Kunyu Lai, Qinghua Xie, Q. Dou
Sep 22, 2021
2021 SAR in Big Data Era (BIGSARDATA)
Polarimetric Synthetic Aperture Radar (PolSAR) is sensitive to vegetation shape and structure, hence it has been frequently used in crop type identification. Many experimental studies in previous literature proved that increasing the number of temporal SAR images (time dimension) or polarimetric feature parameters (feature dimension) can improve the accuracy of SAR crop identification. However, the comparative analysis of the impact of these two factors on the accuracy of temporal SAR crop classification is still a problem. To fill this knowledge gap, this paper collected a total of 11 Sentinel-1 dual-polarization SAR data and 5 RADARSAT-2 full polarization SAR data for testing purposes. The widely used random forest classification method was adopted for crop classification research, and the importance of the number of SAR images and the number of polarimetric features were explored. Compared with dual-polarization SAR data, the overall classification accuracy in a single month with fully polarimetric SAR data can be improved by 17.94% at most. Moreover, the overall classification accuracy using only 5 RADARSAT-2 images can reach at 95.13%, which is 3.05% and 0.84 % higher than that of 5 and 11 Sentinel-1 data, respectively.