Automatic Waterline Extraction of Large-Scale Tidal Flats from SAR Images Based on Deep Convolutional Neural Networks

In this study, we proposed an automatic waterline signature extraction method based on deep convolutional neural networks (DCNNs). Our objective is to provide a rapid and straightforward to use method that can tackle the waterline signature extraction from large-scale tidal flats in Sentinel-1 SAR images without re-training or manual interference. The statistical results show this DCNN-based method has appreciable accuracy for efficient extraction of waterline in SAR images even under complex imaging conditions (the mean precision and recall are 0.81 and 0.88, respectively), implying that this method is potential for rapid analysis of tidal flat topography evolution by using the waterline method.