Comparative Analysis of Artificial Neural Network and XGBoost Algorithm for PolSAR Image Classification

摘要

Image classification has become an important area of research in remote sensing. In this paper, the algorithms Artificial Neural Network (ANN) and Extreme Gradient Boosting (XGBoost) are used to classify compact polarimetric (CP) RISAT-1 cFRS mode data for land cover categorisation over Mumbai region. After preprocessing, Raney decomposition technique was applied to obtain the R, G, B channels of the image. Hyperparameter tuning of ANN was also performed to get the optimal parameters for the classification. Comparative analysis showed that both the algorithms showed almost equal performance on the data in terms of accuracy. However, there was only 1% of the increment found in both the train and test the accuracy of XGBoost classifier. ANN method required tuning, and thus it requires more time for computation while XGBoost algorithm works well without any tuning and thus, XGBoost outperforms the image classification task for CP RISAT-1 data than ANN.