Urban inland inundation monitoring, verification and processing method

Abstract

The invention belongs to the field of urban inland inundation monitoring and early warning, and relates to an urban inland inundation monitoring, verifying and processing method. The method comprises the following steps: S1, arranging and installing ponding monitoring equipment: installing an electronic water gauge and corollary equipment, and standardizing a real-time ponding value calculation method; s2, accumulated water monitoring early warning and reporting: performing personalized and dynamic accumulated water monitoring data collection according to a monitoring data reporting rule; s3, waterlogging standard definition and waterlogging risk identification: automatically recording a waterlogging event according to a waterlogging standard; s4, waterlogging event verification based on the video AI: constructing a waterlogging event verification model, and performing waterlogging event verification in real time based on the waterlogging event verification model; and S5, processing a waterlogging event. According to the invention, the cruising ability of the monitoring equipment can be improved, low-lying ponding monitoring data can be effectively identified, and the data monitoring accuracy is improved; and based on the waterlogging event verification model, the problem of waterlogging misjudgment caused by wrong data reported by the accumulated water monitoring equipment can be efficiently solved in real time.

Description

Technical Field

[0001] The invention belongs to the field of urban waterlogging monitoring and early warning, and relates to an urban waterlogging monitoring, verification and processing method. Background Art

[0002] In the past few decades, during the rapid urbanization process in my country, the long-term investment in urban drainage and flood control facilities was insufficient, the construction standards were low, the maintenance was not in place, the terrain was low-lying, and most cities in my country had concentrated rainfall during the flood season. As a result, many cities experienced serious urban waterlogging caused by short-duration heavy rainfall. Urban waterlogging not only brought serious flooding losses to local areas, caused urban traffic congestion, disrupted urban life and work order, but also posed a huge threat to the lives and property of urban residents.

[0003] In order to reduce the impact of urban waterlogging on urban operations, it is necessary to timely and accurately reflect the current status of waterlogging in flood-prone areas, issue early warnings for waterlogging risks, conduct targeted verification of dangerous situations, and organize reasonable resources to deal with dangerous situations.

[0004] At present, the commonly used means of early warning of waterlogging risks is to deploy electronic water gauges at flood-prone areas to monitor the depth of water in low-lying areas. If the water depth exceeds a certain threshold, a waterlogging alarm will be issued and relevant personnel will be notified to handle the situation. However, there are the following problems:

[0005] (1) Since the waterlogging standards in different regions are related to the real-time waterlogging depth, waterlogging time, and waterlogging range, the waterlogging standard conditions are not clear, which affects the determination of the waterlogging risk level and further affects the allocation of urban emergency resources.

[0006] (2) Due to the unreasonable deployment of single sensor monitoring equipment, equipment failure, insufficient power supply, and obstacles, the accuracy of waterlogging monitoring data is not high, which affects the judgment of urban flooding risk. Summary of the invention

[0007] In view of the deficiencies of the prior art, the present invention provides a method for monitoring, verifying and processing urban waterlogging.

[0008] The main contents of the present invention include:

[0009] (1) Standardize the installation and deployment requirements of water level gauge equipment and the calculation method of water depth, effectively identify water monitoring data in low-lying areas, and improve data monitoring accuracy.

[0010] (2) Define the waterlogging risk level standards, combine them with waterlogging monitoring data, identify waterlogging events and their levels, and effectively guide the allocation of emergency resources.

[0011] (3) Based on the real-time water depth at flood-prone areas, the reporting rules for electronic water gauge water depth data are modified to save water gauge energy consumption and improve data endurance and monitoring accuracy.

[0012] (4) Combining electronic water gauge monitoring data and video AI recognition technology, a waterlogging event verification model is constructed to automatically verify the authenticity of waterlogging, thereby avoiding waterlogging risk judgment and wasting emergency resources.

[0013] The present invention is implemented by the following technical solutions:

[0014] A method for monitoring, verifying and treating urban waterlogging, comprising:

[0015] S1. Installation of waterlogging monitoring equipment: Install electronic water gauge and supporting equipment, and standardize the calculation method of real-time waterlogging value;

[0016] S2. Waterlogging monitoring, early warning and reporting: Collect waterlogging monitoring data in a personalized and dynamic manner according to the monitoring data reporting rules;

[0017] S3. Definition of waterlogging standards and identification of waterlogging risks: Automatically record waterlogging events based on waterlogging standards;

[0018] S4. Verification of waterlogging events based on video AI: Build a waterlogging event verification model, and verify waterlogging events in real time based on the waterlogging event verification model;

[0019] S5. Handling of waterlogging incidents.

[0020] Preferably, in order to ensure the correctness of the waterlogging monitoring data in step S1, the electronic water gauge and supporting equipment are installed according to the following requirements, and the real-time waterlogging value calculation method is standardized:

[0021] (1) Install and deploy electronic water gauges in important urban roads and places, historical urban waterlogging points, high-risk waterlogging sections, low-lying areas, underpasses, underpasses or tunnels, culverts, backflow points along the river, and high-risk areas with potential flooding hazards in underground garages. The installation of electronic water gauges does not require laying lines or damaging the road surface. They can be installed directly on the lower side of the curbstones on both sides of the installation location.

[0022] The real-time water accumulation value of the electronic water gauge = the measured value of the electronic water gauge - the height difference between the installation point and the highest point of the installation section. If the calculated value is less than 0, take 0.

[0023] (2) If there are video surveillance resources at the installation location of the electronic water gauge, the video surveillance resources shall be shared through the national standard GB28181, i.e. GB / T28181-2016 "Technical Requirements for Information Transmission, Exchange and Control of Public Security Video Surveillance Network Systems"; if there are no video surveillance resources at the installation location of the electronic water gauge, they shall be installed separately; each video surveillance camera shall be set with a preset position to align with the installation location of the electronic water gauge.

[0024] Preferably, the monitoring data reporting rules in step S2 include:

[0025] (1) When there is no rainfall, the electronic water gauge senses that there is no water accumulation on site, and the monitoring data of the electronic water gauge is reported to the cloud server at the first frequency;

[0026] (2) When there is water accumulation on site but the water depth does not change or the change is less than the first threshold, the electronic water gauge monitoring data is reported to the cloud server at the second frequency;

[0027] (3) When the water depth on site reaches a first threshold, the electronic water gauge monitoring data is reported to the cloud server at a third frequency.

[0028] The first threshold, the first frequency, the second frequency, and the third frequency are dynamically adjusted based on a comprehensive analysis of the electronic water gauge installation location, the installation point location, and historical waterlogging data of the installation location.

[0029] Preferably, in step S3, based on the waterlogging standards in different areas, when the waterlogging depth exceeds the relevant threshold and the waterlogging time or the waterlogging range exceeds the relevant threshold at the same time, the system automatically records mild waterlogging, moderate waterlogging, and severe waterlogging events; at the same time, the system automatically calls the video surveillance cameras installed around the waterlogging point, automatically takes a first set value of on-site photos at a preset position, and automatically records a video with a second set value of duration.

[0030] Preferably, step S4 includes: using on-site video monitoring data, electronic water gauge water depth data, and water area data to construct a waterlogging range extraction model and a waterlogging depth area analysis model, coupling the waterlogging range extraction model and the waterlogging depth area analysis model to construct a waterlogging event verification model, and performing waterlogging event verification in real time based on the waterlogging event verification model.

[0031] The steps for verifying waterlogging incidents include:

[0032] (1) Collect video surveillance data of flood-prone locations, water depth monitoring data of water gauges, and topographic maps, and perform data preprocessing to obtain a data set for the waterlogging event verification model;

[0033] (2) Construction of waterlogging event verification model: The waterlogging event verification model includes a waterlogging range extraction model and a waterlogging depth and area analysis model. The two models are coupled and work in tandem to verify waterlogging events in real time.

[0034] (3) Based on the waterlogging incident verification model, complete the video AI waterlogging incident verification.

[0035] Furthermore, the steps of obtaining a data set for the waterlogging event verification model include:

[0036] 1) According to the type of flood-prone areas, collect video surveillance data with different historical characteristics of flood-prone areas, including: video data of no waterlogging, video data of mild waterlogging, video data of moderate waterlogging, and video data of severe waterlogging;

[0037] 2) Read the video data, extract the video frames according to the set interval frequency, and uniformly process the pictures into a standard resolution to form a time series picture set E1 of the flood-prone points;

[0038] 3) Screening pictures with water accumulation in the time series picture set E1 to form a water accumulation picture set E2;

[0039] 4) Comparing the water logging depth data of the water gauge in the same period, the abnormal data in the water logging picture set E2 is removed to obtain the processed water logging picture set E3;

[0040] 5) Referring to the historical water gauge waterlogging monitoring depth data, extract the water depth data information corresponding to each picture in the processed waterlogging picture set E3 to form the waterlogging depth data set E4;

[0041] 6) extracting the water accumulation area of ​​each picture in the processed water accumulation picture set E3 under the topographic map to form a water accumulation area data set E5;

[0042] 7) Perform pixel-level semantic annotation on the waterlogged areas in the processed waterlogged image set E3 as the modeling sample data X1 of the semantic segmentation model.

[0043] Furthermore, the process of constructing the waterlogging event verification model includes:

[0044] 1) 80% of the samples in the sample data set X1 are used as the training set of the waterlogging range extraction model, and 20% of the samples are used as the validation set of the waterlogging range extraction model;

[0045] 2) Build a waterlogging range extraction model, input training data and verification data to train the model, and obtain a trained waterlogging range extraction model M1;

[0046] 3) Using the water accumulation range extraction model M1 to predict the processed water accumulation picture set E3, a water accumulation range single pixel data set E6 is obtained;

[0047] 4) Using the single pixel data set E6 of the waterlogging range, the waterlogging depth data set E4, and the waterlogging area data set E5, construct the waterlogging depth and area analysis model modeling sample data X2;

[0048] 5) Build a waterlogging depth and area analysis model. The model input is the waterlogging range single pixel dataset E6, and the output is the waterlogging depth dataset E4 and the waterlogging area dataset E5;

[0049] 6) The waterlogging depth and area analysis model is trained using the sample data set X2 to obtain a trained waterlogging depth and area analysis model M2;

[0050] 7) Couple the waterlogging range extraction model M1 and the waterlogging depth and area analysis model M2 to obtain the waterlogging event verification model.

[0051] Furthermore, the waterlogging range extraction model is implemented based on the semantic segmentation model: the DeeplabV3P function of the paddlex deep learning framework is used to implement the Deeplabv3+ semantic segmentation model to train the training set samples; the Deeplabv3+ network structure includes an encoder and a decoder, the input is the original image in the sample data set X1, and the output is the waterlogging range map in the sample data set X1; the results of the trained semantic segmentation model are saved as the waterlogging range extraction model M1.

[0052] Furthermore, the waterlogging depth and area analysis model is implemented based on the linear network model: the Linear function in the paddlex deep learning framework is used to build a single-layer fully connected linear network model. The model structure includes a single-layer fully connected input, which receives the E6 image data of the single-pixel dataset of the waterlogging range; the output dimension is 2, which are the water depth and waterlogging area respectively.

[0053] Furthermore, the coupled waterlogging range extraction model M1 and the waterlogging depth and area analysis model M2 work in series and collaboratively, the output result "waterlogging range map" of the waterlogging range extraction model M1 serves as the input of the waterlogging depth and area analysis model M2, and the output results of the waterlogging depth and area analysis model M2 include waterlogging depth and waterlogging area.

[0054] Compared with the prior art, the beneficial effects of the present invention include:

[0055] (1) A video AI-based waterlogging incident verification method is proposed to solve the problem of waterlogging misjudgment caused by erroneous data reported by the electronic water gauge front end in real time and efficiently.

[0056] (2) It is proposed to use on-site video surveillance data, electronic water gauge water depth data, and water area data to construct a waterlogging range extraction model and a waterlogging depth and area analysis model. The two models are coupled and work in series to verify waterlogging incidents in real time, which has the advantages of high efficiency and high accuracy.

[0057] (3) Based on the experience of flood prevention work, further standardize the installation and deployment requirements of water level gauge equipment and the calculation method of water accumulation depth, effectively identify water accumulation monitoring data in low-lying areas, and improve the accuracy of data monitoring.

[0058] (4) Based on the experience of flood prevention work, define the waterlogging risk level standard, combine it with the waterlogging monitoring data, identify waterlogging events and their levels, and effectively guide the allocation of emergency resources.

[0059] (5) Based on the existing water gauge monitoring system, the real-time water depth at flood-prone areas is taken into consideration, and the reporting rules for water accumulation data of the electronic water gauge are modified to save the energy consumption of the water gauge and improve the data endurance and monitoring accuracy.

[0060] (6) Compared with the waterlogging identification technology based on electronic water gauge monitoring data, the automatic verification of the authenticity of waterlogging events combined with video AI identification technology can effectively avoid waterlogging risk judgment and waste of emergency resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A flowchart of a method for monitoring, verifying and treating urban waterlogging in one embodiment of the present invention;

[0062] Figure 2 This is an installation layout diagram of an electronic water gauge in one embodiment of the present invention;

[0063] Figure 3 A diagram showing the construction and application of a waterlogging incident verification model in one embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0065] A method for monitoring, verifying and handling urban waterlogging, including: installation and deployment of waterlogging monitoring equipment, waterlogging monitoring warning and reporting, definition of waterlogging standards and waterlogging risk identification, AI-based waterlogging event verification, and waterlogging event handling. The specific workflow is as follows: Figure 1 As shown, including:

[0066] S1. Layout and installation of water accumulation monitoring equipment.

[0067] To ensure the accuracy of waterlogging monitoring data, install electronic water gauges and supporting equipment according to the following requirements, and standardize the calculation method of real-time waterlogging values:

[0068] (1) Install and deploy electronic water gauges in high-risk areas such as important roads and places in the city, historical waterlogging points in the city, high-risk waterlogging sections, low-lying areas, underpasses, underpasses or tunnels, culverts, backflow points along the river, and underground garage flooding risk points. Figure 2 As shown, the installation of the electronic water gauge in this application does not require laying lines or damaging the road surface. It can be installed directly on the lower side of the curbstones on both sides of the installation location, which is convenient for quick installation and maintenance.

[0069] When calculating the real-time water accumulation value of the above-mentioned area, considering the slope of the electronic water gauge installation point and the installation site, it is not accurate to judge the water accumulation situation of the entire installation site only by the electronic water gauge measurement data. Based on the height difference between the electronic water gauge installation point and the highest point of the installation site, it is proposed that: real-time water accumulation value = electronic water gauge measured value - height difference between the installation point and the highest point of the installation site. If the calculated value is less than 0, then 0 is taken. For example, if the highest elevation of the installation site is 10m, and the current water level measured by the electronic water gauge is 10.12m, then the current water accumulation value is 0.12m.

[0070] (2) If there are video surveillance resources at the installation location of the electronic water gauge, the national standard GB28181, i.e. GB / T28181-2016, "Technical Requirements for Information Transmission, Exchange and Control of Public Security Video Surveillance Networking Systems" can be used to share video surveillance resources and save resources; if there are no video surveillance resources at the installation location of the electronic water gauge, it can be installed separately. For each video surveillance camera, a preset position needs to be set to align with the installation location of the electronic water gauge so that the camera can be called to take photos and videos at the preset position later.

[0071] S2. Monitoring, early warning and reporting of waterlogging.

[0072] Based on the real-time water depth of flood-prone points, the present invention proposes a monitoring data reporting rule, which can effectively save the energy consumption of electronic water gauges and improve data endurance and monitoring accuracy. The specific reporting rules for monitoring data include:

[0073] (1) When there is no rainfall, the electronic water gauge senses that there is no water accumulation on site, and the monitoring data of the electronic water gauge is reported to the cloud server at the first frequency.

[0074] (2) When there is water accumulation on site but the water depth does not change or the change is less than the first threshold, the electronic water gauge monitoring data is reported to the cloud server at the second frequency.

[0075] (3) When the water depth on site reaches a first threshold, the electronic water gauge monitoring data is reported to the cloud server at a third frequency.

[0076] In a preferred embodiment, the first threshold is 1 cm, the first frequency is once every 6 hours, the second frequency is once every 5 minutes, and the third frequency is once every 1 minute.

[0077] It should be noted that the first threshold, the first frequency, the second frequency and the third frequency can be dynamically adjusted after comprehensive analysis of the electronic water gauge installation site, the installation point location, the historical waterlogging data of the installation site, etc., while effectively saving the energy consumption of the electronic water gauge, it can collect electronic water gauge monitoring data more effectively and accurately, facilitating subsequent waterlogging analysis.

[0078] S3. Definition of urban flooding standards and identification of urban flooding risks.

[0079] Different depths, durations, and ranges of waterlogging will cause different disaster losses to different land use types such as urban main roads, sunken bridge areas, residential areas, and industrial and commercial areas. In order to reduce the waste of emergency resources and effectively identify waterlogging risks, waterlogging standards for different areas are defined, as described in Table 1:

[0080] Table 1

[0081]

[0082] When the depth of water accumulation exceeds the relevant threshold and the water accumulation time or water accumulation range exceeds the relevant threshold at the same time, the system automatically records mild waterlogging, moderate waterlogging, and severe waterlogging events. At the same time, the system automatically calls the video surveillance cameras installed around the waterlogging point, automatically takes the first set value of on-site photos at the preset position, and automatically records a video with a duration of the second set value for the next step of video AI-based waterlogging event verification, and for later manual verification.

[0083] In a preferred embodiment, the first setting value is 3 sheets, and the second setting value is 10 seconds.

[0084] S4. Verification of urban flooding incidents based on video AI.

[0085] Due to factors such as sensor errors, unreasonable deployment of monitoring equipment, equipment failure, insufficient power supply, and obstacles, the electronic water gauge may report incorrect data, which may lead to misjudgment of waterlogging based on the reported data. In order to solve the problem of misjudgment of waterlogging caused by incorrect data reported by the front end of the electronic water gauge, the present invention proposes a waterlogging event verification method based on video AI.

[0086] The video AI-based waterlogging incident verification method utilizes on-site video surveillance data, electronic water gauge water depth data, and waterlogging area data, couples the waterlogging range extraction model and the waterlogging depth and area analysis model, and constructs a waterlogging incident verification model. Based on the waterlogging incident verification model, waterlogging incidents are verified in real time.

[0087] like Figure 3 As shown in the figure, the steps of constructing the waterlogging event verification model include:

[0088] (1) Collect data such as video surveillance of flood-prone locations, water depth monitoring data, and topographic maps, and perform data preprocessing to obtain a data set for the waterlogging event verification model.

[0089] Specifically:

[0090] 1) According to the type of flood-prone areas, collect video surveillance data with different historical characteristics of flood-prone areas, including video data of no waterlogging, video data of mild waterlogging, video data of moderate waterlogging, and video data of severe waterlogging, with a duration of more than 3 minutes.

[0091] 2) Use opencv to read video data, extract frames from the video according to a set interval frequency, such as 1 second, and uniformly process the images into a standard resolution, such as 256*256, to form a time-series image set E1 of flood-prone points.

[0092] 3) Screening the pictures with water accumulation in the time series picture set E1 to form the water accumulation picture set E2.

[0093] 4) By comparing the water depth monitoring data of the water gauge during the same period, the abnormal data in the water picture set E2 is removed to obtain the processed water picture set E3. The abnormal data includes the obvious deviation data between the water in the picture and the water gauge monitoring data that can be seen by the naked eye.

[0094] 5) With reference to the historical water gauge waterlogging monitoring depth data, the water depth data information corresponding to each picture in the processed waterlogging picture set E3 is extracted to form the waterlogging depth data set E4.

[0095] 6) Under the 1:500 scale topographic map, the waterlogged area of ​​each picture in the processed waterlogged picture set E3 is extracted to form a waterlogged area data set E5.

[0096] 7) Use labelme to perform pixel-level semantic annotation on the water accumulation areas in the processed water accumulation image set E3 as the modeling sample data X1 of the semantic segmentation model.

[0097] Specifically include:

[0098] a. Use labelme to import the processed waterlogged image set E3 dataset and set the water area label;

[0099] b. Mark the water area of ​​the images in the processed water accumulation image set E3;

[0100] c. Export the standard format data of the annotation results, including the annotation coordinate range, original image data and other information;

[0101] d. Perform data processing on the standard format data and output sample data, including original images, water area images and data index files, as sample data set X1.

[0102] (2) Construction of waterlogging incident verification model

[0103] The waterlogging incident verification model includes a waterlogging range extraction model and a waterlogging depth and area analysis model. The two models are coupled and work in series, with the advantages of high efficiency and high precision, and can verify waterlogging incidents in real time.

[0104] 1) 80% of the samples in the sample data set X1 are used as the training set of the waterlogging range extraction model, and 20% of the samples are used as the validation set of the waterlogging range extraction model.

[0105] 2) Build a waterlogging range extraction model, input training data and verification data to train the model, and obtain a trained waterlogging range extraction model M1.

[0106] In a preferred embodiment, the waterlogging range extraction model is implemented based on the semantic segmentation model. Specifically: the DeeplabV3P function of the paddlex deep learning framework is used to implement the Deeplabv3+ semantic segmentation model to train the training set samples. The Deeplabv3+ network structure includes an encoder and a decoder. The input is the original image in the sample data set X1, and the output is the waterlogging range map (water area image) in the sample data set X1. Among them, the key parameters of the model are the number of training iterations of 40, the training batch of 5, and the learning rate of 0.01. The trained semantic segmentation model results are saved as the waterlogging range extraction model M1.

[0107] 3) Use the waterlogging range extraction model M1 to predict the processed waterlogging picture set E3 to obtain the waterlogging range single-pixel data set E6.

[0108] 4) Using the single-pixel dataset E6 of the waterlogging range, the dataset E4 of the waterlogging depth, and the dataset E5 of the waterlogging area, a waterlogging depth and area analysis model modeling sample data X2 is constructed, of which 80% of the samples are used as a training set and 20% of the samples are used as a validation set.

[0109] Specifically: the corresponding image in the single-pixel data set E6 of the waterlogging range, the water depth value of the corresponding image in the waterlogging depth data set E4, and the waterlogging area value of the corresponding image in the waterlogging area data set E5 are taken as a sample, and so on, to form a waterlogging depth and area analysis model sample data set X2.

[0110] 5) Build a waterlogging depth and area analysis model. The model input is the single-pixel dataset E6 of the waterlogging range, and the output is the waterlogging depth dataset E4 and the waterlogging area dataset E5.

[0111] In a preferred embodiment, the water accumulation depth and area analysis model is implemented based on a linear network model. Specifically: a single-layer fully connected linear network model is built using the Linear function in the paddlex deep learning framework. The model structure includes a single-layer fully connected input, receiving the E6 image data of the single-pixel dataset of the water accumulation range; the output dimension is 2, which are the water accumulation depth and water accumulation area.

[0112] 6) The waterlogging depth and area analysis model is trained using the sample data set X2 to obtain a trained waterlogging depth and area analysis model M2.

[0113] 7) Couple the waterlogging range extraction model M1 and the waterlogging depth and area analysis model M2 to obtain the waterlogging event verification model.

[0114] Specifically: the coupled waterlogging range extraction model M1 and the waterlogging depth and area analysis model M2 work in series and collaboratively, the output result "waterlogging range map" of the waterlogging range extraction model M1 serves as the input of the waterlogging depth and area analysis model M2, and the output results of the waterlogging depth and area analysis model M2 include waterlogging depth and waterlogging area.

[0115] (3) Based on the waterlogging incident verification model, complete the video AI waterlogging incident verification.

[0116] During the application stage of the waterlogging event verification model, real-time pictures are used as input parameters of the waterlogging range extraction model M1. The waterlogging range map is input into the waterlogging range analysis model M2 to obtain the waterlogging depth and area, thereby realizing real-time analysis of the waterlogging depth and area.

[0117] Furthermore, the water depth analyzed by the waterlogging event verification model is compared with the real-time water depth of the electronic water gauge. If the difference is greater than ±30%, the system will cancel the recorded mild waterlogging, moderate waterlogging, and severe waterlogging events. If the difference is within ±30%, the system will automatically confirm mild waterlogging, moderate waterlogging, and severe waterlogging events, and send text message notifications of the waterlogging events to the person responsible for rescue.

[0118] S5. Handling of waterlogging incidents. After receiving the SMS notification, the on-site rescue responsible person handles the waterlogging incident on site and reports the waterlogging handling situation through the mobile terminal, including filling in the handling situation, uploading and submitting on-site photos, and the waterlogging incident is handled.

[0119] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for monitoring, verifying and treating urban waterlogging, characterized in that: include: S1. Installation of waterlogging monitoring equipment: Install electronic water gauge and supporting equipment, and standardize the calculation method of real-time waterlogging value; S2. Waterlogging monitoring, early warning and reporting: Collect waterlogging monitoring data in a personalized and dynamic manner according to the monitoring data reporting rules; S3. Definition of waterlogging standards and identification of waterlogging risks: Automatically record waterlogging events based on waterlogging standards; S4. Verification of waterlogging events based on video AI: Build a waterlogging event verification model, and verify waterlogging events in real time based on the waterlogging event verification model; S5. Handling of waterlogging incidents.

2. The method according to claim 1, characterized in that In step S1, to ensure the accuracy of waterlogging monitoring data, install the electronic water gauge and supporting equipment according to the following requirements, and standardize the calculation method of real-time waterlogging value: (1) Install and deploy electronic water gauges in important urban roads and places, historical urban waterlogging sites, high-risk waterlogging sections, low-lying areas, underpasses, underpasses or tunnels, culverts, backflow points along the river, and high-risk areas of flooding in underground garages. The installation of electronic water gauges does not require laying lines or damaging the road surface. It can be installed directly on the lower side of the curbstones on both sides of the installation location; The real-time water accumulation value of the electronic water gauge = the measured value of the electronic water gauge - the height difference between the installation point and the highest point of the installation section. If the calculated value is less than 0, it is taken as 0; (2) If there are video surveillance resources at the installation location of the electronic water gauge, the video surveillance resources shall be shared through the national standard GB28181, i.e. GB / T28181-2016 "Technical Requirements for Information Transmission, Exchange and Control of Public Security Video Surveillance Network Systems"; if there are no video surveillance resources at the installation location of the electronic water gauge, they shall be installed separately; each video surveillance camera shall be set with a preset position to align with the installation location of the electronic water gauge.

3. The method according to claim 1, characterized in that The monitoring data reporting rules in step S2 include: (1) When there is no rainfall, the electronic water gauge senses that there is no water accumulation on site, and the monitoring data of the electronic water gauge is reported to the cloud server at the first frequency; (2) When there is water accumulation on site but the water depth does not change or the change is less than the first threshold, the electronic water gauge monitoring data is reported to the cloud server at the second frequency; (3) When the water depth on site reaches a first threshold, the electronic water gauge monitoring data is reported to the cloud server at a third frequency; The first threshold, the first frequency, the second frequency, and the third frequency are dynamically adjusted based on a comprehensive analysis of the electronic water gauge installation location, the installation point location, and historical waterlogging data of the installation location.

4. The method according to claim 1, characterized in that In step S3, based on the waterlogging standards in different areas, when the waterlogging depth exceeds the relevant threshold and the waterlogging time or the waterlogging range exceeds the relevant threshold at the same time, the system automatically records mild waterlogging, moderate waterlogging, and severe waterlogging events; at the same time, the system automatically calls the video surveillance cameras installed around the waterlogging point, automatically takes the first set value of on-site photos at the preset position, and automatically records a video with a duration of the second set value.

5. The method according to claim 1, characterized in that Step S4 includes: using on-site video monitoring data, electronic water gauge water depth data, and water area data to construct a waterlogging range extraction model and a waterlogging depth area analysis model, coupling the waterlogging range extraction model and the waterlogging depth area analysis model to construct a waterlogging event verification model, and verifying the waterlogging event in real time based on the waterlogging event verification model; The steps for verifying waterlogging incidents include: (1) Collect video surveillance data of flood-prone locations, water depth monitoring data of water gauges, and topographic maps, and perform data preprocessing to obtain a data set for the waterlogging event verification model; (2) Construction of waterlogging event verification model: The waterlogging event verification model includes a waterlogging range extraction model and a waterlogging depth and area analysis model. The two models are coupled and work in tandem to verify waterlogging events in real time. (3) Based on the waterlogging incident verification model, complete the video AI waterlogging incident verification.

6. The method according to claim 5, characterized in that The steps to obtain the dataset for the waterlogging event verification model include: 1) According to the type of flood-prone areas, collect video surveillance data with different historical characteristics of flood-prone areas, including: video data of no waterlogging, video data of mild waterlogging, video data of moderate waterlogging, and video data of severe waterlogging; 2) Read the video data, extract the video frames according to the set interval frequency, and uniformly process the pictures into a standard resolution to form a time series picture set E1 of the flood-prone points; 3) Screening pictures with water accumulation in the time series picture set E1 to form a water accumulation picture set E2; 4) Comparing the water logging depth data of the water gauge in the same period, the abnormal data in the water logging picture set E2 is removed to obtain the processed water logging picture set E3; 5) Referring to the historical water gauge waterlogging monitoring depth data, extract the water depth data information corresponding to each picture in the processed waterlogging picture set E3 to form the waterlogging depth data set E4; 6) extracting the water accumulation area of ​​each picture in the processed water accumulation picture set E3 under the topographic map to form a water accumulation area data set E5; 7) Perform pixel-level semantic annotation on the waterlogged areas in the processed waterlogged image set E3 as the modeling sample data X1 of the semantic segmentation model.

7. The method according to claim 6, characterized in that The process of building the waterlogging event verification model includes: 1) 80% of the samples in the sample data set X1 are used as the training set of the waterlogging range extraction model, and 20% of the samples are used as the validation set of the waterlogging range extraction model; 2) Build a waterlogging range extraction model, input training data and verification data to train the model, and obtain a trained waterlogging range extraction model M1; 3) Using the water accumulation range extraction model M1 to predict the processed water accumulation picture set E3, a water accumulation range single pixel data set E6 is obtained; 4) Using the single pixel data set E6 of the waterlogging range, the waterlogging depth data set E4, and the waterlogging area data set E5, construct the waterlogging depth and area analysis model modeling sample data X2; 5) Build a waterlogging depth and area analysis model. The model input is the waterlogging range single pixel dataset E6, and the output is the waterlogging depth dataset E4 and the waterlogging area dataset E5; 6) The waterlogging depth and area analysis model is trained using the sample data set X2 to obtain a trained waterlogging depth and area analysis model M2; 7) Couple the waterlogging range extraction model M1 and the waterlogging depth and area analysis model M2 to obtain the waterlogging event verification model.

8. The method according to claim 7, characterized in that The waterlogging range extraction model is implemented based on the semantic segmentation model. The input is the original image in the sample data set X1, and the output is the waterlogging range map in the sample data set X1. The trained semantic segmentation model result is saved as the waterlogging range extraction model M1.

9. The method according to claim 7, characterized in that: The waterlogging depth and area analysis model is implemented based on a linear network model. The waterlogging depth and area analysis model receives the single-pixel dataset E6 image data of the waterlogging range; the output dimension is 2, which are the waterlogging depth and waterlogging area respectively.

10. The method according to claim 7, characterized in that The coupled waterlogging range extraction model M1 and the waterlogging depth and area analysis model M2 work in series. The output result "waterlogging range map" of the waterlogging range extraction model M1 serves as the input of the waterlogging depth and area analysis model M2. The output results of the waterlogging depth and area analysis model M2 include waterlogging depth and waterlogging area.

Images