Hail recognition method and device based on radar satellite multi-modal feature fusion

Abstract

The invention discloses a hail identification method and device based on radar satellite multi-modal feature fusion. The method comprises the following steps: acquiring meteorological multi-modal observation data; inputting the multi-modal meteorological observation data into a feature extraction network, and extracting spatial-temporal features of corresponding modals; performing fusion processing on the spatial-temporal features of the corresponding modals to generate fusion feature representation; according to the fusion feature representation, determining the hail particle size, the reflectivity intensity and the duration of the target area; determining a hail grade category of the target area according to a preset hail grade division standard; and outputting a hail grade identification result corresponding to the target area according to the hail grade category. According to the method, the defects of insufficient single modal information and limitation of a static fusion method in the prior art are overcome, the comprehensiveness and accuracy of hail identification are improved, and the prediction reliability of hail particle size, reflectivity strength and duration is improved.

Description

Technical Field

[0001] This invention relates to the field of meteorological disaster monitoring technology, and in particular to a hail identification method and device based on radar satellite multimodal feature fusion. Background Technology

[0002] With advancements in meteorological observation technology, radar and satellite data are increasingly being used in hail monitoring. Radar, through high-resolution scanning, provides spatial dynamic information such as precipitation intensity and velocity fields, making it suitable for capturing the spatial distribution and movement characteristics of hail. Satellites, through multispectral imaging, acquire information such as cloud top brightness temperature and cloud structure, reflecting the macroscopic environment in which hail forms. Currently, hail identification technologies mainly include threshold analysis based on radar echoes, morphological classification based on satellite cloud images, and machine learning-driven parameter prediction models. These methods, by analyzing single or partial modal data, have initially achieved hail detection and classification.

[0003] Existing hail identification technologies have some shortcomings: First, traditional fusion methods often use simple weighting or splicing, lacking a dynamic adjustment mechanism, making it difficult to adapt to the temporal evolution and regional differences of hail events; second, existing technologies often ignore the spatiotemporal correlation between parameters when predicting hail particle size, reflectivity intensity, and duration, affecting prediction accuracy. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the problems existing in the current hail identification method and device based on radar satellite multimodal feature fusion, the present invention is proposed.

[0006] Therefore, the purpose of this invention is to provide a hail identification method and device based on radar satellite multimodal feature fusion, which is suitable for solving the problem that existing technologies are difficult to adapt to the temporal evolution and regional differences of hail events, thus affecting prediction accuracy.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a hail identification method based on radar satellite multimodal feature fusion, comprising: Acquire multimodal meteorological observation data; The multimodal meteorological observation data is input into a feature extraction network to extract the spatiotemporal features of the corresponding modes; The spatiotemporal features of the corresponding modes are fused to generate a fused feature representation; Based on the fused feature representation, the hail particle size, reflectivity intensity, and duration of hail in the target area are determined; The hail level category of the target area is determined according to the preset hail level classification standard; Based on the hail level category, output the hail level identification result corresponding to the target area.

[0008] As a preferred embodiment of the hail identification method based on radar and satellite multimodal feature fusion described in this invention, the feature extraction network includes a radar data processing channel and a satellite data processing channel, wherein the radar data processing channel is used to extract spatial dynamic features, and the satellite data processing channel is used to extract brightness temperature and cloud structure features.

[0009] As a preferred embodiment of the hail identification method based on radar satellite multimodal feature fusion described in this invention, the step of fusing the spatiotemporal features of corresponding modalities to generate a fused feature representation includes: Based on the recognition contribution of the spatiotemporal features of the corresponding modality in historical hail identification samples, a fusion weight factor is assigned to the modality features; Based on the fusion weight factor, the modal features are weighted and superimposed to obtain a fusion feature representation; wherein, the calculation of the fusion weight factor is dynamically completed based on an attention mechanism.

[0010] As a preferred embodiment of the hail identification method based on radar satellite multimodal feature fusion described in this invention, wherein: determining the hail particle size, reflectivity intensity, and duration of the target area based on the fused feature representation includes: The fused feature representation is input into a pre-trained hail parameter prediction model to obtain the predicted hail particle size, predicted reflectivity intensity, and predicted duration of the target area. The specific formula for the hail parameter prediction model is as follows: ; ; ; in, Z represents the predicted hail particle size for the target area, Z represents the predicted radar echo intensity for the target area, and t represents the hail duration. , and These represent features related to particle size, reflectivity, and duration extracted from radar and satellite data fusion, generated by normalizing multimodal data (such as radar velocity fields or satellite brightness temperatures). , and These are all weighting coefficients used to adjust the corresponding features to the physical units of the corresponding parameters, obtained through training with historical hailstone data. , and These are all correlation coefficients, used to capture the spatiotemporal correlation with hail parameters, and are also based on historical training data. , and These are all attention coefficients, representing the importance of features. They are calculated based on the attention mechanism and reflect the contribution of radar and satellite data. , and This is a normalization factor used to prevent the output from being too large. , and Both are exponential functions, used to ensure that the output bits are positive; Spatiotemporal correlation analysis was performed on the predicted hail particle size, predicted reflectivity intensity, and predicted duration to determine the correlation weights between the predicted values. Based on the associated weights, the predicted hail size, predicted reflectivity intensity, and predicted duration are weighted and adjusted to generate the final determination results of hail size, reflectivity intensity, and duration in the target area.

[0011] As a preferred embodiment of the hail identification method based on radar satellite multimodal feature fusion described in this invention, the hail level classification criteria are: no hail, light hail, moderate hail, and heavy hail.

[0012] As a preferred embodiment of the hail identification method based on radar satellite multimodal feature fusion described in this invention, wherein: determining the hail level category of the target area according to a preset hail level classification standard includes: The final determination results of hail particle size, reflectivity intensity and duration in the target area are obtained, and a basic level judgment is made based on the hail particle size, reflectivity intensity and duration. If the hailstone size is less than or equal to the first size threshold D1, the reflectivity intensity is less than or equal to the first intensity threshold Z1, and the duration is less than or equal to the first time threshold T1, then the current state is determined to be no hail, no hail warning is issued, and the target area is continuously monitored. If the hailstone size is greater than the first size threshold D1 but not greater than the second size threshold D2, the reflectivity intensity is greater than the first intensity threshold Z1 but not greater than the second intensity threshold Z2, and the duration is greater than the first time threshold T1 but not greater than the second time threshold T2, then the current state is initially determined to be light hail, and the next step is executed. When the current state is initially determined to be light hail, the level is upgraded based on the particle size change rate, reflectivity intensity deviation and duration weighted value. If the particle size change rate is less than or equal to the preset rate threshold V, the reflectivity intensity deviation is less than or equal to the preset deviation threshold Z' and the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be light hail and a level one warning information is sent to the meteorological warning system. If the rate of change of particle size is greater than the preset rate threshold V, the reflectivity intensity deviation is greater than the preset deviation threshold Z', or the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be moderate hail, a level II warning message is sent to the meteorological warning system and the hail parameters are recorded. If the hailstone size is greater than the second size threshold D2 but not greater than the third size threshold D3, the reflectivity intensity is greater than the second intensity threshold Z2 but not greater than the third intensity threshold Z3, and the duration is greater than the second time threshold T2 but not greater than the third time threshold T3, then the current state is preliminarily determined to be moderate hail, and the next step is executed. When the current state is initially determined to be moderate hail, the level is upgraded based on the particle size change rate, reflectivity intensity deviation and duration weighted value. If the particle size change rate is less than or equal to the preset rate threshold V, the reflectivity intensity deviation is less than or equal to the preset deviation threshold Z' and the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be moderate hail, and a level II warning information is sent to the meteorological warning system and the hail parameters are recorded. If the rate of change of particle size is greater than the preset rate threshold V, the reflectivity intensity deviation is greater than the preset deviation threshold Z', or the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be severe hail, a level 3 warning message is sent to the meteorological warning system and the hail parameters are recorded. If the hailstone size is greater than the third size threshold D3, the reflectivity intensity is greater than the third intensity threshold Z3, or the duration is greater than the third time threshold T3, then the current state is determined to be severe hail, and a level 3 warning message is sent to the meteorological warning system and the hail parameters are recorded.

[0013] As a preferred embodiment of the hail identification method based on radar and satellite multimodal feature fusion described in this invention, wherein: the radar data processing channel and satellite data processing channel of the feature extraction network extract spatiotemporal features, including: Acquire radar observation data input from the radar data processing channel and satellite observation data input from the satellite data processing channel; The radar observation data is preprocessed to extract spatial dynamic features, including generating a spatial velocity field and echo intensity distribution through convolution operations; The satellite observation data is preprocessed to extract brightness temperature and cloud structure features, including generating brightness temperature distribution and cloud top height maps through spectral analysis; Time series analysis is performed on the aforementioned spatial dynamic characteristics, brightness temperature, and cloud structure characteristics to generate spatiotemporal feature vectors for radar modes and satellite modes, respectively.

[0014] Secondly, to further address the problem that existing technologies are ill-suited to the temporal evolution and regional differences of hail events, thus affecting prediction accuracy, this invention provides a hail identification device based on radar satellite multimodal feature fusion, comprising: Data acquisition module: used to acquire meteorological multimodal observation data; Feature extraction module: used to input the multimodal meteorological observation data into the feature extraction network and extract the spatiotemporal features of the corresponding modes; Feature fusion module: used to fuse the spatiotemporal features of the corresponding modality to generate a fused feature representation; Parameter determination module: used to determine the hail particle size, reflectivity intensity, and duration of hail in the target area based on the fused feature representation; Hail level determination module: used to determine the hail level category of the target area according to the preset hail level classification standard; Result output module: used to output the hail level identification result corresponding to the target area according to the hail level category.

[0015] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the hail identification method based on radar satellite multimodal feature fusion as described in the first aspect of the present invention.

[0016] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the hail identification method based on radar satellite multimodal feature fusion as described in the first aspect of the present invention.

[0017] The beneficial effects of this invention are as follows: By extracting spatial dynamic features, brightness temperature, and cloud structure features through radar and satellite data processing channels respectively, and using an attention mechanism to dynamically allocate fusion weights, it overcomes the limitations of insufficient single-modal information and static fusion methods in existing technologies, thereby improving the comprehensiveness and accuracy of hail identification. By using a pre-trained hail parameter prediction model and spatiotemporal correlation analysis, and adjusting the predicted values ​​through correlation weights, it solves the problem of ignoring the spatiotemporal correlation between parameters in existing technologies, and improves the reliability of predicting hail particle size, reflectivity intensity, and duration. Especially in scenarios where hail events change rapidly or have large regional differences, it achieves fine classification from no hail to severe hail through multi-level thresholds and dynamic upgrade judgments, improving the adaptability and sensitivity of hail level determination, reducing misjudgments and omissions, and triggering early warnings, especially in the early or transitional stages of hail events. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the overall process of a hail identification method based on radar satellite multimodal feature fusion proposed in this invention; Figure 2 This is a schematic diagram of the hail level classification judgment framework of a hail identification method based on radar satellite multimodal feature fusion proposed in this invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0023] Example 1 Reference Figures 1-2 As an embodiment of the present invention, a hail identification method based on radar satellite multimodal feature fusion is provided.

[0024] Existing hail identification technologies have some shortcomings: First, traditional fusion methods often use simple weighting or splicing, lacking a dynamic adjustment mechanism, making it difficult to adapt to the temporal evolution and regional differences of hail events; second, existing technologies often ignore the spatiotemporal correlation between parameters when predicting hail particle size, reflectivity intensity, and duration, affecting prediction accuracy.

[0025] This application provides a method and apparatus for hail identification based on radar satellite multimodal feature fusion that can effectively solve the problems mentioned above. The following will describe in detail how to implement this method and apparatus for hail identification based on radar satellite multimodal feature fusion with multiple embodiments.

[0026] Figure 1 A flowchart illustrating the overall process of a hail identification method based on radar satellite multimodal feature fusion is shown, including: S1: Acquire meteorological multimodal observation data.

[0027] Specifically, this invention primarily employs two technologies to acquire multimodal meteorological data: radar acquisition and satellite acquisition. Radar acquisition utilizes ground-based or airborne Doppler weather radar systems to obtain precipitation and motion information of the target area through electromagnetic wave transmission and reception. Satellite acquisition utilizes visible light, infrared, and microwave imagers from meteorological satellites to acquire information such as cloud top brightness temperature and cloud structure through multispectral scanning. Timestamp alignment and geographic coordinate projection techniques (such as the WGS84 coordinate system) are used to ensure the spatiotemporal consistency of radar and satellite data. Data cleaning algorithms (such as denoising and interpolation) are used to handle missing or outlier values ​​to guarantee data quality.

[0028] Radar acquisition technology can acquire radar reflectivity, radial velocity, spectral width, differential reflectivity, and correlation coefficient; satellite acquisition technology can acquire cloud top brightness temperature, visible light reflectivity, water vapor channel brightness temperature, cloud top height, and cloud phase.

[0029] S2: Input multimodal meteorological observation data into the feature extraction network to extract the spatiotemporal features of the corresponding modes.

[0030] Preferably, the feature extraction network includes a radar data processing channel and a satellite data processing channel, wherein the radar data processing channel is used to extract spatial dynamic features, and the satellite data processing channel is used to extract brightness temperature and cloud structure features.

[0031] Furthermore, spatial dynamic features refer to the motion and intensity distribution characteristics of hail extracted from radar data processing channels to characterize the three-dimensional space and its temporal evolution. These features mainly reflect the convective motion (such as updrafts and rotation), spatial distribution and intensity changes of precipitation ions in the hail area. They can directly support the prediction of hail size and duration, provide dynamic information for feature fusion, and enhance the spatiotemporal consistency of hail identification when combined with brightness temperature and cloud structure features. Brightness temperature and cloud structure features are spatiotemporal characteristics extracted from satellite data processing channels to characterize the thermodynamic properties (temperature distribution) and physical structure (cloud morphology, height, and phase) of cloud systems. They reflect the atmospheric environment in which hail forms, provide thermodynamic and structural information, assist in the prediction of hail size and reflectivity intensity, and complement spatial dynamic features in feature fusion, improving the macroscopic environmental analysis capabilities of hail identification.

[0032] The feature extraction network extracts spatiotemporal features through its radar data processing and satellite data processing channels, including: Acquire radar observation data input from the radar data processing channel and satellite observation data input from the satellite data processing channel; Preprocessing of radar observation data to extract spatial dynamic features, including generating spatial velocity fields and echo intensity distributions through convolution operations; Preprocessing of satellite observation data to extract brightness temperature and cloud structure features includes generating brightness temperature distribution and cloud top height maps through spectral analysis; Time series analysis was performed on the spatial dynamic characteristics, brightness temperature, and cloud structure characteristics to generate spatiotemporal feature vectors for radar modes and satellite modes, respectively.

[0033] It should be noted that the feature extraction network adopts a deep learning framework, based on a combination of convolutional neural networks and recurrent neural networks, to process the spatiotemporal characteristics of multimodal meteorological observation data. It includes radar data processing channels and network data processing channels; the overall network structure is as follows: Input layer: Receives synchronized multimodal data. Radar data is a three-dimensional tensor (height × width × time), and satellite data is a multi-channel two-dimensional tensor (height × width × spectral channels). Feature extraction layer: Each channel extracts spatial features through convolution operations and temporal features through time series analysis; Feature integration layer: concatenates spatial and temporal features into a spatiotemporal feature vector; Output layer: Outputs spatiotemporal feature vectors of radar mode and satellite mode for subsequent feature fusion.

[0034] S3: The spatiotemporal features of the corresponding modality are fused to generate a fused feature representation.

[0035] Furthermore, the spatiotemporal features of the corresponding modalities are fused to generate a fused feature representation, including: Based on the recognition contribution of the spatiotemporal features of the corresponding modality in historical hail identification samples, a fusion weight factor is assigned to the modality features; Based on the fusion weight factor, the modal features are weighted and superimposed to obtain the fusion feature representation; the calculation of the fusion weight factor is dynamically completed based on the attention mechanism.

[0036] It should be noted that the goal of feature fusion processing is to integrate the spatial dynamic features extracted by the radar data processing channel and the brightness temperature and cloud structure features extracted by the satellite data processing channel to generate a unified fused feature representation for subsequent prediction of hail particle size, reflectivity intensity and duration. The fusion process is based on the contribution of historical hail identification samples, dynamically allocates weights through an attention mechanism, and generates a fused feature vector by weighted superposition.

[0037] The fusion process is as follows: Input: The spatiotemporal feature vector of the radar mode and the brightness temperature and cloud structure features extracted by the satellite data processing channel are generated by the radar channel and satellite channel of the feature extraction network, respectively; Weighting: Based on the recognition contribution of each modality feature in historical hail identification samples, a fusion weighting factor is calculated, and the weights are dynamically adjusted using an attention mechanism; Weighted superposition and output: Based on the weighting factor, the radar and satellite feature vectors are weighted and fused to generate a fused feature representation, which is used as the input for the subsequent hail parameter prediction model.

[0038] This invention employs a channel-based attention mechanism to assign fusion weight factors to the spatiotemporal feature vectors of radar and satellite modalities. Weight calculation references the contribution of historical hail identification samples, ensuring a dynamic balance between radar features (e.g., strong velocity fields) and satellite features (e.g., low brightness temperature) under different hail scenarios. This invention provides prior knowledge for weight calculation by analyzing the prediction accuracy of each modal feature in historical hail events (e.g., radar reflectivity's contribution to particle size prediction > 0.7), improving the targeting of the fusion. Combined with real-time calculation via the attention mechanism, the weight factors dynamically change with the spatiotemporal characteristics of the input data (e.g., storm intensity, cloud top height), overcoming the limitations of fixed weights in existing technologies.

[0039] S4: Based on the fusion feature representation, determine the hail particle size, reflectivity intensity, and duration of hail in the target area.

[0040] Based on the fused feature representation, the hail particle size, reflectivity intensity, and duration of hail in the target area are determined, including: The fused feature representation is input into a pre-trained hail parameter prediction model to obtain the predicted hail size, reflectivity intensity, and duration of hail in the target area. The specific formula for the hail parameter prediction model is as follows: ; ; ; in, Z represents the predicted hail particle size for the target area, Z represents the predicted radar echo intensity for the target area, and t represents the hail duration. , and These represent features related to particle size, reflectivity, and duration extracted from radar and satellite data fusion, generated by normalizing multimodal data (such as radar velocity fields or satellite brightness temperatures). , and These are all weighting coefficients used to adjust the corresponding features to the physical units of the corresponding parameters, obtained through training with historical hailstone data. , and These are all correlation coefficients, used to capture the spatiotemporal correlation with hail parameters, and are also based on historical training data. , and These are all attention coefficients, representing the importance of features. They are calculated based on the attention mechanism and reflect the contribution of radar and satellite data. , and This is a normalization factor used to prevent the output from being too large. , and Both are exponential functions, used to ensure that the output bits are positive; In the embodiments of this application, , and Normalization factors can be dynamically adjusted to prevent excessively large output values ​​by minimizing prediction errors using historical data, or they can be obtained through data training or optimization algorithms.

[0041] Spatiotemporal correlation analysis was performed on the predicted values ​​of hail particle size, reflectivity intensity, and duration to determine the correlation weights among the predicted values. Based on the correlation weights, the predicted values ​​of hail particle size, reflectivity intensity, and duration are weighted and adjusted to generate the final determination results of hail particle size, reflectivity intensity, and duration in the target area.

[0042] The final determination process for hailstone size, reflectivity intensity, and duration is as follows: Based on historical hail event data, the spatiotemporal correlation between hail particle size, reflectivity intensity and duration is calculated using Pearson correlation coefficient or mutual information method. The correlation weights are dynamically allocated through attention mechanism or restricted weighting model, and the weight coefficients satisfy d+Z+t=100% to ensure that the contribution of each parameter is comprehensively normalized. The adjusted final result is generated by multiplying the association weights by the predicted values, as shown in the following formula: ; ; ; in, This represents the final value for the hailstone diameter. Here, t represents the final value of reflectivity intensity, and t represents the final value of hail duration. , and The main weights for hailstone size, reflectivity intensity, and duration are respectively. , and The cross-influence weights of each parameter.

[0043] In the embodiments of this application, , and Equal weights can be determined by analyzing historical hail event data and using statistical methods (such as principal component analysis or variance decomposition) to calculate the contribution of each parameter to hail severity assessment, thereby allocating initial sovereign weights. Alternatively, machine learning models (such as neural networks) can automatically learn the importance proportions of each parameter during training and constrain their sum to be 1. , and The cross-influence weights are determined based on the spatiotemporal correlation between parameters in historical data (such as Pearson correlation coefficient or mutual information), quantifying the degree of synergistic influence of reflectivity intensity (Z) and duration (t) on particle size (d).

[0044] S5: Determine the hail level category of the target area according to the preset hail level classification standard.

[0045] The hail severity classification standards are no hail, light hail, moderate hail, and heavy hail.

[0046] Based on the preset hail level classification standards, the hail level category of the target area is determined, including: The final determination results of hail particle size, reflectivity intensity, and duration in the target area are obtained, and a basic level judgment is made based on hail particle size, reflectivity intensity, and duration. If the hailstone size is less than or equal to the first size threshold D1, the reflectivity intensity is less than or equal to the first intensity threshold Z1, and the duration is less than or equal to the first time threshold T1, then the current state is determined to be no hail, no hail warning is issued, and the target area is continuously monitored. If the hailstone size is greater than the first size threshold D1 but not greater than the second size threshold D2, the reflectivity intensity is greater than the first intensity threshold Z1 but not greater than the second intensity threshold Z2, and the duration is greater than the first time threshold T1 but not greater than the second time threshold T2, then the current state is initially determined to be light hail, and the next step is executed. When the current state is initially determined to be light hail, the level is upgraded based on the particle size change rate, reflectivity intensity deviation and duration weighted value. If the particle size change rate is less than or equal to the preset rate threshold V, the reflectivity intensity deviation is less than or equal to the preset deviation threshold Z' and the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be light hail and a level one warning information is sent to the meteorological warning system. If the rate of change of particle size is greater than the preset rate threshold V, the reflectivity intensity deviation is greater than the preset deviation threshold Z', or the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be moderate hail, a level II warning message is sent to the meteorological warning system and the hail parameters are recorded. If the hailstone size is greater than the second size threshold D2 but not greater than the third size threshold D3, the reflectivity intensity is greater than the second intensity threshold Z2 but not greater than the third intensity threshold Z3, and the duration is greater than the second time threshold T2 but not greater than the third time threshold T3, then the current state is preliminarily determined to be moderate hail, and the next step is executed. When the current state is initially determined to be moderate hail, the level is upgraded based on the particle size change rate, reflectivity intensity deviation and duration weighted value. If the particle size change rate is less than or equal to the preset rate threshold V, the reflectivity intensity deviation is less than or equal to the preset deviation threshold Z' and the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be moderate hail, and a level II warning information is sent to the meteorological warning system and the hail parameters are recorded. If the rate of change of particle size is greater than the preset rate threshold V, the reflectivity intensity deviation is greater than the preset deviation threshold Z', or the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be severe hail, a level 3 warning message is sent to the meteorological warning system and the hail parameters are recorded. If the hailstone size is greater than the third size threshold D3, the reflectivity intensity is greater than the third intensity threshold Z3, or the duration is greater than the third time threshold T3, then the current state is determined to be severe hail, and a level 3 warning message is sent to the meteorological warning system and the hail parameters are recorded.

[0047] For example, suppose that during a severe convective weather event, the meteorological monitoring system observes via radar that the reflectivity intensity of a target area exceeds the first intensity threshold Z1 but does not reach the second intensity threshold Z2; the hail particle size exceeds the first particle size threshold but does not reach the second particle size threshold D2; and the duration exceeds the first time threshold T1 but does not reach the second time threshold T2. The system initially determines that it is a light hail situation and initiates a Level 1 warning. Subsequently, the system continues to monitor and finds that the rate of change of reflectivity intensity in the area exceeds a preset rate threshold, and the rate of increase of hail particle size also exceeds a preset rate threshold. Based on these dynamic change characteristics, the system automatically upgrades the warning level to Level 2 (moderate hail) and sends a warning message containing specific parameter changes to the meteorological warning center. After receiving the information, the warning center immediately retrieves satellite cloud image data for verification and finds that the cloud top brightness temperature in the area is lower than the third intensity threshold Z3, and the vertical development speed of the cloud body is accelerating. Combining these multimodal data, the system finally determines that the hail situation in the area will develop into severe hail and issues a Level 3 warning 15 minutes in advance. The local disaster prevention department promptly initiates an emergency response based on the warning information, effectively reducing hail disaster losses.

[0048] In this embodiment, D1, D2, and D3 are determined by analyzing the damage to crops caused by hail of different sizes based on historical hail disaster data statistics, combined with the radar detection error range; Z1, Z2, and Z3 establish a reflectivity intensity-hail probability distribution model by analyzing the correspondence between 10 years of radar observation data and ground-measured hail, and select the inflection point of probability change as the grading threshold; T1, T2, and T3 are set according to the life cycle characteristics of severe convective weather and the movement speed of hail clouds. The preset rate threshold V is based on the hail growth physical model (such as the wet growth / dry growth pattern conversion threshold), and the characteristic rate threshold of different growth stages is determined by numerical simulation. The preset deviation threshold Z' is determined by calculating the standard deviation of reflectivity intensity in historical data. The preset weighted threshold Q can be set by experts based on meteorological knowledge, setting the contribution weight of duration to hail level.

[0049] It should be noted that traditional hail warning methods typically rely on a single radar reflectivity threshold (such as 45 dBZ), making it difficult to distinguish between short-duration heavy precipitation and actual hail events. This embodiment overcomes the limitations of a single threshold by establishing a multi-parameter, multi-cascade intelligent judgment system. It constructs a three-dimensional discrimination method that includes particle size, intensity, and duration, introduces a dynamic upgrade mechanism, and achieves adaptive adjustment of the warning level through rate of change monitoring. It also integrates multi-modal data such as satellite cloud images to improve the ability to capture the microscopic physical processes of hail clouds, and establishes a closed-loop system of "monitoring-warning-verification" to optimize threshold parameters through real-time feedback.

[0050] In summary, by extracting spatial dynamic features and brightness temperature and cloud structure features through radar and satellite data processing channels respectively, and using an attention mechanism to dynamically allocate fusion weights, the limitations of insufficient single-modal information and static fusion methods in existing technologies are overcome, improving the comprehensiveness and accuracy of hail identification. By utilizing a pre-trained hail parameter prediction model and spatiotemporal correlation analysis, and adjusting the predicted values ​​through correlation weights, the problem of ignoring the spatiotemporal correlation between parameters in existing technologies is solved, improving the predictive reliability of hail particle size, reflectivity intensity, and duration. Especially in scenarios where hail events change rapidly or have large regional differences, multi-level thresholds and dynamic upgrade judgments are used to achieve fine classification from no hail to severe hail, improving the adaptability and sensitivity of hail level determination, reducing false positives and false negatives, and triggering early warnings, especially in the early or transitional stages of hail events.

[0051] Example 2 is an embodiment of the present invention. This embodiment provides a hail identification device based on radar satellite multimodal feature fusion, comprising: Data acquisition module: used to acquire meteorological multimodal observation data; Feature extraction module: Used to input multimodal meteorological observation data into the feature extraction network and extract the spatiotemporal features of the corresponding modes; Feature fusion module: used to fuse the spatiotemporal features of the corresponding modality to generate a fused feature representation; Parameter determination module: used to determine the hail particle size, reflectivity intensity, and duration of hail in the target area based on the fused feature representation; Hail level determination module: used to determine the hail level category of the target area according to the preset hail level classification standard; The results output module is used to output the hail level identification results corresponding to the target area based on the hail level category.

[0052] Example 3 is an embodiment of the present invention, which differs from the previous embodiment in that: If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0054] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0055] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0056] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A hail identification method based on radar satellite multimodal feature fusion, characterized in that, include: Acquire multimodal meteorological observation data; The multimodal meteorological observation data is input into a feature extraction network to extract the spatiotemporal features of the corresponding modes; The spatiotemporal features of the corresponding modes are fused to generate a fused feature representation; Based on the fused feature representation, the hail particle size, reflectivity intensity, and duration of hail in the target area are determined; The hail level category of the target area is determined according to the preset hail level classification standard; Based on the hail level category, output the hail level identification result corresponding to the target area.

2. The hail identification method based on radar satellite multimodal feature fusion according to claim 1, characterized in that: The feature extraction network includes a radar data processing channel and a satellite data processing channel. The radar data processing channel is used to extract spatial dynamic features, and the satellite data processing channel is used to extract brightness temperature and cloud structure features.

3. The hail identification method based on radar satellite multimodal feature fusion according to claim 1, characterized in that: The step of fusing the spatiotemporal features of the corresponding modality to generate a fused feature representation includes: Based on the recognition contribution of the spatiotemporal features of the corresponding modality in historical hail identification samples, a fusion weight factor is assigned to the modality features; Based on the fusion weight factor, the modal features are weighted and superimposed to obtain a fusion feature representation; wherein, the calculation of the fusion weight factor is dynamically completed based on an attention mechanism.

4. The hail identification method based on radar satellite multimodal feature fusion according to claim 1, characterized in that: The step of determining the hail particle size, reflectivity intensity, and duration of the target area based on the fused feature representation includes: The fused feature representation is input into a pre-trained hail parameter prediction model to obtain the predicted hail particle size, predicted reflectivity intensity, and predicted duration of the target area. Spatiotemporal correlation analysis was performed on the predicted hail particle size, predicted reflectivity intensity, and predicted duration to determine the correlation weights between the predicted values. Based on the associated weights, the predicted hail size, predicted reflectivity intensity, and predicted duration are weighted and adjusted to generate the final determination results of hail size, reflectivity intensity, and duration in the target area.

5. The hail identification method based on radar satellite multimodal feature fusion according to claim 4, characterized in that: The hail severity classification criteria are no hail, light hail, moderate hail, and heavy hail.

6. The hail identification method based on radar satellite multimodal feature fusion according to claim 5, characterized in that: The step of determining the hail level category of the target area according to a preset hail level classification standard includes: The final determination results of hail particle size, reflectivity intensity and duration in the target area are obtained, and a basic level judgment is made based on the hail particle size, reflectivity intensity and duration. If the hailstone size is less than or equal to the first size threshold D1, the reflectivity intensity is less than or equal to the first intensity threshold Z1, and the duration is less than or equal to the first time threshold T1, then the current state is determined to be no hail, no hail warning is issued, and the target area is continuously monitored. If the hailstone size is greater than the first size threshold D1 but not greater than the second size threshold D2, the reflectivity intensity is greater than the first intensity threshold Z1 but not greater than the second intensity threshold Z2, and the duration is greater than the first time threshold T1 but not greater than the second time threshold T2, then the current state is initially determined to be light hail, and the next step is executed. When the current state is initially determined to be light hail, the level is upgraded based on the particle size change rate, reflectivity intensity deviation and duration weighted value. If the particle size change rate is less than or equal to the preset rate threshold V, the reflectivity intensity deviation is less than or equal to the preset deviation threshold Z' and the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be light hail and a level one warning information is sent to the meteorological warning system. If the rate of change of particle size is greater than the preset rate threshold V, the reflectivity intensity deviation is greater than the preset deviation threshold Z', or the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be moderate hail, a level II warning message is sent to the meteorological warning system and the hail parameters are recorded. If the hailstone size is greater than the second size threshold D2 but not greater than the third size threshold D3, the reflectivity intensity is greater than the second intensity threshold Z2 but not greater than the third intensity threshold Z3, and the duration is greater than the second time threshold T2 but not greater than the third time threshold T3, then the current state is preliminarily determined to be moderate hail, and the next step is executed. When the current state is initially determined to be moderate hail, the level is upgraded based on the particle size change rate, reflectivity intensity deviation and duration weighted value. If the particle size change rate is less than or equal to the preset rate threshold V, the reflectivity intensity deviation is less than or equal to the preset deviation threshold Z' and the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be moderate hail, and a level II warning information is sent to the meteorological warning system and the hail parameters are recorded. If the rate of change of particle size is greater than the preset rate threshold V, the reflectivity intensity deviation is greater than the preset deviation threshold Z', or the duration weighted value is less than or equal to the preset weighted threshold Q, then the current state is determined to be severe hail, a level 3 warning message is sent to the meteorological warning system and the hail parameters are recorded. If the hailstone size is greater than the third size threshold D3, the reflectivity intensity is greater than the third intensity threshold Z3, or the duration is greater than the third time threshold T3, then the current state is determined to be severe hail, and a level 3 warning message is sent to the meteorological warning system and the hail parameters are recorded.

7. The hail identification method based on radar satellite multimodal feature fusion according to claim 2, characterized in that: The feature extraction network extracts spatiotemporal features through its radar data processing channel and satellite data processing channel, including: Acquire radar observation data input from the radar data processing channel and satellite observation data input from the satellite data processing channel; The radar observation data is preprocessed to extract spatial dynamic features, including generating a spatial velocity field and echo intensity distribution through convolution operations; The satellite observation data is preprocessed to extract brightness temperature and cloud structure features, including generating brightness temperature distribution and cloud top height maps through spectral analysis; Time series analysis is performed on the aforementioned spatial dynamic characteristics, brightness temperature, and cloud structure characteristics to generate spatiotemporal feature vectors for radar modes and satellite modes, respectively.

8. A hail identification device based on radar satellite multimodal feature fusion, based on the hail identification method according to any one of claims 1-7, characterized in that, The hail identification device includes: Data acquisition module: used to acquire meteorological multimodal observation data; Feature extraction module: used to input the multimodal meteorological observation data into the feature extraction network and extract the spatiotemporal features of the corresponding modes; Feature fusion module: used to fuse the spatiotemporal features of the corresponding modality to generate a fused feature representation; Parameter determination module: used to determine the hail particle size, reflectivity intensity, and duration of hail in the target area based on the fused feature representation; Hail level determination module: used to determine the hail level category of the target area according to the preset hail level classification standard; Result output module: used to output the hail level identification result corresponding to the target area according to the hail level category.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the hail identification method based on radar satellite multimodal feature fusion as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the hail identification method based on radar satellite multimodal feature fusion as described in any one of claims 1-7.

Images