Urban Disaster Emergency Response and Intelligent Management System Based on Unmanned Aerial Vehicles

Abstract

This invention discloses an urban disaster emergency response and intelligent management system based on unmanned aerial vehicles (UAVs), relating to the field of emergency response technology. It includes a drainage flow calculation unit, a monitoring station setting unit, a water depth prediction unit, a recognition model training unit, an actual water accumulation analysis unit, and an emergency response unit. The invention uses the recognition model training unit to process the collected image data to obtain new image data, thereby expanding the number of images in the original sample dataset. Furthermore, it continuously optimizes the parameters of the network model using training, testing, and validation data, further improving the accuracy of the model in recognizing wheel areas. Simultaneously, the actual water accumulation analysis unit calculates the corresponding wheel radius in the image data based on the identified wheel area, further estimating the actual water accumulation height, ensuring that the evacuation plan formulated by the control center is more scientific and effective.

Description

Technical Field

[0001] This invention relates to the field of emergency response technology, specifically to an urban disaster emergency response and intelligent management system based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Flooding is the most significant natural disaster in cities. Utilizing an urban disaster emergency response and intelligent management system allows for real-time monitoring of water depth in various areas and prediction of future water depths based on this data. In the event of flooding, timely evacuation and relocation of people can be organized to ensure public safety. Patent application number 202110277249.7 discloses "a blockchain-based urban natural disaster monitoring and emergency management dispatch platform, comprising: a trusted emergency dispatch system for urban natural disaster monitoring and early warning, and an urban natural disaster monitoring and early warning blockchain management platform connected via a network; multiple independent service systems and blockchain network nodes are distributed and connected via the network for distributed task distribution and dispatch. Under this platform, the local public crisis management command center and other disaster monitoring and early warning systems are connected to the blockchain network. The decentralized nature of the platform allows each department to share data and significantly reduces the possibility of system malfunctions, ensuring performance. The immutability and the smart contracts and consensus mechanisms used in the platform provide credible data verification, thus ensuring accountability. Furthermore, this platform is auxiliary and does not replace or intrude upon existing systems, making it feasible."

[0003] The aforementioned existing technologies have solved problems such as the lack of unified organization and coordination among different departments. However, when the system is in operation, the lack of analysis of the main waterlogged areas in the city leads to unreasonable selection of some waterlogging monitoring station locations. At the same time, the system cannot accurately predict the water depth in the future, making it impossible for operators to detect and intervene in time before urban flooding occurs. Furthermore, the actual water depth of each road needs to be measured manually, which increases a lot of manpower costs. Summary of the Invention

[0004] The purpose of this invention is to provide an urban disaster emergency response and intelligent management system based on unmanned aerial vehicles (UAVs) to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an urban disaster emergency response and intelligent management system based on unmanned aerial vehicles (UAVs), comprising a drainage flow calculation unit and a monitoring station setting unit.

[0006] The water depth prediction unit acquires historical meteorological datasets and water depth datasets for each station. After mapping the meteorological data and water depth data one-to-one in the time series, it segments the meteorological data and water depth data according to the specified sliding window size, combines the two into a multidimensional matrix, and transmits the multidimensional matrix as multiple input vectors to the water depth prediction model for analysis, outputting the predicted values ​​of all water depths in future time intervals.

[0007] The recognition model training unit constructs a sample dataset and divides the image data in the dataset into three categories: test data, training data, and validation data. The training data is input into the feature extraction backbone network model one by one. The network model is used to recognize and segment these vehicle image data and calculate the model weights. After each training cycle, the test data is input into the current network model for analysis to obtain the model's localization accuracy and segmentation accuracy. Then, the validation data is input into the network model to determine the final parameters of the current model and the corresponding weights.

[0008] The actual water accumulation analysis unit transmits image data collected by the UAV to the network model. After the model identifies the wheel area and model, it filters out all arc segments within the area, divides these arc segments into multiple sub-segments, and analyzes these sub-segments using an arc segment determination algorithm. Sub-segments belonging to the same arc segment are stored in the same candidate set. An arbitrary sub-segment is selected, and the distance between the end arc point of the sub-segment and the beginning arc point of another sub-segment, as well as the distance between the beginning arc point and the end arc point of the sub-segment, are calculated. The two distances are compared to determine the value of the corresponding point in the adjacency matrix. After analyzing all arc segments belonging to the same candidate ellipse based on the adjacency matrix, the ellipse is fitted. The fitting result is sampled, and after determining that the current fitting result is the true ellipse, the image data corresponding to the position of the true ellipse is extracted. The depth analysis algorithm is used to calculate the actual water accumulation depth.

[0009] The emergency response unit analyzes the predicted water depth of all stations. When the predicted value exceeds the emergency threshold within a specified time interval, it reports to the control center. The control center then retrieves the actual water depth of the nearest road in the catchment area of ​​the current station, uses deep learning technology to formulate evacuation routes and emergency plans based on the actual water depth, and outputs them through a visual interface.

[0010] Preferably, the drainage flow calculation unit includes a runoff coefficient analysis module, a seasonal factor analysis module, a comprehensive runoff analysis module, a drainage capacity analysis module, and an average runoff analysis module. The runoff coefficient analysis module acquires the duration, interval, rainfall rate, attenuation coefficient, and variation coefficient, calculates the rainfall intensity and total rainfall based on these data, determines the corresponding basic runoff coefficient according to the land cover type, and analyzes the random runoff coefficient for each land cover type based on the current rainfall intensity, duration, and basic runoff coefficient. The seasonal factor analysis module extracts historical rainfall data, calculates the average rainfall for different months, and sets a first transition threshold and a second transition threshold. If the average rainfall for the current month is greater than the first transition threshold and less than the second transition threshold, the current month is marked as a transition month. If the average rainfall is less than or equal to the first transition threshold, the current month is marked as a dry season month. If the average rainfall is greater than or equal to the second transition threshold, the current month is marked as a flood season month. All transition months are sorted according to average rainfall to determine this group. After determining the median of the data, a compensation coefficient is set. The seasonal factor is calculated based on the median and the compensation coefficient. The comprehensive runoff analysis module extracts the risk coefficient, interval coefficient, and impact coefficient. Based on the random runoff coefficient, seasonal factor, risk coefficient, interval coefficient, and impact coefficient, the corresponding comprehensive runoff coefficient is calculated. The drainage capacity analysis module statistically analyzes the data collected from various meteorological stations, identifies the rainfall trend surface based on this data, and divides the designated area into multiple sub-regions according to station location and rainfall trend surface. The surface area, drainage outlet flow rate, and number of outlets in each sub-region are determined. Drainage pipes within each sub-region are numbered, and the length and radius of each pipe are calculated. The drainage capacity analysis algorithm is used to calculate the design drainage flow rate for each sub-region. The average runoff analysis module divides each sub-region into multiple unit regions. Based on the weighting coefficients corresponding to surface cover type and interval time, the comprehensive runoff coefficient of each unit region is analyzed. The average runoff coefficient of each sub-region is then calculated based on this data. The drainage capacity analysis algorithm is as follows:

[0011]

[0012] in, Indicates the first Drainage volume density of each sub-region Indicates the total number of drain pipes. Indicates the first The radius of each drain pipe, Indicates the first The length of each drain pipe Indicates the first The surface area of ​​each sub-region Indicates the first The minimum value of drainage volumetric density for each sub-region. Indicates the first The maximum value of the drainage volumetric density of each sub-region This indicates the flow rate of a single drain outlet. This indicates the total number of drainage outlets within the sub-region. This indicates the design drainage flow rate for the sub-region.

[0013] Preferably, the monitoring station setting unit includes a cumulative flow direction calculation module and a road width determination module. The cumulative flow direction calculation module calculates the altitude and slope of each unit area, analyzes the data using a water ripple analysis algorithm, obtains the flow direction value corresponding to each unit area, traverses each area, determines the building height in each unit area, and analyzes the cumulative flow direction value of each unit area based on this data. The road width determination module calculates the cumulative flow direction value of each unit area, sets areas with cumulative flow direction values ​​lower than a preset value as catchment areas, determines the outlet location of each catchment area and the grade and distribution location of all roads in the designated area, analyzes the grade of the road closest to the outlet based on the current outlet location, and queries the database to find the road width according to the road grade. The water ripple analysis algorithm is specifically a single flow direction algorithm.

[0014] Preferably, the monitoring station setting unit further includes a water accumulation depth output module and a station deployment module. The water accumulation depth output module calculates the corresponding surface flow based on rainfall intensity, the average runoff coefficient of each sub-region, and the surface area. After determining the difference between the surface flow of each sub-region and the designed drainage flow, it uses this difference as the actual overflow flow. It then uses a regional water accumulation analysis algorithm to calculate the corresponding hydraulic radius and surface runoff velocity based on these data. The water accumulation depth of the catchment area is analyzed using the hydraulic radius and road width. The station deployment module analyzes the corresponding flow path using the cumulative flow direction value of each unit region. It performs fusion analysis based on these data using a rainfall model and outputs a real-time water accumulation distribution map of the specified area. Corresponding water accumulation monitoring stations are deployed according to the catchment area location and water accumulation depth in the distribution map. The rainfall model is specifically an SWMM simulation model, and the regional water accumulation analysis algorithm is specifically:

[0015]

[0016] in, Indicates the hydraulic radius. This represents the actual overflow traffic. Indicates the cross-sectional area of ​​the water passage. Represents the correlation coefficient. This represents the average slope value of the sub-region. Indicates surface runoff velocity. Indicates parameters, Indicates the depth of water accumulation in the catchment area. This indicates the average width of the road.

[0017] Preferably, the water depth prediction unit includes a historical data processing module, a matrix construction module, and a predicted value output module. The historical data processing module acquires historical meteorological datasets and water depth datasets for each station. After setting a time interval, it performs interpolation on the water depth data and then resamples it to ensure a one-to-one correspondence between the meteorological data and the water depth data in the time series. The matrix construction module sets four training modes and sets corresponding sliding window sizes according to each training mode. It then segments the meteorological data and water depth data according to the sliding window size and combines them into a multidimensional matrix. The predicted value output module transmits the multidimensional matrix as multiple input vectors to the water depth prediction model for analysis. It replaces the first element in the original input vector with the prediction result, deletes the last element in the input vector, analyzes the new input vector, and repeats the operation until the prediction step size is reached. Finally, it outputs the predicted values ​​for all water depths within future time intervals using a predicted value analysis algorithm. The water depth prediction model is specifically a regressor of a recursive prediction strategy algorithm, and the predicted value analysis algorithm is specifically as follows:

[0018]

[0019] in, The predicted value representing the depth of the water accumulation. This represents the model prediction mapping function. Represents the deviation vector. This indicates the actual depth of the accumulated water.

[0020] Preferably, the recognition model training unit includes a sample data collection module, a parameter update module, a test data input module, and a training termination module. The sample data collection module extracts multiple vehicle side images, delineates the wheel regions in these images, adds corresponding labels based on the wheel model, and processes the images by adjusting brightness and contrast, adding noise, and rotating them to obtain new image data. This new image data, along with the original image data, is stored in a sample dataset. The parameter update module divides the image data in the sample dataset into three categories: test data, training data, and validation data. After setting a training period, the training data is input one by one into the feature extraction backbone network model. This network model is used to recognize and segment these vehicle image data, calculate the model weights, and utilize the loss function. The function calculates the corresponding loss value based on the deviation between the drawn wheel area and the recognition area, and updates the corresponding parameters based on the loss value. After each training cycle, the test data input module inputs the test data into the current network model for analysis. After calculating the model's localization accuracy and segmentation accuracy based on the deviation between the drawn wheel area and the recognition area, it determines whether the localization accuracy and segmentation accuracy have reached the preset values. If they have reached the predicted values, the verification data is input into the network model; otherwise, the training cycle is restarted. When the model receives the verification data, the training end module analyzes the data and determines whether the model should stop training. If the model stops training, the current model parameters and corresponding weights are determined; otherwise, the next training cycle begins. The feature extraction backbone network model is specifically the Mask R-CNN model.

[0021] Preferably, the actual water accumulation analysis unit includes an arc segment screening module and a sub-segment generation module. The arc segment screening module uses a drone to photograph vehicles on the road closest to each water catchment area and transmits the collected image data to a network model. After the model identifies the wheel area and model, it determines the minimum bounding matrix corresponding to the wheel area. An edge extraction algorithm is used to analyze all edge points corresponding to the area. After connecting adjacent edge points, all arc segments in the area are screened out and stored in a set. The relative length of each arc segment is calculated. If the arc segment length is lower than a preset value, the arc segment is deleted; otherwise, the arc segment is retained. The sub-segment generation module divides all arc segments in the set to obtain multiple sub-segments. The angle between the sub-segments is determined, and an arc segment determination algorithm is used to analyze these sub-segments. Sub-segments belonging to the same arc segment are stored in the same candidate set. The edge extraction algorithm is specifically the Canny algorithm.

[0022] Preferably, the actual water accumulation analysis unit further includes a matrix value acquisition module, an ellipse determination module, and an actual value output module. The matrix value acquisition module, after statistically analyzing all segments contained in each candidate set, arbitrarily selects a segment and calculates the distance between the terminal arc point of the segment and the starting arc point of another segment, as well as the distance between the starting arc point and the terminal arc point of the segment. It compares these two distances; if the former is greater than or equal to the latter, the two segments are determined not to be adjacent, and the corresponding point in the adjacency matrix is ​​set to zero. If the former is less than the latter, the two segments are determined to be adjacent. The module then analyzes the curvature and concavity / convexity corresponding to the terminal arc point of the current segment and the starting arc point of another segment to determine whether these two arc points belong to the same arc segment. If they belong to the same arc segment, the corresponding point in the adjacency matrix is ​​set to one; otherwise, the value is zero. If the value of a segment is negative one, the operation is repeated until the value of all points in the adjacency matrix is ​​determined. After the ellipse determination module determines all arc segments belonging to the same candidate ellipse according to the adjacency matrix, it numbers these arc segments one by one, obtains different candidate combinations, fits them, samples the fitting results, analyzes each index of the sampled points, and judges the corresponding verification score based on the index data. If the verification score is higher than the preset value, the current fitting result is determined to be a true ellipse; otherwise, it is a false ellipse. After the actual value output module extracts the image data corresponding to the position of the true ellipse, it analyzes the radius of the wheel in the image according to the center position of the ellipse, queries the database according to the wheel model, determines the actual height of the wheel, and then uses a depth analysis algorithm to analyze and obtain the actual water depth.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] 1. This invention divides a designated area into several sub-regions through a drainage flow calculation unit and calculates the designed drainage flow for each sub-region, providing strong data support for the subsequent deployment of waterlogging monitoring stations. The monitoring station setting unit analyzes the location and depth of the catchment area in each sub-region under different rainfall conditions and outputs a real-time waterlogging distribution map of the entire area. This design not only more intuitively displays the waterlogging situation in the entire area under different rainfall levels but also provides scientific guidance for the manual deployment of waterlogging monitoring stations, making the monitoring data obtained from each station more consistent with actual needs. Simultaneously, the waterlogging depth prediction unit analyzes the meteorological data and waterlogging depth data of each station to predict the waterlogging depth in the future. Furthermore, since meteorological data and waterlogging depth data have a certain correlation, calculating the multi-dimensional matrix jointly constructed by the two will make the predicted values ​​more accurate and reliable.

[0025] 2. This invention uses a recognition model training unit to process the collected image data to obtain new image data, thereby expanding the number of images in the original sample dataset. Furthermore, it continuously optimizes the parameters of the network model using training, testing, and validation data, further improving the accuracy of the model in recognizing wheel areas. Simultaneously, the actual water accumulation analysis unit calculates the corresponding wheel radius in the image data based on the identified wheel areas, further estimating the actual water accumulation height. This ensures that the evacuation plan formulated by the control center is more scientific and effective. The emergency response unit uses human intervention in severely flooded areas to ensure the personal safety of residents during flooding. Attached Figure Description

[0026] Figure 1 A schematic diagram of the overall system flow is provided for embodiments of the present invention;

[0027] Figure 2 This is an internal module block diagram of the drainage flow calculation unit provided in an embodiment of the present invention;

[0028] Figure 3 This is an internal module block diagram of the monitoring site setting unit provided in an embodiment of the present invention;

[0029] Figure 4 This is an internal module block diagram of the water depth prediction unit provided in an embodiment of the present invention;

[0030] Figure 5 This is an internal module block diagram of the recognition model training unit provided in an embodiment of the present invention;

[0031] Figure 6 This is a block diagram of the internal modules of the actual water accumulation analysis unit provided in an embodiment of the present invention.

[0032] In the diagram: 1. Drainage flow calculation unit; 101. Runoff coefficient analysis module; 102. Seasonal factor analysis module; 103. Comprehensive runoff analysis module; 104. Drainage capacity analysis module; 105. Average runoff analysis module; 2. Monitoring station setting unit; 201. Cumulative flow direction calculation module; 202. Road width determination module; 203. Water depth output module; 204. Station layout module; 3. Water depth prediction unit; 301. Historical data processing module; 302. Matrix construction module; 303. Predicted value output module; 4. Recognition model training unit; 401. Sample data collection module; 402. Parameter update module; 403. Test data input module; 404. Training end module; 5. Actual water accumulation analysis unit; 501. Arc segment selection module; 502. Sub-segment generation module; 503. Matrix value retrieval module; 504. Ellipse determination module; 505. Actual value output module; 6. Emergency response unit. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Please see Figures 1-6 The present invention provides a technical solution: an urban disaster emergency response and intelligent management system based on unmanned aerial vehicle (UAV) assistance, including a drainage flow calculation unit 1 and a monitoring station setting unit 2;

[0035] Water depth prediction unit 3 acquires historical meteorological datasets and water depth datasets from each station. After mapping the meteorological data and water depth data one-to-one in the time series, it segments the meteorological data and water depth data according to the specified sliding window size, combines the two into a multidimensional matrix, and transmits the multidimensional matrix as multiple input vectors to the water depth prediction model for analysis, outputting the predicted values ​​of all water depths in future time intervals.

[0036] The recognition model training unit 4 constructs a sample dataset and then divides the image data in the dataset into three categories: test data, training data, and validation data. The training data is input into the feature extraction backbone network model one by one. The network model is used to recognize and segment these vehicle image data and calculate the model weights. After each training cycle, the test data is input into the current network model for analysis. After obtaining the model's localization accuracy and segmentation accuracy, the validation data is input into the network model to determine the final parameters of the current model and the corresponding weights.

[0037] Actual water accumulation analysis unit 5 transmits the image data collected by the UAV to the network model. After the model identifies the wheel area and model, it filters out all arc segments in the area, divides these arc segments into multiple sub-segments, and analyzes these sub-segments using an arc segment determination algorithm. Sub-segments belonging to the same arc segment are stored in the same candidate set. An arbitrary sub-segment is selected, and the distance between the end arc point of the sub-segment and the beginning arc point of another sub-segment, as well as the distance between the beginning arc point and the end arc point of the sub-segment, are calculated. The two distances are compared to determine the value of the corresponding point in the adjacency matrix. After analyzing all arc segments belonging to the same candidate ellipse based on the adjacency matrix, the ellipse is fitted. The fitting result is sampled, and after determining that the current fitting result is the true ellipse, the image data corresponding to the position of the true ellipse is extracted. The depth analysis algorithm is used to calculate the actual water accumulation depth.

[0038] Emergency Response Unit 6 analyzes the predicted water depth values ​​of all stations. When the predicted value exceeds the emergency threshold at a specified time interval, it will report to the control center. The control center will then retrieve the actual water depth of the nearest road in the catchment area where the current station is located, use deep learning technology to formulate evacuation routes and emergency plans based on the actual water depth, and output them through a visual interface.

[0039] The drainage flow calculation unit 1 includes a runoff coefficient analysis module 101, a seasonal factor analysis module 102, a comprehensive runoff analysis module 103, a drainage capacity analysis module 104, and an average runoff analysis module 105. The runoff coefficient analysis module 101 acquires the duration, interval, rainfall rate, attenuation coefficient, and variation coefficient, and calculates the rainfall intensity and total rainfall based on these data. It also determines the corresponding basic runoff coefficient according to the land cover type and analyzes the random runoff coefficient for each land cover type based on the current rainfall intensity, duration, and basic runoff coefficient. The seasonal factor analysis module 102 extracts historical rainfall data, calculates the average rainfall for different months, and sets a first transition threshold and a second transition threshold. If the average rainfall for the current month is greater than the first transition threshold but less than the second transition threshold, the current month is marked as a transition month. All transition months are sorted according to average rainfall, and after determining the median of this data set, a compensation coefficient is set. The calculation is then based on the median and the compensation coefficient. The seasonal factors are calculated, and the comprehensive runoff analysis module 103 extracts the risk coefficient, interval coefficient, and impact coefficient. Based on the random runoff coefficient, seasonal factors, risk coefficient, interval coefficient, and impact coefficient, the corresponding comprehensive runoff coefficient is analyzed. The drainage capacity analysis module 104 statistically analyzes the data collected from various meteorological stations, and then analyzes the rainfall trend surface based on this data. According to the station location and rainfall trend surface, the designated area is divided into multiple sub-regions. The surface area, drainage outlet flow rate, and number of each sub-region are determined, and each drainage pipe within the sub-region is numbered. The length and radius of each drainage pipe are statistically analyzed. The drainage capacity analysis algorithm is used to calculate the design drainage flow rate of the sub-region. The average runoff analysis module 105 divides each sub-region into multiple unit regions. Based on the weight coefficients corresponding to surface cover type and interval time, the comprehensive runoff coefficient of each unit region is analyzed. Based on this data, the average runoff coefficient of each sub-region is calculated. The drainage capacity analysis algorithm is as follows:

[0040]

[0041] in, Indicates the first Drainage volume density of each sub-region Indicates the total number of drain pipes. Indicates the first The radius of each drain pipe, Indicates the first The length of each drain pipe Indicates the first The surface area of ​​each sub-region Indicates the first The minimum value of drainage volumetric density for each sub-region. Indicates the first The maximum value of the drainage volumetric density of each sub-region This indicates the flow rate of a single drain outlet. This indicates the total number of drainage outlets within the sub-region. Indicates the design drainage flow rate of the sub-region;

[0042] The monitoring station setting unit 2 includes a cumulative flow direction calculation module 201 and a road width determination module 202. After the cumulative flow direction calculation module 201 counts the altitude and slope of each unit area, it uses a water ripple analysis algorithm to analyze the data and obtain the flow direction value corresponding to each unit area. It traverses each area to determine the building height in each unit area and analyzes the cumulative flow direction value of each unit area based on the data. After the road width determination module 202 counts the cumulative flow direction value of each unit area, it sets the area with the cumulative flow direction value lower than the preset value as the water catchment area. After determining the outlet location of each water catchment area and the level and distribution of all roads in the specified area, it analyzes the level of the road closest to the outlet based on the current outlet location and queries the database to find the road width according to the road level. The water ripple analysis algorithm is specifically a single flow direction algorithm.

[0043] The monitoring station setting unit 2 also includes a water accumulation depth output module 203 and a station deployment module 204. The water accumulation depth output module 203 calculates the corresponding surface flow based on rainfall intensity, the average runoff coefficient of each sub-region, and the surface area. After determining the difference between the surface flow of each sub-region and the designed drainage flow, it uses this as the actual overflow flow. It then uses a regional water accumulation analysis algorithm to calculate the corresponding hydraulic radius and surface runoff velocity based on these data. The water accumulation depth of the catchment area is analyzed using the hydraulic radius and road width. The station deployment module 204 analyzes the corresponding flow path using the cumulative flow direction value of each unit region. Through a rainfall model, it performs fusion analysis based on these data and outputs a real-time water accumulation distribution map of the specified area. Corresponding water accumulation monitoring stations are deployed according to the catchment area location and water accumulation depth in the distribution map. The rainfall model is specifically an SWMM simulation model, and the regional water accumulation analysis algorithm is as follows:

[0044]

[0045] in, Indicates the hydraulic radius. This represents the actual overflow traffic. Indicates the cross-sectional area of ​​the water passage. Represents the correlation coefficient. This represents the average slope value of the sub-region. Indicates surface runoff velocity. Indicates parameters, Indicates the depth of water accumulation in the catchment area. Indicates the average width of the road;

[0046] The water depth prediction unit 3 includes a historical data processing module 301, a matrix construction module 302, and a prediction value output module 303. The historical data processing module 301 acquires historical meteorological datasets and water depth datasets from various stations. After setting time intervals, it performs interpolation on the water depth data and then resamples it, ensuring a one-to-one correspondence between the meteorological data and the water depth data in the time series. The matrix construction module 302 sets four training modes and sets corresponding sliding window sizes according to each mode. It then segments the meteorological data and water depth data according to the sliding window size and combines them into a multidimensional matrix. The prediction value output module 303 transmits the multidimensional matrix as multiple input vectors to the water depth prediction model for analysis. It replaces the first element of the original input vector with the prediction result, deletes the last element, analyzes the new input vector, and repeats this process until the prediction step size is reached. Finally, it outputs the predicted values ​​for all water depths within future time intervals using a prediction value analysis algorithm. The water depth prediction model is specifically a regressor using a recursive prediction strategy algorithm. The prediction value analysis algorithm is as follows:

[0047]

[0048] in, The predicted value representing the depth of the water accumulation. This represents the model prediction mapping function. Represents the deviation vector. This indicates the actual depth of the accumulated water;

[0049] The recognition model training unit 4 includes a sample data collection module 401, a parameter update module 402, a test data input module 403, and a training termination module 404. The sample data collection module 401 extracts multiple vehicle side images, delineates the wheel regions in these images, adds corresponding labels based on the wheel model, and processes the images by adjusting brightness and contrast, adding noise, and rotating them to obtain new image data. This new image data, along with the original image data, is stored in the sample dataset. The parameter update module 402 divides the image data in the sample dataset into three categories: test data, training data, and validation data. After setting the training period, the training data is input one by one into the feature extraction backbone network model. This network model is used to recognize and segment these vehicle image data and calculate the model weights. The loss function is used to calculate the corresponding loss value based on the deviation between the drawn wheel area and the recognition area. The corresponding parameters are updated according to the loss value. After each training cycle, the test data input module 403 inputs the test data into the current network model for analysis. After calculating the model's localization accuracy and segmentation accuracy based on the deviation between the drawn wheel area and the recognition area, it determines whether the localization accuracy and segmentation accuracy have reached the preset value. If they have reached the predicted value, the verification data is input into the network model. Otherwise, the training cycle is entered again. After the model receives the verification data, the training end module 404 analyzes the data and determines whether the model should stop training. If the model stops training, the current model parameters and corresponding weights are determined. Otherwise, the next training cycle is entered. The feature extraction backbone network model is specifically the Mask R-CNN model.

[0050] The actual water accumulation analysis unit 5 includes an arc segment selection module 501 and a sub-segment generation module 502. The arc segment selection module 501 uses a drone to photograph vehicles on the road closest to each water catchment area and transmits the collected image data to the network model. After the model identifies the wheel area and model, it determines the minimum bounding matrix corresponding to the wheel area. It uses an edge extraction algorithm to analyze all edge points corresponding to the area, connects adjacent edge points, selects all arc segments in the area, stores them in a set, and calculates the relative length of each arc segment. If the arc segment length is lower than a preset value, the arc segment is deleted; otherwise, it is retained. The sub-segment generation module 502 divides all arc segments in the set to obtain multiple sub-segments, determines the angle between the sub-segments, and uses an arc segment determination algorithm to analyze these sub-segments. Sub-segments belonging to the same arc segment are stored in the same candidate set. The edge extraction algorithm is specifically the Canny algorithm, and the arc segment determination algorithm is specifically:

[0051]

[0052] in, Indicates the first From the first arc point to the second The spacing between the arc points Indicates the first From the first arc point to the second The spacing between the arc points Indicates parameters, Indicates the first The arc segment to the first The angle between the arc points;

[0053] The actual water accumulation analysis unit 5 also includes a matrix value acquisition module 503, an ellipse determination module 504, and an actual value output module 505. The matrix value acquisition module 503, after counting all segments in each candidate set, arbitrarily selects a segment and calculates the distance between the terminal arc point of the segment and the starting arc point of another segment, as well as the distance between the starting arc point and the terminal arc point of the segment. It compares these distances; if the former is greater than or equal to the latter, the two segments are determined not to be adjacent, and the corresponding point in the adjacency matrix is ​​set to zero. If the former is less than the latter, the two segments are determined to be adjacent. The module then analyzes the curvature and concavity / convexity corresponding to the terminal arc point of the current segment and the starting arc point of another segment to determine whether these two arc points belong to the same arc segment. If they belong to the same arc segment, the corresponding point in the adjacency matrix is ​​set to one; otherwise, it is set to zero. The value is negative one, and the operation is repeated until the value of all points in the adjacency matrix is ​​determined. After the ellipse determination module 504 determines all arc segments belonging to the same candidate ellipse according to the adjacency matrix, it numbers these arc segments one by one, obtains different candidate combinations, fits them, samples the fitting results, analyzes various indicators of the sampled points, and judges the corresponding verification score according to the indicator data. If the verification score is higher than the preset value, the current fitting result is determined to be a true ellipse; otherwise, it is a false ellipse. After the actual value output module 505 extracts the image data corresponding to the position of the true ellipse, it analyzes the radius of the wheel in the image according to the center position of the ellipse, queries the database according to the wheel model, determines the actual height of the wheel, and then uses a depth analysis algorithm to analyze and obtain the actual water depth. The depth analysis algorithm is as follows:

[0054]

[0055] in, Indicates the actual water depth. This indicates the height at which the wheels are visible in the image. This represents the radius of the wheel in the image. This indicates the actual height of the wheel.

[0056] Working Principle: This invention analyzes the random runoff coefficient for each land cover type using the runoff coefficient analysis module 101 in the drainage flow calculation unit 1, calculates the seasonal factor based on the median and compensation coefficient using the seasonal factor analysis module 102, calculates the comprehensive runoff coefficient using the comprehensive runoff analysis module 103, obtains the design drainage flow for each sub-region using the drainage capacity analysis module 104, and calculates the average runoff coefficient for each sub-region using the average runoff analysis module 105 based on these data. The cumulative flow direction calculation module 201 in the monitoring station setting unit 2 analyzes the cumulative flow direction value for each unit region. The road width determination module 202 analyzes the level of the nearest road to the current outlet based on its location and queries the database for the road width according to the road level. The water depth output module 203 analyzes the water depth of the catchment area. The station deployment module 204 outputs a real-time water distribution map of the designated area, and corresponding water monitoring stations are deployed according to the catchment area location and water depth on the distribution map. The historical data processing module 301 in the water depth prediction unit 3... Meteorological data and water depth data are correlated one-to-one in time series. Matrix construction module 302 combines meteorological data and water depth data into a multidimensional matrix. Prediction output module 303 outputs the predicted values ​​of all water depths within future time intervals. Sample dataset is constructed by sample data collection module 401 in model training unit 4. Parameter update module 402 updates the corresponding model parameters according to the loss value. The model's positioning accuracy and segmentation accuracy are calculated using test data input module 403. Training end module 404 determines the current model parameters and corresponding weights. The relative length of each arc segment is calculated by arc segment screening module 501 in actual water analysis unit 5, and arc segments that meet the conditions are retained. Sub-segment generation module 502 stores sub-segments belonging to the same arc segment into the same candidate set. Matrix value module 503 determines the values ​​of all points in the adjacency matrix. Ellipse determination module 504 determines that the current fitting result is a true ellipse. Actual value output module 505 obtains the actual water depth. Emergency response unit 6 formulates evacuation routes and plans based on these data.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A drone-assisted urban disaster emergency response and intelligent management system, characterized in that, include: The drainage flow calculation unit (1) is used to analyze the runoff coefficient, seasonal factors, comprehensive runoff, drainage capacity and average runoff; The monitoring station setting unit (2) is used for cumulative flow direction calculation, road width determination, water depth output and station layout; Water depth prediction unit (3) acquires historical meteorological datasets and water depth datasets for each station, maps the meteorological data and water depth data one-to-one in the time series, divides the meteorological data and water depth data according to the specified sliding window size, combines the two into a multidimensional matrix, transmits the multidimensional matrix as multiple input vectors to the water depth prediction model for analysis, and outputs the predicted values ​​of all water depths in the future time interval. The recognition model training unit (4) constructs a sample dataset and divides the vehicle image data in the dataset into three categories: test data, training data, and validation data. The training data is input into the feature extraction backbone network model one by one. The network model is used to identify and segment these vehicle image data and calculate the model weights. After each training cycle, the test data is input into the current network model for analysis. After obtaining the model's positioning accuracy and segmentation accuracy, the validation data is input into the network model to determine the final parameters of the current model and the corresponding weights. The actual water accumulation analysis unit (5) transmits the image data collected by the UAV to the network model. After the model identifies the wheel area and model, it filters out all arcs in the area, divides these arcs into multiple sub-segments, analyzes these sub-segments using the arc segment determination algorithm, stores the sub-segments belonging to the same arc segment in the same candidate set, arbitrarily selects a sub-segment, calculates the distance between the end arc point of the sub-segment and the beginning arc point of another sub-segment, and the distance between the beginning arc point and the end arc point of the sub-segment, compares the two to determine the value of the corresponding point in the adjacency matrix, analyzes all arcs belonging to the same candidate ellipse according to the adjacency matrix, fits them, samples the fitting result, determines that the current fitting result is the real ellipse, extracts the image data corresponding to the real ellipse position, calculates it using the depth analysis algorithm, and outputs the actual water accumulation depth. Emergency response unit (6) analyzes the predicted water depth of all stations. When the predicted value of a specified time interval exceeds the emergency threshold, it will report to the control center. The control center will retrieve the actual water depth of the nearest road in the catchment area where the current station is located, use deep learning technology to formulate evacuation routes and emergency plans based on the actual water depth, and output them through a visualization interface.

2. The urban disaster emergency response and intelligent management system based on unmanned aerial vehicle (UAV) assistance according to claim 1, characterized in that: The drainage flow calculation unit (1) includes a runoff coefficient analysis module (101), a seasonal factor analysis module (102), a comprehensive runoff analysis module (103), a drainage capacity analysis module (104), and an average runoff analysis module (105). The runoff coefficient analysis module (101) obtains the duration, interval, rainfall rate, attenuation coefficient, and variation coefficient, calculates the rainfall intensity and total rainfall based on these data, determines the corresponding basic runoff coefficient according to the land cover type, and analyzes the random runoff coefficient for each land cover type according to the current rainfall intensity, duration, and basic runoff coefficient. The seasonal factor analysis module (102) extracts historical rainfall data, calculates the average rainfall for different months, sets a first transition threshold and a second transition threshold. If the average rainfall for the current month is greater than the first transition threshold and less than the second transition threshold, the current month is marked as a transition month. All transition months are sorted according to the average rainfall. After determining the median of this set of data, a compensation coefficient is set. The seasonal factor is calculated based on the median and compensation coefficient. The comprehensive runoff analysis module (103) extracts the risk coefficient, interval coefficient and influence coefficient. Based on the random runoff coefficient, seasonal factor, risk coefficient, interval coefficient and influence coefficient, the corresponding comprehensive runoff coefficient is analyzed. The drainage capacity analysis module (104) counts the data collected by each meteorological station and analyzes the rainfall trend surface based on these data. According to the station location and rainfall trend surface, the designated area is divided into multiple sub-regions. The surface area, drainage outlet flow and number of each sub-region are determined. The drainage pipes in the sub-region are numbered one by one. The length and radius of each drainage pipe are counted. The drainage capacity analysis algorithm is used to calculate these data to obtain the design drainage flow of the sub-region. The average runoff analysis module (105) divides each sub-region into multiple unit regions. After analyzing the comprehensive runoff coefficient of each unit region according to the weight coefficient corresponding to the surface cover type and interval time, the average runoff coefficient of each sub-region is calculated based on these data.

3. The urban disaster emergency response and intelligent management system based on unmanned aerial vehicle (UAV) assistance according to claim 2, characterized in that: The monitoring station setting unit (2) includes a cumulative flow direction calculation module (201) and a road width determination module (202). The cumulative flow direction calculation module (201) calculates the altitude and slope of each unit area, analyzes the data using a water ripple analysis algorithm, obtains the flow direction value corresponding to each unit area, traverses each area, determines the building height in each unit area, and analyzes the cumulative flow direction value of each unit area based on the data. The road width determination module (202) calculates the cumulative flow direction value of each unit area, sets the area with a cumulative flow direction value lower than the preset value as a water catchment area, determines the outlet location of each water catchment area and the grade and distribution location of all roads in the specified area, analyzes the grade of the road closest to the outlet based on the current outlet location, and queries the database for the road width according to the road grade.

4. The urban disaster emergency response and intelligent management system based on unmanned aerial vehicle (UAV) assistance according to claim 3, characterized in that: The monitoring station setting unit (2) further includes a water depth output module (203) and a station layout module (204). The water depth output module (203) calculates the corresponding surface flow based on the rainfall intensity, the average runoff coefficient of each sub-region and the surface area. After determining the difference between the surface flow of each sub-region and the design drainage flow, it takes it as the actual overflow flow. It also uses the regional water accumulation analysis algorithm to calculate the corresponding hydraulic radius and surface runoff velocity based on these data. It uses the hydraulic radius and road width to analyze the water accumulation depth of the catchment area. The station layout module (204) uses the cumulative flow direction value of each unit area to analyze the corresponding flow path. It uses the rainfall model to perform fusion analysis based on these data and outputs a real-time water accumulation distribution map in the specified area. The corresponding water accumulation monitoring stations are set up according to the location of the catchment area and the water accumulation depth in the distribution map.

5. The urban disaster emergency response and intelligent management system based on unmanned aerial vehicle (UAV) assistance according to claim 1, characterized in that: The water depth prediction unit (3) includes a historical data processing module (301), a matrix construction module (302), and a prediction value output module (303). The historical data processing module (301) acquires historical meteorological datasets and water depth datasets for each station. After setting a time interval, it performs interpolation on the water depth data and then performs resampling to make the meteorological data and water depth data correspond one-to-one in the time series. The matrix construction module (302) sets four training modes. After setting the corresponding sliding window size according to the training mode, it divides the meteorological data and water depth data according to the sliding window size and combines the two into a multidimensional matrix. The prediction value output module (303) transmits the multidimensional matrix as multiple input vectors to the water depth prediction model for analysis. After replacing the first element in the original input vector with the prediction result, it deletes the last element in the input vector and analyzes the new input vector. The operation is repeated until the prediction step size is reached. The prediction value analysis algorithm is used to output the prediction value of all water depths in the future time interval.

6. The urban disaster emergency response and intelligent management system based on unmanned aerial vehicle (UAV) assistance according to claim 1, characterized in that: The recognition model training unit (4) includes a sample data collection module (401), a parameter update module (402), a test data input module (403), and a training termination module (404). The sample data collection module (401) extracts multiple vehicle side images, delineates the wheel areas in these side images, adds corresponding label information according to the wheel model, and processes the images by adjusting brightness and contrast, adding noise, and rotating to obtain new image data. These image data and the original image data are stored in the sample dataset. The parameter update module (402) divides the image data in the sample dataset into three categories: test data, training data, and validation data. After setting the training period, the training data is input into the feature extraction backbone network model one by one, and the network model is used to process these vehicle image data. The model identifies and segments the model, calculates the model weights, and uses a loss function to calculate the corresponding loss value based on the deviation between the drawn wheel area and the identified area. The corresponding parameters are updated based on the loss value. After each training cycle, the test data input module (403) inputs the test data into the current network model for analysis. After calculating the model's positioning accuracy and segmentation accuracy based on the deviation between the drawn wheel area and the identified area, it determines whether the positioning accuracy and segmentation accuracy have reached the preset value. If they have reached the predicted value, the verification data is input into the network model. Otherwise, the model enters the training cycle again. After the model receives the verification data, the training end module (404) analyzes the data and determines whether the model should stop training. If the model stops training, the current model parameters and corresponding weights are determined. Otherwise, the model enters the next training cycle.

7. The urban disaster emergency response and intelligent management system based on unmanned aerial vehicle (UAV) assistance according to claim 1, characterized in that: The actual water accumulation analysis unit (5) includes an arc segment screening module (501) and a sub-segment generation module (502). The arc segment screening module (501) uses a drone to take pictures of vehicles on the road closest to each water catchment area and transmits the collected image data to the network model. After the model identifies the wheel area and model, it determines the minimum outer matrix corresponding to the wheel area. It uses an edge extraction algorithm to analyze all edge points corresponding to the area, connects adjacent edge points, filters out all arc segments in the area, stores them in a set, and calculates the relative length of each arc segment. If the arc segment length is lower than a preset value, the arc segment is deleted; otherwise, the arc segment is retained. The sub-segment generation module (502) divides all arc segments in the set to obtain multiple sub-segments, determines the angle between the sub-segments, and uses an arc segment determination algorithm to analyze these sub-segments. Sub-segments belonging to the same arc segment are stored in the same candidate set.

8. The urban disaster emergency response and intelligent management system based on unmanned aerial vehicle (UAV) assistance according to claim 7, characterized in that: The actual water accumulation analysis unit (5) further includes a matrix value module (503), an ellipse determination module (504), and an actual value output module (505). The matrix value module (503) counts all sub-segments contained in each candidate set, then arbitrarily selects a sub-segment, calculates the distance between the end arc point of the sub-segment and the beginning arc point of another sub-segment, and the distance between the beginning arc point and the end arc point of the sub-segment, and compares the two. If the former is greater than or equal to the latter, the two sub-segments are determined not to be adjacent, and the corresponding point in the adjacency matrix is ​​set to zero. If the former is less than the latter, the two sub-segments are determined to be adjacent. The curvature and concavity / convexity of the end arc point of the current sub-segment and the beginning arc point of another sub-segment are analyzed to determine whether the two arc points belong to the same arc segment. If they are determined to belong to the same arc segment, the corresponding point in the adjacency matrix is ​​set to one. If they are determined not to belong to the same arc segment, the value of the corresponding point in the adjacency matrix is ​​set to one. For the same arc segment, the value is negative one. The operation is repeated until the value of all points in the adjacency matrix is ​​determined. After the ellipse determination module (504) determines all arc segments belonging to the same candidate ellipse according to the adjacency matrix, it numbers these arc segments one by one, obtains different candidate combinations, fits them, samples the fitting results, analyzes each index of the sampling points, and judges the corresponding verification score according to the index data. If the verification score is higher than the preset value, the current fitting result is determined to be a real ellipse, otherwise it is a false ellipse. After the actual value output module (505) extracts the image data corresponding to the position of the real ellipse, it analyzes the radius of the wheel in the image according to the center position of the ellipse, queries the database according to the wheel model, determines the actual height of the wheel, and uses the depth analysis algorithm to analyze and obtain the actual water depth.

Images