二偏波合成開口レーダー画像を用いた耕作放棄地の推定

抄録

Dual-polarimetric SAR (Synthetic Aperture Radar) was used for estimating abandoned rice paddies. First, ground validation data was prepared based on the data reported in a previous study and a google earth image. Second, a dualpolarimetric SAR image taken just after water filling in rice paddies was prepared and the lexicographic covariance matrix was calculated in each pixel from the scattering elements of the SAR image. Third, classification of three categories (rice paddies, abandoned rice paddies, residential area) was conducted to the image based on each element of the covariance matrix. For the comparison of the classification, the classification was also conducted using other two satellite images, i.e. the ones of AVNIR-2 and single-polarimetric SAR. As a result, overall accuracy and kappa were 55%, 63% and 22% and 0.32, 0.48, -0.04 for dual-polarimetric SAR, AVNIR-2, and single-polarimetric SAR images, respectively. In the dual polarimetric SAR image, abandoned rice paddies were overestimated, while rice paddies were underestimated.