Machine Learning Based River Segmentation with GF-2 Satellite Imagery

Rivers constitute a crucial element of the natural environment, playing an indispensable role in ecosystem dynamics. Investigating river systems not only facilitates the sustainable management of water resources but also advances ecological preservation, alongside social and economic development. Considering escalating global challenges such as climate change, population pressures, and environmental degradation, acquiring a comprehensive understanding of river resources is imperative for addressing both current and future ecological dilemmas. This study employs high-spatial-resolution remote sensing imagery Gaofen-2 (GF-2) alongside two deep learning models: k-Nearest Neighbor (kNN) and Decision Tree (DT), to conduct river segmentation within the context of Zhuhai, a city characterized by abundant river resources. The research utilizes a 10-fold cross-validation approach and the grid search method to ascertain the optimal parameters for each model, aiming to enhance training accuracy. Comparative analysis indicates that the kNN model outperforms the DT model across various evaluation metrics and segmentation outcomes. The findings of this research contribute valuable theoretical insights and technical support for effective river management strategies, enabling governmental bodies to formulate more scientifically grounded and rational policies and measures for sustainable development.