A method, device, electronic device and computer-readable storage medium for preprocessing storm cell data based on radar images

Abstract

The present invention provides a method, apparatus, electronic device and computer-readable storage medium for preprocessing storm cell data based on radar images. For the first time, the present invention codes the real-time changing radar images, assigns identification addresses to the identified storm cells and stores the coded data of the storm cell changes. By tracking the storm cell data and putting it into an RNN neural network for analysis, it provides a data basis for weather forecasting. Without the need to restore the radar change images, it can be directly referred to by meteorological personnel for meteorological changes, avoiding the waste of radar image information resources, facilitating the preservation and research of meteorological data, and improving the accuracy of meteorological forecasting for small-scale areas.

Description

Technical Field

[0001] The present invention relates to data processing technologies, and more particularly, to a method, apparatus, electronic device, and computer-readable storage medium for data acquisition and coding of storm cell data preprocessing from radar images. Background Art

[0002] As is well known, weather changes will affect people's production operations and daily routines. Therefore, improving the prediction accuracy of weather conditions will bring great convenience to people's daily lives and work. However, most of the current various meteorological prediction models are designed for large-scale area prediction. Its advantage is to reduce the prediction cost, but the disadvantages are also obvious. For different regions and areas with a small area range, it is difficult to predict whether the thunderstorm core area enters the target area. The intuitive result is that the meteorological forecast accuracy of small-scale areas is relatively low.

[0003] In addition, with the change of time in meteorological radar images, meteorological personnel combine the change trend and predict the meteorology based on their own experience, which requires a high level of meteorological personnel and needs to continuously observe the change of meteorology for judgment, thus increasing the workload of meteorological personnel. The existing technology also lacks the intuitive prediction by coding the changing meteorological image data, resulting in serious waste of radar image information. Summary of the Invention

[0004] To solve the deficiencies of the prior art, the present invention provides a method, apparatus, electronic device, and computer-readable storage medium for preprocessing storm cell data based on radar images, which are used to directly obtain coded meteorological data from radar images, improve the accuracy of prediction, and facilitate the preservation of meteorological data and the convenient use of meteorological personnel.

[0005] On the one hand, the present invention provides a method for preprocessing storm cell data based on radar images, including:

[0006] Step S01: Collect real-time radar images at an initial time (T0), and separate the original target mask images of different levels of storm cells in the real-time radar images;

[0007] Step S02: Obtain storm cell information based on each level of the original target mask image, where the storm cell information includes the storm cell contour and the center position. The storm cell contour is obtained by comparing whether the center position is less than or equal to a preset distance. If so, connect the corresponding storm cell contour; if not, the corresponding storm cell has an independent contour;

[0008] Step S03: Configure the prior identification address (ID) based on the storm cell information of different levels obtained in step S02 1), the prior identification address records the positions (x f , y f ) of all the storm cell contours, the central position, the pixel area of the storm cell, and saves the prior identification address to the database;

[0009] Step S04: After a fixed period or manually adjusted acquisition time (T 1 ), in the real-time radar image, the method of Steps S01 - S03 is used to acquire second target mask images of different levels, and a posterior identification address (ID 2 ) is configured;

[0010] Step S05: Compare whether the central positions corresponding to the prior identification address and the posterior identification address are less than or equal to a preset distance. If so, retain the prior identification address and update the central position; if not, retain the posterior identification address and the central position.

[0011] Further, the method further includes: the prior identification address or the posterior identification address also records the radar echo level, and the radar echo level is configured according to the reflectivity factor of the real-time radar image.

[0012] Further, the change data of the storm cell is acquired and coded from the real-time radar image. When the central position of the storm cell is less than or equal to a preset distance from the target prediction location, or any pixel of the contour of the storm cell is less than or equal to a preset distance from the target prediction location, expand the current radar image information data and combine the real-time meteorological data provided by the meteorological station at the target prediction location to construct the number of training layers and training function of a recurrent neural network (RNN), and input the change data and the real-time meteorological data into the recurrent neural network for training respectively to obtain the rainfall prediction parameter values.

[0013] On the other hand, the present invention also provides a storm cell data preprocessing device based on radar images, including: an image acquisition device for acquiring an original radar image and, after a fixed period or manually adjusted acquisition time, acquiring an updated real-time radar image;

[0014] An image processing device for separating the target mask images of each storm cell from the acquired radar image and for obtaining the storm cell information after coding the target mask images;

[0015] A data storage device for configuring an identification address for any of the mask images and for storing the storm cell information and the acquisition time;

[0016] A data processing device is used to compare the changes in the central positions of each storm cell in the mask after each update of the radar image. If the central position of each storm cell in the mask is less than or equal to a preset distance, the identification address is retained and the central position is updated; otherwise, the identification address and the central position are updated.

[0017] In another aspect, the present invention also provides an electronic device and a computer-readable storage medium, including: for executing the above-mentioned method for preprocessing storm cell data of radar images.

[0018] The present invention has the following beneficial effects:

[0019] 1. The present invention first codes the data of the real-time changing radar image. Without the need to restore the changing radar image, the changes of each storm cell are recorded by configuring the identification address, which can be directly referred to by meteorological personnel to avoid waste of radar image information resources, is conducive to the preservation and research of meteorological data, and can automatically track the dynamics of storm cells.

[0020] 2. The present invention strips each storm cell, calculates the central position of each storm cell, analyzes and judges the changes of local storm cells at specific levels, and combines multiple analyses such as moving direction, speed, and the contour pixels of each storm cell, which is conducive to predicting meteorological changes in a small local area and solving the problem of low accuracy of meteorological forecasts in small areas.

[0021] 3. The present invention combines real-time meteorological data, analyzes storm cells with any contour pixels falling within a specific range, summarizes multiple pieces of pre-information and puts them into the RNN neural network for analysis, provides a data basis for weather forecasting, is conducive to quickly and conveniently obtaining information and adjusting and optimizing strategies, the RNN will generate future rainfall data and flood depth, and the system will issue a warning according to the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and do not constitute an improper limitation to the present invention. In the drawings:

[0023] Figure 1 is a schematic diagram of the original radar image;

[0024] Figure 2 is a schematic diagram of the first-level target mask separated from the radar image of the present invention;

[0025] Figure 3 is a schematic diagram of the second-level target mask separated from the radar image of the present invention;

[0026] Figure 4 is a schematic diagram of the third-level target mask separated from the radar image of the present invention;

[0027] Figure 5 Schematic diagram of the integrated first - to - third - level target mask radar image of the present invention;

[0028] Figure 6 Schematic diagram of the independent contour of the storm cell of the present invention;

[0029] Figure 7 Schematic diagram of the merged contour of the storm cell of the present invention;

[0030] Figure 8 Flow chart of the mask image identification address of the present invention;

[0031] Figure 9 Flow chart of the identification address and center position update of the present invention;

[0032] Figure 10 Schematic diagram of the recurrent neural network model of the present invention; Detailed implementation manners

[0033] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0034] Step S01 Information acquisition stage:

[0035] As Figure 1 shown, it is a schematic diagram of the radar image of the original meteorological change. In this embodiment, the first acquisition time is recorded as T 0 time, and the radar image data is updated every 6 minutes thereafter. Each data update is recorded as T n time, where n is the data of the nth update.

[0036] First, directly obtain data from the radar image. Code the data of the radar image obtained from the radar map to obtain the important parameters of the radar image. In this embodiment, the center position of the storm cell, echo classification, and actual area are obtained from the radar image. Specifically, at T 0 time, obtain a radar image with rich information, such as Figure 1As shown, the radar image has real-time storm cell information, recording the approximate area and rainfall intensity of the current storm. In the embodiment of the present invention, the radar image reflectivity factor in the original radar image is configured, that is, the color tone, saturation value, and output tone of the radar image are set in advance to obtain five different levels of target colors with distinct segmentation. Specifically, five-level classification is performed through the radar image color information (dBz), where purple (60 - 72 dBZ) is level five, red (53 - 60 dBZ) is level four, yellow (37 - 53 dBZ) is level three, green (21 - 37 dBZ) is level two, and blue (2 - 18 dBZ) is level one. The classified target colors are the masks that divide the complex original radar image into individual storm cell levels and are also the echo level indicators of the storm cells.

[0037] As Figures 2 - 4 shown is the target mask image after storm cell classification, where Figure 2 is the target mask image of the first level (level 1), which is one of the storm cell level masks segmented from the original radar image at time T 0 . Among them, the object with an obvious contour (i.e., all white areas A in the figure) is the target storm cell. All white areas A are the target storm cell masks, and the black area B is the rest of the non-target research objects, such as other storm cells, oceans, or land and other irrelevant information, to ensure the correct classification of different target objects and exclude redundant and complex information. By setting the preset values for different storm cells at the same time, that is, setting the color tone, saturation value, and output tone according to the original information of the radar image, the masks of different storm cells can be output. Figure 3 is the target mask image of the second level (level 2), Figure 4 is the target mask image of the third level (level 3), forming the corresponding target mask. Figure 5 is a schematic diagram of restoring the first-level to third-level target mask images to the original radar image after actual radar image acquisition, where Figure 5 shows the approximate location, echo level, and corresponding range of the storm cells at the corresponding levels on the map.

[0038] Step S02 Information Acquisition Phase:

[0039] After the storm cell mask is classified and set, various information of the storm cells is obtained, such as data on the storm cell contour and center position. Taking the Figure 2 shown mask as an example, first calculate the area enclosed by the target contour, and obtain the center position according to the following formula.

[0040] (1)

[0041] (2)

[0042] (3)

[0043] (4)

[0044] (5)

[0045] Wherein, mu 00 = m 00 , nu 00 = 1, nu 10 = mu 10 = mu 01 = mu 10 = 0. x, y are the pixel positions of the mask on the radar screen, and are the central pixel positions. The superscripts j, i represent moments, nu ji represents the ratio of the central pixel position to the mask. As Figure 6 shown, compare the distance D between the central positions of the masks at the same level. If the distance D is greater than the preset distance, where the preset distance in this embodiment is 25 real-time radar image pixels, then the corresponding storm cell has an independent storm cell contour. If not, then as Figure 7 shown, connect the storm cell contours of the corresponding masks.

[0046] The central position comparison formula is:

[0047] (6)

[0048] Where Hypot is the output result of the comparison of the two central positions.

[0049] Step S03 Configure identification addresses:

[0050] Configure the prior identification address (ID 1 ) based on the storm cell information at different levels obtained in step S03. The prior identification address records the positions (x f , y f ) of all the storm cell contours, the central positions, and the pixel area of the storm cells, and save the prior identification address to the database.

[0051] Step S04 Storm cell data update

[0052] As Figure 8 shown, in the next acquisition time period ( T nObtain the mask information at a certain moment. In this embodiment, T n The moment is a fixed period of 6 minutes. In other embodiments, the acquisition time can be adjusted manually according to actual needs. At T 1 Based on this, that is, 6 minutes after the initial moment, in the real-time radar image, the method of steps S01 - S03 is used to collect the second target mask image of the corresponding level, and the subsequent identification address (ID 2 ) is obtained.

[0053] Step 05 Storm cell data tracking

[0054] Extract all identification addresses from the database, compare the changes in the central positions of all masks in the identification addresses before and after the acquisition time, such as whether the change in the central position of the corresponding mask in the prior identification address and the subsequent identification address is less than or equal to the preset distance. Refer to Figure 9 . In this embodiment, the preset distance is still set to 25 pixels of the real-time radar image. If so, the corresponding mask is retained in the prior identification address, that is, the prior identification address, and the storm cell center position is updated; if not, the corresponding mask retains the subsequent identification address and the center position. As shown in Table 1.

[0055] Table 1 Coding of radar image information data

[0056]

[0057] As can be seen from Table 1, each independent identification address records the coded mask information of the radar image, including the central position, echo level, mask area, contour pixel coordinates, and recording time. The system refreshes the radar image data every 6 minutes. Combining with Appendix Figure 8 , at T 0 The moment, obtain the mask information, including the central position (x 0 , y 0 ) etc. of the mask at this time, configure the identification address ID and save other data. The radar image refreshes the image after 6 minutes. At this time, it is T 1 The moment, obtain the central position (x 1 , y 1 ) of the mask at this time, configure the new identification address ID and save other data. Compare and judge through the central position comparison formula (6), as Figure 9 shown. If Hypot is less than or equal to 25 pixels, then the identification address ID at T 0 replaces the identification address ID at T 1 (that is, ID 1 records as ID 2 ), (i.e., ID 2 is recorded as ID 1, that is, the identification address ID remains unchanged), the two storm cells are the same storm cell; otherwise, both the ID address and the central position data are updated. As shown in Table 2, it is the record of storm cell data in this embodiment.

[0058] Furthermore, the steps for obtaining, analyzing the parameters of the storm cell and the data for meteorological prediction of the storm cell also include the following:

[0059] Calculate the moving speed of the rain area. The moving speed of the rain area is obtained by the following calculation formula according to the data coded from the radar image by the staff:

[0060] Table 2 Record of the time change of the storm cell

[0061]

[0062] (7)

[0063] (8)

[0064] Among them, Distance represents the distance between the two central positions, and time is every 6 minutes. The moving speed of the storm cell can be obtained according to formula (8).

[0065] Taking the actual data in Table 2 as an example, for storm cell 1 (ID 1 ), its moving speed between 8:00 am and 8:06 am can be known through formulas (7) and (8) to be 30 pixels per minute. Then, according to the radar map scale, the pixels are converted into the real distance and real area. Through the following formulas (9) and (10), the actual moving speed and influence range of the rain area can be obtained.

[0066] Real distance = pixel distance × radar map scale (9)

[0067] Real area = pixel area × radar map scale 2 (10)

[0068] Secondly, the moving wind direction of the rain area is obtained through formula (11):

[0069] (11)

[0070] According to the meteorological prediction of the storm cell, the detailed data of the storm cell are updated, as shown in Table 3.

[0071] Table 3 Extended data of the coded radar image information

[0072]

[0073] Finally, determine the status of the storm cell and the target prediction location. If the center position of the storm cell is within the contour of the storm cell, determine whether the center position of the storm cell is less than or equal to the preset distance from the target prediction location. If the center position of the storm cell is not on the contour of the storm cell, determine whether any pixel of the contour of the storm cell is less than or equal to the preset distance from the target prediction location. When the storm cell falls within the range of the target prediction location, that is, when the above corresponding status is satisfied, expand the current radar image information data (i.e., the data in Table 3) and combine it with the real-time meteorological data provided by the meteorological station at the target prediction location. A total of 15 indicators are summarized (see Figure 10 ) for inputting the data indicators of the artificial network. Among them, the data of each storm cell are stored at the identification address ID n , including the center position (x n , y n ), echo level (Level) n , pixel mask area (Area M) n , actual area of the storm cell (Area R) n , date (Date) n , time (Time) n , moving speed (Speed) n , moving direction (Direction) n , etc., which are obtained by collecting data from the radar image and coding through the above method. The above 8 items of data are the change data of the storm cell obtained by collecting and coding the real-time radar image. Indicators such as the latest air pressure, temperature, humidity, wind speed, wind direction, rainfall in millimeters in the past hour, and total rainfall on the same day are provided by the meteorological department, that is, the meteorological department provides 7 items of real-time meteorological data. See Appendix Figure 10 . Through constructing a recurrent neural network (RNN) for deep learning, setting the number of training layers and training function of the RNN, and inputting the above 15 items of data into the RNN for training respectively, the rain trend prediction parameter values are obtained. The pre-trial prediction parameter values include the rainfall (mm) in the next hour, the duration of heavy rain, and the waterlogging depth in the next hour.

[0074] Furthermore, before performing the recurrent neural network RNN, the data is checked and repaired to ensure that the RNN model can run smoothly. In this embodiment, if it is found during the check that any data in the identification address is missing, the missing data is predicted or generated through the Bayesian inference and prior probability mechanism.

[0075] After obtaining the meteorological prediction data through the recurrent neural network RNN, corresponding warnings are issued according to the indicators of the local meteorological bureau, and a hydrograph is constructed by integrating the above various indicator parameters.

[0076] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims described above.

Claims

1. A storm cell data preprocessing method based on radar images, It is characterized in that include: Step S01: acquiring a real-time radar image at an initial time (T0), and separating original target mask images of storm cells of different levels in the real-time radar image; Step S02: acquiring storm cell information based on the original target mask image of each level, wherein the storm cell information includes a storm cell outline and a center position, wherein the storm cell outline is determined by comparing whether the center position is less than or equal to a preset distance, and if so, the corresponding storm cell outline is connected; if not, the corresponding storm cell is an independent outline; Step S03: Configure a prior identification address (ID 1 ) based on the storm cell information at different levels obtained in Step S02. The prior identification address records the positions (x f , y f ) of all the storm cell contours, the central position, and the pixel area of the storm cell, and saves the prior identification address to the database; Step S04: After a fixed period or manually adjusted acquisition time (T 1 ), in the real-time radar image, use the methods of steps S01 - S03 to collect second target mask images at different levels, and configure the post-identifier address (ID 2 ); Step S05: Compare the center positions corresponding to the previous identification address and the subsequent identification address to see whether they are less than or equal to a preset distance. If so, retain the previous identification address and update the center position; if not, retain the subsequent identification address and the center position; The preceding identification address or the succeeding identification address further records a radar echo level, and the radar echo level is configured according to the real-time radar image reflectivity factor; The change data of the storm cell is acquired by collecting and coding the real-time radar image. When the center position of the storm cell is less than or equal to the preset distance from the target prediction location, or any pixel of the outline of the storm cell is less than or equal to the preset distance from the target prediction location, the current radar image information data is expanded and combined with the real-time meteorological data provided by the meteorological station at the target prediction location, a recurrent neural network (RNN) training layer number and a training function are constructed, and the change data and the real-time meteorological data are respectively input into the recurrent neural network for training to obtain rainfall prediction parameter values, wherein the change data include the center position of the storm cell, the echo level, the pixel mask area, the actual area of ​​the storm cell, the date, the time, the moving speed, and the moving direction.

2. An electronic device, include: processor; a memory for storing instructions executable by the processor; The processor is used to execute the method according to claim 1.

3. A computer-readable storage medium storing a computer program for executing the method of claim 1.

Images