WATSEN: Water Segmentation with Neural Networks¶
WATSEN is a neural network approach to automatically transform CCTV video sequences to information about the local flood level evolution during a flood event. The use of a neural network is expected to be well suited to the low video quality and diversity of appearances that water can have. The method proposed is evaluated with data from controlled flood experiments, in which video and sensor data were collected in parallel.
The network architecture is that of U-Net as proposed by Ronneberger et al.. The CNN architecture code is largely reused from Radio Frequency Interference mitigation using deep convolutional neural networks . See the original GitHub repo for more information.