Meteorological image prediction method, device, computer equipment and storage medium

Abstract

The present application relates to a meteorological image prediction method, device, computer equipment and storage medium. The method comprises: obtaining a meteorological image to be predicted; identifying an image category corresponding to the meteorological image to be predicted; matching an image prediction algorithm of a corresponding type according to the image category; and predicting the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image. The method can improve the accuracy of meteorological image prediction.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a meteorological image prediction method, device, computer equipment and storage medium. Background Art

[0002] With the development of computer technology, it is becoming more and more important to use computer technology to predict weather in the field of meteorology. In traditional technology, the optical flow method in computer vision is used to calculate the motion information of objects between adjacent frames of meteorological images to predict meteorological images.

[0003] However, in the process of predicting meteorological images using prediction algorithms such as the optical flow method in traditional technologies, one of the algorithms is manually selected from multiple types for image prediction. Since the process of manually selecting algorithms is subjective, the accuracy of the algorithm selection is low, which in turn results in low prediction accuracy for meteorological images. Summary of the invention

[0004] Based on this, it is necessary to provide a meteorological image prediction method, device, computer equipment and storage medium that can improve the accuracy of meteorological image prediction in response to the above technical problems.

[0005] A meteorological image prediction method, the method comprising:

[0006] Obtaining the meteorological image to be predicted;

[0007] Identifying the image category corresponding to the meteorological image to be predicted;

[0008] Matching a corresponding type of image prediction algorithm according to the image category;

[0009] The meteorological image to be predicted is predicted according to the matched image prediction algorithm to obtain a predicted image.

[0010] In one embodiment, the method further comprises:

[0011] Acquire initial meteorological images;

[0012] Identifying an initial image category of each of the initial meteorological images;

[0013] Labeling each of the initial meteorological images according to the category of the initial image to obtain a training set;

[0014] Using the training set to perform model training to obtain an image category recognition algorithm;

[0015] The identifying the image category corresponding to the meteorological image to be predicted includes:

[0016] According to the image category recognition algorithm, the category of each of the meteorological images to be predicted is identified to obtain the image category of the meteorological image to be predicted.

[0017] In one embodiment, the initial meteorological image includes at least two initial region sub-images; and the step of labeling each of the initial meteorological images according to the category of the initial image to obtain a training set includes:

[0018] When each of the initial area sub-images of the initial meteorological image corresponds to a different image category, the initial area sub-images are respectively labeled according to the different image categories to obtain initial area labeling data corresponding to each of the initial area sub-images;

[0019] Obtaining initial image annotation data corresponding to the initial meteorological image according to each of the initial region annotation data;

[0020] A training set is obtained according to the initial image annotation data of a plurality of initial meteorological images.

[0021] In one embodiment, the initial meteorological image includes at least two initial region sub-images; and the step of labeling each of the initial meteorological images according to the category of the initial image to obtain a training set includes:

[0022] When the initial area sub-images of the initial meteorological image correspond to different image categories, obtaining a rainfall intensity index corresponding to each of the initial area sub-images;

[0023] Extracting the image category corresponding to the initial area sub-image with the largest rainfall intensity index as the image category of the initial meteorological image;

[0024] Performing category labeling on the initial meteorological image according to the extracted image category to obtain initial image labeling data of the initial meteorological image;

[0025] A labeling set is obtained according to initial image labeling data of a plurality of initial meteorological images.

[0026] In one embodiment, the meteorological image to be predicted includes at least two sub-images of the area to be predicted; the image prediction algorithm of the corresponding type matching according to the image category includes:

[0027] When the sub-images of the to-be-predicted area of ​​the to-be-predicted meteorological image correspond to different prediction image categories, matching corresponding types of image prediction algorithms according to the prediction image categories;

[0028] The step of predicting the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image includes:

[0029] Predicting each of the to-be-predicted region sub-images according to the matched image prediction algorithms to obtain a region prediction sub-image;

[0030] A predicted image corresponding to the meteorological image to be predicted is obtained according to each of the regional prediction sub-images.

[0031] In one embodiment, the method further comprises:

[0032] Acquire meteorological image sequences;

[0033] Using multiple types of image prediction algorithms to predict the next time series meteorological image of the meteorological image sequence to obtain a predicted meteorological image;

[0034] Acquire the next real time-series meteorological image in the meteorological image sequence;

[0035] Calculating the difference between the predicted meteorological image predicted by each of the image prediction algorithms and the real meteorological image;

[0036] The image prediction algorithm with the smallest difference is extracted as the image prediction algorithm corresponding to the meteorological image sequence, and the category of the meteorological image sequence and the matching image prediction algorithm are stored in correspondence.

[0037] In one embodiment, the method further comprises:

[0038] Acquire a simulated image;

[0039] Performing image processing on the simulated image to obtain a simulated image sequence with time sequence changes;

[0040] Using various types of image prediction algorithms to predict the next time-series image of the simulated image sequence to obtain a predicted simulated image;

[0041] Acquire a real simulated image; wherein the real simulated image is the next sequential image in the simulated image sequence;

[0042] Calculating the difference between the predicted simulated image predicted by each of the image prediction algorithms and the real simulated image;

[0043] The image prediction algorithm with the smallest difference is extracted as the image prediction algorithm corresponding to the simulated image, and the image category of the simulated image is stored in correspondence with the matching image prediction algorithm.

[0044] A meteorological image prediction device, comprising:

[0045] An image acquisition module is used to acquire the meteorological image to be predicted;

[0046] A category recognition module, used to identify the image category corresponding to the meteorological image to be predicted;

[0047] An algorithm matching module, used for matching a corresponding type of image prediction algorithm according to the image category;

[0048] The prediction module is used to predict the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image.

[0049] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0050] A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

[0051] The above-mentioned meteorological image prediction method, device, computer equipment and storage medium include: obtaining the meteorological image to be predicted; identifying the image category corresponding to the meteorological image to be predicted; matching the image prediction algorithm of the corresponding type according to the image category; predicting the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image. In the process of predicting the meteorological image, the category of the meteorological image to be predicted is first automatically identified, and then the prediction algorithm of the corresponding type is matched according to the image category, so that the corresponding image prediction can be performed according to the algorithm matching the image category, thereby improving the accuracy of the meteorological image prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A diagram showing an application environment of a meteorological image prediction method in an embodiment;

[0053] Figure 2 A schematic diagram of a flow chart of a meteorological image prediction method in one embodiment;

[0054] Figure 3 A schematic flow chart of a meteorological image prediction step provided in another embodiment;

[0055] Figure 4 A training structure diagram of an image category recognition model provided in one embodiment;

[0056] Figure 5 A schematic diagram of a grayscale handwritten digital image provided in one embodiment;

[0057] Figure 6 is a schematic diagram of an image sequence provided in an embodiment;

[0058] Figure 7 A schematic diagram of a different optical flow method for performing prediction processing provided in one embodiment;

[0059] Figure 8 A schematic diagram of a flow chart of a method for determining a target optical flow method provided in an embodiment;

[0060] Fig. 9 is a structural block diagram of a meteorological image prediction device in one embodiment;

[0061] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0063] The meteorological image prediction method provided in this application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The server 104 obtains the meteorological image to be predicted; identifies the image category corresponding to the meteorological image to be predicted; matches the image prediction algorithm of the corresponding type according to the image category; predicts the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image. Furthermore, the server 104 is also used to push the predicted image to the terminal 102 and display it. Among them, the terminal 102 can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices, and the server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0064] In one embodiment, Figure 2 As shown, a meteorological image prediction method is provided, which is applied to Figure 1 The server in the example is used to illustrate the following steps:

[0065] Step 210: Acquire the meteorological image to be predicted.

[0066] The meteorological image to be predicted may be a radar echo map, which is used to characterize rainfall intensity, and the radar echo map corresponding to the current time may be used to predict the radar echo map within a preset time period in the future, so as to achieve the prediction of future weather. Specifically, at least two meteorological images to be predicted that are continuous in time may be obtained, and the meteorological image within the future time period may be predicted based on the at least two meteorological images to be predicted, so as to perform weather forecasting based on the predicted meteorological images.

[0067] Step 220: Identify the image category corresponding to the meteorological image to be predicted.

[0068] Different meteorological images to be predicted have different image features. According to the image features, the meteorological images to be predicted can be divided into different image categories. For example, according to the motion state of the meteorological images to be predicted, the meteorological images to be predicted can be divided into the meteorological images to be predicted of the translation category, the meteorological images to be predicted of the rotation category, the meteorological images to be predicted of the comprehensive translation and rotation category, the meteorological images to be predicted of the scale change category, and the meteorological images to be predicted of the comprehensive scale change and translation category. Further, the meteorological images to be predicted of the translation category are radar echo change images corresponding to stable layered cloud precipitation; the meteorological images to be predicted of the rotation and rotation plus translation category are radar echo change images corresponding to typhoons; the meteorological images to be predicted of the magnification and magnification plus translation category are radar echo change images corresponding to the occurrence and development stage of convection; the meteorological images to be predicted of the reduction and reduction plus translation category are radar echo change images corresponding to the occurrence and extinction stage of convection. And it should be noted that different image categories correspond to different target weather.

[0069] Step 230: Match the corresponding type of image prediction algorithm according to the image category.

[0070] The image features of different categories of meteorological images to be predicted are different, so the image prediction algorithms applicable to different categories of meteorological images to be predicted are also different. Specifically, since different types of image prediction algorithms have different prediction capabilities and prediction accuracy for different categories of meteorological images to be predicted, the corresponding types of image prediction algorithms can be matched according to the image categories of the meteorological images to be predicted. Specifically, different image categories can be associated and bound with different types of image prediction algorithms in advance and stored in a database, so that the corresponding types of image prediction algorithms can be matched directly from the database according to the image categories. And further, in the specific implementation, the association relationships in the database can be added, deleted, and modified according to the actual situation, so as to realize real-time updating of the association relationships in the database.

[0071] Among them, the image prediction algorithm can be an optical flow method. The optical flow method is a computer vision method that uses the change of pixels in the time domain in the image sequence and the correlation between adjacent frames to obtain the corresponding relationship between the historical frame image and the current frame image, thereby calculating the motion information of the object between adjacent frame images. In addition, the optical flow method is widely used in the field of weather nowcasting, and on the basis of the two most basic methods, more extended methods have been developed through various improvements. The two most basic methods are dense optical flow method and sparse optical flow method. The optical flow fields calculated by dense optical flow method and sparse optical flow method are dense optical flow field and sparse optical flow field respectively. The dense optical flow method calculates the optical flow of each pixel in the meteorological grayscale image, while the sparse optical flow method calculates the optical flow of feature points in the meteorological grayscale image.

[0072] In different weather evolution processes, the image features shown on the radar echo map are also different. For example, the increase and decrease of the radar echo correspond to the occurrence and development stage and the extinction stage of the coherent flow, respectively. The movement of the radar echo corresponds to the movement process of large-scale layered cloud precipitation weather, and the rotation of the radar echo corresponds to the typhoon process, etc. Since there are many optical flow methods developed at present, each optical flow method has its own different characteristics, so each optical flow method may have different applicability to different weather evolution processes. Therefore, this application proposes a method for comparing the applicability of multiple optical flow methods under different radar echo changes. The method is used to obtain the conclusion of which weather evolution process each optical flow method is suitable for. Specifically, in real-time business, a deep learning classification network is used to classify the evolution of real-time radar echo images. For the classified radar meteorological echo images, the most suitable optical flow method is automatically selected for extrapolation, thereby improving the accuracy of the optical flow method in the near-forecast business.

[0073] Specifically, the optical flow method in this application can come from the pysteps system developed by the Royal International Oceanographic Institute of Belgium and the rainymotion system developed by a German meteorological agency. These two systems include different optical flow methods. In this patent, two methods, LK_ex and LK_sprog, are selected in the pysteps system, and four methods, Dense, DenseRotation, Sparse_similarity, and Sparse, are selected in the rainymotion system, as shown in Table 1.

[0074] Table 1 Different optical flow methods and their sources

[0075] method Method source Dense rainymotion DenseRotation rainymotion Sparse_similarity rainymotion Sparse rainymotion LK_ex pysteps Lk_sprog pysteps

[0076] Step 240: predict the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image.

[0077] Specifically, the predicted image can be obtained by predicting the meteorological image to be predicted according to an image prediction algorithm that matches the image category of the meteorological image to be predicted. The predicted image is an image corresponding to the next time frame of the meteorological image to be predicted, which can be a predicted meteorological forecast image used to characterize future weather conditions.

[0078] In this embodiment, in the process of predicting meteorological images, the category of the meteorological image to be predicted is first automatically identified, and then the corresponding type of image prediction algorithm is matched according to the image category, so as to realize the corresponding image prediction according to the prediction algorithm matching the image category, thereby improving the accuracy of meteorological image prediction. Specifically, in this embodiment, by matching the image prediction algorithm of the corresponding type according to the image category, it is possible to obtain a conclusion on which type of weather evolution process different types of image prediction algorithms are suitable for. In real-time services, a mature deep learning classification network is used to classify the evolution of real-time meteorological images to be predicted, such as radar echo images, to obtain image categories, and for the obtained image categories, the most suitable image prediction algorithm, such as the optical flow method, is automatically matched and selected to perform image extrapolation, thereby improving the accuracy of the optical flow method in the near-forecasting service.

[0079] In one of the embodiments, the method further includes: acquiring initial meteorological images; identifying the initial image category of each of the initial meteorological images; labeling each of the initial meteorological images according to the initial image category to obtain a training set; and using the training set to perform model training to obtain an image category recognition algorithm.

[0080] Specifically, the image category of the to-be-predicted meteorological image can be identified using a pre-trained image category recognition algorithm, such as Figure 3 As shown, a flow chart of the meteorological image prediction step in another embodiment is provided, specifically, the training step of the image category recognition algorithm includes: step 310, image category labeling, specifically, pre-labeling the initial meteorological image with the image category to obtain a training set; step 320, image category prediction model training, specifically, using the training set to train the machine learning model to obtain an image category recognition model; step 330, real-time echo classification, specifically, automatically identifying the image category of the meteorological image to be predicted according to the image category recognition model; step 340, automatically matching the corresponding type of optical flow method according to the image category; step 350, performing image prediction according to the matched optical flow method to realize extrapolation of the predicted image.

[0081] The step of obtaining the training set includes: obtaining an initial meteorological image, wherein the initial meteorological image may be a meteorological image of multiple categories, such as a meteorological image of a translation category, a meteorological image of a rotation category, a meteorological image of a comprehensive translation and rotation category, and a meteorological image of a scale change category, etc. Then, the meteorological images of different categories are annotated to obtain the annotated data corresponding to each initial meteorological image, and the annotated data is used as a training set to train the machine learning model to obtain an image category recognition model.

[0082] Specifically, a radar echo map data set within a preset time period can be obtained, and the radar echo map data set can be divided into a training set and a test set at a ratio of 4:1 to obtain a training set and a test set. Then, the initial meteorological images in the training set are labeled by category. For example, for the initial meteorological image at time t0, an image category is marked for each initial meteorological image according to the echo changes of the first 5-10 frames at time t0. For example, in one embodiment, the basis for dividing the image categories may be: if the echo in the initial meteorological image moves in one direction and there is no obvious change in area and strength, the image category of the initial meteorological image is marked as a translation category and can be marked as category 1 at the same time; if the echo in the initial meteorological image corresponds to a rotation echo, the image category of the initial meteorological image can be marked as a rotation category and can be marked as category 2 at the same time; if the echo in the initial meteorological image corresponds to from nothing to something or the area continues to increase and the intensity becomes stronger, the image category of the initial meteorological image can be marked as scale increase and can be marked as category 3 at the same time; if the echo in the initial meteorological image corresponds to scale increase plus movement, the image category of the initial meteorological image is marked as movement plus increase and can be marked as category 4 at the same time; if the echo in the initial meteorological image corresponds to scale reduction and the scale reduction is combined with movement, the echo area becomes smaller or decreases while moving, and the image category of the initial meteorological image can be marked as category 5 at this time.

[0083] It should be noted that, in this embodiment, the category labeling of the initial meteorological image may be performed automatically by a computer or manually based on experience.

[0084] Specifically, the model framework of the image category recognition algorithm in this embodiment can be a CNN network model framework, specifically, AlexNet, InceptionNet, ResNet, VGGNet, etc., which are not limited here. In a specific embodiment, the ResNet model can be selected to classify radar echo images, where RestNet proposes the concept of residual network. In previous network structures, stack layers are used to increase the number of network layers, thereby obtaining a better network structure. Figure 4 As shown, Figure 4 A training structure diagram of an image category recognition model is provided. Figure 4An initial meteorological image of known image category is input, and its image category label is Groundtruth (real label). After the initial meteorological image is calculated and processed by the ResNet classification network, a predicted image is output, and the predicted image carries a category label, where the category label is Predict (prediction label). Then the Loss between Groundtruth and Predict is calculated to update the category recognition parameters in the network model according to the numerical value of the loss degree, and the image category recognition model after the updated recognition parameters is used to continue to recognize the category of the input initial meteorological image again. This iteration is repeated until the initially set number of steps (number of iterations) is completed, and the training of the initial meteorological image is completed. Then the next initial meteorological image is input until the training of all initial meteorological images is completed, and the network training is completed.

[0085] Furthermore, the image category corresponding to the meteorological image to be predicted is identified, including: according to the image category recognition algorithm, the category of each meteorological image to be predicted is identified to obtain the image category of the meteorological image to be predicted. In the specific implementation, the radar echo image acquired in real time is classified and identified using a trained image category recognition algorithm. Specifically, the real-time radar echo image is classified using a pre-trained ResNet, which realizes the automatic recognition of the image category of the meteorological image to be predicted, improves the efficiency and accuracy of image category recognition, and can also update the model in time when adding new image categories or deleting image categories, so that the model has a wider applicability and improves the applicability of the model.

[0086] In one embodiment, the initial meteorological image includes at least two initial area sub-images; the category labeling of each of the initial meteorological images according to the initial image category to obtain a training set includes: when each of the initial area sub-images of the initial meteorological image corresponds to a different image category, the category labeling of the initial area sub-images according to the different image categories is respectively performed to obtain initial area labeling data corresponding to each of the initial area sub-images; initial image labeling data corresponding to the initial meteorological image is obtained according to each of the initial area labeling data; and a training set is obtained according to the initial image labeling data of multiple initial meteorological images.

[0087] When the initial meteorological image corresponds to multiple image categories, the initial meteorological image can be partitioned according to the multiple image categories to obtain at least two initial area sub-maps, and each initial area sub-map can correspond to an image category. In other embodiments, an initial area sub-map can also correspond to more than one image category. In this case, one image category can be dominant in each initial area sub-map.

[0088] Furthermore, when the initial meteorological image corresponds to multiple image categories, it also includes: labeling each initial regional sub-image according to the image category corresponding to each initial regional sub-image, obtaining initial regional labeling data corresponding to each initial regional sub-image, and further, obtaining initial image labeling data corresponding to the initial meteorological image according to the initial regional labeling data of each initial regional sub-image. Furthermore, a training set is obtained according to the initial image labeling data of multiple initial meteorological images.

[0089] In this embodiment, when the initial meteorological image corresponds to multiple image categories, it also includes partitioning the initial meteorological image according to the image categories to obtain initial area sub-images corresponding to different image categories, and then labeling the initial area sub-images according to the different image categories to obtain initial area labeling data corresponding to each initial area sub-image, thereby achieving more accurate image category labeling for the same initial meteorological image. When the category of the initial meteorological image is identified using a trained image category recognition algorithm, the regional features of each initial meteorological image can be obtained to obtain an image category containing at least one, thereby improving the recognition accuracy of the image category.

[0090] In one of the embodiments, the meteorological image to be predicted includes at least two regional sub-images to be predicted; the image prediction algorithm of the corresponding type is matched according to the image category, including: when the regional sub-images to be predicted of the meteorological image to be predicted correspond to different prediction image categories, the image prediction algorithm of the corresponding type is matched according to each prediction image category; the meteorological image to be predicted is predicted according to the matched image prediction algorithm to obtain a predicted image, including: predicting each regional sub-image to be predicted according to each matched image prediction algorithm to obtain a regional prediction sub-image; obtaining the predicted image corresponding to the meteorological image to be predicted according to each regional prediction sub-image.

[0091] Specifically, when different initial region sub-images in the initial meteorological image correspond to different image categories, the image categories of the different initial region sub-images can be annotated to obtain a training set, and then the training set can be used to train the model to obtain an image category recognition model. When the image category recognition model is used to perform category recognition on the meteorological image to be predicted, the image categories of the multiple region sub-images to be predicted of the meteorological image to be predicted can be annotated to obtain a meteorological image to be predicted containing multiple image categories.

[0092] In one embodiment, image prediction algorithms corresponding to different image categories can be obtained, so as to perform image prediction on sub-images of different categories of regions to be predicted according to the matched multiple image prediction algorithms, obtain prediction sub-images of each region, and finally merge the prediction sub-images of each region to obtain the prediction image corresponding to the meteorological image to be predicted. In another embodiment, image prediction algorithms corresponding to different image categories can also be obtained, and then rainfall intensity indexes corresponding to different sub-images of regions to be predicted can be obtained, and the image prediction algorithm of the sub-image of the region to be predicted with the largest rainfall intensity index is used as the image prediction algorithm corresponding to the meteorological image to be predicted, and then the image prediction algorithm is used to predict the meteorological image to be predicted, and the prediction image corresponding to the meteorological image to be predicted is obtained.

[0093] In another embodiment, the initial meteorological image includes at least two initial area sub-images; the category labeling of each of the initial meteorological images according to the initial image category to obtain a training set includes: when the initial area sub-images of the initial meteorological image correspond to different image categories, obtaining a rainfall intensity index corresponding to each of the initial area sub-images; extracting the image category corresponding to the initial area sub-image with the largest rainfall intensity index as the image category of the initial meteorological image; category labeling of the initial meteorological image according to the extracted image category to obtain initial image labeling data of the initial meteorological image; and obtaining a labeling set based on the initial image labeling data of multiple initial meteorological images.

[0094] The rainfall intensity index is used to characterize the probability of rainfall, and the higher the rainfall intensity index, the greater the probability of rainfall, and the lower the rainfall intensity index, the smaller the probability of rainfall. In another embodiment, when different initial area sub-images of the initial meteorological image correspond to different image categories, it also includes obtaining the rainfall intensity index corresponding to each initial area sub-image, and taking the image category of the initial area sub-image with the largest rainfall intensity index as the image category corresponding to the initial meteorological image. In one embodiment, the rainfall intensity index can be the grayscale value of the initial meteorological image, and generally the larger the grayscale value, the larger the rainfall intensity index.

[0095] In this embodiment, when the initial meteorological image corresponds to multiple image types, it also includes determining the image category corresponding to the initial meteorological image according to the rainfall intensity index corresponding to each initial regional sub-image, so as to take the image category corresponding to the initial regional sub-image with the highest probability of rainfall as the image category corresponding to the initial meteorological image, thereby achieving the purpose of accurately determining the image category.

[0096] In one of the embodiments, the method also includes: obtaining a meteorological image sequence; predicting the next time-series meteorological image of the meteorological image sequence using multiple categories of image prediction algorithms to obtain a predicted meteorological image; obtaining the actual next time-series meteorological image in the meteorological image sequence; calculating the difference between the predicted meteorological image predicted by each of the image prediction algorithms and the actual meteorological image; extracting the image prediction algorithm with the smallest difference as the image prediction algorithm corresponding to the meteorological image sequence, and storing the category of the meteorological image sequence and the matching image prediction algorithm in correspondence.

[0097] Specifically, the meteorological image sequence includes a plurality of meteorological images corresponding to a preset time period, and specifically may be a radar echo map corresponding to the evolution process of the weather. Furthermore, the meteorological images included in the meteorological image sequence may be images of various types, such as translation-type meteorological images, rotation-type meteorological images, etc. In order to obtain the prediction capabilities of different image prediction algorithms for different categories of meteorological images, it also includes using multiple types of image prediction algorithms to predict different categories of meteorological images respectively to obtain predicted images, and then calculating the difference between the predicted images and the real images, and judging the prediction accuracy of each image prediction algorithm for different types of meteorological images according to the size of the difference value.

[0098] In this embodiment, different types of image prediction algorithms are used to predict different categories of meteorological images to obtain predicted images, and the image prediction algorithm that matches each image category is determined based on the difference between the predicted image and the actual image, so as to match the corresponding algorithms for different image categories and achieve reasonable allocation of prediction algorithms.

[0099] In one of the embodiments, the method further includes: acquiring a simulated image; performing image processing on the simulated image to obtain a simulated image sequence with time-series changes; predicting the next time-series image of the simulated image sequence using multiple types of image prediction algorithms to obtain a predicted simulated image; acquiring a real simulated image; wherein the real simulated image is the next time-series image in the simulated image sequence; calculating the difference between the predicted simulated image predicted by each of the image prediction algorithms and the real simulated image; extracting the image prediction algorithm with the smallest difference as the image prediction algorithm corresponding to the simulated image, and storing the image category of the simulated image in correspondence with the matching image prediction algorithm.

[0100] Specifically, the simulated image is an image used to simulate the weather evolution process, and the simulated image is processed to obtain a simulated image sequence, wherein the simulated image sequence is an image sequence with time-series changes, and the simulated image sequence includes multiple images for simulating the evolution process of the target weather. Further, multiple simulated images with continuous time sequence in the simulated image sequence can be used to replace the radar echo map corresponding to the weather evolution process, so as to realize the simulation of the weather evolution process.

[0101] Specifically, the computer device performs image processing on the simulated image to obtain multiple simulated images with time-series changes derived from the simulated image based on the image processing, and the derived multiple simulated images constitute a simulated image sequence. For example, the simulated image is displayed, such as changing at least one display parameter of the image brightness, grayscale, and contrast, and the multiple simulated images obtained by changing the display parameters of the simulated image over time constitute a simulated image sequence, which can be used to simulate the evolution process of cloudy weather. The computer device obtains the simulated image sequence derived from the simulated image, and then uses a variety of different image prediction algorithms such as optical flow method to perform forecast processing on the simulated image sequence to obtain a forecast image. Specifically, the computer device uses the images in the image sequence as input samples, and inputs them into multiple optical flow prediction models respectively. Each optical flow prediction model is a machine learning model, which can correspond to an optical flow method. Each optical flow prediction model performs forecast processing (prediction processing) on ​​the input image and outputs a forecast image (prediction image).

[0102] In this embodiment, a variety of different image prediction algorithms such as optical flow methods include at least six different types of optical flow methods, including LK_ex and LK_sprog in the pysteps system and Dense, DenseRotation, Sparse and SparseSD in the rainymotion system.

[0103] Compare the predicted simulated image with the real simulated image, and determine the image prediction algorithm suitable for predicting the image type according to the comparison result. The real simulated image is an image in the simulated image sequence, specifically the next sequential simulated image corresponding to the last sequential simulated image to be predicted. For example, the obtained simulated image sequence is images 1, 2, 3, 4, 5, and the computer device predicts and processes the simulated images 2 and 3 in the simulated image sequence, and outputs the predicted image 4'. The last sequential image to be predicted is image 3, and the real simulated image here is the next sequential image of image 3, that is, image 4.

[0104] Specifically, the computer device compares the predicted simulated image obtained by each image prediction algorithm, such as each optical flow method, with the corresponding real simulated image to obtain multiple comparison results, each of which corresponds to an optical flow method. The computer device determines the predicted simulated image that is closest to the corresponding real simulated image among the multiple comparison results, and uses the optical flow method obtained by the predicted simulated image as the target optical flow method suitable for predicting the image type.

[0105] In this embodiment, the computer equipment uses a variety of different optical flow methods to perform forecast processing on a simulated image sequence that simulates the evolution process of the target weather to obtain a forecast image. The simulated image sequence that simulates the evolution process of the target weather has a smaller data volume than the radar echo image, thereby reducing the amount of calculation for subsequent comparison and determination of the target optical flow method, and improving the overall efficiency of determining the target optical flow method. The forecast simulated image and the real simulated image are then compared, and the target optical flow method suitable for forecasting the target weather is determined based on the comparison results. This is conducive to the targeted selection of an optical flow method suitable for the target weather to be forecasted for extrapolated forecasting in meteorological forecasting, thereby ultimately improving the accuracy of the forecast.

[0106] In one embodiment, the simulated image can also be processed by movement to simulate the evolution of more types of target weather, and the acquisition of the simulated image sequence includes: performing image processing on the simulated image to obtain the simulated image sequence with time-series changes; wherein the image processing includes at least one of translation, rotation and scaling.

[0107] like Figure 5 , which is a schematic diagram of a grayscale handwritten digital image in one embodiment. In this embodiment, an image in the MNIST dataset is used as a simulation image, and the MNIST dataset is a grayscale handwritten digital image with a size of 28*28 pixels.

[0108] Specifically, the computer device performs movement processing on the simulated image, such as at least one of translation, rotation and scaling, and multiple simulated images obtained by the movement processing of the simulated image over time constitute a simulated image sequence. Figure 6 As shown, Figure 6 is a schematic diagram of an image sequence in an embodiment, where a picture of a handwritten number "7" in the MNIST dataset is used as a simulated image and the simulated image sequence is obtained by performing translation over time.

[0109] Among them, the translation simulated image sequence obtained by translating the simulated image can be used to simulate the radar echo changes corresponding to stable layered cloud precipitation; the rotation simulated image sequence and the rotation + translation simulated image sequence obtained by rotating and rotating and translating the simulated image can be used to simulate the radar echo changes corresponding to the typhoon; the enlarged simulated image sequence and the enlarged + translation simulated image sequence obtained by enlarging and enlarging and translating the simulated image can be used to simulate the radar echo changes corresponding to the occurrence and development stage of convection; the reduced simulated image sequence and the reduced + translation simulated image sequence obtained by reducing and reducing and translating the simulated image can be used to simulate the radar echo changes corresponding to the occurrence and disappearance stage of convection.

[0110] In this embodiment, the computer device performs image processing including at least one of translation, rotation and scaling on the simulated image to simulate more types of target weather, so as to determine the applicable target optical flow method for more types of target weather through the above-mentioned target optical flow method determination method, thereby expanding the scope of application of the target optical flow method determination method and improving the applicability of the target optical flow method determination method.

[0111] In one embodiment, in order to further improve the accuracy of the optical flow method prediction, the simulated image sequence is predicted using a plurality of different optical flow methods to obtain a predicted image, including: predicting at least two simulated images that are time-sequentially consecutive in the simulated image sequence using the plurality of different optical flow methods to obtain the predicted image of the next time-series simulated image; predicting at least two images that are time-sequentially consecutive in the image sequence using a plurality of different optical flow methods to obtain a predicted image of the next time-series image.

[0112] Among them, different optical flow methods require different numbers of images for prediction processing, and the existing optical flow method requires at least two images for prediction processing.

[0113] Specifically, Figure 7 As shown, Figure 7The schematic diagram of forecast processing by different optical flow methods in one embodiment, the simulated image sequence used to simulate the target weather evolution process includes 40 images of time-series change, which are numbered 1, 2, 3, 4...39, 40 in the order of time-series change. The optical flow A method needs 2 images for forecast processing, and the time-series continuous images 1 and 2 are input for forecast processing to obtain the forecast image 3' of the next time-series image 3, and the time-series continuous images 2 and 3 are input for forecast processing to obtain the forecast image 4' of the next time-series image 4... and so on, and finally the forecast images 3'~40' of images 3~40 are obtained. Optical flow method B requires 20 images for prediction processing. Similarly, continuous time-series images 1 to 20 are input for prediction processing to obtain a prediction image 21' of the next time-series image 21. Continuous time-series images 2 to 21 are input for prediction processing to obtain a prediction image 22' of the next time-series image 22... and so on, and finally the predicted images 21' to 40' of images 21 to 40 are obtained.

[0114] Optionally, the computer device may select at least two images before time t0, such as images at time t0-1 and time t0-2, which are continuous with time t0, as input images for prediction processing using different optical flow methods, to obtain a prediction image at time t0, and similarly, a prediction image for images after time t0 may be obtained. Optionally, for the aforementioned six optical flow methods in this embodiment, each prediction processing requires at least 2 images and at most 20 images.

[0115] Accordingly, the predicted image obtained by each optical flow method is compared with the next temporal image in the image sequence.

[0116] Specifically, the computer device obtains the image similarity between the forecast image obtained by each optical flow method and the next time-series image, and compares the image similarity obtained by each optical flow method. For example, the computer device obtains the image similarity between the forecast image 3'~40' obtained by the above optical flow method A and the corresponding next time-series image 3~40, and the image similarity between the forecast image 21'~40' obtained by the above optical flow method B and the corresponding next time-series image 21~40, compares the image similarity obtained by the optical flow method A and the optical flow method B, and selects the optical flow method that obtains the highest image similarity as the target optical flow method suitable for forecasting the target weather. Among them, the image similarity can be represented by the pixel matching percentage between images. The greater the pixel matching percentage between images, the greater the image similarity between images.

[0117] In this embodiment, the computer device uses at least two images that are time-sequentially continuous in the image sequence as input images for forecasting processing by multiple different optical flow methods. The time-sequentially continuous images can accurately reflect the correlation of changes between images. Forecasting processing based on images with accurate change correlation is conducive to obtaining a predicted image that is closer to the real image, that is, the next time-series image, and improves the accuracy of optical flow method forecasting. Further, the image similarity obtained according to each optical flow method is used as the basis for determining the target optical flow method from multiple optical flow methods. The image similarity can intuitively reflect whether the adopted optical flow method is accurate in forecasting the target weather, and improves the certainty of determining the target optical flow method.

[0118] In one embodiment, the degree of image similarity can be characterized by a target critical success index and a target structural similarity value. Specifically, the degree of image similarity between each predicted image obtained by the optical flow method and the next time series image is obtained, including: calculating the target critical success index and the target structural similarity value between each predicted image obtained by the optical flow method and the next time series image.

[0119] Specifically, the computer device can calculate the critical success index and the structural similarity value between each of the multiple forecast images obtained by the optical flow method and the corresponding multiple next time series images through the critical success index formula (1) and the structural similarity value formula (2), and obtain multiple critical success indexes and multiple structural similarity values. The computer device further averages the multiple critical success indexes obtained to obtain the target critical success index, and averages the multiple structural similarity values ​​obtained to obtain the target structural similarity value.

[0120] In the two grayscale images, one is the actual image and the other is the forecast image. It is stipulated that if an echo appears on the grid point at a certain position in the actual situation and the forecast also appears, the point is judged to be successful; if an echo appears on the grid point in the actual situation and no echo appears in the forecast, the point is judged to be a failure; if an echo does not appear on the grid point in the actual situation and an echo appears in the forecast, the point is judged to be a false alarm. The critical success index (CSI) is the ratio of the number of successful grid points to the sum of the number of successful grid points, the number of failed grid points and the number of false alarm grid points.

[0121] CSI = (success) / (success + failure + false alarm) (1)

[0122] In this embodiment, the forecast image is a forecast image (predicted image) obtained by using the optical flow method, the actual image is the next time series image corresponding to the forecast image, the grid points are the areas pre-divided on the image, and the echoes are the pixels of the handwritten numbers.

[0123] Among them, structural similarity (SSIM) is an indicator to measure the similarity between two images (x and y) (the range of SSIM is 0 to 1), which consists of three parts: brightness contrast l(x, y), contrast contrast c(x, y) and structure contrast s(x, y).

[0124]

[0125]

[0126]

[0127]

[0128] Among them, C 3 =C 2 / 2Simplified to get formula (2). μ x is the average grayscale of image x, μ y is the average grayscale of image y, σ x 2 is the grayscale variance of image x, σ y 2 is the grayscale variance of image y, σ xy is the covariance of the grayscale of image x and y, C 1 =(k 1 L) 2 , C 2 =(k 2 L) 2 is a constant used to maintain stability, and L is the dynamic range of the grayscale value. 1 =0.01, k 2 =0.03.

[0129] The target critical success index and the target structure similarity value are weightedly summed to obtain the image similarity degree obtained by each optical flow method.

[0130] Specifically, the computer device sets a first weight w1 for the target critical success index and a second weight w2 for the target structural similarity value, w1+w2=1. The target critical success index can more accurately reflect the similarity between images, so w1>w2 is set, and then the target critical success index and the target structural similarity value are weightedly summed to obtain the image similarity between the predicted image obtained by the corresponding optical flow method and the corresponding next time series image, which is used as the final score of the corresponding optical flow method.

[0131] As shown in Table 2 below, in this embodiment, the following 6 optical flow methods are used to forecast the image sequence simulating the weather evolution process of a certain target, and the CSI and SSIM are obtained. The CSI and SSIM are weighted and summed to obtain the final score (SCORE).

[0132] Table 1 Scoring table of 6 optical flow methods

[0133] CSI SSIM SCORE Dense A1 B1 C1 DenseRotation A2 B2 C2 Sparse A3 B3 C3 SparseSD A4 B4 C4 lk_ex A5 B5 C5 Lk_sprog A6 B6 C6

[0134] In this embodiment, the computer device calculates the target critical success index and the target structure similarity value between the forecast image obtained by each optical flow method and the corresponding next time series image. The target critical success index can reflect the local similarity between the forecast image and the corresponding next time series image, and the target structure similarity value can reflect the overall similarity between the forecast image and the corresponding next time series image. The computer device then performs a weighted summation of the target critical success index and the target structure similarity value to take into account both the local similarity and the overall similarity, and obtains the image similarity as the final score of each optical flow method, thereby achieving a comprehensive score for each optical flow method, so as to improve the accuracy of determining the target optical flow method suitable for predicting the target weather among multiple optical flow methods based on the image similarity.

[0135] In a specific embodiment, Figure 8 As shown, a method for determining a target optical flow method is provided, comprising:

[0136] S810, performing image processing on the simulated image to obtain a simulated image sequence with time-series changes; wherein the image processing includes at least one of translation, rotation and scaling.

[0137] S820: Perform prediction processing on at least two simulated images that are sequentially continuous in the simulated image sequence using a plurality of different optical flow methods to obtain a prediction image of the next sequential image.

[0138] S830, calculating a critical success index and a structural similarity value between each of the multiple prediction images obtained by the optical flow method and the corresponding multiple next time series images.

[0139] S840. Average the critical success indexes to obtain a target critical success index.

[0140] S850, averaging the structural similarity values ​​to obtain a target structural similarity value.

[0141] S860, performing weighted summation on the target critical success index and the target structure similarity value to obtain the image similarity degree obtained by each optical flow method.

[0142] S870, comparing the image similarities obtained by each optical flow method, and selecting the optical flow method that obtains the highest image similarity as the target optical flow method.

[0143] In this embodiment, the computer device performs at least one of translation, rotation and scaling on the initial image to obtain a simulated image sequence that can be used to simulate the target weather evolution process, thereby reducing the amount of calculation of the entire target optical flow method and improving efficiency. The computer device then uses a variety of different optical flow methods to forecast at least two simulated images that are sequential in time in the simulated image sequence, and obtains the forecast image of the next sequential image. The image similarity between the forecast image obtained by each optical flow method and the next sequential image is calculated. The image similarity is characterized by the weighted sum of the target critical success index and the target structure similarity value, taking into account the local similarity and the overall similarity, and achieving a comprehensive score for each optical flow method. Finally, the computer device selects the optical flow method that obtains the highest image similarity as the target optical flow method. The image similarity can intuitively reflect whether the adopted optical flow method is accurate in forecasting the target weather, thereby improving the certainty of determining the target optical flow method.

[0144] It should be understood that although Figure 2-8 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2-8 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0145] In one embodiment, Fig. 9 As shown, a meteorological image prediction device is provided, comprising:

[0146] An image acquisition module 910 is used to acquire a meteorological image to be predicted;

[0147] A category identification module 920 is used to identify the image category corresponding to the meteorological image to be predicted;

[0148] An algorithm matching module 930 is used to match a corresponding type of image prediction algorithm according to the image category;

[0149] The prediction module 940 is used to predict the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image.

[0150] For the specific definition of the meteorological image prediction device, please refer to the definition of the meteorological image prediction method above, which will not be repeated here. Each module in the above-mentioned meteorological image prediction device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0151] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store meteorological image prediction data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a meteorological image prediction method is implemented.

[0152] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0153] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0154] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0155] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0156] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0157] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A meteorological image prediction method, characterized in that: The method comprises: Obtain the meteorological image to be predicted; Identify the image category corresponding to the meteorological image to be predicted; wherein the image category is divided according to the motion state of the meteorological image to be predicted, and the image category includes the meteorological image to be predicted of the translation category, the meteorological image to be predicted of the rotation category, the meteorological image to be predicted of the comprehensive translation and rotation category, the meteorological image to be predicted of the scale change category, and the meteorological image to be predicted of the comprehensive scale change and translation category; Matching a corresponding type of image prediction algorithm according to the image category; wherein the image prediction algorithm includes a target optical flow method; Predicting the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image; Before matching the image prediction algorithm of the corresponding type according to the image category, the method further includes: Acquire a simulated image; Performing image processing on the simulated image to obtain a simulated image sequence with time sequence changes; Using a plurality of different optical flow methods to perform prediction processing on at least two sequentially consecutive simulated image sequences in the simulated image sequence to obtain a prediction image of the next sequential image; Acquire a real simulated image; wherein the real simulated image is the next sequential image in the simulated image sequence; Calculate the critical success index and structural similarity value between the multiple forecast images obtained by each optical flow method and the corresponding multiple next time series images; wherein the structural similarity value is determined based on brightness contrast, contrast contrast and structural contrast; the critical success index is the ratio of the number of successful grid points to the total number of grid points, wherein the total number of grid points includes the number of successful grid points, the number of failed grid points and the number of false alarm grid points; average the critical success index to obtain the target critical success index; average the structural similarity value to obtain the target structural similarity value; perform weighted summation on the target critical success index and the target structural similarity value to obtain the degree of image similarity obtained by each optical flow method; The image similarities obtained by each optical flow method are compared, the optical flow method that obtains the highest image similarity is selected as the target optical flow method, and the image category of the simulated image and the matching target optical flow method are stored in correspondence.

2. The method according to claim 1, characterized in that The method further comprises: Acquire initial meteorological images; Identifying an initial image category of each of the initial meteorological images; Labeling each of the initial meteorological images according to the category of the initial image to obtain a training set; Using the training set to perform model training to obtain an image category recognition algorithm; The identifying the image category corresponding to the meteorological image to be predicted includes: According to the image category recognition algorithm, the category of each of the meteorological images to be predicted is identified to obtain the image category of the meteorological image to be predicted.

3. The method according to claim 2, characterized in that The initial meteorological image includes at least two initial area sub-images; the step of labeling each of the initial meteorological images according to the category of the initial image to obtain a training set includes: When each of the initial area sub-images of the initial meteorological image corresponds to a different image category, the initial area sub-images are respectively labeled according to the different image categories to obtain initial area labeling data corresponding to each of the initial area sub-images; Obtaining initial image annotation data corresponding to the initial meteorological image according to each of the initial region annotation data; A training set is obtained according to the initial image annotation data of a plurality of initial meteorological images.

4. The method according to claim 2, characterized in that: The initial meteorological image includes at least two initial region sub-images; the step of labeling each of the initial meteorological images according to the category of the initial image to obtain a training set includes: When the initial area sub-images of the initial meteorological image correspond to different image categories, obtaining a rainfall intensity index corresponding to each of the initial area sub-images; Extracting the image category corresponding to the initial area sub-image with the largest rainfall intensity index as the image category of the initial meteorological image; Performing category labeling on the initial meteorological image according to the extracted image category to obtain initial image labeling data of the initial meteorological image; A labeling set is obtained according to initial image labeling data of a plurality of initial meteorological images.

5. The method according to claim 1 or 3, characterized in that: The meteorological image to be predicted includes at least two sub-images of the area to be predicted; the image prediction algorithm of the corresponding type is matched according to the image category, including: When the sub-images of the to-be-predicted area of ​​the to-be-predicted meteorological image correspond to different prediction image categories, matching corresponding types of image prediction algorithms according to the prediction image categories; The step of predicting the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image includes: Predicting each of the to-be-predicted region sub-images according to the matched image prediction algorithms to obtain a region prediction sub-image; A predicted image corresponding to the meteorological image to be predicted is obtained according to each of the regional prediction sub-images.

6. The method according to claim 1, characterized in that The method further comprises: Acquire meteorological image sequences; Using multiple types of image prediction algorithms to predict the next time series meteorological image of the meteorological image sequence to obtain a predicted meteorological image; Acquire the next real time-series meteorological image in the meteorological image sequence; Calculating the difference between the predicted meteorological image predicted by each of the image prediction algorithms and the real meteorological image; The image prediction algorithm with the smallest difference is extracted as the image prediction algorithm corresponding to the meteorological image sequence, and the category of the meteorological image sequence and the matching image prediction algorithm are stored in correspondence.

7. The method according to claim 1, characterized in that The image processing includes at least one of translation, rotation and scaling; a translation simulation image sequence obtained by translating the simulation image is used to simulate the radar echo changes corresponding to stable layered cloud precipitation; a rotation simulation image sequence and a rotation plus translation simulation image sequence obtained by rotating and rotating and translating the simulation image are used to simulate the radar echo changes corresponding to the typhoon; an enlarged simulation image sequence and an enlarged translation simulation image sequence obtained by magnifying and magnifying and translating the simulation image are used to simulate the radar echo changes corresponding to the occurrence and development stage of convection; a reduced simulation image sequence and a reduced simulation image sequence obtained by reducing and reducing and translating the simulation image are used to simulate the radar echo changes corresponding to the occurrence and disappearance stage of convection.

8. A meteorological image prediction device, characterized in that: The device comprises: An image acquisition module is used to acquire the meteorological image to be predicted; A category recognition module, used to identify the image category corresponding to the meteorological image to be predicted; wherein the image category is divided according to the motion state of the meteorological image to be predicted, and the image category includes the meteorological image to be predicted of the translation category, the meteorological image to be predicted of the rotation category, the meteorological image to be predicted of the comprehensive translation and rotation category, the meteorological image to be predicted of the scale change category, and the meteorological image to be predicted of the comprehensive scale change and translation category; An algorithm matching module, used for matching a corresponding type of image prediction algorithm according to the image category; wherein the image prediction algorithm includes a target optical flow method; A prediction module, used for predicting the meteorological image to be predicted according to the matched image prediction algorithm to obtain a predicted image; The device is also used for: Acquire a simulated image; Performing image processing on the simulated image to obtain a simulated image sequence with time sequence changes; Using a plurality of different optical flow methods to perform prediction processing on at least two sequentially consecutive simulated image sequences in the simulated image sequence to obtain a prediction image of the next sequential image; Acquire a real simulated image; wherein the real simulated image is the next sequential image in the simulated image sequence; Calculate the critical success index and structural similarity value between the multiple forecast images obtained by each optical flow method and the corresponding multiple next time series images; wherein the structural similarity value is determined based on brightness contrast, contrast contrast and structural contrast; the critical success index is the ratio of the number of successful grid points to the total number of grid points, wherein the total number of grid points includes the number of successful grid points, the number of failed grid points and the number of false alarm grid points; average the critical success index to obtain the target critical success index; average the structural similarity value to obtain the target structural similarity value; perform weighted summation on the target critical success index and the target structural similarity value to obtain the degree of image similarity obtained by each optical flow method; The image similarities obtained by each optical flow method are compared, the optical flow method that obtains the highest image similarity is selected as the target optical flow method, and the image category of the simulated image and the matching target optical flow method are stored in correspondence.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images