River Extraction from High Spatial Resolution Remote Sensing Image Using Machine Learning Approaches

The river is a critical freshwater resource for human civilization and plays a pivotal role in sustaining diverse plant and animal species within the ecosystem. Recognizing the paramount importance of rivers, it is imperative to preserve their healthy ecosystems to ensure sustainable utilization. Using machine learning models, this study uses high-resolution remote sensing data from China's Gaofen-2 satellite (GF-2) to promptly and accurately extract real-time river distribution data. The city of Zhuhai, situated downstream of the Pearl River Basin and characterized by an extensive network of rivers, was selected as the study area. This paper employed the Random Forest (RF) and Support Vector Machine (SVM) models for river extraction. The findings revealed superior performance of the RF model over the SVM model, with an overall classification accuracy (OA) of 97.88% and a Kappa coefficient of 0.9577, relative to SVM's 87.55% and 0.7511, respectively. This study demonstrated a substantial enhancement in classification accuracy compared to prior research utilizing SVM with lower-resolution multispectral remote sensing images. Applying machine learning models based on high spatial resolution images holds promise for providing precise information essential for efficient management and preservation of river resources, thereby serving as a powerful instrument to ensure their scientific exploitation and sustainable development.