Real-time Disaster Reduction Of Urban Flood Using Computer Vision Coupled With High-precision Rain-flood Numerical Simulation

In recent years,with global warming and accelerated urbanization,urban flooding has become a prominent issue that severely affects social production and residents’ lives.The stable operation of urban drainage infrastructure is crucial for reducing the impact of urban flooding and ensuring urban drainage safety.However,the current maintenance of urban surface drainage facilities mainly relies on manual inspections,which are inefficient and prone to missing or incorrect inspections.This often leads to drainage facility failures during heavy rainfall,such as blocked storm drains,storm drains covered by leaves,and poor drainage at discharge outlets.These issues cause urban drainage facilities to lose their drainage capacity,preventing timely removal of surface water and resulting in urban flooding.Additionally,after heavy rainfall,flood control personnel are unable to monitor the operational status of drainage facilities in realtime,making it difficult to accurately assess the flood risk level.This leads to uneven distribution of flood control resources and delayed risk management,severely threatening residents’ property safety and disrupting normal social operations.To address these issues,this paper adopts an interdisciplinary approach to research realtime urban flood disaster mitigation technologies by coupling computer vision and high-precision rainstorm numerical simulations.Before a heavy rainfall event,computer vision recognition technology is employed to intelligently identify the status of drainage facilities,significantly enhancing the operational efficiency of daily urban drainage systems.Concurrently,it is essential to conduct localized urban flood risk assessments to improve disaster prevention and control capabilities.Following heavy rainfall,one-dimensional and two-dimensional hydrological and hydraulic simulations are conducted to model the real-time operation of drainage facilities.By coupling intelligent recognition technology for drainage facility status,high-risk areas of drainage facility operation can be quickly identified,and flood risk levels can be scientifically assessed.This enables efficient and rational allocation of flood control resources,achieving real-time flood disaster mitigation.The main conclusions of this paper are as follows:(1)Given the lack of publicly available annotated datasets for urban surface drainage facilities both domestically and internationally,this study enhances the quality of such datasets.Based on traditional industry standards for classifying defects in surface drainage facilities and leveraging visual features,the YOLOV7 object detection algorithm is used.By optimizing defect classification,iterating model optimization,annotating images,and augmenting data to expand the image dataset,a comprehensive image sample dataset for surface drainage facilities has been constructed.This dataset includes three types of surface drainage facilities(inspection wells,storm drains,and discharge outlets)and 15 types of facility defect states,totaling 34,367 images.This dataset provides a foundation for intelligent identification of drainage facility defect target states.(2)For the first time,a target detector suitable for surface drainage facilities has been constructed.By testing the detection performance of nine detection algorithms from the YOLO series on drainage facility defects,the YOLOV7 object detection algorithm demonstrated significant advantages in recall(0.837)and model detection precision(m AP50 of 0.872).It also meets the performance requirements for real-time defect detection in drainage facilities(FPS of 294.118).By integrating the SE attention module into the backbone network,the model’s detection performance improved further,with an m AP50 increase of 1.03%,recall enhancement of 1.69%,and FPS improvement of 3.03%.Comprehensive analysis led to the selection of YOLOV7+SE as the drainage facility defect target detector,providing algorithmic support for real-time intelligent inspection of drainage facilities.(3)A video stream-based drainage facility target tracking and spatial matching analysis algorithm is introduced.Using YOLOV7+SE as the defect target detector and the Bo T-SORT target tracking algorithm,the defect detection results from YOLOV7+SE serve as inputs for the Bo T-SORT algorithm.This enables the integration of defect target detection across multiple consecutive frames of images.By employing a spatial matching algorithm that limits to the nearest neighbor within a specified radius,the output results of the target tracking algorithm are matched with GIS data of surface drainage facilities.A real-time defect detection system for drainage facilities is developed,achieving real-time upload of vehicle-mounted video,real-time defect detection by the system,and real-time matching of defects with GIS data of facilities.It further advances the replacement of manual inspections with intelligent inspections of drainage facilities.(4)Based on the SWMM and TELEMAC-2D models,a one-and two-dimensional coupled hydrological and hydrodynamic model for urban flooding is established.This model couples the one-dimensional pipeline network and the two-dimensional surface runoff model to simulate and analyze urban flooding scenarios under specific rainfall conditions.An urban flooding risk assessment indicator system is developed,covering three dimensions: disaster,exposure,and vulnerability,and including both physical and statistical mechanisms,with a total of 12 indicators.A new method is proposed for calculating weights by coupling K-means clustering with an improved projection pursuit technique.This method applies layered projection pursuit based on the continuity characteristics of indicator data to obtain more scientifically accurate indicator weights.The comprehensive application of this method in the Suqian region of Jiangsu achieves a scientific assessment of urban flooding risk.(5)Using historical monitoring data and future monitoring scenarios of drainage facility status,three-stage real-time monitoring results under specific rainfall scenarios are constructed.The occlusion rate is used as an indicator to determine the blockage and occlusion levels of drainage facilities,providing the degree of siltation for different drainage facilities.For realtime monitoring scenarios of drainage facility status,a real-time simulation method for urban flooding risk is proposed.This method uses a coupled one-and two-dimensional hydrological and hydrodynamic model to simulate regional inundation scenarios and flooding risk.Considering the economic population distribution patterns,different stages of monitoring scenarios,and flooding risk identification results,a dual-stage real-time disaster mitigation strategy is proposed.This strategy analyzes the impact of different real-time mitigation strategies on the benefits of urban flooding disaster reduction from key areas,edge areas,and the entire region.It does this based on the positive(negative)value ratio of key road risks and comprehensive disaster reduction benefit indicators.