Deep Learning-Driven Culvert Monitoring: A Novel Approach for Flash Flood Mitigation through Visual Blockage Analysis

The primary cause of urban flash floods is often cited as trash clogging culverts. Flash floods can be avoided with the help of intelligent video analytic (IVA) methods that can extract information about blockages in order to make maintenance-related decisions in a timely manner. Knowing the percentage of culverts that are visually blocked can help prioritise maintenance at heavily blocked locations. In this paper, we introduce a deep learning-powered segmentation-classification pipeline for automatically detecting and segmenting culvert openings and then classifying them into one of four categories based on the degree to which they are visually blocked. The deep learning models underwent training using datasets sourced from the Visual Hydraulics Blockage Dataset (VHD) and Images of Culverts Blockage (ICOB). The outcomes revealed that Mask R-CNN attains the highest performance in segmentation, achieving an mAP@75 score of 77.2%. In contrast, NASNet excelled in classification tasks, with a remarkable 81.2% accuracy on the test data. To underscore the practical importance of these findings, one potential application is demonstrated, which involves using the models for monitoring visual obstructions.