachi District, Hokkaido

This paper describes the classification of agricultural land use based on multi-temporal TerraSAR-X images taken during the vegetation season in Tokachi District, Hokkaido. Support Vector Machine (SVM) is becoming a popular alternative to traditional image classification methods because it performs accurate classifications using small training samples. Nevertheless, some parameter tunings are required. Thereby the approach was adopted after retrieving the most appropriate parameters in this study. Using the sixteen TerraSAR-X images acquired between May 2 and November 5, 2009, the overall accuracies for all classes are 92.9% and 91.7% for HH and VV polarization respectively. In addition, the almost perfect agreements ( Kappa >0.80) were achieved by using the 4 scenes (acquired on June 4, June 26, July 7 and July 18) for both polarizations. However, more scenes (6 scenes for HH, 8 scenes for VV) were required in order to discriminate maize fields.