Performance Analysis of Machine Learning Techniques for Water body Extraction

The demand for day-to-day agricultural needs is fulfilled by water conservation in lakes. The level of water content in the lake must be measured in order to forecast yield-to-consumption tradeoff. Therefore, precise extraction and continued monitoring of water resources is very important and urgently needed for agricultural planning. In this paper, pixel-level features are extracted using satellite derived spectral indices and gabor filters. Water and non-water pixels classified based on the derived feature vectors. Four medium resolution multispectral images of Veeranam Lake which captured by Resourcesat-2 satellite in four different years considered for analysis. Machine learning algorithms like Naive Bayes (NB), Random Forest (RF), Support Vector Machine (SVM) and eXtreme Gradient Boosting (XGB) are analyzed for water body extraction. An extensive qualitative and quantitative evaluation of machine learning classifiers are done and results are compared. The comparison shows that XGB classifier outperforms the other machine learning algorithms considered.