Urban Waterlogged Road Detection under Rainfall Scenarios

Urban road flooding caused by rainfall can severely disrupt traffic as citis expand. Detecting waterlogged road sections is crucial for maintaining smooth traffic and artificial drainage preparation. However, traditional machine learning methods are not effective in achieving high accuracy for waterlogged road recognition, while deep learning methods require significant time to build datasets or deploy multiple waterlogged depth sensors on each road. The present study employs unsupervised learning methods to promptly construct a dataset by detecting data anomalies and tagging them as abnormal results. Then, four hypotheses regarding waterlogged roads are proposed based on the China National Standards on road congestion standards, everyday experiences, and social sensing data. A labeled dataset is constructed from these hypotheses. Finally, a deep-learning model based on RNN (DRNN) is employed to intelligently identify urban waterlogged roads. The experimental findings indicate that on the urban waterlogged road dataset, the model has a recall rate of 93% for waterlogged state recognition. This study provides a method for quickly constructing a road waterlogging dataset based on anomaly detection and helps city managers identify waterlogged urban roads, which can significantly reduce hardware resource costs and keep traffic flowing.