A rainfall observation and assessment method for full coverage of high-density cities

Abstract

The present invention relates to the field of rainfall data forecasting technology, and in particular to a rainfall observation and evaluation method for full coverage of high-density cities. The method comprises the following steps: performing monitoring point location analysis on a target city to obtain a camera monitoring point set; selecting monitoring cameras for the target city based on the camera monitoring point set to obtain a raindrop monitoring camera identification set; wherein the raindrop monitoring camera identification set includes a number of raindrop monitoring camera identifications; performing real-time rainfall image stream acquisition on the corresponding camera based on the raindrop monitoring camera identification to obtain a real-time rainfall image sequence; and performing automatic raindrop edge recognition on raindrops based on the real-time rainfall image sequence to obtain a raindrop edge recognition result map. The present invention overcomes the limitations of traditional rainfall station distribution, realizes rainfall monitoring without blind spots in the entire city, and has the advantage of high spatial resolution.

Description

Technical Field

[0001] The present invention relates to the technical field of rainfall data forecasting, and in particular to a rainfall observation and evaluation method for full coverage of high-density cities. Background Art

[0002] Currently, rainfall monitoring in high-density cities mainly relies on fixed rain gauges, rain stations and other equipment. However, this approach has some problems: the monitoring equipment is mostly distributed in a point-like manner, with a limited number, and cannot fully cover the entire urban area. In particular, when rainfall is unevenly distributed, sparse point monitoring makes it difficult to reflect the detailed rainfall changes. Interpolation using sparse rainfall station data often leads to large errors, especially in areas with complex rainfall changes, making it difficult to provide accurate rainfall information. In addition, the installation and maintenance of rain gauges require a large amount of capital investment, especially in high-density cities. In order to improve the coverage of rainfall monitoring, a large number of rain gauges need to be deployed, which increases the cost of equipment installation and maintenance. In addition, existing monitoring methods have limitations in the deployment of rain gauges. Due to the complexity of urban space and the limited number of equipment, many areas are still in monitoring blind spots, and accurate rainfall data cannot be effectively obtained.

[0003] Furthermore, the current large variations in models, focal lengths, and mounting angles of outdoor cameras in cities make it difficult to directly standardize the spatial dimensions of rainfall images captured by different cameras. This results in inconsistent calculations of raindrop size and density when analyzing rainfall characteristics captured by different cameras. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a rainfall observation and evaluation method with full coverage for high-density cities to solve at least one of the above technical problems.

[0005] To achieve the above objectives, a rainfall observation and assessment method for full coverage of high-density cities is proposed, which includes the following steps:

[0006] Step S1: Analyze the locations of monitoring points in the target city to obtain a camera monitoring point set; select monitoring cameras for the target city based on the camera monitoring point set to obtain a raindrop monitoring camera identification set; wherein the raindrop monitoring camera identification set includes a plurality of raindrop monitoring camera identifications;

[0007] Step S2: according to the raindrop monitoring camera identifier, a corresponding camera is used to collect a real-time rainfall image stream to obtain a real-time rainfall image sequence; based on the real-time rainfall image sequence, raindrop edges are automatically recognized to obtain a raindrop edge recognition result image;

[0008] Step S3: Calculating the falling velocity of raindrops based on the raindrop edge recognition result image to obtain raindrop falling velocity data; extracting raindrop particle features based on the raindrop falling velocity data based on the real-time rainfall image sequence to obtain rainfall particle feature data;

[0009] Step S4: constructing a rainfall value prediction model; dynamically adjusting the rainfall value prediction model to obtain an updated rainfall value prediction model;

[0010] Step S5: Based on the raindrop monitoring camera identification, the raindrop edge recognition result map and the rainfall particle feature data, the updated rainfall value prediction model is used to predict the rainfall at the camera location to obtain a real-time rainfall prediction result;

[0011] Step S6: Traverse the raindrop monitoring camera identification set, execute steps S2-S3 and S5 for each camera corresponding to the raindrop monitoring camera identification, and obtain the real-time rainfall prediction result set; visualize the real-time rainfall prediction result set to obtain the city predicted rainfall distribution map to realize rainfall observation evaluation.

[0012] By leveraging the city's existing camera network, this method achieves comprehensive coverage of the entire urban area. This allows for capturing subtle changes in rainfall, particularly in situations of uneven rainfall distribution, thereby improving monitoring coverage and data comprehensiveness. By capturing real-time rainfall image sequences and automatically identifying raindrop edges, errors caused by interpolation analysis can be reduced, providing more accurate rainfall data, particularly in areas with drastic rainfall variations. Compared to increasing the number of rain gauges, this method eliminates the need for deploying a large number of additional monitoring sites, thereby reducing equipment installation and maintenance costs. Unconstrained by the complexity of urban space and the number of devices required, this method can adapt to the monitoring needs of diverse regions, particularly those where traditional rain gauges are difficult to deploy. By analyzing real-time rainfall image sequences, it provides real-time rainfall forecasts and dynamically adjusts the forecast model based on the latest rainfall data, improving the timeliness and accuracy of the forecasts. By extracting raindrop velocity and particle characteristics, it provides rainfall data with high spatial resolution, facilitating a more detailed understanding and prediction of rainfall distribution. By combining historical and real-time rainfall data, the rainfall prediction model can be incrementally trained and adjusted, continuously optimizing its performance. By visualizing the real-time rainfall forecast results, a city-wide predicted rainfall distribution map can be generated. This method is unaffected by the physical damage or technical failures of traditional rain gauges, improving the stability and reliability of the monitoring system. In summary, this method overcomes the limitations of traditional station-based rainfall gauges, enabling city-wide rainfall monitoring without blind spots and offering the advantage of high spatial resolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Other features, objects and advantages of the present invention will become more apparent from reading the detailed description made with reference to the following drawings:

[0014] Figure 1 A schematic flow chart of the steps of a rainfall observation and assessment method for full coverage of a high-density city according to an embodiment is shown.

[0015] Figure 2 A detailed flowchart of step S16 of an embodiment is shown.

[0016] Figure 3 A detailed flowchart of step S17 of an embodiment is shown. DETAILED DESCRIPTION

[0017] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0018] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0019] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0020] To achieve this, please refer to Figures 1 to 3 The present invention provides a rainfall observation and assessment method for full coverage of high-density cities, comprising the following steps:

[0021] Step S1: Analyze the locations of monitoring points in the target city to obtain a camera monitoring point set; select monitoring cameras for the target city based on the camera monitoring point set to obtain a raindrop monitoring camera identification set; wherein the raindrop monitoring camera identification set includes a plurality of raindrop monitoring camera identifications;

[0022] Step S2: according to the raindrop monitoring camera identifier, a corresponding camera is used to collect a real-time rainfall image stream to obtain a real-time rainfall image sequence; based on the real-time rainfall image sequence, raindrop edges are automatically recognized to obtain a raindrop edge recognition result image;

[0023] Step S3: Calculating the falling velocity of raindrops based on the raindrop edge recognition result image to obtain raindrop falling velocity data; extracting raindrop particle features based on the raindrop falling velocity data based on the real-time rainfall image sequence to obtain rainfall particle feature data;

[0024] Step S4: constructing a rainfall value prediction model; dynamically adjusting the rainfall value prediction model to obtain an updated rainfall value prediction model;

[0025] Step S5: Based on the raindrop monitoring camera identification, the raindrop edge recognition result map and the rainfall particle feature data, the updated rainfall value prediction model is used to predict the rainfall at the camera location to obtain a real-time rainfall prediction result;

[0026] Step S6: Traverse the raindrop monitoring camera identification set, execute steps S2-S3 and S5 for each camera corresponding to the raindrop monitoring camera identification, and obtain the real-time rainfall prediction result set; visualize the real-time rainfall prediction result set to obtain the city predicted rainfall distribution map to realize rainfall observation evaluation.

[0027] In this embodiment, GIS software is first used to analyze the locations of monitoring points in a city. Based on a city map and historical rainfall data, areas prone to heavy rain and areas with low rainfall are identified, and monitoring points are deployed in these areas. Using the spatial analysis tools in the GIS software, a camera monitoring point set is obtained. This point set includes all locations in the city where cameras need to be installed. Next, appropriate monitoring cameras are selected based on the camera monitoring point set to form a raindrop monitoring camera identification set. Using the database of the city's public safety system, cameras whose locations, angles, and resolutions meet the rainfall monitoring requirements are screened and their identifiers are recorded. A network video recorder (NVR) or video surveillance management system (VCMS) is used to capture real-time rainfall image streams from the cameras in this identification set to obtain a real-time rainfall image sequence. Image processing software, such as OpenCV, is used to automatically identify raindrop edges on these image sequences. An edge detection algorithm is used to generate a raindrop edge identification result map. Based on the raindrop edge identification result map, the falling velocity of raindrops is calculated to obtain raindrop falling velocity data. The motion estimation tool in the image processing software is used to calculate the velocity by analyzing the position changes of raindrops in consecutive frames. Based on the falling velocity data, raindrop particle features are extracted to obtain rainfall particle feature data. Raindrop characteristics such as size, shape, and density are analyzed to provide input data for the rainfall prediction model. A rainfall value prediction model is constructed, using a machine learning model such as a random forest or neural network. The model is trained and tested using historical rainfall data and rainfall particle feature data. The rainfall value prediction model is dynamically adjusted. The latest rainfall data and particle feature data are collected, and model parameters are continuously updated to adapt to weather changes. Finally, the updated rainfall value prediction model is combined with the raindrop monitoring camera identifiers, the raindrop edge recognition results map, and the rainfall particle feature data to predict rainfall at each camera location, generating real-time rainfall prediction results. The above steps are repeated for each camera in the raindrop monitoring camera identifier set to obtain a real-time rainfall prediction result set. The real-time rainfall prediction result set is visualized using GIS software. A city-wide predicted rainfall distribution map is generated, showing rainfall amounts and intensities in different areas of the city, enabling rainfall observation and evaluation.

[0028] Preferably, step S1 includes the following steps:

[0029] Step S11: Collect rainfall measurement points in the target city to obtain a rainfall measurement point distribution map;

[0030] Specifically, data from rainfall monitoring stations provided by the city's meteorological bureau can be used as a foundation. These stations are located in different parts of the city and provide real-time and historical rainfall data. Using GIS (Geographic Information System) software, the locations of these monitoring stations and rainfall data are imported into the system. GIS software accurately maps each monitoring station using longitude and latitude coordinates. By correlating the rainfall data for each monitoring station with its geographic location, a distribution map of the actual rainfall measurements is generated.

[0031] Step S12: collecting camera geographic coordinates and generating a distribution map of the target city to obtain a camera geographic distribution map;

[0032] Specifically, a database of city surveillance cameras can be obtained from the city's public security department. The database contains information such as the unique identifier of each camera, the latitude and longitude coordinates of the installation location, and the shooting direction. Using GIS software, these data can be imported. GIS software can accurately mark the location of each camera on the map based on the latitude and longitude coordinates. Using the symbolization function of GIS software, icons and labels can be added to each camera. Different layers can be set up, for example, to distinguish high-resolution cameras from ordinary-resolution cameras, or to classify cameras according to their direction and monitoring range. Ultimately, a detailed camera geographic distribution map is generated, which shows the location and related information of all surveillance cameras in the city.

[0033] Step S13: collecting camera monitoring parameters according to the camera geographic distribution map to obtain a camera monitoring parameter table; wherein the camera monitoring parameter table includes a plurality of camera monitoring parameter items, each camera monitoring parameter item including at least a camera identifier, a camera image resolution, a camera installation height, a camera image orientation, and the camera's geographical coordinates;

[0034] Specifically, you can work with the city's public safety department to obtain a detailed list of all surveillance cameras in the city. This list records parameters such as each camera's unique identifier, image resolution, installation height, image orientation, and geographic coordinates. Through on-site inspections and reviewing installation records, verify and record the specific parameters of each camera. For example, a camera installed on a main street has the identifier "CAMERA-123," a 1080p image resolution, an installation height of 8 meters, an image orientation facing south, and geographic coordinates of latitude 34.0522 and longitude -118.2437. Manually enter this information into an Excel spreadsheet to create a table of camera surveillance parameters. This table contains a row for each camera, containing all the above parameters.

[0035] Step S14: historical rainfall records are collected for the target city to obtain a historical rainfall record set, and spatiotemporal feature recognition and rainfall pattern mining are performed on the historical rainfall record set to obtain a rainfall pattern recognition result set;

[0036] Specifically, rainfall data for the target city over the past five years can be obtained from the Meteorological Bureau. These data include daily rainfall, rainfall start and end times, rainfall intensity, etc. By manually reading meteorological reports and rainfall records, these data are organized into a historical rainfall record set. For example, it is recorded that rainfall is generally high every summer, especially in certain areas of the city, such as mountainous areas and coastal areas, where rainfall intensity and frequency are significantly higher than in other areas. Statistical analysis methods are used, such as calculating the average rainfall and number of rainy days in each season, as well as the distribution of rainfall intensity. GIS software is used to identify the spatiotemporal characteristics of rainfall. For example, through layer analysis, it is found that rainfall in the north of the city is generally less than in the south, while the eastern coastal areas are more susceptible to heavy rainfall brought by typhoons. Through these analyses, rainfall patterns, such as seasonal rainfall patterns and regional rainfall patterns, are discovered, and this information is organized into a rainfall pattern recognition result set.

[0037] Step S15: dividing and labeling the rainfall measurement point distribution map according to the rainfall pattern recognition result set to obtain a divided rainfall measurement point distribution map; wherein the rainfall area division includes the division of heavy rain prone areas and the division of low rainfall areas;

[0038] Specifically, you can use GIS software to open a previously generated rainfall measurement point distribution map. This map contains the locations and rainfall data for each monitoring station in the city. Combined with the rainfall pattern recognition result set, this creates a report containing historical data analysis results, which identifies the characteristics of areas prone to heavy rainfall and areas with low rainfall. For example, the report shows that the mountainous area to the north of the city experiences a high frequency of heavy rainfall in summer due to its topography, while the southern part of the city, far from the ocean, experiences relatively little rainfall. Based on this information, the GIS software can use different colors or patterns to mark different areas on the map. In the mountainous area, red markers are used to indicate areas prone to heavy rainfall, and a legend is added to the map to explain the color representation. In the southern part, low rainfall areas are marked with blue markers, which are also indicated in the legend. You can also adjust the transparency or size of the markers to indicate different levels of rainfall intensity. In the GIS software, use the buffer analysis tool to create buffer zones within a certain radius, centered on the monitoring stations in the heavy rainfall-prone areas. This will more clearly identify the extent of high-risk areas. The result is a new rainfall measurement point distribution map, which not only shows the location and rainfall of each monitoring station, but also clearly marks the areas prone to heavy rain and low rainfall areas.

[0039] Step S16: Based on the division of the rainfall measurement point distribution map and the camera geographical distribution map and according to the camera monitoring parameter table, the monitoring point location analysis is performed on the target city to obtain a camera monitoring point set;

[0040] Specifically, please refer to the sub-steps of step S16 for the detailed implementation process of this embodiment.

[0041] Step S17: Select surveillance cameras for the target city according to the camera surveillance point set to obtain a raindrop surveillance camera identification set; wherein the raindrop surveillance camera identification set includes a plurality of raindrop surveillance camera identifications.

[0042] Specifically, please refer to the sub-steps of step S17 for the detailed implementation process of this embodiment.

[0043] The present invention integrates the rainfall measurement points and the camera geographic coordinate data, combined with the camera monitoring parameters, to more accurately identify and select cameras suitable for raindrop monitoring, thereby improving the efficiency of data utilization and the pertinence of monitoring. By identifying the spatiotemporal characteristics of historical rainfall records and mining rainfall patterns, it is possible to identify areas prone to heavy rain and areas with low rainfall, so that monitoring resources can be deployed in these key areas, improving the accuracy and response speed of rainfall forecasts. By dividing and labeling rainfall areas, it helps to optimize the allocation of monitoring resources, ensuring that there are enough monitoring points in key areas, and avoiding over-deployment of resources in low rainfall areas. Through accurate monitoring point location analysis, it can be ensured that the selected cameras can represent the actual conditions of different rainfall areas, enhancing the representativeness and comprehensiveness of the monitoring data. By conducting detailed monitoring point location analysis and camera selection in the target city, it is possible to quickly respond to rainfall conditions in different areas.

[0044] Preferably, step S16 includes the following steps:

[0045] Step S161: assigning rainfall area identifiers to the divided rainfall measurement point distribution map to obtain a rainfall area identifier set;

[0046] Specifically, you can use GIS software to open the rainfall measurement point distribution map, which has been divided into different rainfall areas based on historical rainfall data and rainfall pattern recognition result sets. Define an identification rule, for example, use a combination of letters and numbers to represent different rainfall areas, such as "A1" represents the first rainstorm-prone area, and "B1" represents the first low rainfall area. Use the annotation tool of GIS software to manually or automatically add corresponding identifiers to each divided area. For example, in the mountainous area in the north of the city, it was determined to be a rainstorm-prone area based on previous analysis, and it was marked as "A1". An attribute table was created in GIS to record each identifier and its corresponding regional characteristics and location information. Finally, a rainfall area identification set is obtained, which includes the name, location and characteristic description of each rainfall area.

[0047] Step S162: Calculating the rainfall area of ​​the rainfall area based on the distribution map of rainfall measurement points to obtain a rainfall area set;

[0048] Specifically, GIS software can be used to calculate the area of ​​rainfall areas that have been divided and identified. Using the spatial analysis tools of the GIS software, select the polygonal boundary of each rainfall area and use the "area calculation" function to calculate the area of ​​each area. For example, select the rainstorm-prone area identified as "A1", which is represented as a closed polygon in GIS. By applying the area calculation tool, the GIS software will calculate the area of ​​the area based on the coordinate points of the polygon and display the results in square kilometers or hectares. Record the area of ​​each area and organize this data into a rainfall area set, including the identification, location and calculated area of ​​each area.

[0049] Step S163: determining the monitoring camera requirements for the rainfall area according to the preset rainfall area type rule, the rainfall area identifier set, and the rainfall area area set, to obtain a rainfall area monitoring requirement set;

[0050] Specifically, a set of rainfall area type rules can be developed. This set of rules determines the number and type of surveillance cameras required for each area based on rainfall, area, and the frequency of historical rainfall events. For example, for a rainstorm-prone area identified as "A1," the rules stipulate that at least 2 high-resolution cameras are required for every 5 square kilometers, and 1 camera with night vision capability is required for every 10 square kilometers. According to the rainfall area identification set, the area of ​​the "A1" area is 15 square kilometers. According to the rules, it is calculated that the "A1" area requires at least 3 high-resolution cameras and 1 night vision camera. These requirements are recorded in a new table to form a rainfall area monitoring requirement set. This set details the identification, area, and required camera type and number of each rainfall area.

[0051] Step S164: Screening camera monitoring points in the rainfall area according to the rainfall area monitoring requirement set based on the camera geographic distribution map to obtain a camera monitoring point set, wherein the camera monitoring point set includes a plurality of camera monitoring points.

[0052] Specifically, you can open a city camera geographic distribution map in GIS software. This map details the locations of all available cameras in the city, with each camera location identified by a geographic coordinate. The rainfall area monitoring requirements set includes the number and distribution characteristics of monitoring points required for different rainfall areas. Determine which geographic locations should house cameras to meet rainfall monitoring requirements. For example, according to the rainfall area monitoring requirements set, five monitoring points are required in the mountainous areas to the north of the city and five in the low-lying areas to the south, as these areas are prone to heavy rain. In the GIS software, use the query tool to select geographic points within the target rainfall area. Set query conditions, such as "geographic points located in the mountainous areas to the north of the city" and "geographic points located in the low-lying areas to the south of the city." The GIS software will automatically select geographic points that meet these conditions. For example, ten geographic points meeting these conditions are found in the mountainous areas to the north of the city and eight in the low-lying areas to the south of the city. Based on the area, shape, and rainfall intensity distribution of the rainfall areas, select the optimal five locations in the mountainous areas to the north and five in the low-lying areas to the south. For example, we select the coordinates (34.052235, -118.243684), (34.053345, -118.244789), (34.054456, -118.245890), (34.055567, -118.246901), and (34.056678, -118.247902) in the mountainous area to the north of the city as camera monitoring points. Similarly, we select five geographic coordinates for the low-lying area to the south of the city. Ultimately, we obtain a set of several camera monitoring points.

[0053] The present invention can understand and manage the rainfall distribution in a city in more detail by allocating regional identification and calculating the area of ​​rainfall measurement points. Based on the type rules and area of ​​the rainfall area, the specific requirements of each rainfall area for monitoring cameras can be determined, thereby achieving more reasonable camera deployment and ensuring that key areas are effectively monitored. By screening camera monitoring points that match the monitoring requirements of the rainfall area, the efficiency of monitoring resource utilization can be improved, ensuring that monitoring cameras can cover key rainfall areas and improving monitoring effects. By combining the geographical distribution of cameras with the characteristics of rainfall areas, the correlation between monitoring data and actual rainfall conditions is enhanced. Through precise demand analysis and camera screening, the deployment of cameras in unnecessary areas is avoided, thereby reducing costs.

[0054] Preferably, step S17 includes the following steps:

[0055] Step S171: extracting corresponding camera parameter items from the camera monitoring parameter table according to the camera monitoring point to obtain a candidate camera monitoring parameter item set;

[0056] Specifically, the camera monitoring parameter table contains parameters such as identifiers, resolutions, mounting heights, camera orientations, and geographic coordinates for all cameras in the city. These parameters are extracted for each camera monitoring point within the "A1" rainfall area. Using an Excel spreadsheet, the camera monitoring parameter table is opened and, using the filtering function, the geographic coordinates are used to select only cameras within the "A1" area. For example, all cameras with coordinates within a specific latitude and longitude range are selected. The parameters of these cameras are copied into a new worksheet to form a set of candidate camera monitoring parameter items. Each camera parameter in this set is represented by a row, including the identifier "CAMERA-101," the resolution "1080p," the mounting height "10 meters," the orientation "North," and the geographic coordinates "34.0522, -118.2437." This results in a set containing the parameters of all candidate cameras within the "A1" area.

[0057] Step S172: If there is only one candidate camera monitoring parameter item in the candidate camera monitoring parameter item set, the corresponding camera is directly used as the raindrop monitoring camera to obtain the raindrop monitoring camera identifier;

[0058] Specifically, suppose that at a specific monitoring point in rainfall area "A1," the candidate camera monitoring parameter item set contains only one camera, "CAMERA-101." Since this monitoring point has only this one candidate camera, it is directly identified as the raindrop monitoring camera. The camera's identifier, "CAMERA-101," is recorded as the raindrop monitoring camera identifier and marked as an active monitoring point in the monitoring system.

[0059] Step S173: if there are two or more candidate camera monitoring parameter items in the candidate camera monitoring parameter item set, deduplication is performed on the candidate camera monitoring parameter item set according to the camera shooting orientation to obtain a raindrop monitoring camera identifier;

[0060] Specifically, we can assume that at another monitoring point in the "A1" rainfall area, the candidate camera monitoring parameter item set includes three cameras: "CAMERA-102," "CAMERA-103," and "CAMERA-104," all facing north. Since these cameras have the same shooting direction, only one of them needs to be selected as the raindrop monitoring camera. Based on other camera parameters, such as resolution and installation height, the most suitable camera is selected. For example, "CAMERA-103," which has the highest resolution and a moderate installation height, is selected as the raindrop monitoring camera. The identifier of "CAMERA-103" is recorded as the raindrop monitoring camera identifier and marked as the active monitoring point in the system, while the other two cameras are temporarily used as backup monitoring points.

[0061] Step S174: traverse the camera monitoring point set, and for each camera monitoring point, execute steps S171 to S173 to obtain a raindrop monitoring camera identification set; wherein the raindrop monitoring camera identification set includes several raindrop monitoring camera identifications.

[0062] Specifically, traverse the camera monitoring point set and process these monitoring points one by one. For each monitoring point, first execute step S171 to extract the parameter item set of the candidate camera from the camera monitoring parameter table. Check the number of cameras in this set. If there is only one camera, directly identify it as the raindrop monitoring camera identifier (step S172). If there are two or more cameras, deduplicate according to the camera shooting direction and select the most suitable camera as the raindrop monitoring camera identifier (step S173). Repeat this process until all monitoring points in the "A1" area are processed. Finally, a raindrop monitoring camera identifier set is obtained, which contains the identifiers of all active monitoring points in the "A1" area, such as "CAMERA-101" and "CAMERA-103".

[0063] The present invention can accurately screen out cameras suitable for raindrop monitoring by extracting parameter items that match the camera monitoring points, ensuring that the selected cameras can provide high-quality rainfall data. When there are multiple candidate camera monitoring parameter items, the deduplication step avoids repeated monitoring of the same direction in the same area, thereby reducing resource waste and improving monitoring efficiency. By considering the shooting direction of the camera for deduplication, it is ensured that the monitoring points can cover rainfall conditions in different directions, enhancing the comprehensiveness and three-dimensional perception of monitoring. Selecting the correct camera for raindrop monitoring can reduce monitoring data quality issues caused by improper camera position or poor orientation, and improve the accuracy and reliability of rainfall data. By traversing the camera monitoring point set and performing parameter extraction and deduplication steps, an efficient raindrop monitoring camera network can be quickly constructed, improving the response speed and operational efficiency of the monitoring system.

[0064] Preferably, step S2 includes the following steps:

[0065] Step S21: collecting a real-time rainfall image stream from a corresponding camera according to the raindrop monitoring camera identifier to obtain an original real-time rainfall image stream;

[0066] Specifically, all cameras in the "A1" area that are identified as raindrop monitoring, such as "CAMERA-101", can be determined. These cameras have been selected and configured to monitor rainfall through previous steps. Using a network video recorder (NVR) or a video surveillance management system (VCMS), these systems can remotely access and control these cameras. Through these systems, cameras such as "CAMERA-101" are configured to transmit video streams in real time at a specific resolution and frame rate. The video stream is transmitted to the data center via a wired or wireless network. In the data center, video stream acquisition software, such as FFmpeg, is used to receive and store these video streams. FFmpeg can process continuous video signals from the camera and save them as video files, such as MP4 or AVI formats. Ultimately, the original real-time rainfall image stream is obtained.

[0067] Step S22: reconstructing the video frame time sequence of the original real-time rainfall image stream to obtain an initial real-time rainfall image sequence, and removing unqualified images from the initial real-time rainfall image sequence to obtain a filtered real-time rainfall image sequence;

[0068] Specifically, you can use video processing software such as Adobe Premiere Pro or DaVinci Resolve to open the original real-time rainfall image stream file. First, reconstruct the video frame timing to ensure that each frame is arranged in the correct time sequence. Check the video stream for lost or duplicate frames and correct them. For example, if the timestamp of a frame is found to be significantly different from the timestamp of the previous frame, identify it as a lost frame and attempt to recover or fill it from the previous and next frames. If duplicate frames are found, delete them. Perform a quality assessment on the images in the initial real-time rainfall image sequence. Use image quality assessment tools such as ImageJ or custom Python scripts to automatically detect image clarity, brightness, and contrast. For example, set a threshold to automatically remove blurry images caused by camera shake, sudden changes in lighting, or obstructions. You can also manually inspect the image sequence to remove unqualified images caused by camera stains or damage. After this series of processing, a filtered real-time rainfall image sequence can be obtained.

[0069] Step S23: performing image format conversion and standardization on each image in the screened real-time rainfall image sequence to obtain a standard real-time rainfall image sequence, and gray-scaling each image in the standard real-time rainfall image sequence to obtain a gray-scale real-time rainfall image sequence;

[0070] Specifically, you can use image processing software, such as Photoshop or ImageMagick, to open and filter the real-time rainfall image sequence. Select the entire image sequence and set the conversion parameters to convert each image to a uniform format and size. For example, convert all images to 800x600 pixels and unify the file format to PNG. This format can reduce the file size while maintaining image quality. Grayscale each image. In Photoshop, this can be achieved by selecting "Grayscale" from the "Mode" option under the "Image" menu. After these steps, you will finally get a grayscale real-time rainfall image sequence.

[0071] Step S24: performing image noise filtering and smoothing processing on each image in the grayscale real-time rainfall image sequence to obtain a real-time rainfall image sequence;

[0072] Specifically, image processing software can be used to further process each image in the grayscale real-time rainfall image sequence. First, use the Gaussian blur technique to smooth the image. In Photoshop, select "Gaussian Blur" from the "Blur" option under the "Filter" menu and set an appropriate radius. In ImageMagick, the blur command can be used to achieve a similar effect. For example, a Gaussian blur with a radius of 1 can effectively remove fine noise while preserving image edge information. A median filter is used to further remove salt and pepper noise from the image. In Photoshop, this can be achieved by selecting "Median" from the "Noise" option under the "Filter" menu. In ImageMagick, the morphology command can be used to apply a median filter. Selecting a 3x3 or 5x5 window size helps eliminate isolated white or black spots without blurring important image features. After these processing steps, the real-time rainfall image sequence is finally obtained.

[0073] Step S25: Perform clarity evaluation and optimal image selection on the real-time rainfall image sequence to obtain the optimal real-time rainfall image, and perform image gradient calculation on the optimal real-time rainfall image to obtain rainfall image gradient data; perform edge detection on the optimal real-time rainfall image based on the rainfall image gradient data to obtain a raindrop edge recognition result image.

[0074] Specifically, image processing software can be used to assess the clarity of each image in a real-time rainfall image sequence. A common method for assessing clarity is to calculate the image's sharpness. This sharpness can be calculated using the image's Laplacian operator, which highlights edges and details in an image, thereby reflecting image clarity. The Laplacian operator value is calculated for each image and compared to determine which image is the sharpest. For example, image numbered "image_07" has the highest Laplacian operator value in the sequence, so it is selected as the optimal real-time rainfall image. Next, the image gradient is calculated for this optimal real-time rainfall image. The Sobel operator in the image processing software is used to calculate the image's horizontal and vertical gradients. The Sobel operator determines the magnitude and direction of the gradient by calculating the difference in grayscale value between each pixel in the image and its neighbors. Applying the Sobel operator yields a gradient image in which each pixel value represents the gradient magnitude at the corresponding location. For example, applying the Sobel operator to "image_07" yields a grayscale image showing the gradient magnitude, highlighting edges and details within the image. Finally, edge detection is performed based on the gradient data to identify raindrop edges. The Canny edge detection algorithm is used, which identifies edges within an image based on high gradient values. The Canny algorithm's threshold is set to determine which gradient values ​​are high enough to be considered an edge. For example, the low threshold is set to 50 and the high threshold is set to 150. After applying the Canny algorithm, the resulting image shows raindrop edge detection, which clearly illustrates the edges of the raindrops.

[0075] The present invention removes low-quality images that affect the analysis results by reconstructing the video frame time sequence and eliminating unqualified images. By converting and standardizing the image format, the consistency of the image sequence is ensured. The image information is simplified through grayscale processing, making the rainfall characteristics in the image more prominent. The image noise filtering and smoothing steps reduce the noise and anomalies in the image and improve the image clarity. By performing image gradient calculation and edge detection on the optimal real-time rainfall image, the raindrop edge can be identified more accurately. Through clarity evaluation and optimal image selection, it is possible to quickly focus on the most representative and informative image, thereby improving the efficiency and accuracy of the analysis. The final raindrop edge recognition result image provides a basis for the visualization of rainfall data, making the changes in rainfall distribution and intensity more intuitive and easy to understand.

[0076] Preferably, step S3 includes the following steps:

[0077] Step S31: removing incomplete raindrops from the raindrop edge recognition result map to obtain a regular raindrop edge recognition map, and assigning raindrop numbers to the regular raindrop edge recognition map to obtain a raindrop number set;

[0078] Specifically, image processing software, such as OpenCV, can be used to process the edge detection result map. Binarize the image, converting the edge-detected image into a black-and-white image containing only the raindrop edges, with the raindrop edges in white and the background in black. Use morphological operations within the image processing software, such as opening (erosion followed by dilation), to remove noise and fine broken edges, thereby obtaining more complete raindrop edges. Use the connected component labeling function within the image processing software to identify and label each individual raindrop in the image. This function scans the image, labels groups of touching pixels as a single connected component, and assigns each connected component a unique number. For example, if there are five complete raindrops in the image, the software will label them 1, 2, 3, 4, and 5. Incomplete raindrops are identified based on their morphological characteristics, such as those that are too small or have an irregular shape, and are removed from the image. This results in a regular raindrop edge recognition map and a corresponding set of raindrop numbers.

[0079] Step S32: Calculating the area of ​​raindrops based on the regular raindrop edge recognition map and the raindrop number set to obtain a raindrop area set;

[0080] Specifically, image processing software, such as OpenCV or MATLAB, can be used to process regular raindrop edge recognition images. Use the software's regional attribute calculation function to calculate the area of ​​each marked connected component (that is, each raindrop). This function calculates the number of pixels inside each connected component to obtain the area of ​​the raindrop. For example, if the raindrop numbered 1 contains 12 white pixels, then the area of ​​this raindrop is 12 pixel units. Record the number and corresponding area of ​​each raindrop to form a raindrop area set. For example, the following data is obtained: the area of ​​raindrop 1 is 12 pixel units, the area of ​​raindrop 2 is 10 pixel units, the area of ​​raindrop 3 is 8 pixel units, and so on. This set provides information about the size of each raindrop.

[0081] Step S33: Calculate the pixel size of the raindrops according to the raindrop area set to obtain a raindrop pixel size data set; wherein the specific calculation formula for the pixel size calculation is as follows:

[0082]

[0083] Among them, π is the ratio of circumference to diameter;

[0084] Specifically, you can use image processing software, such as MATLAB or Python's OpenCV library, to process the raindrop area set. First, extract the area of ​​each raindrop from the area set. For example, suppose a raindrop has an area of ​​12 pixels. Substitute the area of ​​the raindrop into the formula: get: After calculation, the radius of the raindrop is approximately 1.24 pixels. This process is repeated for all raindrops to obtain the pixel size of each raindrop and record it in the raindrop pixel size dataset. This dataset contains the number of each raindrop and its corresponding radius.

[0085] Step S34: randomly selecting adjacent raindrops according to the raindrop number set and the regular raindrop edge recognition map to obtain a first raindrop number and a second raindrop number; wherein the raindrops corresponding to the first raindrop number and the second raindrop number are adjacent raindrops;

[0086] Specifically, image processing software, such as MATLAB or Python's OpenCV library, can be used to process the regular raindrop edge identification map and the raindrop number set. First, extract the number and position information of all raindrops from the number set. For example, suppose there are raindrops numbered 1, 2, 3, and 4, and their positions are (x1, y1), (x2, y2), (x3, y3), and (x4, y4), respectively. Use the distance formula to calculate the distance between each pair of raindrops; calculate the distance between all pairs of raindrops, and select the pair with the closest distance as adjacent raindrops. For example, the distance between raindrops numbered 1 and 2 is the shortest, and these two numbers are selected as the first raindrop number and the second raindrop number. Record these two numbers and mark the corresponding raindrops on the regular raindrop edge identification map.

[0087] Step S35: determining the center points of the raindrop corresponding to the first raindrop number and the raindrop corresponding to the second raindrop number based on the regular raindrop edge recognition map to obtain the first raindrop center point and the second raindrop center point, and calculating the pixel distance between adjacent raindrops based on the first raindrop center point and the second raindrop center point to obtain adjacent raindrop pixel interval data;

[0088] Specifically, image processing software, such as MATLAB or Python's OpenCV library, can be used to process the regular raindrop edge recognition map. From step S34, the numbers of two adjacent raindrops are obtained, assuming they are number 1 and number 2. Use the connected component analysis function of the image processing software to determine the center point of each raindrop. This function calculates the center of mass of each connected component (raindrop), that is, the center point of the raindrop. For example, the center point of the raindrop numbered 1 is located at the coordinates (50, 60), and the center point of the raindrop numbered 2 is located at the coordinates (58, 65). Next, use the distance formula to calculate the distance between the two center points; record this distance to obtain the pixel spacing data of adjacent raindrops.

[0089] Step S36: extracting corresponding sizes from the raindrop pixel size dataset according to the first raindrop number and the second raindrop number to obtain first raindrop size data and second raindrop size data, and performing mean calculation on the first raindrop size data and the second raindrop size data to obtain average size data of adjacent raindrops;

[0090] Specifically, spreadsheet software, such as Microsoft Excel, can be used to process the raindrop pixel size dataset. The radius of each raindrop is obtained from step S33. Assume that the radius of raindrop number 1 is 7 pixels, and the radius of raindrop number 2 is 8 pixels. Extract the radius of these two raindrops from the raindrop pixel size dataset, calculate their average, and record this average radius to obtain the average size data of adjacent raindrops.

[0091] Step S37: Calculating the ratio of the average size of adjacent raindrops to the pixel spacing of adjacent raindrops to obtain a raindrop size-raindrop spacing ratio. Inferring the rainfall intensity of the target city based on the raindrop size-raindrop spacing ratio and a preset rainfall intensity table to obtain urban rainfall intensity data. Matching the urban rainfall intensity data with a preset rainfall intensity-velocity table to obtain raindrop falling velocity data.

[0092] Specifically, assuming that the average size of adjacent raindrops obtained from step S36 is 7.5 pixels, and the pixel spacing between adjacent raindrops obtained from step S35 is 9.43 pixels, calculate the ratio of these two values: The ratio is ≈0.795, and the rainfall intensity corresponding to this ratio is determined according to the preset rainfall intensity table. According to the information in the table, a ratio between 0.06 and 0.1 corresponds to heavy rain. Therefore, it is inferred that the current rainfall intensity is heavy rain. This rainfall intensity is matched with the preset rainfall intensity-speed table to determine the falling speed of the raindrops. According to the information in the table, the falling speed of raindrops corresponding to heavy rain is 5 to 7 meters per second. The median value of this range, i.e. 6 meters per second, can be selected as the falling speed of the current raindrops. Record this falling speed to obtain the raindrop falling speed data. Among them, the preset rainfall intensity table contains the following information:

[0093] Light rain: ratio (less than 0.02) raindrop falling speed is: 2-3m / s,

[0094] Moderate rain: ratio (0.02-0.6) raindrop falling speed is: 3-5m / s,

[0095] Heavy rain: ratio (0.06-0.1) raindrop falling speed is: 5-7m / s,

[0096] Heavy rain: The ratio (greater than 0.1) raindrop falling speed is: 7-8m / s.

[0097] Step S38: extracting raindrop particle features based on the real-time rainfall image sequence and the raindrop falling velocity data to obtain rainfall particle feature data.

[0098] Specifically, please refer to the sub-steps of step S38 for the detailed implementation process of this embodiment.

[0099] The present invention ensures the integrity and accuracy of the analyzed data by removing incomplete raindrops and assigning them numbers. By systematically calculating and analyzing raindrop area, pixel size, and the spacing between adjacent raindrops, it provides detailed data support for inferring rainfall intensity and falling velocity. By matching the ratio of raindrop size to spacing with a preset rainfall intensity table, it is possible to refine the inference of urban rainfall intensity and improve the accuracy of rainfall forecasts. At the same time, by matching rainfall intensity data with the rainfall intensity-velocity table, the falling velocity of raindrops can be accurately calculated, providing key parameters for extracting rainfall particle characteristics. By providing accurate raindrop size, spacing, and falling velocity data, the performance of the rainfall prediction model is enhanced, improving the accuracy and reliability of the forecast.

[0100] Preferably, step S38 includes the following steps:

[0101] Step S381: detecting the motion trajectory of raindrops based on the real-time rainfall image sequence to obtain raindrop motion trajectory data;

[0102] Specifically, you can use video analysis software, such as Adobe Premiere Pro or professional motion tracking tools, such as Mocha Pro, to process real-time rainfall image sequences. Open the video file and use the software's motion tracking function to identify and track the movement of raindrops between consecutive frames. For example, select a raindrop that is clearly visible in the first frame and let the software automatically track its position in subsequent frames. The software records the position of the raindrop in each frame and generates a motion trajectory dataset containing the coordinates of the raindrop in each frame. Check these trajectories to make sure they are accurate and manually adjust any misidentified parts. For example, if a raindrop jumps too much between two frames because it merges with another raindrop, mark it as invalid data. Finally, the raindrop motion trajectory data is obtained.

[0103] Step S382: Based on the regular raindrop edge recognition map, actual distance calculation is performed on a single pixel in the regular raindrop edge recognition map according to the raindrop falling velocity data and the raindrop motion trajectory data to obtain the actual distance data of the single pixel. The specific calculation formula for the actual distance calculation is as follows:

[0104]

[0105] Among them, v rain is the raindrop velocity, Δt is the time interval, Δd is the number of pixel displacements of raindrops between video frames, and S is the actual distance corresponding to a single pixel;

[0106] Specifically, image processing and data analysis software, such as MATLAB or Python's OpenCV library, can be used to process the raindrop motion trajectory data and raindrop falling speed data. Assume that the raindrop motion trajectory obtained from step S381 shows that a raindrop moves from coordinates (50, 60) to (55, 65) between two consecutive frames. The frame rate is 100 frames per second, so the time interval is 0.01 seconds. Assume that the raindrop falling speed obtained from step S37 is 6 meters per second. Substitute the above data into the formula: To calculate the actual distance of a single pixel, Record this value to get the actual distance data of a single pixel.

[0107] Step S383: Calculating the actual size of raindrops based on the regular raindrop edge recognition map and the actual distance data of individual pixels to obtain actual size data of raindrops. The specific calculation formula for calculating the actual size is as follows:

[0108] L = d rain ×S;

[0109] Among them, L is the actual size of the raindrop, d rain is the number of pixels occupied by a raindrop in the regular raindrop edge recognition image, and S is the actual distance represented by a single pixel;

[0110] Specifically, image processing software, such as MATLAB or Python's OpenCV library, can be used to process the regular raindrop edge recognition image and the actual distance data of a single pixel. Assume that the actual distance of a single pixel obtained from step S382 is 0.0085 meters / pixel. Knowing that from the regular raindrop edge recognition image, the number of pixels occupied by a raindrop is 5 pixels (this is obtained by calculating the number of pixels within the raindrop edge). Substitute the above data into the formula: L = d rain ×S, we get L = 5 × 0.0085 = 0.0425m; record this value to get the actual size data of the raindrops.

[0111] Step S384: Calculate the raindrop spacing for the target city based on the regular raindrop edge recognition map and the actual distance data of each pixel to obtain average raindrop spacing data. The specific calculation formula for calculating the raindrop spacing is as follows:

[0112]

[0113] Among them, P is the average raindrop spacing, i is the index, N is the total number of raindrops, x i is the horizontal coordinate of the center point of the i-th raindrop, y i is the vertical coordinate of the center point of the i-th raindrop, x i+1 is the horizontal coordinate of the center point of the i+1th raindrop, yi+1 is the vertical coordinate of the center point of the i+1th raindrop, and S is the actual distance data of a single pixel;

[0114] Specifically, image processing software, such as MATLAB or Python's OpenCV library, can be used to process the regular raindrop edge recognition image and the actual distance data of a single pixel. Assume that the actual distance of a single pixel obtained from step S382 is 0.0085 meters / pixel. From the regular raindrop edge recognition image, there are 5 raindrops, and their center point coordinates are (50, 60), (58, 65), (65, 70), (72, 75) and (80, 80). Use the following formula to calculate the average spacing of raindrops: By substituting the above data into the formula, we can calculate P≈0.0766m. Record this value to obtain the average raindrop spacing data.

[0115] Step S385: Based on the regular raindrop edge recognition map, the rainfall density of the target city is calculated according to the actual distance data of a single pixel to obtain the raindrop spacing density data, and the actual raindrop size data, the average raindrop spacing data and the raindrop spacing density data are recorded as rainfall particle feature data.

[0116] Specifically, image processing and data analysis software, such as MATLAB or Python's OpenCV library, can be used to process regular raindrop edge recognition images and single pixel actual distance data. Assume that the actual size of the raindrops obtained from step S383 is 0.0425 meters, and the average spacing of the raindrops obtained from step S384 is 0.0766 meters. The average spacing of the raindrops is directly used to calculate the density of raindrops per square meter. The rainfall density can be estimated by dividing the total number of raindrops by the total area they occupy. Assuming that 100 raindrops are identified in the image and the area of ​​the image is 1 square meter, the rainfall density can be estimated as: Record this value to obtain the raindrop spacing density data. This data provides the number of raindrops per unit area. Finally, the actual raindrop size data, the average raindrop spacing data, and the raindrop spacing density data are combined to form a rainfall particle feature dataset.

[0117] The present invention significantly enhances the ability to monitor rainfall dynamics by detecting the trajectory of raindrop motion. By calculating the actual distance of a single pixel, the pixel data is converted into physical distance, thereby greatly improving the spatial resolution of rainfall data. By using actual distance data to accurately measure raindrop size, this is crucial for analyzing rainfall intensity and type. By calculating the distance between raindrops and rainfall density, the rainfall density analysis is optimized and the spatial distribution characteristics of rainfall are revealed. By converting pixel data into actual physical dimensions, the practicality and compatibility of the data are enhanced, making the data easier to compare and integrate with data from ground monitoring stations. The present invention improves the system's adaptability to different rainfall conditions, and can provide accurate raindrop characteristic data regardless of whether it is light rain or heavy rain, thereby enhancing the stability and reliability of the system in diverse environments.

[0118] Preferably, step S4 includes the following steps:

[0119] Step S41: Obtain camera monitoring parameter table;

[0120] Specifically, the camera monitoring parameter table obtained in step S13 may be obtained by consulting a database.

[0121] Step S42: collecting historical rainfall videos and historical actual rainfall data for the target city according to the camera monitoring parameter table to obtain a historical rainfall video stream set and a historical actual rainfall data set; wherein the historical rainfall video stream set includes a plurality of historical rainfall video streams;

[0122] Specifically, the location information in the camera monitoring parameter table can be used to retrieve historical rainfall video streams from the city's public security department's monitoring system. For example, if a rainfall event occurred during a specific date and time period, the video streams that captured the rainfall can be retrieved based on this time window and the camera's installation location. Five video streams from cameras are found, each located in different parts of the city and pointing to the area affected by the rainfall. These video streams are exported from the monitoring system. Each video stream contains continuous frames that record the details of the rainfall event. At the same time, historical actual rainfall data is obtained from the meteorological department. This data includes rainfall amount, rainfall start and end times, rainfall intensity, and so on. This historical rainfall data is associated with the corresponding video streams to form a set of historical rainfall video streams and a set of historical actual rainfall data.

[0123] Step S43: traverse the historical rainfall video stream set, and for each historical rainfall video stream, execute steps S22 to S25 and steps S31 to S38 to obtain a historical raindrop edge recognition result atlas and a historical rainfall particle feature dataset;

[0124] Specifically, you can use video processing software, such as Adobe Premiere Pro, to open a video stream from the historical rainfall video stream set. Each frame in the video stream is processed according to the methods described in steps S22 to S25 and steps S31 to S38 to obtain a historical raindrop edge recognition result atlas and a historical rainfall particle feature dataset.

[0125] Step S44: using the historical actual rainfall dataset, the historical raindrop edge recognition result atlas, and the historical rainfall particle feature dataset to perform model training and error adjustment on the preset machine learning model to obtain a rainfall value prediction model;

[0126] Specifically, a deep learning model, specifically a convolutional neural network (CNN), is constructed and trained using a historical actual rainfall dataset, a historical raindrop edge recognition result atlas, and a historical rainfall particle feature dataset. The model architecture is first defined, consisting of multiple convolutional layers, pooling layers, depthwise convolutional layers, fully connected layers, and an output layer. In the convolutional layers, 3x3 convolution kernels are used to extract low-level features from the image, such as raindrop edges and outlines, and noise is removed using the ReLU activation function. A 2x2 max pooling layer is used to reduce feature dimensionality while retaining key information. In the depthwise convolutional layers, larger convolution kernels (e.g., 5x5) are used to extract more complex rainfall features, such as the grouping and density of raindrops. In the fully connected layers, the feature maps output by the convolutional layers are flattened into one-dimensional vectors and fused with device parameters (such as camera focal length, tilt angle, orientation, image resolution, and frame rate). Finally, in the output layer, a linear activation function is used to map features to specific rainfall intensity values. The number of neurons in the output layer is 1, and the actual rainfall (mm / hour) in the image is predicted. During the model training phase, the image dataset is divided into a 70% training set and a 30% validation set. Mean squared error (MSE) is used as the loss function, and Adam is used as the optimizer for parameter updates. The model calculates the loss value through forward propagation and then adjusts the weights through backpropagation. This process is iterated until the loss function converges. During the validation and adjustment phase, the rainfall results predicted by the model based on the image data are compared with the measured data from the rain gauge, and the error index (mean squared error (MSE)) is calculated. Based on the error, the model parameters are adjusted, the model output is calibrated, and the prediction error is reduced. This process is iterated multiple times until the error between the model and the measured data is reduced to an acceptable range. The result is a trained and error-adjusted rainfall prediction model.

[0127] Step S45: Dynamically adjust the rainfall value prediction model to obtain an updated rainfall value prediction model.

[0128] Specifically, please refer to the sub-steps of step S45 for the detailed implementation process of this embodiment.

[0129] The present invention improves data integration capabilities by integrating camera monitoring parameter tables with collected historical rainfall videos and actual rainfall data, providing training material for rainfall prediction models. Through in-depth analysis of historical rainfall video streams, key raindrop edge recognition results and rainfall particle characteristics are extracted, thereby enhancing the analysis capabilities of past rainfall events. Utilizing these historical data sets, it is possible to train and optimize preset machine learning models, construct a highly accurate rainfall value prediction model, and improve the initial accuracy of the prediction model. By dynamically adjusting the rainfall value prediction model, it is possible to adapt to new rainfall conditions and trends, maintain the timeliness and accuracy of the model, and improve the reliability of the prediction. At the same time, training with historical data enhances the model's ability to recognize different rainfall patterns and improves the model's generalization capabilities. Continuous error adjustment helps reduce prediction errors and improve prediction accuracy.

[0130] Preferably, step S45 includes the following steps:

[0131] Step S451: Obtaining a raindrop monitoring camera identification set;

[0132] Specifically, the raindrop monitoring camera identification set obtained in step S17 may be obtained by consulting a database.

[0133] Step S452: Rainfall event data is collected for the target city based on the raindrop monitoring camera identification set to obtain a new rainfall video stream set and a new actual rainfall dataset; step S43 is then executed on the new rainfall video stream set to obtain a new raindrop edge recognition result atlas and a new rainfall particle feature dataset;

[0134] Specifically, raindrop monitoring camera identification sets can be used to remotely access these cameras through the data interface of the urban public security monitoring system. The system is configured to automatically record video streams during rainfall events. For example, CAMERA-101, CAMERA-202, and CAMERA-303 are set to start recording when rainfall is detected, and the video streams are saved to a central server. Corresponding actual rainfall data, including rainfall amount, rainfall intensity, rainfall duration, etc., are collected from the weather station. These data can be automatically obtained using the weather station's API interface, or obtained from the weather station by manual recording. After collecting new rainfall video streams and actual rainfall data, the processing flow described in step S43 is executed on these video streams. Through these processes, a new raindrop edge recognition result atlas and a new rainfall particle feature dataset are obtained. For example, in the video recorded by CAMERA-101, the average diameter of the raindrops is 0.5 mm, the average spacing is 80 mm, and the rainfall density is 120 raindrops / square meter.

[0135] Step S453: Recording the historical actual rainfall dataset, the historical raindrop edge recognition result atlas, and the historical rainfall particle feature dataset as a historical training dataset, and recording the new actual rainfall dataset, the new raindrop edge recognition result atlas, and the new rainfall particle feature dataset as a new training dataset;

[0136] Specifically, the historical actual rainfall dataset, the historical raindrop edge recognition result atlas, and the historical rainfall particle feature dataset can be marked as the historical training dataset, and the new actual rainfall dataset, the new raindrop edge recognition result atlas, and the new rainfall particle feature dataset can be marked as the new training dataset;

[0137] Step S454: Perform weighted data fusion on the historical training data set and the new training data set to obtain a fused training data set. The specific calculation formula for weighted data fusion is as follows:

[0138] W r =α×W l +(1―α)×W h ;

[0139] Among them, W r is the fusion training data set, W l is a new training dataset, W h is the historical training data set, α is the weight coefficient;

[0140] Specifically, data analysis software, such as Python's Pandas library, can be used to process the historical training dataset and the new training dataset. First, weights are assigned to the two parts of data. These weights are based on the reliability, timeliness, or other relevant factors of the data. For example, the new training dataset is considered to better reflect the current climate conditions, so it is given a higher weight. The weight coefficient α is set to 0.6, which means that the new training dataset will account for 60% of the fused training dataset, while the historical training dataset will account for 40%. The following weighted average formula is used to fuse the dataset: W r =α×W l +(1―α)×W h ; Substitute the specific weight value into the formula to obtain: Fusion training data set = 0.6× l New training dataset + 0.4 × historical training dataset; in this way, the historical and new training datasets are merged into a fused training dataset, which combines the stability of historical data and the timeliness of new data.

[0141] Step S455: perform incremental training and adjustment on the rainfall value prediction model using the fused training data set to obtain an updated rainfall value prediction model.

[0142] Specifically, a machine learning platform, such as Python's scikit-learn library or a deep learning framework like TensorFlow or PyTorch, can be used to load a previously trained rainfall prediction model. This model is a pre-trained convolutional neural network (CNN) used to predict rainfall based on image features. The fused training dataset contains historical and new rainfall feature data, along with corresponding actual rainfall labels. This dataset is divided into a feature set and a label set. The feature set includes features such as raindrop size, spacing, and density, while the label set contains actual rainfall. The model is incrementally trained using the feature set. Instead of training from scratch, the fused training dataset is added to the model training process. Incremental training allows the model to learn from new data while retaining the knowledge learned from historical data. During training, the model's performance is monitored using the mean squared error (MSE) as the loss function and the Adam optimizer for parameter updates. For example, if the model performs worse on new rainfall feature data than on historical data, fine-tuning the model's depthwise convolutional layers can be performed by increasing or decreasing the number of convolution kernels or adjusting the number of neurons in the fully connected layers. Through repeated iterative training and validation, the model parameters are continuously adjusted until the model achieves satisfactory performance on the fused training dataset. Ultimately, an updated rainfall prediction model is obtained, which combines historical and new rainfall data to more accurately predict rainfall under different conditions.

[0143] The present invention significantly improves the timeliness of the prediction model by updating the raindrop monitoring camera identification set and collecting new rainfall event data in real time, ensuring that the model can reflect the latest rainfall conditions. By combining historical and new rainfall data, the comprehensiveness of the data is enhanced, and the model's understanding of rainfall patterns under different times and conditions is improved. Through weighted data fusion, the scheme optimizes model performance, balances the importance of historical and new data, improves prediction accuracy, and enhances the generalization ability of the model. By using the fused training data set for incremental training, the adaptability of the model is improved, enabling the model to adapt to new rainfall trends and patterns and maintain continuous accuracy and relevance. By continuously updating the model to include the latest rainfall data, the risk of prediction accuracy being affected by outdated data is reduced, and the flexibility of data processing is improved, allowing the model to adjust its learning focus based on the latest rainfall conditions.

[0144] Preferably, step S6 includes the following steps:

[0145] Step S61: traverse the raindrop monitoring camera identification set, and execute steps S2 to S3 and S5 for each camera corresponding to the raindrop monitoring camera identification, thereby obtaining a real-time rainfall prediction result set;

[0146] Specifically, a raindrop monitoring camera identifier set can be obtained from a database. This set contains the identifiers of all cameras used for rainfall monitoring. For example, this set includes identifiers such as "CAMERA-101," "CAMERA-102," and "CAMERA-103." Use video processing software such as Adobe Premiere Pro or professional video processing tools like FFmpeg to access the real-time video streams from these cameras. For each camera corresponding to a raindrop monitoring camera identifier, perform steps S2, S3, and S5 to obtain a set of real-time rainfall prediction results.

[0147] Step S62: performing spatiotemporal data alignment and integration on the real-time rainfall prediction result set to obtain a spatiotemporal consistent rainfall dataset;

[0148] Specifically, geographic information system (GIS) software, such as ArcGIS or QGIS, can be used to process the real-time rainfall forecast result set. This software is capable of processing spatial data and allows rainfall forecast results to be combined with geographic location information. The data in the real-time rainfall forecast result set is imported into the GIS software and matched with the geographic coordinates of the camera. In this way, each rainfall forecast result is associated with a specific geographic location. These data are time-aligned to ensure that the rainfall forecast results of different cameras are compared and integrated at the same time point. For example, the rainfall forecast results every 5 minutes are integrated together to reduce the impact of random fluctuations and provide a more stable rainfall estimate. The result is a spatiotemporally consistent rainfall dataset that contains the rainfall distribution of the entire urban area at a specific time point.

[0149] Step S63: Detect and remove outliers on the spatiotemporally consistent rainfall dataset to obtain an outlier-free rainfall dataset;

[0150] Specifically, you can use data analysis software, such as Python's Pandas and SciPy libraries, to process a temporally consistent rainfall dataset. Load the temporally consistent rainfall dataset and perform statistical analysis on it to identify any outliers. Use the Z-score method to detect outliers, which calculates the number of standard deviations each data point has from the mean. Data points with a Z-score greater than 3 or less than -3 are considered anomalous. For example, if rainfall data in a particular area suddenly surges to 200 mm / hour, while rainfall in surrounding areas is only 50 mm / hour, this data point will have a Z-score significantly greater than 3, so it will be marked as an outlier. Manually examine this data point to determine whether it is due to sensor error, data entry error, or other non-rainfall factors. Once confirmed as an anomaly, remove it from the dataset. Use a box plot to further identify outliers. Box plots identify outliers in the data by displaying the quartiles and outliers. Examine the points outside the "whiskers" of the box plot and remove any obvious outliers. After outlier detection and removal, the abnormal rainfall dataset is obtained.

[0151] Step S64: spatially interpolate the abnormal rainfall dataset and generate a rainfall intensity prediction map to obtain a rainfall intensity prediction map, and perform contour drawing and rainfall distribution visualization on the rainfall intensity prediction map to obtain a city-level predicted rainfall distribution map to achieve rainfall observation evaluation.

[0152] Specifically, you can use GIS software, such as ArcGIS or QGIS, to process the anomaly-free rainfall dataset. Import the anomaly-free rainfall dataset into the GIS software and combine it with the city's geographic information. Use kriging interpolation to spatially interpolate the data. This is a statistical method based on geographic location and rainfall values ​​that can predict rainfall at unknown locations. For example, suppose you have rainfall data for 50 monitoring points, but you want to obtain rainfall forecasts for every 100-meter grid for the entire city. Use the kriging interpolation tool in the GIS software to input the rainfall and geographic location of the monitoring points. The software will automatically calculate and generate a rainfall forecast map for every 100-meter grid for the entire city. Next, generate a rainfall intensity forecast map and plot contour lines on it in the GIS software. Contour lines show the boundaries of areas with uniform rainfall. Set the contour line interval, for example, one contour line every 10 mm / hour, to clearly demonstrate variations in rainfall intensity. Finally, visualize the rainfall distribution on the rainfall intensity forecast map. Use the GIS software's visualization tools and select appropriate color schemes and symbols to visually display the rainfall distribution.

[0153] By traversing the raindrop monitoring camera identification set and executing the relevant steps, the present invention significantly enhances the timeliness of rainfall forecasts and can quickly generate a set of real-time rainfall forecast results. By aligning and integrating these results with spatiotemporal data, the consistency of the data in time and space is ensured. By detecting and removing outliers, the accuracy of the dataset is improved, and abnormal data that affects the accuracy of the forecast is removed. By performing spatial interpolation on the de-anomaly rainfall dataset and generating rainfall intensity forecast maps, a refined analysis of rainfall intensity is achieved. By drawing contour lines and visualizing rainfall distribution, data visualization is optimized, making the rainfall distribution intuitive and easy to understand.

[0154] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0155] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A rainfall observation and assessment method for high-density urban areas, characterized by: The following steps are involved: Step S1: Analyze the location of monitoring points in the target city to obtain a set of camera monitoring points; Selecting surveillance cameras for the target city based on the camera surveillance point set to obtain a raindrop surveillance camera identification set; wherein the raindrop surveillance camera identification set includes a plurality of raindrop surveillance camera identifications; Step S2: according to the raindrop monitoring camera identifier, a corresponding camera is used to collect a real-time rainfall image stream to obtain a real-time rainfall image sequence; based on the real-time rainfall image sequence, raindrop edges are automatically recognized to obtain a raindrop edge recognition result image; Step S3: Calculating the falling velocity of raindrops based on the raindrop edge recognition result image to obtain raindrop falling velocity data; extracting raindrop particle features based on the raindrop falling velocity data based on the real-time rainfall image sequence to obtain rainfall particle feature data; wherein step S3 includes the following steps: Step S31: removing incomplete raindrops from the raindrop edge recognition result map to obtain a regular raindrop edge recognition map, and assigning raindrop numbers to the regular raindrop edge recognition map to obtain a raindrop number set; Step S32: Calculating the area of ​​raindrops based on the regular raindrop edge recognition map and the raindrop number set to obtain a raindrop area set; Step S33: Calculate the pixel size of the raindrops according to the raindrop area set to obtain a raindrop pixel size dataset; wherein the specific calculation formula for the pixel size calculation is as follows: ； in, is pi; Step S34: randomly selecting adjacent raindrops according to the raindrop number set and the regular raindrop edge recognition map to obtain a first raindrop number and a second raindrop number; wherein the raindrops corresponding to the first raindrop number and the second raindrop number are adjacent raindrops; Step S35: determining the center points of the raindrop corresponding to the first raindrop number and the raindrop corresponding to the second raindrop number based on the regular raindrop edge recognition map to obtain the first raindrop center point and the second raindrop center point, and calculating the pixel distance between adjacent raindrops based on the first raindrop center point and the second raindrop center point to obtain adjacent raindrop pixel interval data; Step S36: extracting corresponding sizes from the raindrop pixel size dataset according to the first raindrop number and the second raindrop number to obtain first raindrop size data and second raindrop size data, and performing mean calculation on the first raindrop size data and the second raindrop size data to obtain average size data of adjacent raindrops; Step S37: Calculating the ratio of the average size of adjacent raindrops to the pixel spacing of adjacent raindrops to obtain a raindrop size-raindrop spacing ratio. Inferring the rainfall intensity of the target city based on the raindrop size-raindrop spacing ratio and a preset rainfall intensity table to obtain urban rainfall intensity data. Matching the urban rainfall intensity data with a preset rainfall intensity-velocity table to obtain raindrop falling velocity data. Step S38: extracting raindrop particle features based on the real-time rainfall image sequence and the raindrop falling velocity data to obtain rainfall particle feature data; Step S4: constructing a rainfall value prediction model; dynamically adjusting the rainfall value prediction model to obtain an updated rainfall value prediction model; Step S5: Based on the raindrop monitoring camera identification, the raindrop edge recognition result map and the rainfall particle feature data, the updated rainfall value prediction model is used to predict the rainfall at the camera location to obtain a real-time rainfall prediction result; Step S6: Traverse the raindrop monitoring camera identification set, and execute steps S2-S3 and step S5 for the camera corresponding to each raindrop monitoring camera identification to obtain a real-time rainfall prediction result set; visualize the real-time rainfall prediction result set to obtain a city predicted rainfall distribution map to realize rainfall observation evaluation.

2. The method for rainfall observation and assessment with full coverage of high-density cities according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Collect rainfall measurement points in the target city to obtain a rainfall measurement point distribution map; Step S12: collecting camera geographic coordinates and generating a distribution map of the target city to obtain a camera geographic distribution map; Step S13: collecting camera monitoring parameters according to the camera geographic distribution map to obtain a camera monitoring parameter table; wherein the camera monitoring parameter table includes a plurality of camera monitoring parameter items, each camera monitoring parameter item including at least a camera identifier, a camera image resolution, a camera installation height, a camera image orientation, and the camera's geographical coordinates; Step S14: historical rainfall records are collected for the target city to obtain a historical rainfall record set, and spatiotemporal feature recognition and rainfall pattern mining are performed on the historical rainfall record set to obtain a rainfall pattern recognition result set; Step S15: dividing and labeling the rainfall measurement point distribution map according to the rainfall pattern recognition result set to obtain a divided rainfall measurement point distribution map; wherein the rainfall area division includes the division of heavy rain prone areas and the division of low rainfall areas; Step S16: Based on the division of the rainfall measurement point distribution map and the camera geographical distribution map and according to the camera monitoring parameter table, the monitoring point location analysis is performed on the target city to obtain a camera monitoring point set; Step S17: Select surveillance cameras for the target city according to the camera surveillance point set to obtain a raindrop surveillance camera identification set; wherein the raindrop surveillance camera identification set includes a plurality of raindrop surveillance camera identifications.

3. The method for rainfall observation and assessment with full coverage of high-density cities according to claim 1 is characterized in that: Step S16 includes the following steps: Step S161: assigning rainfall area identifiers to the divided rainfall measurement point distribution map to obtain a rainfall area identifier set; Step S162: Calculating the rainfall area of ​​the rainfall area based on the distribution map of rainfall measurement points to obtain a rainfall area set; Step S163: determining the monitoring camera requirements for the rainfall area according to the preset rainfall area type rule, the rainfall area identifier set, and the rainfall area area set, to obtain a rainfall area monitoring requirement set; Step S164: Screening camera monitoring points in the rainfall area according to the rainfall area monitoring requirement set based on the camera geographic distribution map to obtain a camera monitoring point set, wherein the camera monitoring point set includes a plurality of camera monitoring points.

4. The method for rainfall observation and assessment with full coverage of high-density cities according to claim 2 is characterized in that: Step S17 includes the following steps: Step S171: extracting corresponding camera parameter items from the camera monitoring parameter table according to the camera monitoring point to obtain a candidate camera monitoring parameter item set; Step S172: If there is only one candidate camera monitoring parameter item in the candidate camera monitoring parameter item set, the corresponding camera is directly used as the raindrop monitoring camera to obtain the raindrop monitoring camera identifier; Step S173: if there are two or more candidate camera monitoring parameter items in the candidate camera monitoring parameter item set, deduplication is performed on the candidate camera monitoring parameter item set according to the camera shooting orientation to obtain a raindrop monitoring camera identifier; Step S174: traverse the camera monitoring point set, execute steps S171 to S173 for each camera monitoring point, and obtain a raindrop monitoring camera identification set; wherein the raindrop monitoring camera identification set includes several raindrop monitoring camera identifications.

5. The method for rainfall observation and assessment with full coverage of high-density cities according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: collecting a real-time rainfall image stream from a corresponding camera according to the raindrop monitoring camera identifier to obtain an original real-time rainfall image stream; Step S22: reconstructing the video frame time sequence of the original real-time rainfall image stream to obtain an initial real-time rainfall image sequence, and removing unqualified images from the initial real-time rainfall image sequence to obtain a filtered real-time rainfall image sequence; Step S23: performing image format conversion and standardization on each image in the screened real-time rainfall image sequence to obtain a standard real-time rainfall image sequence, and gray-scaling each image in the standard real-time rainfall image sequence to obtain a gray-scale real-time rainfall image sequence; Step S24: performing image noise filtering and smoothing processing on each image in the grayscale real-time rainfall image sequence to obtain a real-time rainfall image sequence; Step S25: Perform clarity evaluation and optimal image selection on the real-time rainfall image sequence to obtain the optimal real-time rainfall image, and perform image gradient calculation on the optimal real-time rainfall image to obtain rainfall image gradient data; perform edge detection on the optimal real-time rainfall image based on the rainfall image gradient data to obtain a raindrop edge recognition result image.

6. The method for rainfall observation and assessment with full coverage of high-density cities according to claim 1 is characterized in that: Step S38 includes the following steps: Step S381: detecting the motion trajectory of raindrops based on the real-time rainfall image sequence to obtain raindrop motion trajectory data; Step S382: Based on the regular raindrop edge recognition map, actual distance calculation is performed on a single pixel in the regular raindrop edge recognition map according to the raindrop falling velocity data and the raindrop motion trajectory data to obtain the actual distance data of the single pixel. The specific calculation formula for the actual distance calculation is as follows: ； in, is the raindrop velocity, is the time interval, is the number of pixel displacements of raindrops between video frames, is the actual distance corresponding to a single pixel; Step S383: Calculating the actual size of raindrops based on the regular raindrop edge recognition map and the actual distance data of individual pixels to obtain actual size data of raindrops. The specific calculation formula for calculating the actual size is as follows: ； in, is the actual size of raindrops, is the number of pixels occupied by raindrops in the regular raindrop edge recognition image, is the actual distance represented by a single pixel; Step S384: Calculate the raindrop spacing for the target city based on the regular raindrop edge recognition map and the actual distance data of each pixel to obtain average raindrop spacing data. The specific calculation formula for calculating the raindrop spacing is as follows: ； in, is the average raindrop spacing, is the index, is the total number of raindrops, For the The horizontal coordinate of the center point of the raindrop, For the The vertical coordinate of the center point of the raindrop, For the The horizontal coordinate of the center point of the raindrop, For the The vertical coordinate of the center point of the raindrop, The actual distance data of a single pixel; Step S385: Based on the regular raindrop edge recognition map, the rainfall density of the target city is calculated according to the actual distance data of a single pixel to obtain the raindrop spacing density data, and the actual raindrop size data, the average raindrop spacing data and the raindrop spacing density data are recorded as rainfall particle feature data.

7. The method for rainfall observation and assessment with full coverage of high-density cities according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: Obtain camera monitoring parameter table; Step S42: collecting historical rainfall videos and historical actual rainfall data for the target city according to the camera monitoring parameter table to obtain a historical rainfall video stream set and a historical actual rainfall data set; wherein the historical rainfall video stream set includes a plurality of historical rainfall video streams; Step S43: traverse the historical rainfall video stream set, and execute steps S22 to S25 and steps S31 to S38 for each historical rainfall video stream, thereby obtaining a historical raindrop edge recognition result atlas and a historical rainfall particle feature dataset; Step S44: using the historical actual rainfall dataset, the historical raindrop edge recognition result atlas, and the historical rainfall particle feature dataset to perform model training and error adjustment on the preset machine learning model to obtain a rainfall value prediction model; Step S45: Dynamically adjust the rainfall value prediction model to obtain an updated rainfall value prediction model.

8. The method for rainfall observation and assessment with full coverage of high-density cities according to claim 7 is characterized in that: Step S45 includes the following steps: Step S451: Obtaining a raindrop monitoring camera identification set; Step S452: Rainfall event data is collected for the target city based on the raindrop monitoring camera identification set to obtain a new rainfall video stream set and a new actual rainfall dataset; step S43 is then executed on the new rainfall video stream set to obtain a new raindrop edge recognition result atlas and a new rainfall particle feature dataset; Step S453: Recording the historical actual rainfall dataset, the historical raindrop edge recognition result atlas, and the historical rainfall particle feature dataset as a historical training dataset, and recording the new actual rainfall dataset, the new raindrop edge recognition result atlas, and the new rainfall particle feature dataset as a new training dataset; Step S454: Perform weighted data fusion on the historical training data set and the new training data set to obtain a fused training data set. The specific calculation formula for weighted data fusion is as follows: ； in, Fusion training dataset, New training dataset, Historical training dataset, is the weight coefficient; Step S455: perform incremental training and adjustment on the rainfall value prediction model using the fused training data set to obtain an updated rainfall value prediction model.

9. The method for rainfall observation and assessment with full coverage of high-density cities according to claim 1 is characterized in that: Step S6 includes the following steps: Step S61: traverse the raindrop monitoring camera identification set, and execute steps S2 to S3 and S5 for each camera corresponding to the raindrop monitoring camera identification, thereby obtaining a real-time rainfall prediction result set; Step S62: performing spatiotemporal data alignment and integration on the real-time rainfall prediction result set to obtain a spatiotemporal consistent rainfall dataset; Step S63: Detect and remove outliers on the spatiotemporally consistent rainfall dataset to obtain an outlier-free rainfall dataset; Step S64: spatially interpolate the abnormal rainfall dataset and generate a rainfall intensity prediction map to obtain a rainfall intensity prediction map, and perform contour drawing and rainfall distribution visualization on the rainfall intensity prediction map to obtain a city-level predicted rainfall distribution map to achieve rainfall observation evaluation.

Images