Classification of cultivated rice fields in northern Vietnam using polarimetric RADARSAT-2 data

摘要

This paper presents two classification schemes to identify rice fields in a complex land-use watershed in northern Vietnam: Thresholding and Support Vector Machine (SVM) classification algorithm. The data used are C-band dual-polarization (HH, HV) and polarimetric (quad-pol) data from RADARSAT-2. Two questions are posed: firstly, what is the usefulness of different polarizations (HH and HV) for rice detection? Secondly, between different polarimetric parameters, which ones are the most suitable to separate the rice fields from others? The analysis shows that the rice intensity values increase in the middle of the crop season and are different from other vegetation types in HH polarization. In parallel, the coherence (T) matrix seems a suitable polarimetric parameter for rice extraction. Results suggest that both HH polarization intensity values and quadpol data could identify rice fields at regional scale with a precision of 71% and 80%.