Çok zamanlı polarimetrik SAR verileri ile tarımsal ürünlerin sınıflandırılması

Abstract

This study evaluates the use of multi-temporal Polarimetric SAR (Polarimetric Synthetic Aperture Radar, PolSAR) images for crop classification. Multi-temporal polarimetric SAR images could be very advantageous for crop classification especially in time-critical agricultural projects. Within this research, three types of machine learning algorithms (light gradient boosting machines (LightGBM), random forest (RF) and support vector machines (SVM)) were utilized for the classification of five crops (maize, potato, wheat, sunflower, and alfalfa). From the multi-temporal PolSAR data, the original features (i.e. linear backscatter coefficients) of Radarsat-2 were extracted and incorporated into the classification step. The overall classification accuracies were obtained as 0.857 (±0.026), 0.855 (±0.033) and 0.834 (±0.039) for LightGBM, RF and SVM, respectively. The difference between the accuracies obtained by LightGBM and random forest (RF) was found to be statistically non-significant based upon the McNemar’s test. K-fold cross validation was used to assess the classification results. Furthermore, these results showed the added benefits of multi-temporal PolSAR data for crop classification.