Flood early warning information acquisition method and system based on geographic information

Abstract

The invention relates to the technical field of hydrological monitoring, and particularly discloses a flood early warning information acquisition method and system based on geographic information, and the method comprises the steps: constructing a three-dimensional scene; a rainfall detection point is selected based on the three-dimensional scene, and scene states of the rainfall detection point under different rainfall amount conditions are acquired; creating a sample set from precipitation conditions to scene states, and training a neural network model based on the sample set; reading precipitation in future time from the weather forecast information, determining a scene state based on the trained neural network model, and generating early warning information based on the scene state; according to the method, the three-dimensional scene is generated based on the geographic information, the precipitation information is obtained based on the three-dimensional scene, the scene state at the corresponding moment is obtained, the mapping relation from the precipitation information to the scene state is trained, the scene state can be directly predicted according to the predicted precipitation information during prediction, the fitting degree with the actual situation is higher, and the early warning accuracy is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological monitoring, and in particular to a method and system for collecting flood warning information based on geographic information. Background Art

[0002] Flood refers to the phenomenon that low-lying areas are flooded or waterlogged due to heavy rain, rainstorm or continuous rainfall. Flood mainly harms the growth of crops, causing crop yield reduction or crop failure, and disrupting the normal development of agricultural production and other industries.

[0003] The flood warning process is essentially an analysis of rainfall information. The existing flood warning process mostly relies on some fixed indicators. For example, how much rainfall is reached and what is the probability of flooding. This one-to-one correspondence in values ​​may differ from the actual situation. In actual scenarios, even if the rainfall in two areas is the same, there is a high possibility that one area will experience flooding while the other will not. Therefore, how to improve the fit between flood warning technology and actual conditions and improve the accuracy of warnings is a technical problem that this application wants to solve. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for collecting flood warning information based on geographic information to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: A method and system for collecting flood warning information based on geographic information, the method comprising: Acquire point cloud data of the area based on a drone carrying a LiDAR device, perform three-dimensional modeling of the area based on the point cloud data, and construct a three-dimensional scene; Selecting a precipitation detection point based on a three-dimensional scene, and obtaining a scene state of the precipitation detection point under different precipitation conditions; wherein the precipitation condition is the precipitation within a preset period, the scene state is a parameter change, and the parameter type corresponding to the parameter change includes at least an image; Create a sample set from precipitation conditions to scene states, and train a neural network model based on the sample set; Reading the precipitation in the future from the weather forecast information, determining the scene state based on the trained neural network model, and generating warning information based on the scene state; updating the neural network model according to the scene state; Among them, the mapping relationship between the scene state and the warning information is preset. After the scene state is determined by applying the trained neural network model, the scene state is input into the mapping relationship as an independent variable, and the output is used as the warning information.

[0006] As a further solution of the present invention: the step of acquiring point cloud data of a region based on a drone carrying a LiDAR device, performing three-dimensional modeling of the region based on the point cloud data, and constructing a three-dimensional scene comprises: Receive boundary coordinates input by the user and fit the boundary contour; Query the detection radius of the drone carrying the LiDAR device, array the contour based on the detection radius, and obtain equidistant contour lines; wherein the array distance is the product of the collection radius and a preset ratio; the ratio ranges from 0.5 to 1; Select the starting point and the end point in the equidistant contour line to obtain the acquisition path; Sending a collection path to at least one drone to obtain point cloud data of the area; the drone carries a LiDAR device; The region is three-dimensionally modeled based on the point cloud data to construct a three-dimensional scene.

[0007] As a further solution of the present invention: the step of selecting a precipitation detection point based on a three-dimensional scene and obtaining the scene state of the precipitation detection point under different precipitation conditions includes: Get the highest and lowest points of the 3D scene and calculate the height difference. The height difference is divided according to a preset vertical step size to obtain a height set, and a plane set parallel to the horizontal plane is generated based on the height set; Get the intersection of each plane and the three-dimensional scene, and select precipitation detection points based on the complement of the intersection; The precipitation conditions are determined according to the preset rain gauge, the scene status at the precipitation detection point is obtained, and the precipitation conditions and the scene status are aligned in the time domain.

[0008] As a further solution of the present invention: the step of determining the precipitation condition according to the preset rain gauge, obtaining the scene state at the precipitation detection point, and performing time domain registration on the precipitation condition and the scene state comprises: Obtain precipitation with location tag and time tag according to the preset rain gauge; Query the three-dimensional scene at the most recent moment according to the time tag, and insert the precipitation with similar time tags into the three-dimensional scene according to the location tag as the precipitation boundary condition; similar time tags mean that the time difference of the time tags is less than a preset threshold; Based on the hydrological modeling software, a hydrological simulation is performed on the three-dimensional scene containing precipitation boundary conditions to obtain the precipitation at other locations; wherein the distance between the locations is a preset value; Count the precipitation at all locations as the precipitation condition; Acquire scene images with time tags based on image collectors installed at precipitation detection points to determine scene states; Temporal registration of precipitation conditions and scene states.

[0009] As a further solution of the present invention: the step of reading the precipitation in the future time from the weather forecast information, determining the scene state based on the trained neural network model, and generating warning information based on the scene state; and updating the neural network model according to the scene state includes: Read the precipitation amount in the future time in the weather forecast information; Simulate precipitation conditions based on current up-to-date 3D scenarios and precipitation; The precipitation conditions are input into the neural network model, and the scene status is output; When the scene state is acquired through the image collector, the scene state acquired by the image collector is compared with the scene state output by the neural network model to determine the error rate; When the error rate meets the preset update conditions, the sample set is updated, and then the neural network model is updated.

[0010] As a further solution of the present invention: the scene state comparison process uses image similarity; the error rate update condition is: ; In the formula, For the current moment, is the preset maximum time length, express Unit time length; express The error rate corresponding to the moment, is the preset threshold.

[0011] The technical solution of the present invention also provides a flood warning information collection system based on geographic information, the system comprising: A three-dimensional scene construction module is used to obtain point cloud data of a region based on a drone carrying a LiDAR device, perform three-dimensional modeling of the region based on the point cloud data, and construct a three-dimensional scene; A data acquisition module, used to select precipitation detection points based on a three-dimensional scene, and obtain scene states of the precipitation detection points under different precipitation conditions; wherein the precipitation condition is the precipitation within a preset period, the scene state is the parameter change, and the parameter type corresponding to the parameter change includes at least an image; The sample set application module is used to create a sample set from precipitation conditions to scene states, and train a neural network model based on the sample set; A model application module is used to read the precipitation in the future from the weather forecast information, determine the scene state based on the trained neural network model, generate warning information based on the scene state; and update the neural network model according to the scene state; Among them, the mapping relationship between the scene state and the warning information is preset. After the scene state is determined by applying the trained neural network model, the scene state is input into the mapping relationship as an independent variable, and the output is used as the warning information.

[0012] As a further solution of the present invention: the three-dimensional scene construction module includes: A contour fitting unit, used to receive boundary coordinates input by a user and fit the boundary contour; An equidistant line generation unit is used to query the detection radius of the drone carrying the LiDAR device, array the contour based on the detection radius, and obtain an equidistant contour line; wherein the array distance is the product of the acquisition radius and a preset ratio; the ratio ranges from 0.5 to 1; A path generation unit is used to select a starting point and an end point in the equidistant contour line to obtain a collection path; A data acquisition unit, used to send a collection path to at least one drone to acquire point cloud data of an area; the drone carries a LiDAR device; The modeling unit is used to perform three-dimensional modeling on the area based on the point cloud data to construct a three-dimensional scene.

[0013] As a further solution of the present invention: the data acquisition module includes: A height difference calculation unit, used to obtain the height of the highest point and the lowest point of the three-dimensional scene and calculate the height difference; A plane set generating unit, used for dividing the height difference according to a preset vertical step length to obtain a height set, and generating a plane set parallel to the horizontal plane based on the height set; A point selection unit is used to obtain the intersection of each plane and the three-dimensional scene, and select precipitation detection points based on the complement of the intersection; The time domain registration unit is used to determine the precipitation condition according to the preset rain gauge, obtain the scene state at the precipitation detection point, and perform time domain registration on the precipitation condition and the scene state.

[0014] As a further solution of the present invention: the model application module includes: A forecast information reading unit, used to read the precipitation amount in the future time from the weather forecast information; A condition simulation unit for simulating precipitation conditions based on the current latest three-dimensional scene and precipitation; A state output unit, used to input precipitation conditions into the neural network model and output the scene state; An error rate calculation unit, used for comparing the scene state acquired by the image collector with the scene state output by the neural network model to determine the error rate when the scene state is acquired by the image collector; The updating unit is used to update the sample set and then update the neural network model when the error rate meets the preset updating condition.

[0015] Compared with the prior art, the beneficial effects of the present invention are: the present invention generates a three-dimensional scene based on geographic information, obtains precipitation information based on the three-dimensional scene, and simultaneously obtains the scene state at the corresponding moment, and trains the mapping relationship from precipitation information to scene state. During prediction, the scene state can be directly predicted according to the predicted precipitation information, which has a higher fit with the actual situation and a higher warning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0017] Figure 1 The overall flow chart of the flood warning information collection method based on geographic information is shown.

[0018] Figure 2 The structural diagram of the flood warning information collection system based on geographic information is shown. DETAILED DESCRIPTION

[0019] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, 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.

[0020] Figure 1 The following is a general flow chart of a method and system for collecting flood warning information based on geographic information. In an embodiment of the present invention, a method for collecting flood warning information based on geographic information includes: Step S100: acquiring point cloud data of a region based on a drone carrying a LiDAR device, performing three-dimensional modeling of the region based on the point cloud data, and constructing a three-dimensional scene; LiDAR is an active radar, which is a technology that measures the distance between an object and a sensor by emitting laser pulses and receiving reflected signals. LiDAR equipment carried by aircraft or drones can be used to scan the entire city to generate detailed elevation data and building outlines. In the technical solution of the present invention, it is necessary to model an area, which is generally a sub-area of ​​a city, and the 3D modeling process is not complicated.

[0021] Step S200: selecting a precipitation detection point based on a three-dimensional scene, and obtaining a scene state of the precipitation detection point under different precipitation conditions; wherein the precipitation condition is the precipitation within a preset period, the scene state is a parameter change, and the parameter type corresponding to the parameter change includes at least an image; After the three-dimensional modeling is completed, a three-dimensional scene is obtained. By analyzing the three-dimensional scene, some points can be selected as precipitation detection points. Detectors are installed at the precipitation detection points to obtain the status under different precipitation conditions. This actually reflects the dependent variable, that is, what changes will the precipitation detection points undergo due to the precipitation.

[0022] Furthermore, regarding the precipitation condition, this is obtained by an existing rain gauge and represents the independent variable, that is, how much precipitation there is in a period of time.

[0023] It is worth mentioning that the corresponding period of the rain gauge is generally hours, and each acquisition is the precipitation for one hour, which is equivalent to increasing the precipitation on the state one hour ago. Therefore, the output, that is, the scene state is limited to the change amount, called the parameter change amount; the scene state is a superordinate concept, which can include multiple types. In this application, at least it includes an image, which reflects the visual state at the precipitation detection point. In fact, various soil parameters can also be introduced. Accordingly, a detector for detecting soil parameters should be installed at the precipitation detection point.

[0024] Step S300: creating a sample set from precipitation conditions to scene states, and training a neural network model based on the sample set; A sample set of precipitation conditions to scene states is created, with precipitation conditions as features and scene states as labels, and a neural network model can be trained. This is a conventional technical means in the prior art. Specifically, in practical applications, precipitation conditions are often a numerical matrix, and scene states are a bunch of images. The trained neural network model is a mapping of numerical matrices to images. A numerical matrix is ​​input and an image corresponding to the numerical matrix is ​​output. In fact, using a mapping table is also a feasible solution. The precipitation conditions and scene states are counted to build a one-to-one relationship. When in use, the data closest to the input precipitation conditions is queried to read the scene state. Although this method is logically simple, if you want to achieve a more accurate reading effect, the length of the data table will be very large, and the traversal process is very troublesome. Therefore, the present application uses a neural network model to establish a mapping relationship from precipitation conditions to images.

[0025] Step S400: Read the precipitation in the future from the weather forecast information, determine the scene state based on the trained neural network model, generate warning information based on the scene state; and update the neural network model according to the scene state.

[0026] By reading the precipitation in the future from the weather forecast information and inputting the precipitation in the future into the neural network model, the scene state can be obtained. The scene state represents the parameter change at the precipitation detection point under the precipitation conditions in the future. This is the application process of the neural network model. Specifically, by analyzing the scene state output by the neural network model, it can be determined whether floods will occur, and then early warning information is generated. The relationship between the scene state and the early warning information is essentially the existing judgment rules, and can also be completed with the help of existing evaluation models. It is worth mentioning that existing evaluation models are often also neural network models.

[0027] As time goes by, the equipment at the precipitation detection point will actually obtain the actual data, compare it with the neural network model, judge the accuracy of the neural network model, and then update the neural network model.

[0028] Regarding step S100, the steps of acquiring point cloud data of a region based on a drone carrying a LiDAR device, performing three-dimensional modeling of the region based on the point cloud data, and constructing a three-dimensional scene include: Receive boundary coordinates input by the user and fit the boundary contour; Query the detection radius of the drone carrying the LiDAR device, array the contour based on the detection radius, and obtain equidistant contour lines; wherein the array distance is the product of the collection radius and a preset ratio; the ratio ranges from 0.5 to 1; Select the starting point and the end point in the equidistant contour line to obtain the acquisition path; Sending a collection path to at least one drone to obtain point cloud data of the area; the drone carries a LiDAR device; The region is three-dimensionally modeled based on the point cloud data to construct a three-dimensional scene.

[0029] The boundary coordinates are input by the user. By fitting the boundary coordinates, the boundary contour can be obtained. Based on the boundary contour, multiple equidistant lines can be constructed, and based on multiple equidistant lines, a collection path can be constructed. Among them, the distance between the equidistant lines is determined by the working parameters of the drone. The collection height and collection angle of the drone can determine a collection range (circle). The radius of the collection range can be obtained as the interval between the equidistant lines.

[0030] It is worth mentioning that in the technical solution of the present invention, the interval of the equidistant lines can be 90% of the radius of the acquisition range. At this time, when the drone collects close-up images, there will be a small overlap. This overlap can self-verify the image acquisition process of the drone. The self-verification method is that for three adjacent equidistant lines, a part of the image on both sides of the middle equidistant line is collected during the acquisition process on the adjacent equidistant lines. If the overlapping areas are the same, it can be determined that the images are all correct images. If one of the overlapping areas is wrong (the two images are different in the overlapping areas), then the images on the adjacent equidistant lines can be regarded as abnormal images. If both overlapping areas are wrong, then the image on the middle equidistant line can be regarded as an abnormal image. Since the concept of error in the recognition process is very low (but not zero), the above scheme can be used to extremely quickly and accurately determine abnormal images, and they can be eliminated in subsequent use.

[0031] It should be noted that the starting point and the end point in the collection path are the point closest to the drone device as the starting point, and the last point after traversing the collection path is called the end point. There is a certain degree of randomness in the setting, so there is no limit.

[0032] Regarding step S200, the step of selecting a precipitation detection point based on the three-dimensional scene and obtaining the scene state of the precipitation detection point under different precipitation conditions includes: Get the highest and lowest points of the 3D scene and calculate the height difference. The height difference is divided according to a preset vertical step size to obtain a height set, and a plane set parallel to the horizontal plane is generated based on the height set; Get the intersection of each plane and the three-dimensional scene, and select precipitation detection points based on the complement of the intersection; The precipitation conditions are determined according to the preset rain gauge, the scene status at the precipitation detection point is obtained, and the precipitation conditions and the scene status are aligned in the time domain.

[0033] A plane is created at a certain height, and the plane and the three-dimensional scene will have an intersection. The intersection is an entity, such as the interior of a building. The complement of the intersection corresponds to the external space. It is necessary to select precipitation detection points in the complement of the intersection. The selection method is random. In addition, the lower the height of the plane, the easier it is to accumulate water, and then form floods. Therefore, the lower the height, the more precipitation detection points there are. On this basis, the precipitation conditions and scene states within the same time (the time difference is small enough) are obtained as sample data. Among them, the precipitation conditions are obtained by the rain gauge.

[0034] Furthermore, the step of determining the precipitation condition according to a preset rain gauge, obtaining the scene state at the precipitation detection point, and performing time domain registration on the precipitation condition and the scene state includes: Obtain precipitation with location tag and time tag according to the preset rain gauge; Query the three-dimensional scene at the most recent moment according to the time tag, and insert the precipitation with similar time tags into the three-dimensional scene according to the location tag as the precipitation boundary condition; similar time tags mean that the time difference of the time tags is less than a preset threshold; Based on the hydrological modeling software, a hydrological simulation is performed on the three-dimensional scene containing precipitation boundary conditions to obtain the precipitation at other locations; wherein the distance between the locations is a preset value; Count the precipitation at all locations as the precipitation condition; Acquire scene images with time tags based on image collectors installed at precipitation detection points to determine scene states; Temporal registration of precipitation conditions and scene states.

[0035] The above content limits the process of obtaining precipitation conditions and scene states. When the rain gauge obtains precipitation, the location tag is determined according to the location of the rain gauge. When obtaining precipitation, the time point is recorded as a time tag. The three-dimensional scene at the most recent time is queried according to the time tag as a basis, and all precipitation within the same time is queried. The location tag is inserted into the basis as a precipitation boundary condition. Then, based on the hydrological modeling software, a hydrological simulation is performed on the three-dimensional scene containing the precipitation boundary condition to obtain the precipitation at all locations in the three-dimensional scene. Among them, the location refers to a group of points, and the distance between points is a preset value. There is a grid display function in the hydrological modeling software. The grid nodes of the grid are the above-mentioned locations. The precipitation at all locations is counted to obtain a matrix, which is called the precipitation condition.

[0036] An image collector is installed at the precipitation detection point, and the image collector obtains the scene image with a time tag as the scene state; finally, the precipitation condition and the scene state are aligned according to the time tag (the time difference can be small enough).

[0037] Regarding step S400, the step of reading the precipitation in the future time in the weather forecast information, determining the scene state based on the trained neural network model, and generating warning information based on the scene state; and updating the neural network model according to the scene state includes: Read the precipitation amount in the future time in the weather forecast information; Simulate precipitation conditions based on current up-to-date 3D scenarios and precipitation; The precipitation conditions are input into the neural network model, and the scene status is output; When the scene state is acquired through the image collector, the scene state acquired by the image collector is compared with the scene state output by the neural network model to determine the error rate; When the error rate meets the preset update conditions, the sample set is updated, and then the neural network model is updated.

[0038] Under existing technical conditions, the accuracy of weather forecast information is very high. The precipitation in the future is read from the weather forecast information, which is generally the precipitation within an hour. The precipitation in the future is converted into precipitation conditions using the above-mentioned precipitation condition generation scheme, and the precipitation conditions are input into the neural network model to output the scene state. As time goes by, when the scene state is obtained through the image collector, the scene state obtained by the image collector and the scene state output by the neural network model are compared to determine the error rate. When the error rate is large enough, the sample set needs to be updated by eliminating relatively old data. After the sample set is updated, the neural network model will also be updated accordingly.

[0039] In a preferred embodiment of the technical solution of the present invention, the scene state comparison process adopts image similarity; the image similarity can be calculated using an existing method.

[0040] The update condition of the error rate is: ; In the formula, For the current moment, is the preset maximum time length, express Unit time length; express The error rate corresponding to the moment, is the preset threshold.

[0041] The updating condition of the error rate means that the sum of multiple error rates in the recent period is large enough.

[0042] Figure 2 The structure diagram of the flood warning information collection system based on geographic information is shown. In a preferred embodiment of the technical solution of the present invention, a flood warning information collection system based on geographic information is also provided. The system 10 includes: A three-dimensional scene construction module 11 is used to obtain point cloud data of a region based on a drone carrying a LiDAR device, perform three-dimensional modeling of the region based on the point cloud data, and construct a three-dimensional scene; The data acquisition module 12 is used to select a precipitation detection point based on the three-dimensional scene, and obtain the scene state of the precipitation detection point under different precipitation conditions; wherein the precipitation condition is the precipitation within a preset period, the scene state is the parameter change, and the parameter type corresponding to the parameter change includes at least an image; A sample set application module 13, used to create a sample set from precipitation conditions to scene states, and train a neural network model based on the sample set; The model application module 14 is used to read the precipitation in the future time in the weather forecast information, determine the scene state based on the trained neural network model, generate warning information based on the scene state; and update the neural network model according to the scene state; The scene status application module 15 is used to generate warning information based on the scene status.

[0043] Furthermore, the three-dimensional scene construction module 11 includes: A contour fitting unit, used to receive boundary coordinates input by a user and fit the boundary contour; An equidistant line generation unit is used to query the detection radius of the drone carrying the LiDAR device, array the contour based on the detection radius, and obtain an equidistant contour line; wherein the array distance is the product of the acquisition radius and a preset ratio; the ratio ranges from 0.5 to 1; A path generation unit is used to select a starting point and an end point in the equidistant contour line to obtain a collection path; A data acquisition unit, used to send a collection path to at least one drone to acquire point cloud data of an area; the drone carries a LiDAR device; The modeling unit is used to perform three-dimensional modeling on the area based on the point cloud data to construct a three-dimensional scene.

[0044] Specifically, the data acquisition module 12 includes: A height difference calculation unit, used to obtain the height of the highest point and the lowest point of the three-dimensional scene and calculate the height difference; A plane set generating unit, used for dividing the height difference according to a preset vertical step length to obtain a height set, and generating a plane set parallel to the horizontal plane based on the height set; A point selection unit is used to obtain the intersection of each plane and the three-dimensional scene, and select precipitation detection points based on the complement of the intersection; The time domain registration unit is used to determine the precipitation condition according to the preset rain gauge, obtain the scene state at the precipitation detection point, and perform time domain registration on the precipitation condition and the scene state.

[0045] Furthermore, the model application module 14 includes: A forecast information reading unit, used to read the precipitation amount in the future time from the weather forecast information; A condition simulation unit for simulating precipitation conditions based on the current latest three-dimensional scene and precipitation; A state output unit, used to input precipitation conditions into the neural network model and output the scene state; An error rate calculation unit, used for comparing the scene state acquired by the image collector with the scene state output by the neural network model to determine the error rate when the scene state is acquired by the image collector; The updating unit is used to update the sample set and then update the neural network model when the error rate meets the preset updating condition.

[0046] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for collecting flood warning information based on geographic information, characterized in that: The method comprises: Acquire point cloud data of the area based on a drone carrying a LiDAR device, perform three-dimensional modeling of the area based on the point cloud data, and construct a three-dimensional scene; Selecting a precipitation detection point based on a three-dimensional scene, and obtaining a scene state of the precipitation detection point under different precipitation conditions; wherein the precipitation condition is the precipitation within a preset period, the scene state is a parameter change, and the parameter type corresponding to the parameter change includes at least an image; Create a sample set from precipitation conditions to scene states, and train a neural network model based on the sample set; Reading the precipitation in the future from the weather forecast information, determining the scene state based on the trained neural network model, and generating warning information based on the scene state; updating the neural network model according to the scene state; Among them, the mapping relationship between the scene state and the warning information is preset. After the scene state is determined by applying the trained neural network model, the scene state is input into the mapping relationship as an independent variable, and the output is used as the warning information.

2. The method for collecting flood warning information based on geographic information according to claim 1, characterized in that: The steps of acquiring point cloud data of a region based on a drone carrying a LiDAR device, performing three-dimensional modeling of the region based on the point cloud data, and constructing a three-dimensional scene include: Receive boundary coordinates input by the user and fit the boundary contour; Query the detection radius of the drone carrying the LiDAR device, array the contour based on the detection radius, and obtain equidistant contour lines; wherein the array distance is the product of the collection radius and a preset ratio; the ratio ranges from 0.5 to 1; Select the starting point and the end point in the equidistant contour line to obtain the acquisition path; Sending a collection path to at least one drone to obtain point cloud data of the area; the drone carries a LiDAR device; The region is three-dimensionally modeled based on the point cloud data to construct a three-dimensional scene.

3. The method for collecting flood warning information based on geographic information according to claim 1, characterized in that: The step of selecting a precipitation detection point based on the three-dimensional scene and obtaining the scene state of the precipitation detection point under different precipitation conditions includes: Get the highest and lowest points of the 3D scene and calculate the height difference. The height difference is divided according to a preset vertical step size to obtain a height set, and a plane set parallel to the horizontal plane is generated based on the height set; Get the intersection of each plane and the three-dimensional scene, and select precipitation detection points based on the complement of the intersection; The precipitation conditions are determined according to the preset rain gauge, the scene status at the precipitation detection point is obtained, and the precipitation conditions and the scene status are aligned in the time domain.

4. The method for collecting flood warning information based on geographic information according to claim 3 is characterized in that: The step of determining the precipitation condition according to the preset rain gauge, obtaining the scene state at the precipitation detection point, and performing time domain registration on the precipitation condition and the scene state comprises: Obtain precipitation with location tag and time tag according to the preset rain gauge; Query the three-dimensional scene at the most recent time according to the time tag, and insert the precipitation of similar time tags into the three-dimensional scene according to the location tag as the precipitation boundary condition; similar time tags mean that the time difference of the time tags is less than a preset threshold; Based on the hydrological modeling software, a hydrological simulation is performed on the three-dimensional scene containing precipitation boundary conditions to obtain the precipitation at other locations; wherein the distance between the locations is a preset value; Count the precipitation at all locations as the precipitation condition; Acquire scene images with time tags based on image collectors installed at precipitation detection points to determine scene states; Temporal registration of precipitation conditions and scene states.

5. The method for collecting flood warning information based on geographic information according to claim 1, characterized in that: The method comprises reading the precipitation in the future from the weather forecast information, determining the scene state based on the trained neural network model, and generating warning information based on the scene state; The steps to update the neural network model according to the scene state include: Read the precipitation amount in the future time in the weather forecast information; Simulate precipitation conditions based on current up-to-date 3D scenarios and precipitation; The precipitation conditions are input into the neural network model, and the scene status is output; When the scene state is acquired through the image collector, the scene state acquired by the image collector is compared with the scene state output by the neural network model to determine the error rate; When the error rate meets the preset update conditions, the sample set is updated, and then the neural network model is updated.

6. The method for collecting flood warning information based on geographic information according to claim 5, characterized in that: The scene state comparison process uses image similarity; the error rate update condition is: ; In the formula, For the current moment, is the preset maximum time length, express Unit time length; express The error rate corresponding to the moment, is the preset threshold.

7. A flood warning information collection system based on geographic information, characterized in that: The system comprises: A three-dimensional scene construction module is used to obtain point cloud data of a region based on a drone carrying a LiDAR device, perform three-dimensional modeling of the region based on the point cloud data, and construct a three-dimensional scene; A data acquisition module, used to select precipitation detection points based on a three-dimensional scene, and obtain the scene state of the precipitation detection points under different precipitation conditions; wherein the precipitation condition is the precipitation within a preset period, the scene state is the parameter change, and the parameter type corresponding to the parameter change includes at least an image; The sample set application module is used to create a sample set from precipitation conditions to scene states, and train a neural network model based on the sample set; A model application module is used to read the precipitation in the future from the weather forecast information, determine the scene state based on the trained neural network model, generate warning information based on the scene state; and update the neural network model according to the scene state; Among them, the mapping relationship between the scene state and the warning information is preset. After the scene state is determined by applying the trained neural network model, the scene state is input into the mapping relationship as an independent variable, and the output is used as the warning information.

8. The flood warning information collection system based on geographic information according to claim 7 is characterized in that: The three-dimensional scene construction module includes: A contour fitting unit, used to receive boundary coordinates input by a user and fit the boundary contour; An equidistant line generation unit is used to query the detection radius of the drone carrying the LiDAR device, array the contour based on the detection radius, and obtain an equidistant contour line; wherein the array distance is the product of the acquisition radius and a preset ratio; the ratio ranges from 0.5 to 1; A path generation unit is used to select a starting point and an end point in the equidistant contour line to obtain a collection path; A data acquisition unit, used to send a collection path to at least one drone to acquire point cloud data of an area; the drone carries a LiDAR device; The modeling unit is used to perform three-dimensional modeling on the area based on the point cloud data to construct a three-dimensional scene.

9. The flood warning information collection system based on geographic information according to claim 7 is characterized in that: The data acquisition module comprises: A height difference calculation unit, used to obtain the height of the highest point and the lowest point of the three-dimensional scene and calculate the height difference; A plane set generating unit, used for dividing the height difference according to a preset vertical step length to obtain a height set, and generating a plane set parallel to the horizontal plane based on the height set; A point selection unit is used to obtain the intersection of each plane and the three-dimensional scene, and select precipitation detection points based on the complement of the intersection; The time domain registration unit is used to determine the precipitation condition according to the preset rain gauge, obtain the scene state at the precipitation detection point, and perform time domain registration on the precipitation condition and the scene state.

10. The flood warning information collection system based on geographic information according to claim 7, characterized in that: The model application module includes: A forecast information reading unit, used to read the precipitation amount in the future time from the weather forecast information; A condition simulation unit for simulating precipitation conditions based on the current latest three-dimensional scene and precipitation; A state output unit, used to input precipitation conditions into the neural network model and output the scene state; An error rate calculation unit, used for comparing the scene state acquired by the image collector with the scene state output by the neural network model to determine the error rate when the scene state is acquired by the image collector; The updating unit is used to update the sample set and then update the neural network model when the error rate meets the preset updating condition.

Images