A Novel Method for Monitoring River Level Changes under Bridges with Time Series SAR Images

River level monitoring is crucial for hydrological studies, providing essential information for flood forecasting, water resource management, and environmental protection. In this article, we present a novel method for monitoring river level under bridges using time-series Synthetic Aperture Radar images. First, we transfer a DeepLab V3+ network model for road segmentation to bridge segmentation, fine-tuning it with bridge scattering signal data, while a new loss supervision function CentroidLoss, has been added to the model to improve the integrity of the bridge signal segmentation. Furthermore, the Energy Accumulation Algorithm (EAA) is proposed to improve the accuracy of river level measurements in areas of low signal-to-noise ratio with noise such as ships and waves. Leveraging deep learning and EAA, the proposed approach accurately extracts bridge scattering signals and precisely estimates the peak positions of the bridge's multiple scattering signals, enabling precise river level monitoring. Sentinel-1A and COSMO-SkyMed data were applied as inputs to our method, and the comparison between the river levels measured by the proposed method and those of local hydrological stations reveals submeter level estimation accuracy.