Image recognition method and system based on weather modification

Abstract

This invention provides an image recognition method and system based on weather modification. First, a meteorological image dataset containing multiple time-series original meteorological images of a target area is acquired. Then, feature extraction processing is performed on the original meteorological images to obtain the distribution characteristics and dynamic evolution characteristics of meteorological elements. Next, based on a preset weather impact analysis model, weather impact analysis is performed on the aforementioned distribution characteristics and dynamic evolution characteristics to generate weather intervention decision features. Finally, an artificial weather modification operation strategy is generated based on the weather intervention decision features and sent to the operation execution equipment to trigger meteorological intervention operations. This allows for comprehensive and in-depth analysis of meteorological image features, scientifically and accurately generating operation strategies, and improving the effectiveness and efficiency of artificial weather modification operations.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to an image recognition method and system based on artificial weather influence. Background Technology

[0002] In the meteorological field, weather modification is an important technology that uses human intervention to control local weather processes in order to achieve goals such as disaster prevention and mitigation, and the rational use of climate resources. Traditional weather impact decision-making relies primarily on the experience of meteorological experts and some relatively simple meteorological data statistical methods. Meteorological experts need to spend a significant amount of time manually analyzing and judging massive amounts of meteorological image data, which is not only inefficient but also easily influenced by subjective factors, making it difficult to guarantee the accuracy and timeliness of decisions. At the same time, existing meteorological data analysis methods often only extract relatively superficial information from meteorological images, failing to comprehensively and deeply explore the distribution characteristics and dynamic evolution patterns of meteorological elements, thus failing to provide accurate and effective decision-making basis for weather modification operations. This greatly limits the effectiveness of weather modification operations and prevents them from fully playing their role in responding to extreme weather and ensuring agricultural production. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide an image recognition method based on weather modification, the method comprising:

[0004] Acquire a set of meteorological image data for the target area, wherein the set of meteorological image data includes multiple raw meteorological images arranged in time sequence;

[0005] The original meteorological images are subjected to feature extraction processing to obtain the distribution characteristics and dynamic evolution characteristics of meteorological elements in each original meteorological image;

[0006] Based on a preset weather impact analysis model, the distribution characteristics and dynamic evolution characteristics of the meteorological elements are analyzed to generate weather intervention decision characteristics of the original meteorological image.

[0007] Based on the weather intervention decision characteristics, an operational strategy for artificial weather modification is generated, and the operational strategy is sent to the operational execution equipment to trigger meteorological intervention operations.

[0008] In another aspect, embodiments of the present invention also provide an image recognition system based on weather modification, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0009] Based on the above, this embodiment of the invention acquires a meteorological image data set containing multiple time-series original meteorological images of a target area, and performs feature extraction processing on these original meteorological images. This allows for a comprehensive and in-depth acquisition of the distribution characteristics and dynamic evolution characteristics of meteorological elements in each original meteorological image. Based on a preset weather impact analysis model, the distribution characteristics and dynamic evolution characteristics of meteorological elements are analyzed to generate weather intervention decision features for the original meteorological images. This makes the decision-making process more scientific and accurate, eliminating excessive reliance on the experience of meteorological experts and improving the timeliness and accuracy of decisions. Finally, based on the weather intervention decision features, an artificial weather modification operation strategy is generated and sent to the operation execution equipment to trigger meteorological intervention operations. This enables rapid and accurate intervention decisions based on real-time meteorological changes, significantly improving the effectiveness and efficiency of artificial weather modification operations and better leveraging the role of artificial weather modification in disaster prevention and mitigation, and the rational utilization of climate resources. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the execution flow of the image recognition method based on artificial weather modification provided in an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of exemplary hardware and software components of an image recognition system based on weather modification provided in an embodiment of the present invention. Detailed Implementation

[0012] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an image recognition method based on weather modification according to an embodiment of the present invention. The following is a detailed description of the image recognition method based on weather modification.

[0013] Step S110: Obtain a meteorological image data set of the target area, wherein the meteorological image data set includes multiple raw meteorological images arranged in time sequence.

[0014] In this embodiment, the meteorological image dataset consists of multiple raw meteorological images arranged in chronological order. These raw meteorological images are acquired from a wide range of sources, commonly including meteorological satellites, ground meteorological monitoring stations, and meteorological radars. Meteorological satellites can perform macroscopic monitoring of target areas from space, acquiring large-scale meteorological images. Their advantage lies in their large coverage area, providing global meteorological information. Ground meteorological monitoring stations, on the other hand, focus on meteorological observations at specific locations, acquiring high-resolution local meteorological images that more accurately reflect local meteorological conditions. Meteorological radars mainly monitor the distribution of water vapor and precipitation in clouds by emitting and receiving electromagnetic waves, generating corresponding meteorological images.

[0015] In this embodiment, the raw meteorological images collected from different data sources differ in several aspects. In terms of time, the image acquisition frequencies of different devices vary. Meteorological satellites may perform global scans at fixed time intervals, resulting in relatively long image acquisition intervals; while ground-based meteorological monitoring stations may acquire images more frequently to capture rapid changes in local weather conditions. Spatially, the image resolutions of different devices also differ. Meteorological satellite images have relatively low resolution but wide coverage; images from ground-based meteorological monitoring stations and weather radars have higher resolution but limited coverage. Furthermore, images from different data sources also differ in color modes, data formats, and other aspects.

[0016] To ensure the integrity and accuracy of the acquired meteorological image data set, a data acquisition system needs to be constructed. This system should possess data receiving, storage, and preprocessing functions. In the data receiving stage, image data from multiple data sources needs to be received simultaneously, and preliminary verification and filtering should be performed to eliminate invalid or erroneous data. For data storage, a suitable database should be used to store this image data, ensuring data security and accessibility. In data preprocessing, operations such as format conversion and color correction are required for images from different data sources to ensure consistency in data format and color mode.

[0017] Step S120: Perform feature extraction processing on the original meteorological images to obtain the distribution characteristics and dynamic evolution characteristics of meteorological elements for each original meteorological image.

[0018] After obtaining the raw meteorological image dataset, feature extraction processing is required to obtain the distribution characteristics and dynamic evolution characteristics of meteorological elements in each raw meteorological image. The distribution characteristics of meteorological elements reflect the spatial distribution of meteorological elements, while the dynamic evolution characteristics reflect the changes of meteorological elements over time.

[0019] Step S121: Perform time alignment processing on the original meteorological images, and perform spatial registration processing on the set of original meteorological images after time alignment processing to obtain spatially aligned original meteorological images.

[0020] Because raw meteorological images from different data sources exhibit temporal and spatial inconsistencies, time alignment is the first step. The purpose of time alignment is to ensure temporal consistency across all raw meteorological images, facilitating accurate subsequent analysis of dynamic changes in meteorological elements. This process can be performed based on the timestamp information from different data sources. For image data acquired at different frequencies, interpolation or resampling methods can be used to unify them to the same time interval. For example, if meteorological satellite images are acquired at one-hour intervals while ground-based meteorological monitoring station images are acquired at half-hour intervals, interpolation algorithms can be used to adjust the time interval of the meteorological satellite images to half-hour intervals, achieving time synchronization.

[0021] After time alignment, the original meteorological image set needs to be spatially registered. Spatial registration aligns different images in space, ensuring they correspond to the same geographic region. A common spatial registration method is feature matching. This method first extracts feature points, such as corner points and edge points, from different images, then calculates the similarity between these feature points to find corresponding feature point pairs in different images. Finally, based on the correspondence of these feature point pairs, affine transformations or perspective transformations are used to spatially align the different images.

[0022] To improve the accuracy of spatial registration, multi-scale feature matching methods can be employed. For example, coarse matching can be performed on low-resolution images to find approximate correspondences, followed by fine matching on high-resolution images to further optimize the correspondences. Furthermore, geographic information system (GIS) data can be combined to utilize geographic coordinate information to assist in spatial registration and improve its accuracy.

[0023] Step S122: Perform noise suppression processing on the spatially aligned original meteorological image, using an adaptive filtering algorithm to reduce random noise interference in the image, while retaining detailed information of key meteorological elements, to obtain a denoised meteorological image.

[0024] After time alignment and spatial registration, spatially aligned original meteorological images are obtained. However, these original meteorological images may contain various types of noise, such as salt-and-pepper noise and Gaussian noise, which can affect subsequent feature extraction and analysis. Therefore, noise suppression processing is required. This embodiment uses an adaptive filtering algorithm to reduce random noise interference in the image while preserving the detailed information of key meteorological elements.

[0025] The core of adaptive filtering algorithms is to adaptively adjust filtering parameters based on the local features of an image to achieve the best noise reduction effect. Common adaptive filtering algorithms include adaptive median filtering and adaptive Wiener filtering. The adaptive median filtering algorithm dynamically adjusts the size of the filtering window based on the neighborhood information of pixels in the image. For noisy points, median filtering is used; for non-noisy points, the filtering remains unchanged, thus removing noise while preserving image details. The adaptive Wiener filtering algorithm calculates the optimal filtering coefficients for each pixel based on the local statistical characteristics of the image and removes noise through linear filtering.

[0026] In practical applications, it is necessary to select an appropriate adaptive filtering algorithm based on the specific characteristics of the image. For images with a lot of salt-and-pepper noise, the adaptive median filtering algorithm may be more effective; for images with a lot of Gaussian noise, the adaptive Wiener filtering algorithm may be more suitable. Furthermore, the parameters of the filtering algorithm need to be adjusted according to factors such as image resolution and noise intensity to achieve the best denoising effect.

[0027] In addition, noise suppression can be achieved by combining multi-scale analysis methods. For example, the image can be decomposed into sub-images of different scales; then, adaptive filtering can be applied to each sub-image; finally, the processed sub-images can be reconstructed to obtain the denoised meteorological image. The multi-scale analysis method described above can better preserve the image's detailed information and improve the denoising effect.

[0028] Step S123: Perform contrast enhancement processing on the denoised meteorological image to obtain an enhanced meteorological image.

[0029] In this embodiment, after obtaining the denoised meteorological image, contrast enhancement processing is required to make the meteorological elements in the denoised meteorological image clearer and more distinguishable. The purpose of contrast enhancement processing is to increase the brightness difference between different regions in the image, making the image details more obvious. Common contrast enhancement methods include histogram equalization, adaptive histogram equalization, and gamma correction.

[0030] Histogram equalization is a simple and effective contrast enhancement method that improves image contrast by redistributing the grayscale histogram of an image, making the grayscale values ​​more evenly distributed across the entire grayscale range. However, histogram equalization may lead to the loss of local details in the image, especially when there are large, uniform regions in the image.

[0031] Adaptive histogram equalization is an improvement on histogram equalization. It divides the image into multiple small blocks, performs histogram equalization on each block separately, and then merges the processed blocks into a complete image through interpolation. This method can improve image contrast while better preserving local image details.

[0032] Gamma correction adjusts the brightness and contrast of an image by performing a non-linear transformation on its grayscale values. By selecting an appropriate gamma value, bright areas of the image can be made brighter and dark areas darker, thereby enhancing the image's contrast.

[0033] In practical applications, the appropriate contrast enhancement method should be selected based on the specific characteristics of the denoised meteorological image. For images with low overall contrast, histogram equalization or adaptive histogram equalization may be more effective; for images with low local contrast, gamma correction may be more suitable. Furthermore, multiple contrast enhancement methods can be combined, using one method for initial enhancement followed by another for further optimization to achieve better enhancement results.

[0034] Step S124: Extract features from the enhanced meteorological image, extracting static distribution features reflecting the current meteorological state and dynamic evolution features reflecting changes over time. The static distribution features include cloud density distribution parameters, precipitation area boundary continuity parameters, and temperature gradient direction consistency parameters. The dynamic evolution features include cloud movement direction offset parameters between adjacent times, precipitation area change rate parameters, and temperature gradient value fluctuation amplitude parameters.

[0035] Step S1241: Perform cloud region segmentation processing on the enhanced meteorological image, identify pixels in the enhanced meteorological image whose gray values ​​are in the cloud feature range through a threshold segmentation algorithm, and generate a cloud mask image.

[0036] This step involves segmenting the cloud regions in the enhanced meteorological image to identify them. A threshold segmentation algorithm is used, based on the characteristic range of grayscale values ​​for clouds in the image. By setting an appropriate grayscale threshold range, pixels in the enhanced meteorological image whose grayscale values ​​fall within this cloud region's characteristic range are identified. For example, in practice, through extensive meteorological image analysis and experimentation, it has been determined that the grayscale values ​​of clouds generally fall within a specific range. When processing the enhanced meteorological image, the grayscale value of each pixel is compared with this range. If the pixel's grayscale value falls within this range, it is considered to belong to a cloud region. After identifying the pixels in the cloud regions, a cloud mask image is generated. The cloud mask image is a binary image, where pixels belonging to cloud regions are marked with a specific value (e.g., 1), and pixels not belonging to cloud regions are marked with another value (e.g., 0). Thus, the cloud mask image clearly delineates the range of cloud regions in the enhanced meteorological image.

[0037] Step S1242: Calculate the pixel density distribution of the cloud region based on the cloud mask image, count the number of cloud pixels per unit area as the cloud density distribution parameter, perform edge detection processing on the enhanced meteorological image, extract the boundary contour lines between precipitation and non-precipitation areas, calculate the continuous length ratio of the boundary contour lines as the precipitation area boundary continuity parameter, and perform gradient calculation processing on the enhanced meteorological image, extract the temperature gradient direction of each pixel, and count the proportion of gradient points in the same direction as the temperature gradient direction consistency parameter.

[0038] After obtaining the cloud mask image, the pixel density distribution of the cloud region is calculated. The cloud mask image is divided into several small regional units, and the number of cloud pixels in each regional unit is counted. Then, the number of cloud pixels in each regional unit is divided by the area of ​​that regional unit to obtain the cloud pixel density of that regional unit. Combining the cloud pixel densities of all regional units forms the pixel density distribution of the cloud region. The number of cloud pixels per unit area is used as the cloud density distribution parameter, which reflects the spatial density of the cloud distribution.

[0039] To calculate the continuity parameter of the precipitation region boundary, edge detection processing is performed on the enhanced meteorological image. Edge detection algorithms can identify locations in the image where grayscale values ​​change abruptly; these locations often correspond to the boundaries of different regions. Through edge detection, the boundary contour lines between precipitation and non-precipitation regions are extracted. Then, the length of the continuous portion of the boundary contour line is calculated, and its proportion to the total boundary contour line length is determined; this proportion is the precipitation region boundary continuity parameter. This parameter describes the degree of continuity between the precipitation and non-precipitation regions.

[0040] When calculating the temperature gradient direction consistency parameter, gradient calculation processing is performed on the enhanced meteorological image. Gradient calculation yields the temperature gradient direction for each pixel. Then, the number of gradient points with the same direction in the image is counted, and their proportion to the total number of gradient points is calculated. This proportion is the temperature gradient direction consistency parameter, which measures the degree of spatial consistency of the temperature gradient direction and helps in understanding the distribution and changes of the temperature field.

[0041] Step S1243: Obtain the enhanced meteorological image of the previous moment as a reference image, perform optical flow field calculation processing on the enhanced meteorological image of the current moment and the reference image to obtain the displacement vector set of the pixels in the cloud system region, and calculate the angle between the overall movement direction of the cloud system and the movement direction of the previous moment based on the displacement vector set as the offset parameter of the cloud system movement direction.

[0042] To calculate the cloud system's movement direction offset parameter, an enhanced meteorological image from the previous moment is needed as a reference image. The current enhanced meteorological image and the reference image are then processed using optical flow field calculation. Optical flow field calculation involves analyzing the motion of pixels in adjacent frames to obtain the displacement vector of each pixel. In this embodiment, the focus is on the displacement vectors of pixels within the cloud system region. Through optical flow field calculation, a set of displacement vectors for pixels in the cloud system region is obtained. Then, based on this set of displacement vectors, the overall movement direction of the cloud system is calculated. This can be determined by methods such as weighted averaging of the displacement vectors of all pixels within the cloud system region. Next, the angle between the current overall movement direction of the cloud system and the movement direction from the previous moment is calculated; this angle is the cloud system movement direction offset parameter. This parameter reflects the change in the cloud system's movement direction between two adjacent moments.

[0043] Step S1244: Calculate the ratio of the difference between the current precipitation area and the previous precipitation area to the area at the previous time as the precipitation area change rate parameter.

[0044] To calculate the precipitation area change rate parameter, it is first necessary to determine the area of ​​the precipitation region at the current and previous moments. For example, this can be achieved by segmenting the precipitation region in the enhanced meteorological images of the current and previous moments, using a threshold segmentation algorithm similar to cloud region segmentation to identify the pixels of the precipitation region. Then, the number of pixels in the precipitation region is counted, and the pixel count is converted into actual area based on the image resolution. After obtaining the areas of the precipitation region at the current and previous moments, their difference is calculated. This difference is then divided by the area of ​​the precipitation region at the previous moment; the resulting ratio is the precipitation area change rate parameter. This parameter describes the change in the area of ​​the precipitation region between two adjacent moments, reflecting the development trend of precipitation, such as whether the precipitation region is expanding or shrinking.

[0045] Step S1245: Calculate the average of the absolute differences between the current temperature gradient value and the temperature gradient value at the corresponding position at the previous time as the temperature gradient value fluctuation amplitude parameter.

[0046] When calculating the temperature gradient fluctuation amplitude parameter, it is necessary to obtain the temperature gradient value of each pixel in the enhanced meteorological images at the current and previous times. Gradient calculation processing is performed on the enhanced meteorological images at the current and previous times to obtain the temperature gradient value of each pixel. Then, the temperature gradient value of each pixel at the current time is compared with the temperature gradient value of the corresponding pixel at the previous time, and their absolute differences are calculated. The absolute differences of all pixels are summarized, and then the average of these absolute differences is calculated. This average value is the temperature gradient fluctuation amplitude parameter. This temperature gradient fluctuation amplitude parameter reflects the fluctuation of the temperature gradient value between two adjacent times, which helps in analyzing the stability and trend of the temperature field.

[0047] Step S125: Perform feature integration processing on the static distribution features and dynamic evolution features to generate meteorological element distribution features and dynamic evolution features that contain spatial dimension distribution information and temporal dimension evolution information.

[0048] For example, a weighted stitching method can be used to assign different weights to static distribution features and dynamic evolution features. The weight allocation is determined based on actual needs and the importance of different features in the analysis. For instance, if more attention is paid to the spatial distribution of meteorological elements, a larger weight can be assigned to static distribution features; if more attention is paid to the temporal changes of meteorological elements, a larger weight can be assigned to dynamic evolution features. The static distribution features and dynamic evolution features are stitched together in a predetermined order to generate meteorological element distribution features and dynamic evolution features that contain both spatial and temporal evolution information. During the stitching process, it is necessary to ensure that the dimensions of the static distribution features and dynamic evolution features match. If their dimensions are different, interpolation or dimensionality reduction methods can be used to make their dimensions consistent. Through feature integration processing, the resulting meteorological element distribution features and dynamic evolution features can more comprehensively and accurately reflect the state and changes of meteorological elements in the original meteorological image.

[0049] Step S130: Based on the preset weather impact analysis model, perform weather impact analysis on the distribution characteristics and dynamic evolution characteristics of the meteorological elements to generate weather intervention decision features of the original meteorological image.

[0050] Step S131: Input the distribution characteristics of the meteorological elements into the spatial feature processing module of the weather impact analysis model, and extract the local detail features and global distribution features of the meteorological elements in the spatial dimension through a convolutional neural network to generate a spatial feature representation.

[0051] Step S1311: Perform spatial domain decomposition on the distribution characteristics of the meteorological elements, and divide them into cloud structure sub-features, precipitation area sub-features and temperature field distribution sub-features according to the meteorological element type.

[0052] The distribution characteristics of meteorological elements are spatially decomposed, dividing them into different sub-features based on their type. The cloud structure sub-feature mainly describes the structural features of cloud systems, such as morphology and density; the precipitation region sub-feature focuses on the boundaries and internal pixel distribution of precipitation regions; and the temperature field distribution sub-feature emphasizes the spatial distribution of temperature. This division allows for more targeted feature extraction and analysis of different meteorological elements. For example, the cloud structure sub-feature reflects the formation, development, and dissipation processes of cloud systems, which is important for analyzing the impact of clouds on weather; the precipitation region sub-feature helps determine the extent and intensity of precipitation; and the temperature field distribution sub-feature helps understand the distribution patterns and trends of temperature.

[0053] Step S1312: Input the cloud structure sub-features into the first branch of the convolutional neural network, and extract the continuity features of the cloud edge contour and the gradient distribution features of the internal pixel density through morphological convolution kernels.

[0054] The cloud structure features are input into the first branch of the convolutional neural network. In this first branch, morphological convolutional kernels are used for feature extraction. These kernels process the cloud structure features, extracting the continuity of the cloud edge contours. The continuity of the cloud edge contours reflects the integrity and stability of the cloud system, playing a crucial role in analyzing its development and evolution. Simultaneously, the gradient distribution features of pixel density within the cloud system can also be extracted. The gradient distribution of pixel density reflects the structural changes within the cloud system, such as whether the density is uniform and whether there are regions with large density gradients.

[0055] Step S1313: Input the sub-features of the precipitation area into the second branch of the convolutional neural network, use an adaptive threshold convolution kernel to process the boundary transition zone between the precipitation area and the non-precipitation area, and extract the boundary clarity features and the uniformity features of the pixel values ​​inside the area.

[0056] The sub-features of the precipitation region are input into the second branch of the convolutional neural network. In this second branch, an adaptive thresholding convolutional kernel is used to process the boundary transition zone between precipitation and non-precipitation regions. The adaptive thresholding convolutional kernel can automatically adjust the threshold based on local image features, better handling boundary transition zones. Through this processing, the sharpness features of the precipitation region boundary can be extracted; a sharp boundary helps to accurately determine the range and intensity of precipitation. Simultaneously, the uniformity features of pixel values ​​within the precipitation region can also be extracted. The uniformity of pixel values ​​reflects the consistency of precipitation intensity within the precipitation region. High pixel value uniformity within the precipitation region indicates relatively stable precipitation intensity; low uniformity may indicate significant variations in precipitation intensity within the precipitation region, potentially suggesting the existence of localized areas of heavy or light precipitation.

[0057] Step S1314: Input the temperature field distribution sub-features into the third branch of the convolutional neural network, process the spatial orientation features of the temperature gradient through directional convolution kernels, and extract the clustering features of the same gradient region and the dispersion features of the opposite gradient region.

[0058] The temperature field distribution features are input into the third branch of the convolutional neural network. This third branch uses directional convolutional kernels to process the spatial orientation features of the temperature gradient. The spatial orientation of the temperature gradient reflects the direction of temperature change in space, which is crucial for understanding the laws of heat transfer and distribution. Through processing with directional convolutional kernels, the clustering characteristics of regions with the same temperature change direction can be extracted. The clustering degree of regions with the same temperature change direction indicates the degree to which regions in the temperature field are clustered together. Higher clustering degree means that a large area has similar temperature change trends, which may indicate that a large-scale meteorological process is occurring. Simultaneously, the dispersion features of regions with opposite temperature change directions can also be extracted. The dispersion degree of regions with opposite temperature change directions reflects the degree to which regions with different temperature change directions are dispersed. Higher dispersion degree indicates that there are many temperature changes in different directions in the temperature field, which may reflect the complexity and variability of local meteorological conditions.

[0059] Step S1315: Perform cross-feature association processing on the local detail features output by the first branch, the second branch, and the third branch, and calculate the spatial overlap parameter between cloud edge continuity and precipitation boundary clarity, as well as the matching parameter between temperature gradient aggregation and cloud internal density gradient.

[0060] Cross-feature correlation processing of the local detail features output by the three branches of a convolutional neural network helps to discover the intrinsic connections between different meteorological elements. First, the spatial overlap parameter between cloud system edge continuity and precipitation boundary clarity is calculated. Cloud system edge continuity reflects the integrity and stability of the cloud system, while precipitation boundary clarity is related to the extent and intensity of precipitation. By analyzing their spatial overlap, the degree of correlation between clouds and precipitation can be understood. For example, a high degree of spatial overlap between cloud system edge continuity and precipitation boundary clarity may indicate that cloud system development and evolution have a significant impact on precipitation formation and distribution. Next, the matching degree parameter between temperature gradient clustering and cloud system internal density gradient is calculated. Temperature gradient clustering reflects the aggregation of unidirectional gradient regions in the temperature field, while cloud system internal density gradient reflects the density changes within the cloud system. The matching degree parameter can reveal the relationship between temperature changes and cloud system structure. A high matching degree indicates that temperature changes may cause corresponding adjustments in the cloud system's internal structure; conversely, a low matching degree may indicate that other factors influence cloud system development.

[0061] Step S1316: Perform element-wise multiplication of the cross-feature association processing result with the original meteorological element distribution characteristics to generate enhanced local features containing spatial synergistic relationships between elements.

[0062] In this embodiment, the element-wise multiplication operation highlights the parts of the original meteorological element distribution characteristics that are related to the cross-feature association processing results, thereby enhancing the expressive power of the features. Through the above feature enhancement processing, the generated enhanced local features contain the spatial synergistic relationships between elements. For example, in the relationship between cloud systems and precipitation, enhanced local features can more clearly show how the development of cloud systems affects the distribution of precipitation, as well as the synergistic changes between the two.

[0063] Step S1317: Perform spatial dimension compression processing on the enhanced local features through global pooling operation to extract global distribution features that reflect the distribution ratio and centroid position of cloud systems, precipitation, and temperature fields in the overall region.

[0064] In this embodiment, global pooling integrates enhanced local features spatially, extracting global distribution features that reflect the proportion and centroid location of cloud systems, precipitation, and temperature fields across the entire region. The distribution proportion represents the relative size of cloud systems, precipitation, and temperature fields within the target area, which helps understand the importance and influence of different meteorological elements in the overall region. The centroid location reflects the spatial concentration of cloud systems, precipitation, and temperature fields; changes in the centroid location allow analysis of the movement trends and patterns of meteorological elements. Global pooling can employ methods such as average pooling or max pooling; the appropriate pooling method is selected based on actual needs to obtain more accurate global distribution features.

[0065] Step S1318: Input the enhanced local features and global distribution features into the spatial feature synthesis layer, and assign weight coefficients related to the global distribution to the local features of different elements through the attention mechanism to generate a spatial feature representation that simultaneously contains the internal details of the elements and the collaborative relationships between the elements.

[0066] In this embodiment, an attention mechanism is used in the spatial feature synthesis layer to assign weight coefficients related to the global distribution to the local features of different elements. The attention mechanism can automatically adjust the importance of local features of different elements based on the global distribution characteristics. For example, if the global distribution characteristics show that cloud systems have a large distribution ratio in the overall region, then the local features of cloud systems will be assigned a higher weight coefficient during the synthesis process, thus being more prominently reflected in the final spatial feature representation. Through this method, the generated spatial feature representation simultaneously includes detailed information within elements and the collaborative relationships between elements, enabling a more comprehensive and accurate description of the spatial characteristics of meteorological elements.

[0067] Step S132: Input the dynamic evolution features into the time feature processing module of the weather impact analysis model, and extract the continuous evolution features and trend prediction features of meteorological elements in the time dimension through a recurrent neural network to generate a time feature representation.

[0068] For example, step S1321: Arrange the dynamic evolution features in chronological order into time series data including cloud movement trajectory, precipitation area expansion rate, and temperature gradient fluctuation.

[0069] In this embodiment, the cloud system movement trajectory records the positional changes of the cloud system at different times, reflecting the movement path and direction of the cloud system; the precipitation area expansion rate reflects the speed of area change of the precipitation area over time, which helps to understand the development trend of precipitation; and the temperature gradient fluctuation describes the change of the temperature gradient value over time, reflecting the stability and dynamic changes of the temperature field. Arranging these features in chronological order into time series data can clearly show the evolution of meteorological elements over time.

[0070] Step S1322: Input the time series data into the initial state unit of the recurrent neural network to initialize the initial memory state containing the cloud system movement direction, precipitation area baseline value, and temperature gradient baseline fluctuation amount of the previous time step.

[0071] In this embodiment, the initial state unit initializes an initial memory state containing the cloud system movement direction, the baseline value of the precipitation area, and the baseline fluctuation of the temperature gradient from the previous time step. The cloud system movement direction from the previous time step provides a reference for subsequent judgments on changes in the cloud system movement direction; the baseline value of the precipitation area can serve as the basis for measuring changes in the precipitation area; and the baseline fluctuation of the temperature gradient is used to compare the current temperature gradient fluctuation. Setting the initial memory state allows the recurrent neural network to consider previous state information when processing time-series data, thereby better capturing the continuous evolution characteristics of meteorological elements.

[0072] Step S1323: Perform state update operation at each time step: calculate the directional deviation between the cloud system movement trajectory at the current time step and the cloud system movement direction in the initial memory state to generate cloud system motion stability characteristics; calculate the rate matching between the precipitation area expansion rate at the current time step and the area benchmark value in the initial memory state to generate precipitation development coordination characteristics; calculate the fluctuation consistency between the temperature gradient fluctuation at the current time step and the benchmark fluctuation amount in the initial memory state to generate temperature field evolution persistence characteristics.

[0073] In this embodiment, for cloud systems, the directional deviation between the current time step's cloud system movement trajectory and the initial memory state's cloud system movement direction is calculated. By calculating this directional deviation, a cloud system motion stability characteristic can be generated. A small directional deviation indicates relatively stable cloud system motion; a large directional deviation indicates that the cloud system motion may be significantly affected by external factors. For precipitation areas, the rate of precipitation area expansion at the current time step is matched with the area baseline value in the initial memory state to obtain a precipitation development coordination characteristic. This precipitation development coordination characteristic reflects whether the expansion of the precipitation area is coordinated with the previous area baseline value, determining whether precipitation development conforms to the expected trend. For the temperature field, the fluctuation consistency between the current time step's temperature gradient fluctuation and the baseline fluctuation amount in the initial memory state is calculated to generate a temperature field evolution persistence characteristic. High fluctuation consistency indicates good persistence in the temperature field evolution; conversely, low consistency may indicate abnormal temperature changes.

[0074] Step S1324: Combine the cloud system motion stability characteristics, precipitation development coordination characteristics, and temperature field evolution persistence characteristics into intermediate evolution characteristics for the current time step.

[0075] In this embodiment, the splicing operation integrates the evolution characteristics of different meteorological elements, so that these characteristics can be correlated and complement each other. Thus, the intermediate evolution characteristics include the comprehensive evolution information of cloud system, precipitation, and temperature field at the current time step.

[0076] Step S1325: The intermediate evolution features are fused with the hidden state of the previous time step through the state transition function of the recurrent neural network to generate an updated hidden state containing historical evolution information.

[0077] In this embodiment, the state transition function can generate an updated hidden state containing historical evolution information based on intermediate evolution characteristics and the hidden state of the previous time step. The updated hidden state not only includes the evolution information of meteorological elements at the current time step but also incorporates historical information from previous time steps, enabling the network to learn and memorize the long-term evolution process of meteorological elements. This fusion process helps capture the continuous changes and trends of meteorological elements over time.

[0078] Step S1326: Perform time-weighted accumulation processing on the updated hidden states of continuous time steps. The weight values ​​are set according to the time step and the current time using an exponential decay rule to generate trend prediction features that reflect long-term evolution trends.

[0079] In this embodiment, during the weighted accumulation process, the weight values ​​are set using an exponential decay rule based on the interval between the time step and the current time. The closer the time step is to the current time, the higher its weight value for updating the hidden state; the farther the time step is from the current time, the lower its weight value. Through the above-mentioned exponential decay weighted accumulation process, the importance of recent meteorological element evolution can be highlighted, while also taking into account historical evolution information. The resulting trend prediction feature reflects the long-term evolution trend of meteorological elements.

[0080] Step S1327: Input the updated hidden state and trend prediction features into the time feature integration layer, and extract the difference features between the current evolution state and the long-term trend through time difference convolution operation to generate a time feature representation that simultaneously contains immediate evolution details and long-term trend correlation.

[0081] The updated hidden state and trend prediction features are input into a temporal feature integration layer. This layer then uses temporal difference convolution to extract the difference features between the current evolution state and the long-term trend. This temporal difference convolution operation allows for comparison and analysis of the updated hidden state and trend prediction features, identifying the differences between the current evolution state and the long-term trend of meteorological elements. These differences reveal anomalous changes in meteorological elements in the short term, which is crucial for the timely detection of meteorological abrupt changes and abnormal weather phenomena. By integrating the updated hidden state, trend prediction features, and difference features, the generated temporal feature representation simultaneously includes immediate evolution details and long-term trend correlations, enabling a more comprehensive and accurate description of the characteristics of meteorological elements in the time dimension.

[0082] Step S133: Perform feature fusion processing on the spatial feature representation and the temporal feature representation, and assign weight parameters to features of different dimensions through an attention mechanism to generate a fused feature vector.

[0083] In this embodiment, an attention mechanism is used to assign weight parameters to features of different dimensions. Spatial feature representation reflects the spatial distribution and characteristics of meteorological elements, while temporal feature representation reflects the temporal evolution and trends of meteorological elements. The attention mechanism can automatically adjust the weights of different features based on their importance in the current weather impact analysis. For example, if the current weather situation focuses more on spatial distribution changes, the weight of spatial feature representation will be relatively high; if more attention is paid to temporal evolution trends, the weight of temporal feature representation will increase. Through this method, spatial and temporal feature representations are fused to generate a fused feature vector. The fused feature vector integrates information about meteorological elements in both spatial and temporal dimensions.

[0084] Step S134: Input the fused feature vector into the decision reasoning module of the weather impact analysis model, and perform nonlinear transformation processing on the fused feature vector through a fully connected neural network to generate a set of potential intervention features that reflect the effects of different artificial weather modification measures.

[0085] Step S1341: Input the fused feature vector into the first fully connected layer of the decision reasoning module, and generate an intermediate feature vector containing basic intervention features through linear transformation and activation function processing.

[0086] In this embodiment, in the first fully connected layer, a linear transformation is first performed, multiplying the fused feature vector by the layer's weight matrix and adding a bias term. This linear transformation provides preliminary feature combination and transformation of the fused feature vector. Then, an activation function is applied, introducing non-linearity and enabling the network to learn more complex feature relationships. After the linear transformation and activation function processing, an intermediate feature vector containing basic intervention features is generated. These basic intervention features are fundamental features obtained after the initial processing of the fused feature vector.

[0087] Step S1342: Input the intermediate feature vector into the second fully connected layer of the decision reasoning module, and generate a high-level feature vector containing extended intervention features through feature recombination and nonlinear mapping processing.

[0088] In this embodiment, a feature recombination operation is performed in the second fully connected layer, recombining and arranging the features in the intermediate feature vector to uncover new correlations and combinations between features. Simultaneously, nonlinear mapping is used to further introduce nonlinear factors, enabling more complex feature representations. After feature recombination and nonlinear mapping, a high-level feature vector containing extended intervention features is generated. These extended intervention features expand and deepen the basic intervention features, providing a more comprehensive reflection of the potential effects of different weather modification measures.

[0089] Step S1343: Input the high-level feature vector into the third fully connected layer of the decision reasoning module, and generate a key feature vector containing core intervention features through feature filtering and dimensionality compression.

[0090] In this embodiment, feature filtering is performed in the third fully connected layer. Based on the importance and relevance of features, the features most influential on the effectiveness of weather modification measures are selected. Simultaneously, dimensionality compression is performed to reduce the dimensionality of the feature vectors, remove redundant information, and improve feature representation efficiency. After feature filtering and dimensionality compression, a key feature vector containing core intervention features is generated. These core intervention features are the most important features obtained after layers of filtering and processing.

[0091] Step S1344: Normalize the key feature vector, input the normalized key feature vector into the output layer of the decision reasoning module, and generate the probability distribution vector corresponding to different artificial weather modification measures through classification activation function.

[0092] The key feature vectors are normalized. Normalization maps each element of the key feature vector to a defined range, ensuring that all elements have the same scale and preventing excessive differences in feature scale from affecting subsequent processing. The normalized key feature vectors are then input into the output layer of the decision reasoning module, where they are processed by a classification activation function. The classification activation function transforms the key feature vectors into probability distribution vectors corresponding to different weather modification measures. Each element in the probability distribution vector represents the probability of achieving the desired effect by taking a certain weather modification measure. The probability distribution vector provides a clear understanding of the likelihood and effectiveness of different weather modification measures.

[0093] Step S1345: Extract the measure types whose probability values ​​exceed a preset threshold as potential intervention features based on the probability distribution vector, and generate a set of potential intervention features containing measure type identifiers and corresponding probability values.

[0094] In this embodiment, the preset threshold is a boundary set based on actual needs and experience, used to filter out weather modification measures with a high probability and effectiveness. Measures with probability values ​​exceeding the preset threshold are designated as potential intervention features, and their identifiers and corresponding probability values ​​are recorded. This information is combined to generate a set of potential intervention features containing measure type identifiers and corresponding probability values. This set of potential intervention features provides specific weather modification measure options for subsequent feasibility assessments and decision-making.

[0095] Step S135: Perform a feasibility assessment on the set of potential intervention features, and select effective intervention features that meet the current meteorological conditions by combining the correlation between different intervention measures and meteorological responses in historical operation data.

[0096] Step S1351: Extract historical meteorological sample data from the preset historical operation database that matches the current meteorological element distribution characteristics and dynamic evolution characteristics.

[0097] Historical meteorological sample data matching the current meteorological element distribution characteristics and dynamic evolution characteristics are extracted from a pre-set historical operation database. This database stores a large amount of meteorological data and operation records from past weather modification operations. By comparing the current meteorological element distribution characteristics and dynamic evolution characteristics, similar meteorological sample data is searched within the historical operation database. Methods such as feature similarity calculation, including Euclidean distance and cosine similarity between current features and historical sample features, can be used to identify historical meteorological sample data with high similarity. These matched historical meteorological sample data contain meteorological response information after different intervention measures were taken under similar meteorological conditions in the past.

[0098] Step S1352: Perform statistical processing on the effects of intervention measures on the historical meteorological sample data, and calculate the meteorological response index corresponding to each intervention measure. The meteorological response index includes precipitation increment parameter, cloud dissipation rate parameter, and temperature adjustment amplitude parameter.

[0099] In this embodiment, for each intervention measure recorded in the historical sample data, the corresponding meteorological response index is calculated. The meteorological response index includes a precipitation increment parameter, a cloud dissipation rate parameter, and a temperature regulation amplitude parameter. The precipitation increment parameter represents the increase in precipitation in the precipitation area after taking a certain intervention measure relative to the situation without intervention; the cloud dissipation rate parameter reflects the speed at which the cloud system dissipates under the influence of the intervention measure; and the temperature regulation amplitude parameter measures the degree to which the intervention measure regulates the temperature field. Through statistical analysis of the historical meteorological sample data, the specific values ​​of the meteorological response index corresponding to each intervention measure can be obtained.

[0100] Step S1353: Calculate the similarity value between the current potential intervention feature and the historical sample features, and perform a weighted average of the meteorological response indicators of the historical meteorological sample data based on the similarity value to generate the predicted response indicator corresponding to the current potential intervention feature. First, calculate the similarity value between the current potential intervention feature and the historical sample features. Various similarity calculation methods can be used, such as Euclidean distance and cosine similarity. Taking Euclidean distance as an example, the current potential intervention feature and the historical sample features are considered as vectors in the feature space, and the Euclidean distance between them is calculated. The smaller the Euclidean distance, the higher the similarity between the two features. For each historical sample feature, calculate its similarity value with the current potential intervention feature.

[0101] In this embodiment, historical sample data with higher similarity values ​​have a greater weight in the weighted average for their meteorological response indicators. The specific calculation process involves multiplying the meteorological response indicator (including precipitation increment parameter, cloud dissipation rate parameter, and temperature regulation amplitude parameter) of each historical sample data by its corresponding similarity value, then summing all the products, and finally dividing by the sum of the similarity values. This yields the predicted response indicator corresponding to the current potential intervention feature. The predicted response indicator reflects the possible meteorological response under current meteorological conditions when implementing artificial weather modification measures corresponding to the potential intervention feature.

[0102] Step S1354: Select potential intervention features whose predicted response indicators meet the preset response indicator threshold range as candidate intervention features.

[0103] In this embodiment, the preset response index threshold range is set based on actual needs and meteorological conditions to assess the feasibility of potential intervention features. For the precipitation increment parameter, cloud dissipation rate parameter, and temperature regulation amplitude parameter in the predicted response indicators, corresponding threshold ranges are set respectively. If the predicted response indicators of a potential intervention feature all fall within the preset threshold range, the potential intervention feature is considered feasible and is selected as a candidate intervention feature. For example, if the preset precipitation increment parameter threshold range is a certain interval, when the predicted precipitation increment parameter of a potential intervention feature falls within this interval, the threshold requirement for the precipitation increment parameter is met. Only when all predicted response indicators meet their corresponding threshold ranges can a potential intervention feature become a candidate intervention feature.

[0104] Step S1355: Perform conflict detection processing on the candidate intervention features, exclude intervention feature combinations that would cause the meteorological effects to cancel each other out when implemented simultaneously, and retain the candidate intervention features without conflict as effective intervention features.

[0105] In this embodiment, during the process of weather modification, different intervention measures may affect each other, and the simultaneous implementation of some intervention features may cause the meteorological effects to cancel each other out. For example, one intervention measure is to increase precipitation, while another is to dissipate cloud systems. If these two measures are implemented simultaneously, their effects may conflict, failing to achieve the expected weather modification goals.

[0106] A specific method for conflict detection and processing can be to establish a conflict rule base, which records the conflict relationships between different intervention features. For each candidate combination of intervention features, the conflict rule base is queried to determine whether a conflict exists. If a conflict exists, the intervention feature combination is excluded; if no conflict exists, the candidate intervention features in the combination are retained. After conflict detection and processing, the retained conflict-free candidate intervention features are considered effective intervention features. Effective intervention features are weather modification measures that, under current meteorological conditions, have both good predictive response effects and do not conflict with each other.

[0107] Step S136: Perform feature encoding on the effective intervention features to generate weather intervention decision features that include intervention type identifier, area of ​​effect, and timing of effect identifier.

[0108] In this embodiment, effective intervention features are encoded to transform them into a format that is easy to store and transmit, while also containing key decision-making information. During the feature encoding process, an intervention type identifier is first determined. This identifier clarifies the type of weather modification measure taken, such as rain enhancement, cloud dispersal, and temperature regulation. Based on the specific measure corresponding to the effective intervention feature, a corresponding intervention type identifier is assigned.

[0109] Next, the area of ​​effect is determined, which refers to the geographical region where weather modification measures need to be implemented. This area can be represented using geographic coordinates, such as recording the boundary and center coordinates. This coordinate information allows for accurate location of the areas requiring intervention.

[0110] Finally, the timing identifier was determined. This identifier, combined with timestamp information from the meteorological image data set, identified the start time and duration of the intervention. The timing identifier is crucial for the rational scheduling of weather modification operations, ensuring that interventions are implemented at the appropriate time to achieve the best results.

[0111] By combining and encoding the intervention type identifier, the area of ​​effect, and the timing of the intervention identifier, a weather intervention decision feature is generated. This weather intervention decision feature is a comprehensive feature that contains the key information required to implement weather modification measures.

[0112] Step S140: Generate an artificial weather modification operation strategy based on the weather intervention decision characteristics, and send the operation strategy to the operation execution equipment to trigger meteorological intervention operations.

[0113] Step S141: Analyze the intervention type identifier in the weather intervention decision features to determine the type of artificial weather modification measures to be implemented. The types of measures include rain enhancement operations, cloud dissipation operations, and temperature regulation operations.

[0114] In this embodiment, the intervention type identifier in the weather intervention decision features is parsed to determine the type of artificial weather modification measures that need to be implemented. The intervention type identifier is assigned during the feature encoding process; parsing this identifier clarifies the specific artificial weather modification measures. In this embodiment, the measure types include rain enhancement operations, cloud dissipation operations, and temperature regulation operations. Different intervention type identifiers correspond to different measure types. For example, when the intervention type identifier is a specific code, it indicates that rain enhancement operations need to be implemented; when the intervention type identifier is another code, it indicates that cloud dissipation operations or temperature regulation operations need to be carried out.

[0115] Step S142: Analyze the effective area range in the weather intervention decision features, extract the geographical boundary coordinates and center coordinates of the area where intervention measures need to be implemented, and generate regional positioning information.

[0116] In this embodiment, the effective area range in the weather intervention decision features is analyzed to extract the geographical boundary coordinates and center coordinates of the area requiring intervention. The effective area range is recorded in the form of geographical coordinates during feature encoding; this coordinate information can be obtained through analysis. The geographical boundary coordinates clearly define the specific extent of the effective area, while the center coordinates serve as a reference point for the operation. Combining the boundary coordinates and the center coordinates generates regional positioning information. This regional positioning information is crucial for the accurate arrival of operational equipment in the operational area, ensuring that weather modification measures are implemented within the correct geographical region.

[0117] Step S143: Analyze the timing identifier in the weather intervention decision features, combine it with the timestamp information of the meteorological image data set, determine the start time and duration of the intervention measures, and generate time planning information.

[0118] In this embodiment, the timing identifier in the weather intervention decision features is analyzed and combined with the timestamp information of the meteorological image dataset to determine the start time and duration of the intervention measures. The timing identifier contains time-related information, and its analysis provides a general time range. The timestamp information of the meteorological image dataset records the time of meteorological image acquisition; combined with the timing identifier, the start time of the intervention measures can be determined more accurately. Based on the information provided in the timing identifier, the duration of the intervention measures can also be determined. Combining the start time and duration generates time planning information. Time planning information is crucial for rationally scheduling the working hours of operational equipment and ensuring that artificial weather modification measures are implemented within an appropriate timeframe.

[0119] Step S144: Perform operation parameter configuration processing on the measure type, regional positioning information and time planning information, and match corresponding operation equipment parameters for each measure type. The operation equipment parameters include catalyst spreading amount parameters, operation height parameters and operation route parameters.

[0120] Step S1441: For the type of rain enhancement operation, query the preset rain enhancement operation parameter configuration table, and select the corresponding catalyst type and single seeding amount parameters according to the size of the area to be affected and the cloud development stage at the time of the operation.

[0121] In this embodiment, for each type of rain enhancement operation, a preset rain enhancement operation parameter configuration table is queried. This table is based on extensive experimental and practical experience, and records the catalyst type and single-sowing quantity parameters corresponding to different effective area sizes and different cloud development stages. Based on the analyzed effective area size and the cloud development stage at the time of the operation, the corresponding record is searched in the configuration table, and an appropriate catalyst type and single-sowing quantity parameter are selected. For example, if the effective area is large and the cloud system is in a vigorous development stage, it may be necessary to select a catalyst with a strong catalytic effect and appropriately increase the single-sowing quantity; if the effective area is small and the cloud system development is relatively stable, a catalyst with a moderate catalytic effect and a smaller single-sowing quantity can be selected.

[0122] Step S1442: Based on the center coordinates and boundary coordinates of the operational area, and combined with the maximum range parameters of the operational aircraft, plan the optimal flight route from the airport to the operational area, and calculate the turning point coordinates and flight altitude parameters in the optimal flight route.

[0123] In this embodiment, the optimal flight route from the airport to the work area is planned based on the center and boundary coordinates of the work area and the maximum range parameters of the operating aircraft. Planning the optimal flight route requires consideration of multiple factors, such as flight distance, flight time, and weather conditions. By comprehensively analyzing these factors, an optimal route is selected that ensures the operating aircraft can safely reach the work area while remaining within its maximum range. During the planning process, the coordinates of turning points in the optimal flight route are calculated, taking into account both the route's rationality and flight safety. Simultaneously, flight altitude parameters are calculated based on the weather conditions of the work area and the requirements for catalyst dissemination. Flight altitude parameters are crucial for effective catalyst dissemination and operational results and require precise calculation based on actual conditions.

[0124] Step S1443: For the cloud suppression operation type, query the preset cloud suppression operation parameter configuration table, and determine the catalyst application interval and application concentration parameters based on the cloud density distribution parameters and temperature gradient direction consistency parameters within the action area.

[0125] In this embodiment, the cloud suppression operation parameter configuration table records the catalyst application interval and application concentration parameters corresponding to the cloud density distribution parameters and temperature gradient direction consistency parameters within different action areas. Based on the analyzed cloud density distribution parameters and temperature gradient direction consistency parameters within the action area, the corresponding records are searched in the configuration table to determine the appropriate catalyst application interval and application concentration parameters. If the cloud density is high, it may be necessary to shorten the application interval and increase the application concentration; if the cloud density is low and the temperature gradient direction consistency is high, the application interval and application concentration can be appropriately extended and the application concentration reduced.

[0126] Step S1444: Combine the start time and duration of the operation to calculate the time difference between the aircraft taking off from the airport and arriving at the operation area, and adjust the takeoff time parameters.

[0127] In this embodiment, the flight time can be calculated based on the planned optimal flight route and the flight speed of the operational aircraft. The takeoff time of the operational aircraft is obtained by subtracting the flight time from the start time of the operation. If the calculated takeoff time is unreasonable, such as conflicting with airport scheduling, the takeoff time needs to be adjusted. When adjusting the takeoff time parameters, various factors need to be considered comprehensively to ensure that the operational aircraft can arrive at the operational area at the appropriate time and carry out cloud clearing operations on schedule.

[0128] Step S1445: For the type of temperature regulation operation, query the preset temperature regulation operation parameter configuration table, and select the corresponding heat dissipant type and release rate parameter according to the temperature gradient value fluctuation amplitude parameter of the action area and the precipitation area change rate parameter.

[0129] For each temperature regulation operation type, consult the preset temperature regulation operation parameter configuration table. This table records the heat dissipation agent type and release rate parameters corresponding to the temperature gradient fluctuation range and precipitation area change rate parameters for different affected areas. Based on the analyzed temperature gradient fluctuation range and precipitation area change rate parameters for the affected area, locate the corresponding records in the configuration table and select the appropriate heat dissipation agent type and release rate parameters. If the temperature gradient fluctuation range is large, it may be necessary to select a heat dissipation agent with a strong heat diffusion effect and appropriately increase the release rate; if the temperature gradient fluctuation range is small and the precipitation area change rate is relatively stable, a heat dissipation agent with a moderate heat diffusion effect and a lower release rate can be selected.

[0130] Step S1446: Combine the equipment parameters of different measure types to generate a set of operational equipment parameters, including catalyst type and single application rate parameters, optimal flight path, takeoff time parameters, heat dissipant type and release rate parameters.

[0131] In this embodiment, the set of operational equipment parameters includes information such as catalyst type and single-batch dispersal parameters (applicable to rain enhancement and cloud dissipation operations), optimal flight path, takeoff time parameters, and heat dissipating agent type and release rate parameters (applicable to temperature control operations). This set of operational equipment parameters provides detailed guidance for the operation of the equipment, ensuring that weather modification operations can be carried out accurately and effectively.

[0132] Step S145: Perform strategy integration processing on the measure type, area positioning information, time planning information and operation equipment parameters to generate an operation strategy text containing complete operation instructions.

[0133] In this embodiment, firstly, the basic objectives and operational methods of the operation are determined based on the type of measure. Then, regional location information, time planning information, and operational equipment parameters are integrated into the specific operational process. For example, the operation strategy text clearly specifies the time, location, equipment, and parameters for implementing weather modification operations. Through strategy integration processing, an operation strategy text containing complete operational instructions is generated. The operation strategy text is a detailed operational guide covering all aspects of weather modification operations, providing clear and explicit operational guidelines for the operation execution equipment.

[0134] Step S146: The job strategy text is sent to the target job execution device, and the job execution device automatically triggers device startup, parameter setting and job execution process according to the job strategy text.

[0135] In this embodiment, the target operation execution equipment can be an operational aircraft, a ground-based launching device, etc. After receiving the operation strategy text, the operation execution equipment can automatically trigger equipment startup, parameter setting, and operation execution procedures according to the operation instructions in the text. For example, after receiving the operation strategy text, the operational aircraft can automatically start its engine and prepare for takeoff according to the takeoff time parameters in the text; during flight, it navigates according to the optimal flight route; upon reaching the operation area, it performs catalyst and heat dissipation operations according to catalyst type and single-batch dispersal parameters, heat dissipant type and release rate parameters, etc. The ground-based launching device will also perform corresponding parameter settings and launch operations according to the operation strategy text to ensure the smooth implementation of weather modification operations.

[0136] Throughout the process, various privacy protection and leak prevention technologies were employed for the collection and processing of privacy-sensitive data. For example, meteorological image data was encrypted during collection to prevent theft during transmission and storage. Anonymization and desensitization techniques were used to remove sensitive information, such as personally identifiable information, from the meteorological image data during analysis and processing. Simultaneously, a strict access control mechanism was established, ensuring that only authorized personnel could access and process the meteorological image data, guaranteeing its security and privacy. Similar protective measures were implemented for data in the historical operational database to prevent data leakage and misuse.

[0137] Figure 2 Schematic diagrams are shown of exemplary hardware and software components of an image recognition system 100 based on weather modification, which can implement the ideas of this application, according to some embodiments of this application. For example, processor 120 can be used in the image recognition system 100 based on weather modification and to perform the functions in this application.

[0138] The image recognition system 100 based on weather modification can be a general-purpose server or a special-purpose server; both can be used to implement the image recognition method based on weather modification of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0139] For example, the image recognition system 100 based on weather modification may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the image recognition system 100 based on weather modification may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The image recognition system 100 based on weather modification also includes an I / O interface 150 between the computer and other input / output devices.

[0140] For ease of explanation, only one processor is described in the image recognition system 100 based on weather modification. However, it should be noted that the image recognition system 100 based on weather modification in this application may also include multiple processors, and therefore the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the image recognition system 100 based on weather modification performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0141] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned image recognition method based on artificial weather modification is implemented.

[0142] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. An image recognition method based on weather modification, characterized in that, The method includes: Acquire a set of meteorological image data for the target area, wherein the set of meteorological image data includes multiple raw meteorological images arranged in time sequence; The original meteorological images are subjected to feature extraction processing to obtain the distribution characteristics and dynamic evolution characteristics of meteorological elements in each original meteorological image; Based on a preset weather impact analysis model, the distribution characteristics and dynamic evolution characteristics of the meteorological elements are analyzed to generate weather intervention decision characteristics of the original meteorological image. Based on the weather intervention decision characteristics, an artificial weather modification operation strategy is generated, and the operation strategy is sent to the operation execution equipment to trigger meteorological intervention operations; The weather impact analysis model, based on a preset model, performs weather impact analysis on the distribution characteristics and dynamic evolution characteristics of the meteorological elements, generating weather intervention decision features for the original meteorological image, including: The distribution characteristics of the meteorological elements are input into the spatial feature processing module of the weather impact analysis model. The local detail features and global distribution features of the meteorological elements in the spatial dimension are extracted by the convolutional neural network to generate a spatial feature representation. The dynamic evolution features are input into the time feature processing module of the weather impact analysis model, and the continuous evolution features and trend prediction features of meteorological elements in the time dimension are extracted through a recurrent neural network to generate a time feature representation. The spatial and temporal feature representations are fused, and weight parameters are assigned to features of different dimensions through an attention mechanism to generate a fused feature vector. The fused feature vector is input into the decision reasoning module of the weather impact analysis model. The fused feature vector is then processed by a fully connected neural network to perform nonlinear transformation, generating a set of potential intervention features that reflect the effects of different artificial weather modification measures. The feasibility assessment of the potential intervention feature set is performed, and the correlation between different intervention measures and meteorological response in historical operation data is combined to screen out effective intervention features that meet the current meteorological conditions. The effective intervention features are subjected to feature encoding to generate weather intervention decision features that include intervention type identifier, area of ​​effect, and timing of effect identifier; The step of inputting the dynamic evolution features into the time feature processing module of the weather impact analysis model, and extracting the continuous evolution features and trend prediction features of meteorological elements in the time dimension through a recurrent neural network to generate a time feature representation includes: The dynamic evolution features are arranged in chronological order as time series data including cloud movement trajectories, precipitation area expansion rates, and temperature gradient fluctuations. The time series data is input into the initial state unit of the recurrent neural network to initialize the initial memory state, which includes the cloud system movement direction, precipitation area baseline value, and temperature gradient baseline fluctuation amount of the previous time step. At each time step, perform a state update operation: calculate the directional deviation between the cloud system movement trajectory at the current time step and the cloud system movement direction in the initial memory state to generate cloud system motion stability characteristics; calculate the rate matching between the precipitation area expansion rate at the current time step and the area benchmark value in the initial memory state to generate precipitation development coordination characteristics; and calculate the fluctuation consistency between the temperature gradient fluctuation at the current time step and the benchmark fluctuation amount in the initial memory state to generate temperature field evolution persistence characteristics. The cloud system motion stability characteristics, precipitation development coordination characteristics, and temperature field evolution persistence characteristics are spliced ​​together to form the intermediate evolution characteristics of the current time step; The intermediate evolution features are fused with the hidden state of the previous time step using the state transition function of a recurrent neural network to generate an updated hidden state containing historical evolution information. The hidden state of continuous time steps is updated and processed by time-weighted accumulation. The weight value is set according to the interval between the time step and the current time using an exponential decay rule to generate trend prediction features that reflect long-term evolution trends. The updated hidden state and trend prediction features are input into the time feature integration layer. The difference features between the current evolution state and the long-term trend are extracted through time difference convolution operation to generate a time feature representation that simultaneously contains real-time evolution details and long-term trend correlation.

2. The image recognition method based on weather modification according to claim 1, characterized in that, The step of performing feature extraction processing on the original meteorological images to obtain the distribution characteristics and dynamic evolution characteristics of meteorological elements in each original meteorological image includes: The original meteorological images are time-aligned, and the time-aligned set of original meteorological images is spatially registered to obtain spatially aligned original meteorological images. The original spatially aligned meteorological image is subjected to noise suppression processing. An adaptive filtering algorithm is used to reduce random noise interference in the image while retaining the detailed information of key meteorological elements, resulting in a denoised meteorological image. The denoised meteorological image is subjected to contrast enhancement processing to obtain an enhanced meteorological image; Feature extraction is performed on the enhanced meteorological image to extract static distribution features reflecting the current meteorological state and dynamic evolution features reflecting changes over time. The static distribution features include cloud density distribution parameters, precipitation area boundary continuity parameters, and temperature gradient direction consistency parameters. The dynamic evolution features include cloud movement direction offset parameters between adjacent time points, precipitation area change rate parameters, and temperature gradient value fluctuation amplitude parameters. The static distribution features and dynamic evolution features are integrated to generate meteorological element distribution features and dynamic evolution features that contain spatial dimension distribution information and temporal dimension evolution information.

3. The image recognition method based on weather modification according to claim 2, characterized in that, The step of extracting features from the enhanced meteorological image includes extracting static distribution features reflecting the current meteorological state and dynamic evolution features reflecting changes over time, including: The enhanced meteorological image is subjected to cloud system region segmentation processing. A threshold segmentation algorithm is used to identify pixels in the enhanced meteorological image whose gray values ​​are in the cloud system feature range, and a cloud system mask image is generated. The pixel density distribution of the cloud region is calculated based on the cloud mask image. The number of cloud pixels per unit area is counted as the cloud density distribution parameter. Edge detection processing is performed on the enhanced meteorological image to extract the boundary contour lines between precipitation and non-precipitation areas. The continuous length ratio of the boundary contour lines is calculated as the boundary continuity parameter of the precipitation area. Gradient calculation processing is performed on the enhanced meteorological image to extract the temperature gradient direction of each pixel. The number ratio of gradient points with the same direction is counted as the temperature gradient direction consistency parameter. The enhanced meteorological image of the previous moment is used as a reference image. The enhanced meteorological image of the current moment and the reference image are processed by optical flow field calculation to obtain the displacement vector set of the pixels in the cloud system region. Based on the displacement vector set, the angle between the overall movement direction of the cloud system and the movement direction of the previous moment is calculated as the offset parameter of the cloud system movement direction. The ratio of the difference between the current precipitation area and the previous precipitation area to the area at the previous time is used as the parameter of the rate of change of precipitation area. The average of the absolute differences between the current temperature gradient value and the temperature gradient value at the corresponding position at the previous time is used as the parameter for the fluctuation range of the temperature gradient value.

4. The image recognition method based on weather modification according to claim 1, characterized in that, The step involves inputting the fused feature vector into the decision reasoning module of the weather impact analysis model, and performing nonlinear transformation processing on the fused feature vector through a fully connected neural network to generate a set of potential intervention features reflecting the effects of different artificial weather modification measures, including: The fused feature vector is input into the first fully connected layer of the decision reasoning module, and processed by linear transformation and activation function to generate an intermediate feature vector containing basic intervention features; The intermediate feature vector is input into the second fully connected layer of the decision reasoning module, and through feature recombination and nonlinear mapping processing, a high-level feature vector containing extended intervention features is generated. The high-level feature vector is input into the third fully connected layer of the decision reasoning module. Through feature filtering and dimensionality compression, a key feature vector containing core intervention features is generated. The key feature vectors are normalized, and the normalized key feature vectors are input into the output layer of the decision reasoning module. Through the classification activation function, probability distribution vectors corresponding to different artificial weather modification measures are generated. Based on the probability distribution vector, measure types with probability values ​​exceeding a preset threshold are extracted as potential intervention features, generating a set of potential intervention features containing measure type identifiers and corresponding probability values.

5. The image recognition method based on weather modification according to claim 1, characterized in that, The feasibility assessment of the potential intervention feature set, combined with the correlation between different intervention measures and meteorological responses in historical operational data, filters out effective intervention features that meet the current meteorological conditions, including: Extract historical meteorological sample data from a pre-set historical operation database that matches the current distribution characteristics and dynamic evolution characteristics of meteorological elements; The historical meteorological sample data is subjected to statistical processing of the effects of intervention measures, and meteorological response indicators corresponding to each intervention measure are calculated. The meteorological response indicators include precipitation increment parameters, cloud dissipation rate parameters, and temperature regulation amplitude parameters. Calculate the similarity value between the current potential intervention feature and the historical sample feature, and perform a weighted average of the meteorological response index of the historical meteorological sample data based on the similarity value to generate the predicted response index corresponding to the current potential intervention feature; Potential intervention features whose predicted response indicators meet the preset response indicator threshold range are selected as candidate intervention features; The candidate intervention features are subjected to conflict detection processing to exclude combinations of intervention features that would cause the meteorological effects to cancel each other out when implemented simultaneously, and retain the non-conflicting candidate intervention features as effective intervention features.

6. The image recognition method based on weather modification according to claim 1, characterized in that, The step of generating an artificial weather modification strategy based on the weather intervention decision characteristics and sending the strategy to the execution equipment to trigger meteorological intervention includes: The intervention type identifier in the weather intervention decision features is analyzed to determine the types of artificial weather modification measures that need to be implemented. These measures include rain enhancement operations, cloud dispersal operations, and temperature regulation operations. The effective area range in the weather intervention decision features is analyzed, and the geographical boundary coordinates and center coordinates of the area where intervention measures need to be implemented are extracted to generate regional positioning information; The timing identifier in the weather intervention decision features is analyzed, and combined with the timestamp information of the meteorological image data set, the start time and duration of the intervention measures are determined to generate time planning information. The operation parameters are configured for the aforementioned measure type, regional positioning information, and time planning information. Corresponding operation equipment parameters are matched for each measure type. The operation equipment parameters include catalyst spreading rate parameters, operation height parameters, and operation route parameters. The measures type, area location information, time planning information, and work equipment parameters are integrated and processed to generate a work strategy text containing complete operation instructions; The job strategy text is sent to the target job execution device, which automatically triggers device startup, parameter settings, and job execution process based on the job strategy text.

7. The image recognition method based on weather modification according to claim 6, characterized in that, The process of configuring operational parameters for the measure type, regional location information, and time planning information, and matching corresponding operational equipment parameters for each measure type, includes: For each type of rain enhancement operation, consult the preset rain enhancement operation parameter configuration table, and select the corresponding catalyst type and single seeding amount parameters based on the size of the area to be affected and the cloud development stage at the time of the operation. Based on the center coordinates and boundary coordinates of the operational area, and combined with the maximum range parameters of the operational aircraft, the optimal flight route from the airport to the operational area is planned, and the turning point coordinates and flight altitude parameters in the optimal flight route are calculated. For each type of cloud suppression operation, the preset cloud suppression operation parameter configuration table is queried, and the catalyst application interval and application concentration parameters are determined based on the cloud density distribution parameters and temperature gradient direction consistency parameters within the action area. Based on the start time and duration of the operation, calculate the time difference between the aircraft taking off from the airport and arriving at the operation area, and adjust the takeoff time parameters accordingly. For each type of temperature regulation operation, consult the preset temperature regulation operation parameter configuration table, and select the corresponding heat dissipant type and release rate parameter based on the temperature gradient fluctuation range parameter of the affected area and the area change rate parameter of the precipitation area. By combining equipment parameters from different measures, a set of operational equipment parameters is generated, including catalyst type and single-batch application rate parameters, optimal flight path, takeoff time parameters, heat dissipant type and release rate parameters.

8. The image recognition method based on weather modification according to claim 1, characterized in that, The step of inputting the distribution characteristics of the meteorological elements into the spatial feature processing module of the weather impact analysis model, and extracting the local detail features and global distribution features of the meteorological elements in the spatial dimension through a convolutional neural network to generate a spatial feature representation includes: The distribution characteristics of the meteorological elements are spatially decomposed and divided into cloud structure sub-features, precipitation area sub-features and temperature field distribution sub-features according to the meteorological element type. The cloud structure features are input into the first branch of the convolutional neural network, and the continuity features of the cloud edge contour and the gradient distribution features of the internal pixel density are extracted through morphological convolution kernels. The sub-features of the precipitation area are input into the second branch of the convolutional neural network. An adaptive threshold convolutional kernel is used to process the boundary transition zone between the precipitation area and the non-precipitation area, and to extract the boundary clarity features and the uniformity features of the pixel values ​​inside the area. The temperature field distribution sub-features are input into the third branch of the convolutional neural network. The spatial orientation features of the temperature gradient are processed by the directional convolution kernel to extract the clustering features of the same gradient region and the dispersion features of the opposite gradient region. Cross-feature association processing is performed on the local detail features output by the first branch, the second branch, and the third branch to calculate the spatial overlap parameter between cloud edge continuity and precipitation boundary clarity, as well as the matching parameter between temperature gradient aggregation and cloud internal density gradient. The cross-feature association processing result is multiplied element-wise with the original meteorological element distribution characteristics to generate enhanced local features that contain spatial synergistic relationships between elements. The enhanced local features are spatially compressed by global pooling to extract global distribution features that reflect the distribution ratio and centroid position of cloud systems, precipitation, and temperature fields in the overall region. The enhanced local features and global distribution features are input into the spatial feature synthesis layer. An attention mechanism is used to assign weight coefficients related to the global distribution to the local features of different elements, thereby generating a spatial feature representation that simultaneously contains the internal details of the elements and the collaborative relationships between the elements.

9. An image recognition system based on weather modification, characterized in that, The image recognition system based on weather modification includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the image recognition method based on weather modification as described in any one of claims 1-8.

Images