PolSAR Data Classification Using Extreme Gradient Boosting Classifier

This work explored the extreme gradient boosting classifier (XGBoost) to classify land cover using PolSAR (Polarimetric Synthetic aperture radar) data. The XGBoost is a type of Ensemble learning algorithms. These algorithms have attracted many researchers in the classification of remote sensing data since they outperform other Machine Learning ML methods in terms of accuracy. In this work, the compensated Kenaugh matrix's diagonal elements and the Six Components Decomposition Method (6SDM) components were combined and used as a feature vector to feed the classifier. ALOS-2 L band data over the San Francisco, USA was used to test the used classifier's performance. The results show that XGBoost outperformed SVM, KNN, and RF classifiers in terms of accuracy.