Method for refining low-resolution radar reflectivity factor based on phased array radar

Abstract

The application discloses a low-resolution radar reflectivity factor refinement method based on a phased array radar, relates to the radio technology field, and comprises the following steps: time alignment is performed on the cleaned phased array radar linear domain reflectivity factor and the cleaned low-resolution radar linear domain reflectivity factor, and simulation is performed based on an interpolation model; space geometry matching is performed on the cleaned phased array radar linear domain reflectivity factor and the cleaned low-resolution radar linear domain reflectivity factor, and simulation is performed on the phased array radar linear domain reflectivity factor dataset based on a multi-base radar detection data conversion model and a coarse beam simulation operator; a residual network model based on a self-attention mechanism is trained and iteratively optimized to obtain an optimized model, the cleaned low-resolution radar linear domain reflectivity factor data are input into the optimized model, and low-resolution radar spatial refined reflectivity factor data are output. The application can improve the authenticity, physical consistency and business availability of the refined reconstruction result.

Description

Technical Field

[0001] This invention relates to the field of radio technology, and more particularly to a method for refining the reflectivity factor of low-resolution radar based on phased array radar. Background Technology

[0002] Reflectivity factor from weather radar is a crucial observational parameter characterizing the spatial distribution and intensity variations of precipitation particles, and it is also a key foundational data for monitoring severe convection and short-term nowcasting. However, existing operational weather radars, limited by antenna beamwidth, range resolution, and volumetric scanning systems, typically acquire spatially coarse reflectivity factor data that struggles to accurately depict the small-scale structural features of rapidly evolving weather processes. In contrast, phased-array radars offer higher spatial resolution and faster scanning capabilities, providing more refined information on precipitation echo structure. Therefore, how to leverage high-spatial-resolution phased-array radar data to enhance the refinement of low-spatial-resolution operational radar reflectivity factor data has become an important research direction for the in-depth application of weather radar data.

[0003] Currently, methods for spatial refinement of reflectivity factors based on radar observation data mainly include interpolation resampling methods, degradation model-based inversion recovery methods, multi-temporal or multi-source observation fusion methods, and machine learning or deep learning-based super-resolution reconstruction methods. Interpolation resampling methods primarily fill in numerical values ​​between existing pixels using bilinear interpolation, bicubic interpolation, spline interpolation, etc. While simple to implement, they do not fundamentally restore new physical details, easily leading to smoothing of echo boundaries and weakening of extrema. Degradation model-based inversion recovery methods, while considering factors such as radar beam ambiguity and range response to some extent, typically rely on idealized degradation assumptions, making it difficult to accurately describe the complex observation process of real operational radar, and are highly sensitive to parameter settings. Multi-temporal or multi-source observation fusion methods can utilize more information for enhancement, but when precipitation systems evolve rapidly or different radar observation geometric conditions vary significantly, temporal and spatial mismatches easily occur, affecting the reliability of the refinement results. In recent years, with the development of deep learning methods, existing technologies have attempted to use structures such as convolutional neural networks, residual networks, or generative adversarial networks to achieve super-resolution reconstruction of radar reflectivity factor data. While these methods improve spatial detail recovery to some extent, they still have the following shortcomings: First, many methods construct training samples by manually downsampling the same radar data, resulting in significant differences between the obtained samples and real heterogeneous radar observations, making it difficult to accurately reflect the intrinsic differences between different radars in beamwidth, range resolution, elevation coverage, and scanning systems. Second, existing methods often fail to fully consider the time synchronization issues and spatial detection volume differences between different radars, leading to spatiotemporal registration errors between coarse input observations and fine output labels. Third, existing methods typically lack physical consistency constraints for degradation from high-resolution to low-resolution observations, meaning that although the model output has higher numerical resolution, it may not be consistent with real operational radar observations after degradation back to a coarse scale, resulting in insufficient physical reliability. Fourth, some methods use auxiliary information that cannot be obtained during the inference phase during the training phase, causing inconsistencies between training and deployment conditions, making it difficult to directly apply to actual operational scenarios.

[0004] Therefore, this invention develops a method for refining the low-resolution radar reflectivity factor based on phased array radar to solve the above problems. Summary of the Invention

[0005] This invention proposes a method for refining the low-resolution radar reflectivity factor based on phased array radar, in order to solve the problems of existing technologies that are difficult to accurately reflect the intrinsic differences between different radars, have spatiotemporal registration errors, lack physical reliability, and are difficult to directly apply to actual business scenarios.

[0006] The present invention achieves the above objectives through the following technical solutions:

[0007] This invention provides a method for refining the reflectivity factor of low-resolution radar based on phased array radar, comprising:

[0008] Data on reflectivity factors, operational status, and ground clutter occurrence probability of two geographically adjacent radars are collected. The two radars are a low-resolution radar and a phased array radar, respectively.

[0009] Quality control was performed on the reflectivity factor data of the two radars to obtain the cleaned linear domain reflectivity factors of the two radars.

[0010] Based on the operational status data of the two radars, the linear domain reflectivity factor of the phased array radar is time-aligned with that of the low-resolution radar. Then, simulation is performed based on the interpolation model to obtain the simulated value of the linear domain reflectivity factor of the phased array radar that is time-aligned with the linear domain reflectivity factor of the low-resolution radar.

[0011] Based on the operational status data of the two radars, the simulated values ​​of the linear domain reflectivity factor of the phased array radar are spatially geometrically matched with the linear domain reflectivity factor of the low-resolution radar to obtain a phased array radar linear domain reflectivity factor dataset that spatially matches the low-resolution radar linear domain reflectivity factor. Based on the multi-static radar detection data conversion model and coarse beam simulation operator, the phased array radar linear domain reflectivity factor dataset is simulated to obtain a phased array radar linear domain reflectivity factor simulation dataset with the same spatial position and resolution as the low-resolution radar linear domain reflectivity factor.

[0012] Low-resolution radar linear domain reflectivity factor data is used as training input, and the phased array radar linear domain reflectivity factor dataset is used as training output. The residual network model based on the self-attention mechanism is trained and iteratively optimized in combination with the phased array radar linear domain reflectivity factor simulation dataset to obtain an optimized model. The cleaned low-resolution radar linear domain reflectivity factor data is input into the optimized model, and the low-resolution radar spatially refined reflectivity factor data is output.

[0013] Furthermore, the reflectivity factor data includes radial distance, azimuth angle, and antenna elevation angle; the operational status data includes the polar coordinate spatial location of the radar site, which includes the longitude, latitude, and altitude of the antenna feed of the radar site; the radar's low radial distance resolution and low beamwidth; the radar's volume scan period; and the ground clutter occurrence probability data refers to the frequency data of ground clutter occurrence collected at the polar coordinate spatial location.

[0014] Furthermore, quality control was performed on the reflectivity factor data of the two radars to obtain the cleaned linear domain reflectivity factors of the two radars, including:

[0015] Based on the reflectivity factor data of the two radars, and based on the reliable threshold range of weather echoes, signal-to-noise ratio calculation and signal-to-noise ratio threshold, the basic mask matrix and signal-to-noise ratio decision mask matrix of the weather echoes of the two radars are calculated respectively.

[0016] Based on the reflectivity factor data of the two radars, and based on the clutter texture decision threshold and the clutter vertical consistency decision threshold, the clutter texture decision mask matrix and the clutter vertical consistency decision mask matrix of the two radars are calculated respectively.

[0017] Based on the clutter occurrence probability data of the two radars, and based on the clutter probability judgment threshold, the clutter probability map identification mask matrix of the two radars is calculated respectively.

[0018] Based on the clutter texture decision mask matrix, clutter vertical consistency decision mask matrix, and clutter probability map identification mask matrix of the two radars, the clutter combination decision mask matrix is ​​obtained by logical operation combination calculation.

[0019] Based on the ground clutter combination decision mask matrix, signal-to-noise ratio decision mask matrix, and weather echo basic mask matrix of the two radars, the effective echo decision mask matrices of the two radars are constructed respectively.

[0020] Based on the effective echo decision mask matrix and reflectivity factor data of the two radars, the cleaned linear domain reflectivity factors of the two radars are generated respectively.

[0021] Furthermore, based on the reflectivity factor data of the two radars, and considering the reliable threshold range of weather echoes, signal-to-noise ratio (SNR) calculation, and SNR threshold, the basic mask matrix and SNR decision mask matrix of the weather echoes for the two radars are calculated respectively, including:

[0022] Set the confidence threshold range for weather echoes, which includes the minimum and maximum confidence echo thresholds;

[0023] Determine whether the value of each spatial location point in the reflectivity factor data is within the reliable threshold range of weather echoes. Mark spatial location points that do not fall within the threshold range or have empty values ​​as 0, and mark the rest as 1, thereby generating the basic mask matrix of weather echoes.

[0024] The total number of distances required for statistical radial noise base is set to N_noise, and the quiet zone threshold is 0. Then, N_noise consecutive data points are extracted from the farthest radial distance of the reflectivity factor data, and all data points less than the quiet zone threshold in these N_noise data points are extracted to form logarithmic domain noise data.

[0025] Each reflectance factor in the logarithmic domain noise data is transformed using the logarithmic-to-linear formula to form linear domain noise data. The arithmetic mean of the linear domain noise data is then taken as the noise floor value.

[0026] The reflectivity factor data is converted using a logarithmic-to-linear formula to form linear domain radar reflectivity factor data. The linear domain signal-to-noise ratio data is then calculated based on the linear domain radar reflectivity factor data and the noise floor value.

[0027] Set a signal-to-noise ratio (SNR) threshold, mark the spatial locations of data with SNR values ​​greater than the threshold in the linear domain as 1, and mark the rest as 0, thereby generating an SNR decision mask matrix.

[0028] Furthermore, based on the reflectivity factor data of the two radars, and using the clutter texture decision threshold and the clutter vertical consistency decision threshold, the clutter texture decision mask matrix and the clutter vertical consistency decision mask matrix of the two radars are calculated respectively, including:

[0029] For reflectivity factor data, set the window width and construct the local analysis window size. Take a certain spatial location point in the reflectivity factor data as the center, extract the reflectivity factor data in the local analysis window around it in the echo of the same antenna elevation layer, and calculate the texture feature value of the center point according to the texture calculation formula.

[0030] The center point in the above processing is used to traverse all spatial location points in the reflectivity factor data in turn to form the texture data of the reflectivity factor.

[0031] Set a ground clutter texture decision threshold, mark the spatial location points corresponding to the data in the texture data that are greater than the ground clutter texture decision threshold as 1, and mark the rest as 0, and generate a ground clutter texture decision mask matrix.

[0032] Then, based on the reflectivity factor data, the reflectivity factor data of adjacent lower antenna elevation angle layers and higher antenna elevation angle layers in the same volume scanning period are selected, and the difference between the reflectivity factor data of the lower antenna elevation angle layer and the higher antenna elevation angle layer is calculated. The reflectivity factor data of all adjacent antenna elevation angle layers are processed in sequence according to this method to generate a dataset of differences in reflectivity factors of adjacent elevation angle layers.

[0033] Set a vertical consistency decision threshold for ground clutter. Mark the spatial location points corresponding to the data in the difference dataset that are greater than the vertical consistency decision threshold for ground clutter as 1, and mark the rest as 0, thereby generating a vertical consistency decision mask matrix for ground clutter.

[0034] Furthermore, based on the clutter occurrence probability data from the two radars and the clutter probability judgment threshold, the clutter probability map identification mask matrices for the two radars are calculated respectively. Based on the clutter texture decision mask matrix, clutter vertical consistency decision mask matrix, and clutter probability map identification mask matrix of the two radars, a clutter combination decision mask matrix is ​​obtained through logical operation combination calculation. Based on the clutter combination decision mask matrix, signal-to-noise ratio decision mask matrix, and weather echo basic mask matrix of the two radars, the effective echo decision mask matrices for the two radars are constructed respectively, including:

[0035] Set a ground clutter probability judgment threshold based on ground clutter occurrence probability data;

[0036] The spatial location points corresponding to the ground clutter probability determination threshold in the ground clutter occurrence probability data are marked as 1, and the rest are marked as 0, thereby generating a ground clutter probability map identification mask matrix.

[0037] Based on the clutter texture decision mask matrix, clutter vertical consistency decision mask matrix, and clutter probability map identification mask matrix from the two radars, a combined clutter decision mask matrix is ​​obtained through logical operations. The formula for the combined logical operations is as follows:

[0038] Ground clutter combination decision mask matrix = [Ground clutter probability map recognition mask matrix ∨ (Ground clutter texture decision mask matrix ∧ Ground clutter vertical consistency decision mask matrix)], where ∨ represents a logical OR operation and ∧ represents a logical AND operation;

[0039] The effective echo decision mask matrix is ​​constructed based on the ground clutter combined decision mask matrix, the signal-to-noise ratio decision mask matrix, and the weather echo fundamental mask matrix. The calculation formula is as follows:

[0040] Effective echo decision mask matrix = Echo base mask matrix × Signal-to-noise ratio decision mask matrix × (1 - Ground clutter combination decision mask matrix).

[0041] Furthermore, based on the operational status data of the two radars, the linear domain reflectivity factor of the phased array radar is time-aligned with that of the low-resolution radar. Then, simulations are performed using an interpolation model to obtain simulated values ​​of the phased array radar linear domain reflectivity factor that are time-aligned with the low-resolution radar's linear domain reflectivity factor. These simulated values ​​include:

[0042] Extract a complete timestamp sequence corresponding to a continuous volume scan cycle from the timestamp sequence obtained from the timestamp sequence of the antenna elevation angle echo of each layer within each volume scan cycle.

[0043] In the timestamp sequence of the antenna elevation angle echo acquired in each volume scanning cycle, the two time points with the smallest time difference between the start and end times of the timestamp sequence corresponding to the continuous volume scanning cycle are selected respectively to construct a scanning timestamp sequence that is time-aligned with the timestamp sequence corresponding to the continuous volume scanning cycle.

[0044] For a given antenna elevation angle, select a time value corresponding to the specified antenna elevation angle scan from the timestamp sequence corresponding to the continuous volume scan period, and extract the low-resolution radar linear domain reflectivity factor corresponding to the time value.

[0045] At the same time, multiple time values ​​corresponding to multiple scans at the same specified antenna elevation angle are extracted, and the linear domain reflectivity factor of the phased array radar corresponding to the multiple time values ​​is extracted.

[0046] The phased array radar linear domain reflectivity factors corresponding to the multiple time values ​​and the low-resolution radar linear domain reflectivity factor corresponding to one time value are input into the Lagrange interpolation model, and the simulated value of the phased array radar linear domain reflectivity factor that is precisely time-aligned with the low-resolution radar linear domain reflectivity factor at the specified antenna elevation angle is output.

[0047] Applying the above steps to the echoes of all antenna elevation layers of the low spatial resolution radar, we obtain simulated values ​​of the linear domain reflectivity factor of the phased array radar that are time-aligned with the linear domain reflectivity factor of the low-resolution radar at all antenna elevation angles.

[0048] Furthermore, based on the operational status data of the two radars, the simulated values ​​of the linear domain reflectivity factor of the phased array radar are spatially geometrically matched with the linear domain reflectivity factor of the low-resolution radar to obtain a dataset of linear domain reflectivity factors of the phased array radar that spatially matches the linear domain reflectivity factor of the low-resolution radar, including:

[0049] Take any spatial location point U as the center point from the linear domain reflectivity factor data of low-resolution radar, and extract the radial distance, azimuth angle and antenna elevation angle corresponding to the spatial location point by combining the polar coordinate spatial location of the low-resolution radar echo. Combine the collected radial distance resolution and beamwidth of the low-resolution radar to construct the low spatial resolution volume of the low spatial resolution radar centered on point U.

[0050] Based on the radial range resolution and beamwidth of the phased array radar, the phased array radar echo volume contained within the low spatial resolution volume is initially screened out.

[0051] Take any point in the phased array radar echo volume, and combine it with the polar coordinate spatial position information of the phased array radar echo to extract the radial distance, azimuth angle and antenna elevation angle of the arbitrary point, and calculate the distance difference, azimuth difference and elevation angle difference between the spatial position point and the arbitrary point.

[0052] Based on the set range half-width threshold, azimuth half-width threshold, and elevation half-width threshold, the region half-width overlap determination conditions are set, including the absolute value of the range difference not being greater than the range half-width threshold, the absolute value of the azimuth difference not being greater than the azimuth half-width threshold, and the elevation difference not being greater than the elevation half-width threshold. When any point simultaneously satisfies the above three inequality determination conditions, and the mask matrix of the effective echo decision corresponding to any point is 1, then it is determined that any point has effectively overlapped with the low spatial resolution volume.

[0053] The above steps traverse all points in the phased array radar echo volume and filter out the effective echo data of the phased array radar contained in the low spatial resolution volume.

[0054] By traversing all spatial location points in the low-resolution radar linear domain reflectivity factor data, a phased array radar linear domain reflectivity factor dataset that spatially matches the low-resolution radar linear domain reflectivity factor is constructed.

[0055] Furthermore, based on the multi-static radar detection data conversion model and coarse beam simulation operator, the linear domain reflectivity factor dataset of the phased array radar is simulated to obtain a simulated dataset of the linear domain reflectivity factor of the phased array radar with the same spatial location and resolution as the low-resolution radar linear domain reflectivity factor, including:

[0056] Based on the linear domain reflectivity factor data of low-resolution radar, any spatial point is selected from the polar coordinate spatial location of the low-resolution radar echo.

[0057] Based on low-resolution radar linear domain reflectivity factor data, a data set corresponding to the spatial geometric matching of the spatial point is extracted from the phased array radar linear domain reflectivity factor dataset. Combined with the polar coordinate spatial position of the phased array radar echo, the polar coordinate spatial position information corresponding to the data set corresponding to the spatial geometric matching of the spatial point is extracted.

[0058] Then, the spatial position information of the two radars, as well as the polar coordinate spatial position information corresponding to the data set that corresponds to the spatial geometric matching of the spatial point, are input into the multi-static radar detection data conversion model. The model is run to obtain the spatial position information of the data set that corresponds to the spatial geometric matching of the spatial point in the low-resolution radar polar coordinate system.

[0059] Based on the spatial position information of the data set corresponding to the spatial geometric matching of the spatial point in the low-resolution radar polar coordinate system, a coarse beam simulation operator based on three dimensions—radial range, azimuth, and elevation—is constructed. Specifically, in the low spatial resolution radar polar coordinate system, the degree of deviation of each data point in the data set corresponding to the spatial geometric matching of the spatial point from the spatial point in the three dimensions of radial range, azimuth, and antenna elevation is calculated. The degree of deviation in radial range is substituted into the radial range response function to obtain the weighted value in the range direction. The degree of deviation in azimuth and antenna elevation is substituted into the antenna main lobe and side lobe radii. In the radiation energy distribution function, weighted values ​​are obtained in the azimuth and antenna elevation directions, respectively. Then, the total weighted values ​​in the radial distance, azimuth, and elevation are calculated and normalized to obtain the normalized total weighted value. The product of the total weighted value and the data set corresponding to the spatial geometric matching of the spatial point is used as the low-resolution radar linear domain reflectivity factor simulation data at the spatial point. By traversing all points of the cleaned low-resolution radar linear domain reflectivity factor data according to the above processing flow, a phased array radar linear domain reflectivity factor simulation dataset with the same spatial location and spatial resolution as the cleaned low-resolution radar linear domain reflectivity factor data can be obtained.

[0060] Furthermore, low-resolution radar linear domain reflectivity factor data is used as training input, and the phased array radar linear domain reflectivity factor dataset is used as training output. The residual network model based on a self-attention mechanism is then trained and iteratively optimized using the simulated phased array radar linear domain reflectivity factor dataset to obtain an optimized model, including:

[0061] Low-resolution radar linear domain reflectivity factor data is used as the input dataset for model training, and phased array radar linear domain reflectivity factor data is used as the output dataset for training the residual network model based on the attention mechanism. The residual network model based on the self-attention mechanism is then trained.

[0062] The total loss function is constructed based on the low-resolution radar linear domain reflectivity factor data, the phased array radar linear domain reflectivity factor dataset, the model prediction of the residual network model based on the self-attention mechanism, and the simulated dataset of the phased array radar linear domain reflectivity factor. The total loss function includes a weighted sum of fine-scale supervision loss, coarse-scale consistency loss function, physical consistency loss, and real coarse observation consistency loss.

[0063] The parameters of the residual network model based on the attention mechanism are iteratively optimized and trained using the total loss function to obtain the optimized model.

[0064] The beneficial effects of this invention are as follows:

[0065] This invention proposes a method for refining low-resolution radar reflectivity factors based on phased array radar. It utilizes collaborative observation data from spatially adjacent high-spatial-resolution phased array radars and low-spatial-resolution operational radars. First, high-quality coarse-to-fine beam pairing samples are constructed through quality control, precise temporal matching, and spatial geometric correspondence between coarse and fine beams. Then, a spatial geometric mapping strategy for coarse and fine beams is designed, and a coarse beam simulation operator based on radial distance, azimuth, and elevation is constructed, achieving simulation of low-spatial-resolution data from high-spatial-resolution data. Finally, by combining a deep learning model and physical consistency constraints, a refined reconstruction of low-resolution radar reflectivity factor data from high-spatial-resolution reflectivity factor data is achieved. Compared with existing technologies, this invention provides a method for refining low-resolution radar reflectivity factors that is suitable for real heterogeneous radar observation conditions, while also considering spatiotemporal matching accuracy, spatial geometric correspondence, radar beam physical degradation processes, and the advantages of deep learning modeling. This improves the realism, physical consistency, and operational usability of the refined reconstruction results. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the low-resolution radar reflectivity factor refinement method based on phased array radar proposed in this application.

[0067] Figure 2 This is a flowchart illustrating the implementation of the phased array radar and low-resolution radar signal-to-noise ratio and echo fundamental mask matrix generation of the present invention.

[0068] Figure 3 This is a flowchart illustrating the implementation of the present invention for generating linear domain reflectivity factor data for cleaned phased array radar and low-resolution radar.

[0069] Figure 4 This is a flowchart illustrating the implementation of the present invention for generating simulated values ​​of the linear domain reflectivity factor of a phased array radar that are precisely time-aligned with the linear domain reflectivity factor of a low-resolution radar.

[0070] Figure 5 This is a flowchart illustrating the implementation of the present invention for generating a phased array radar linear domain reflectivity factor dataset that is spatially matched with the linear domain reflectivity factor of a low-resolution radar.

[0071] Figure 6 This is a flowchart illustrating the implementation process of generating a simulated dataset of linear domain reflectivity factors for phased array radars with the same spatial location and resolution as those for low-resolution radars.

[0072] Figure 7 This is a flowchart illustrating the implementation of the overall estimation model training set construction and loss function setting for this invention. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0074] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0075] like Figure 1 As shown, this invention specifically relates to a method for refining the reflectivity factor of low-resolution radar based on phased array radar, which specifically includes the following steps:

[0076] Step 1: Collect reflectivity factor data, operational status data, and ground clutter occurrence probability data of two geographically adjacent radars, namely a low-resolution radar and a phased array radar.

[0077] Collect low-resolution radar echo datasets, operational status datasets, and ground clutter occurrence probability datasets; collect mobile high spatial resolution phased array radar echo datasets, radar operational status datasets, and ground clutter occurrence probability datasets that are geographically adjacent to the low-resolution radars.

[0078] The low-resolution radar echo dataset refers to the radar reflectivity factor data dBZ0_wea detected at polar coordinate spatial locations (r0_wea, phi0_wea, seat0_wea), where r0_wea is the radial distance, phi0_wea is the azimuth angle, seat0_wea is the antenna elevation angle, and the timestamp sequence t0_wea of ​​the antenna elevation angle echo acquired in each layer within each volume scanning cycle.

[0079] The low-resolution radar operation status dataset mainly includes: the spatial location information of the radar site as P_wea = (Lon_wea, Lat_wea, H_wea), where Lon_wea, Lat_wea, and H_wea are the longitude, latitude, and altitude of the antenna feed, respectively; the radar's low radial range resolution delt_R_wea = 250 meters and low beamwidth delt_sita_wea = 1.0 degree; and the radar volume scan period delt_T_wea = 6 minutes.

[0080] The low-resolution radar clutter occurrence probability dataset refers to the frequency dataset P_Clutter_wea of ​​clutter occurrences collected at polar coordinate spatial locations (r0_wea, phi0_wea, sita0_wea).

[0081] The aforementioned mobile high spatial resolution phased array radar echo dataset refers to the radar reflectivity factor data dBZ0_par detected at polar coordinate spatial locations (r0_par, phi0_par, sita0_par), where r0_par is the radial distance, phi0_par is the azimuth angle, sita0_par is the antenna elevation angle, and the timestamp sequence t0_par obtained from the antenna elevation angle echo of each layer within each volume scanning cycle;

[0082] The aforementioned mobile high spatial resolution phased array radar operation status dataset mainly includes: the spatial location information of the radar site as P_par = (Lon_par, Lat_par, H_par), where Lon_par, Lat_par, and H_par are the longitude, latitude, and altitude of the antenna feed, respectively; the radar's high radial range resolution delt_R_par = 30 meters and high beamwidth delt_sita_par = 0.5 degrees; and the radar volume scanning period delt_T_par = 50 seconds.

[0083] The aforementioned mobile high spatial resolution phased array radar ground clutter occurrence probability dataset refers to the ground clutter frequency dataset P_Clutter_par established at polar coordinate spatial locations (r0_par, phi0_par, sita0_par);

[0084] Step 2: Based on the reflectivity factor data of the two radars, and based on the reliable threshold range of weather echoes, signal-to-noise ratio calculation, and signal-to-noise ratio threshold, calculate the basic mask matrix and signal-to-noise ratio decision mask matrix of the weather echoes for the two radars respectively.

[0085] First, the confidence threshold range for weather echoes is set to dBZ_rang = [dBZmin = -5, dBZmax = 75], where dBZmin and dBZmax are the lowest and highest confidence echo thresholds, respectively. Second, it is determined whether the value of each spatial location point in the phased array radar reflectivity factor data dBZ0_par collected in step (1) is within the range of dBZ_rang. Spatial location points that do not fall within the threshold range or have empty values ​​are marked as 0, and the rest are marked as 1, thereby generating the phased array radar weather echo basic mask matrix M_basic_par. Then, the total number of distance libraries required for statistical radial noise base is set to N_noise = 20, and the quiet zone threshold is set to dBZ_quiet = 0. Then, continuous N_noise is extracted from the farthest radial distance of the phased array radar reflectivity factor data dBZ0_par collected in step (1). N_noise data points are collected, and all data points less than dBZ_quiet are extracted to form a logarithmic domain noise dataset dBZ_noise_set. Then, each reflectivity factor data in the logarithmic domain noise dataset dBZ_noise_set is transformed according to the logarithmic-to-linear formula to form a linear domain noise dataset Z_noise_set. The arithmetic mean of the dataset Z_noise_set is then taken as the noise floor value, denoted as Z_noise. After that, the phased array radar reflectivity factor data dBZ0_par collected in step (1) is transformed according to the logarithmic-to-linear formula to form a linear domain radar reflectivity factor dataset Z0_par. Then, SNR_par is calculated according to SNR_par=Z0_par / (Z_noise+ε), where ε is a very small positive number, to obtain the linear domain signal-to-noise ratio dataset SNR_par. Next, set the signal-to-noise ratio threshold to Thre_SNR=2, mark the spatial location points corresponding to the data in the dataset SNR_par that are greater than the signal-to-noise ratio threshold Thre_SNR as 1, and mark the rest as 0, thereby generating the phased array radar signal-to-noise ratio decision mask matrix M_SNR_par; finally, replace the phased array radar reflectivity factor data dBZ0_par in this step with the low-resolution radar reflectivity factor data dBZ0_wea collected in step (1), and then execute this step again to obtain the low-resolution radar weather echo basic mask matrix M_basic_wea and the low-resolution radar signal-to-noise ratio decision mask matrix M_SNR_wea.

[0086] The formula for converting logarithmic to linear is: Z = 10(dBZ / 10), where Z and dBZ are the reflectance factor data in the linear and logarithmic domains, respectively.

[0087] Step 3: Based on the reflectivity factor data of the two radars, and using the clutter texture decision threshold and the clutter vertical consistency decision threshold, calculate the clutter texture decision mask matrix and the clutter vertical consistency decision mask matrix for the two radars respectively. Based on the clutter occurrence probability data of the two radars, and using the clutter probability judgment threshold, calculate the clutter probability map recognition mask matrix for the two radars respectively. Based on the clutter texture decision mask matrix, the clutter vertical consistency decision mask matrix, and the clutter probability map recognition mask matrix of the two radars, calculate the clutter combination decision mask matrix using logical operations. Based on the clutter combination decision mask matrix, the signal-to-noise ratio decision mask matrix, and the weather echo basic mask matrix of the two radars, construct the effective echo decision mask matrix for the two radars respectively. Based on the effective echo decision mask matrix and the reflectivity factor data of the two radars, generate the cleaned linear domain reflectivity factor for the two radars respectively.

[0088] like Figure 3 First, based on the clutter probability data P_Clutter_par collected in step (1), the clutter probability judgment threshold Thre_P_Clutter=0.4 is set. The spatial location points corresponding to the data in P_Clutter_par that are greater than the judgment threshold Thre_P_Clutter are marked as 1, and the rest are marked as 0, thereby generating the clutter probability map recognition mask matrix M_ClutterMap_par for the phased array radar. Then, for the reflectivity factor data dBZ0_par collected in step (1), the window width is set to N=5, and the local analysis window size is N×N. Taking a certain spatial location point in dBZ0_par as the center, the reflectivity factor data in the local analysis window around it is extracted from the echo at the same antenna elevation layer. According to the texture calculation formula, the texture feature value Tex_point of the center point is calculated. Then, the center point in the above processing is traversed through all spatial location points of dBZ0_par in turn to finally form the texture dataset Tex_par of the reflectivity factor of the phased array radar. Next, the ground clutter texture decision threshold Thre_Tex=5 is set, and the spatial location points corresponding to the data in the dataset Tex_par that are greater than Thre_Tex are marked as 1, and the rest are marked as 0, thereby generating the phased array radar ground clutter texture decision mask matrix M_Tex_par.

[0089] Next, based on the phased array radar reflectivity factor data dBZ0_par collected in step (1), the reflectivity factor data of adjacent lower antenna elevation angle layers and higher antenna elevation angle layers in the same volume scanning period are selected and denoted as dBZ0_par_L1 and dBZ0_par_L2, respectively. The difference between the reflectivity factors of these two layers is calculated as delta_dBZ0_L1=dBZ0_par_L1-dBZ0_par_L2. The reflectivity factor data of all adjacent antenna elevation angle layers are processed in sequence according to this method to generate a dataset of differences in reflectivity factors between adjacent elevation angle layers, denoted as delta_dBZ0_L1. After setting delta_dBZ0, the vertical consistency decision threshold Thre_Vert is set to 15. Spatial location points corresponding to data in delta_dBZ0 that are greater than Thre_Vert are marked as 1, and the rest are marked as 0, thereby generating the phased array radar ground clutter vertical consistency decision mask matrix M_Vert_par. Then, the three mask matrices M_ClutterMap_par, M_Tex_par, and M_Vert_par are combined by logical operations to generate the ground clutter combination decision mask matrix M_Clutter_par.

[0090] Finally, based on M_basic_par, M_SNR_par obtained in step (2) and M_Clutter_par obtained in this step, the mask matrix for the effective echo decision of the phased array radar is constructed as M_par = M_basic_par × M_SNR_par × (1-M_Clutter_par). Then, M_par and the original reflectivity factor data dBZ0_par are used to generate the cleaned phased array radar reflectivity factor dBZ1_par = dBZ0_par × M_par. dBZ1_par is then converted into the cleaned phased array radar linear domain reflectivity factor Z1_par using a logarithmic-to-linear formula. Replace the phased array radar reflectivity factor data dBZ0_par and P_Clutter_par in this step with the low-resolution radar reflectivity factor data dBZ0_wea and P_Clutter_wea collected in step (1), and then combine them with M_basic_wea and M_SNR_wea obtained in step (2). Execute the processing flow of this step to obtain the mask matrix M_wea of ​​the effective echo of the low-resolution radar and the cleaned low-resolution radar linear domain reflectivity factor Z1_wea.

[0091] The texture calculation formula is as follows: ,in This refers to the reflectance factor at a specific spatial location within the local analysis window. This represents the average reflectance factor within the local analysis window.

[0092] The logical operation combination is: M_Clutter_par=[M_ClutterMap_par∨(M_Tex_par∧M_Vert_par)], where ∨ represents the logical OR operation and ∧ represents the logical AND operation.

[0093] Step 4: As Figure 4 As shown, based on the operational status data of the two radars, the linear domain reflectivity factor of the phased array radar is time-aligned with that of the low-resolution radar. Then, simulation is performed using an interpolation model to obtain the simulated value of the phased array radar's linear domain reflectivity factor, which is time-aligned with the low-resolution radar's linear domain reflectivity factor. Specifically, this includes:

[0094] First, based on the timestamp sequences t0_par and t0_wea of ​​the antenna elevation angle echoes of each layer of the phased array radar and low-resolution radar obtained in step (1), and the volume scan periods delt_T_par and delt_T_wea of ​​the two radars, a complete continuous volume scan period corresponding to the timestamp sequence is extracted from the t0_wea sequence, denoted as t0_wea_V. Then, in the t0_par sequence, the two time points with the smallest time difference with the start and end times of t0_wea_V are selected respectively, thereby constructing a phased array radar scanning timestamp sequence that is time-aligned with t0_wea_V, denoted as t0_par_V; second, for a specified antenna elevation angle, a time value corresponding to the antenna elevation angle scan is selected from the sequence t0_wea_V, denoted as t0_wea_1, and combined with Z1_wea obtained in step (3), the reflectivity factor corresponding to the time value t0_wea_1 is extracted, denoted as Z1_wea_t1. Meanwhile, multiple time values ​​corresponding to multiple scans at the same antenna elevation angle are extracted from the sequence t0_par_V, denoted as t0_par_1, and combined with Z1_par obtained in step (3) to extract the reflectivity factor Z1_par_t1 corresponding to multiple time values ​​t0_par_1. Then, Z1_par_t1, t0_wea_1, and t0_par_1 are input into the Lagrange interpolation model, the model is run, and the output is the simulated value of the phased array radar reflectivity factor precisely aligned with the time of Z1_wea_t1. The above steps are applied to the echoes of all antenna elevation angle layers of the low-resolution radar to obtain the simulated value of the phased array radar linear domain reflectivity factor on all elevation angle layers precisely aligned with the time of Z1_wea, denoted as Z2_par.

[0095] Step 5: As Figure 5As shown, based on the operational status data of the two radars, the simulated values ​​of the linear domain reflectivity factor of the phased array radar are spatially geometrically matched with the linear domain reflectivity factor of the low-resolution radar. This yields a dataset of the phased array radar linear domain reflectivity factors spatially matched with the low-resolution radar linear domain reflectivity factor, including:

[0096] First, based on Z1_wea obtained in step (3), the polar coordinate spatial location information (r1_wea, phi1_wea, sita1_wea) corresponding to Z1_wea is extracted from the polar coordinate spatial location (r0_wea, phi0_wea, sita0_wea) of the low-resolution radar echo collected in step (1). Here, r1_wea, phi1_wea, and sita1_wea represent the radial distance, azimuth angle, and antenna elevation angle of the reflectivity factor data Z1_wea, respectively. Simultaneously, based on... The Z2_par obtained in step (4) is combined with the polar coordinate spatial position (r0_par, phi0_par, sita0_par) of the phased array radar echo collected in step (1) to extract the polar coordinate spatial position information (r2_par, phi2_par, sita2_par) corresponding to Z2_par. Here, r2_par, phi2_par and sita2_par represent the radial distance, azimuth angle and antenna elevation angle of the simulated value Z2_par of the linear domain reflectivity factor of the phased array radar, respectively.

[0097] Next, taking any spatial location point U in Z1_wea as the center point, and combining the information of (r1_wea, phi1_wea, sita1_wea), the radial distance, azimuth angle and antenna elevation angle corresponding to point U are extracted as (r_u, phi_u, sita_u), respectively. Then, combined with the radial range resolution delt_R_wea and beamwidth delt_sita_wea of ​​the low-resolution radar collected in step (1), the low spatial resolution volume delta_V of the low-resolution radar centered on point U is constructed, where the range coverage range of delta_V is delta_V_R. ang = [r_u - delt_R_wea / 2, r_u + delt_R_wea / 2], the coverage range in azimuth is delta_V_AZ = [phi_u - delt_sita_wea / 2, phi_u + delt_sita_wea / 2], and the coverage range in elevation is delta_V_EL = [sita_u - delt_sita_wea / 2, sita_u + delt_sita_wea / 2]; then, based on the radial range resolution delt_R_par and collected in step (1) of the phased array radar, The beamwidth delt_sita_par is used to initially screen out the phased array radar echo volume V_par_match contained within the low spatial resolution volume delta_V. The index values ​​of this echo volume in range, azimuth, and elevation are index_Rang=round(delta_V_Rang / delt_R_par), index_AZ=round(delta_V_AZ / delt_sita_par), and index_EL=round(delta_V_EL / delt_sita_par), respectively, where r `ound()` is a rounding function; then, any point of the phased array radar echo volume V_par_match is taken and denoted as point K. Then, combined with the information of (r2_par, phi2_par, seata2_par), the radial distance, azimuth angle and antenna elevation angle of point K are extracted and denoted as (r_k, phi_k, seata_k) respectively. The distance difference between point K and point U is calculated as delta_R_KU=r_k-r_u, the azimuth difference is delta_AZ_KU=phi_k-phi_u, and the elevation difference is delta_EL_KU=sita_k-sita_u.Next, a dual-threshold attribution rule for "overlap between center point and spatial resolution unit" is constructed. Specifically, the distance half-width threshold, azimuth half-width threshold, and elevation half-width threshold are set as follows: Thr_Rang=(delt_R_wea+delt_R_par) / 2, Thr_AZ=(delt_sita_wea+delt_sita_par) / 2, Thr_EL=(delt_sita_wea+delt_sita_par) / 2. The region half-width overlap determination conditions are set as follows: |delta_R_KU|≤Thr_Rang, |delta_AZ_KU|≤Thr_AZ, |delta_EL_KU|≤Thr_EL. When point K simultaneously satisfies the above three inequality determination conditions, and the corresponding M_par at point K is 1, then point K is determined to have effectively overlapped with the low spatial resolution volume delta_V. By iterating through all points in the phased array radar echo volume V_par_match using point K from the above steps, the effective echo data of the phased array radar contained in the low spatial resolution volume delta_V can be finely filtered out. Finally, by iterating through all spatial location points in Z1_wea using point U from this step and performing the processing in this step, a phased array radar linear domain reflectivity factor dataset corresponding to the low resolution radar linear domain reflectivity factor Z1_wea, after precise spatial matching of high and low beam and range resolution, can be constructed, denoted as Z3_par.

[0098] Step 6: As Figure 6 As shown, based on the multi-static radar detection data conversion model and coarse beam simulation operator, the linear domain reflectivity factor dataset of phased array radar is simulated, resulting in a simulated dataset of linear domain reflectivity factor of phased array radar with the same spatial location and resolution as that of low-resolution radar. This dataset includes:

[0099] First, from the low-resolution radar linear domain reflectivity factor Z1_wea generated in step (3) and the spatial location information (r1_wea, phi1_wea, sita1_wea) generated in step (5), select any spatial point M, and denote the linear domain reflectivity factor data of this point as Z1_wea_M. The spatial information of this point M is denoteed as (r1_wea_M, phi1_wea_M, sita1_wea_M), where r1_wea_M, phi1_wea_M and sita1_wea_M represent the radial distance, azimuth angle and antenna elevation angle of point M, respectively.

[0100] Secondly, based on the low-resolution radar linear domain reflectivity factor Z1_wea generated in step (3) and the phased array radar linear domain reflectivity factor dataset Z3_par corresponding to the low-resolution radar linear domain reflectivity factor Z1_wea after spatial geometric matching, the data set corresponding to the spatial geometric matching of point M is extracted from the phased array radar linear domain reflectivity factor dataset Z3_par, denoted as Z3_par_M. Combined with the polar coordinate spatial position (r0_par, phi0_par, sita0_par) of the phased array radar echo collected in step (1), the polar coordinate spatial position information (r3_par_M, phi3_par_M, sita3_par_M) corresponding to Z3_par_M is extracted, where r3_par_M, phi3_par_M and sita3_par_M represent the radial distance, azimuth angle and antenna elevation angle of the reflectivity factor dataset Z3_par_M, respectively.

[0101] Then, P_wea = (Lon_wea, Lat_wea, H_wea) and P_par = (Lon_par, Lat_par, H_par) collected in step (1), along with (r3_par_M, phi3_par_M, sita3_par_M), are input into the multi-static radar detection data conversion model. The model is run to obtain the spatial location information of the Z3_par_M dataset in the low-resolution radar polar coordinate system, denoted as (r3_wea_M, phi3_wea_M, sita3_wea_M), where r3_wea_M, phi3_wea_M and sita3_wea_M represent the radial distance, azimuth angle and antenna elevation angle of the reflectivity factor dataset Z3_par_M, respectively.

[0102] Subsequently, a coarse beam simulation operator based on three dimensions—radial range, azimuth, and elevation—is constructed. Specifically, in the low-resolution radar polar coordinate system, the degree of deviation of each data point in Z3_par_M from point M in the three dimensions of radial range, azimuth, and antenna elevation is calculated, denoted as delt_r_M=|r3_wea_M-r1_wea_M|, delt_phi_M=|phi3_wea_M-phi1_wea_M|, and delt_sita_M=|sita3_wea_M-sita1_wea_M). Then, delt_r_M is substituted into the radial range response function to obtain the weighted value delt_r_weight in the range direction. Similarly, delt_phi_M and delt_sita_M are substituted into the antenna main lobe and sidelobe radiation energy distribution function to obtain the weighted values ​​delt_phi_weight and delt_sita_M in the azimuth and antenna elevation directions, respectively. The total weighted value delt_weight is calculated by first calculating lt_sita_weight, then delt_r_weight × delt_phi_weight × delt_sita_weight, and then normalized to obtain the normalized total weighted value delt_weight_nor = delt_weight / max(delt_weight), where max() is the maximum value function. Finally, the result of delt_weight_nor × Z3_par_M is used as the low spatial resolution linear domain reflectivity factor simulation data at point M. By iterating through all points in the Z1_wea dataset from point M in the above processing flow, the phased array radar linear domain reflectivity factor simulation dataset Z1_wea_sim with the same spatial location and spatial resolution as the low resolution radar linear domain reflectivity factor Z1_wea can be obtained.

[0103] The radial distance response function is: delt_r_weight=exp(-0.5×delt_r_M2 / delt_R_wea2);

[0104] The antenna radiation energy distribution function is: delt_phi_weight=(1-s_w1)×exp(-0.5×delt_phi_M2 / delt_sita_wea2)+s_w1×exp(-s_w2×delt_phi_M) and delt_sita_weight=(1-s_w1)×exp(-0.5×delt_sita_M2 / delt_sita_wea2)+s_w1×exp(-s_w2×delt_sita_M), where s_w1 is the sidelobe weight, taking values ​​between [0,1], and s_w2 is the sidelobe attenuation coefficient.

[0105] Step 7: As Figure 7 As shown, low-resolution radar linear domain reflectivity factor data is used as training input, and the phased array radar linear domain reflectivity factor dataset is used as training output. The residual network model based on a self-attention mechanism is trained and iteratively optimized using the simulated phased array radar linear domain reflectivity factor dataset to obtain an optimized model, including:

[0106] First, based on Z1_wea obtained in step (3), Z3_par generated in step (5), and Z1_wea_sim generated in step (6), a residual network model MODEL1 based on an attention mechanism is constructed. Z1_wea is used as the input dataset for model training, and Z3_par is used as the output dataset for model training. The high spatial resolution reflectivity factor predicted by the model is defined as Z1_out. Second, a fine-scale supervision loss Loss_fine=|Z1_out-Z3_par| is constructed, a coarse-scale consistency loss function Loss_coar=|Z1_wea-Z1_wea_sim| is constructed, a physical consistency loss Loss_sim=|Z1_out-Z1_wea_sim| is constructed, and a true coarse observation consistency function is constructed. The loss is calculated as Loss_real = |Z1_out - Z1_wea|, which leads to the total loss function Loss = apha1 × Loss_fine + apha2 × Loss_coar + apha3 × Loss_sim + apha4 × Loss_real, where apha1, apha2, apha3, and apha4 are the weight coefficients of each loss term. Then, the parameters of the model MODEL1 are iteratively optimized and trained using the above total loss function, and the trained model is denoted as MODEL2. Finally, in the inference stage, Z1_wea obtained in step (3) is input into the model MODEL2, and the model is run to generate low-resolution radar reflectivity factor data with the same spatial refinement as the high spatial resolution of the phased array radar.

[0107] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for refining the reflectivity factor of low-resolution radar based on phased array radar, characterized in that, include: Data on reflectivity factors, operational status, and ground clutter occurrence probability of two geographically adjacent radars are collected. The two radars are a low-resolution radar and a phased array radar, respectively. Quality control was performed on the reflectivity factor data of the two radars to obtain the cleaned linear domain reflectivity factors of the two radars. Based on the operational status data of the two radars, the linear domain reflectivity factor of the phased array radar is time-aligned with that of the low-resolution radar. Then, simulation is performed based on the interpolation model to obtain the simulated value of the linear domain reflectivity factor of the phased array radar that is time-aligned with the linear domain reflectivity factor of the low-resolution radar. Based on the operational status data of the two radars, the simulated values ​​of the linear domain reflectivity factor of the phased array radar are spatially geometrically matched with the linear domain reflectivity factor of the low-resolution radar to obtain a phased array radar linear domain reflectivity factor dataset that spatially matches the low-resolution radar linear domain reflectivity factor. Based on the multi-static radar detection data conversion model and coarse beam simulation operator, the phased array radar linear domain reflectivity factor dataset is simulated to obtain a phased array radar linear domain reflectivity factor simulation dataset with the same spatial position and resolution as the low-resolution radar linear domain reflectivity factor. Low-resolution radar linear domain reflectivity factor data is used as training input, and the phased array radar linear domain reflectivity factor dataset is used as training output. The residual network model based on the self-attention mechanism is trained and iteratively optimized in combination with the phased array radar linear domain reflectivity factor simulation dataset to obtain an optimized model. The cleaned low-resolution radar linear domain reflectivity factor data is input into the optimized model, and the low-resolution radar spatially refined reflectivity factor data is output.

2. The method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 1, characterized in that, Reflectivity factor data includes radial distance, azimuth, and antenna elevation. Operational status data includes the polar coordinate spatial location of the radar site, which includes the longitude, latitude, and altitude of the antenna feed. The radar also includes low radial distance resolution, low beamwidth, and volumetric scanning period. Ground clutter occurrence probability data refers to the frequency data of ground clutter occurrence collected at the polar coordinate spatial location.

3. The method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 2, characterized in that, Quality control was performed on the reflectivity factor data of the two radars to obtain the cleaned linear domain reflectivity factors of the two radars, including: Based on the reflectivity factor data of the two radars, and based on the reliable threshold range of weather echoes, signal-to-noise ratio calculation and signal-to-noise ratio threshold, the basic mask matrix and signal-to-noise ratio decision mask matrix of the weather echoes of the two radars are calculated respectively. Based on the reflectivity factor data of the two radars, and based on the clutter texture decision threshold and the clutter vertical consistency decision threshold, the clutter texture decision mask matrix and the clutter vertical consistency decision mask matrix of the two radars are calculated respectively. Based on the clutter occurrence probability data of the two radars, and based on the clutter probability judgment threshold, the clutter probability map identification mask matrix of the two radars is calculated respectively. Based on the clutter texture decision mask matrix, clutter vertical consistency decision mask matrix, and clutter probability map identification mask matrix of the two radars, the clutter combination decision mask matrix is ​​obtained by logical operation combination calculation. Based on the ground clutter combination decision mask matrix, signal-to-noise ratio decision mask matrix, and weather echo basic mask matrix of the two radars, the effective echo decision mask matrices of the two radars are constructed respectively. Based on the effective echo decision mask matrix and reflectivity factor data of the two radars, the cleaned linear domain reflectivity factors of the two radars are generated respectively.

4. The method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 3, characterized in that, Based on the reflectivity factor data from the two radars, and considering the reliable threshold range of weather echoes, signal-to-noise ratio (SNR) calculation, and SNR threshold, the basic mask matrix and SNR decision mask matrix for the weather echoes from the two radars are calculated, including: Set the confidence threshold range for weather echoes, which includes the minimum and maximum confidence echo thresholds; Determine whether the value of each spatial location point in the reflectivity factor data is within the reliable threshold range of weather echoes. Mark spatial location points that do not fall within the threshold range or have empty values ​​as 0, and mark the rest as 1, thereby generating the basic mask matrix of weather echoes. The total number of distances required for statistical radial noise base is set to N_noise, and the quiet zone threshold is 0. Then, N_noise consecutive data points are extracted from the farthest radial distance of the reflectivity factor data, and all data points less than the quiet zone threshold in these N_noise data points are extracted to form logarithmic domain noise data. Each reflectance factor in the logarithmic domain noise data is transformed using the logarithmic-to-linear formula to form linear domain noise data. The arithmetic mean of the linear domain noise data is then taken as the noise floor value. The reflectivity factor data is converted using a logarithmic-to-linear formula to form linear domain radar reflectivity factor data. The linear domain signal-to-noise ratio data is then calculated based on the linear domain radar reflectivity factor data and the noise floor value. Set a signal-to-noise ratio (SNR) threshold, mark the spatial locations of data with SNR values ​​greater than the threshold in the linear domain as 1, and mark the rest as 0, thereby generating an SNR decision mask matrix.

5. The method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 4, characterized in that, Based on the reflectivity factor data from the two radars, and using the clutter texture decision threshold and the clutter vertical consistency decision threshold, the clutter texture decision mask matrix and the clutter vertical consistency decision mask matrix for the two radars are calculated, including: For reflectivity factor data, set the window width and construct the local analysis window size. Take a certain spatial location point in the reflectivity factor data as the center, extract the reflectivity factor data in the local analysis window around it in the echo of the same antenna elevation layer, and calculate the texture feature value of the center point according to the texture calculation formula. The center point in the above processing is used to traverse all spatial location points in the reflectivity factor data in turn to form the texture data of the reflectivity factor. Set a ground clutter texture decision threshold, mark the spatial location points corresponding to the data in the texture data that are greater than the ground clutter texture decision threshold as 1, and mark the rest as 0, and generate a ground clutter texture decision mask matrix. Then, based on the reflectivity factor data, the reflectivity factor data of adjacent lower antenna elevation angle layers and higher antenna elevation angle layers in the same volume scanning period are selected, and the difference between the reflectivity factor data of the lower antenna elevation angle layer and the higher antenna elevation angle layer is calculated. The reflectivity factor data of all adjacent antenna elevation angle layers are processed in sequence according to this method to generate a dataset of differences in reflectivity factors of adjacent elevation angle layers. Set a vertical consistency decision threshold for ground clutter. Mark the spatial location points corresponding to the data in the difference dataset that are greater than the vertical consistency decision threshold for ground clutter as 1, and mark the rest as 0, thereby generating a vertical consistency decision mask matrix for ground clutter.

6. The method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 5, characterized in that, Based on the clutter occurrence probability data from the two radars, and using the clutter probability judgment threshold, the clutter probability map identification mask matrices for both radars are calculated. Based on the clutter texture decision mask matrix, clutter vertical consistency decision mask matrix, and clutter probability map identification mask matrix from both radars, a clutter combination decision mask matrix is ​​obtained through logical operation combination calculation. Based on the clutter combination decision mask matrix, signal-to-noise ratio decision mask matrix, and weather echo basic mask matrix from both radars, the effective echo decision mask matrices for both radars are constructed, including: Set a ground clutter probability judgment threshold based on ground clutter occurrence probability data; The spatial location points corresponding to the ground clutter probability determination threshold in the ground clutter occurrence probability data are marked as 1, and the rest are marked as 0, thereby generating a ground clutter probability map identification mask matrix. Based on the clutter texture decision mask matrix, clutter vertical consistency decision mask matrix, and clutter probability map identification mask matrix from the two radars, the clutter combination decision mask matrix is ​​obtained through logical operation combination calculation. The logical operation combination calculation formula is as follows: Ground clutter combination decision mask matrix = [Ground clutter probability map recognition mask matrix ∨ (Ground clutter texture decision mask matrix ∧ Ground clutter vertical consistency decision mask matrix)], where ∨ represents a logical OR operation and ∧ represents a logical AND operation; The effective echo decision mask matrix is ​​constructed based on the ground clutter combined decision mask matrix, the signal-to-noise ratio decision mask matrix, and the weather echo fundamental mask matrix. The calculation formula is as follows: Effective echo decision mask matrix = Echo base mask matrix × Signal-to-noise ratio decision mask matrix × (1 - Ground clutter combination decision mask matrix).

7. The method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 2, characterized in that, Based on the operational status data of the two radars, the linear domain reflectivity factor of the phased array radar is time-aligned with that of the low-resolution radar. Then, simulations are performed using an interpolation model to obtain simulated values ​​of the phased array radar linear domain reflectivity factor that are time-aligned with the low-resolution radar's linear domain reflectivity factor. These values ​​include: Extract a complete timestamp sequence corresponding to a continuous volume scan cycle from the timestamp sequence obtained from the timestamp sequence of the antenna elevation angle echo of each layer within each volume scan cycle. In the timestamp sequence of the antenna elevation angle echo acquired in each volume scanning cycle, the two time points with the smallest time difference between the start and end times of the timestamp sequence corresponding to the continuous volume scanning cycle are selected respectively to construct a scanning timestamp sequence that is time-aligned with the timestamp sequence corresponding to the continuous volume scanning cycle. For a given antenna elevation angle, select a time value corresponding to the specified antenna elevation angle scan from the timestamp sequence corresponding to the continuous volume scan period, and extract the low-resolution radar linear domain reflectivity factor corresponding to the time value. At the same time, multiple time values ​​corresponding to multiple scans at the same specified antenna elevation angle are extracted, and the linear domain reflectivity factor of the phased array radar corresponding to the multiple time values ​​is extracted. The phased array radar linear domain reflectivity factors corresponding to the multiple time values ​​and the low-resolution radar linear domain reflectivity factor corresponding to one time value are input into the Lagrange interpolation model, and the simulated value of the phased array radar linear domain reflectivity factor that is precisely time-aligned with the low-resolution radar linear domain reflectivity factor at the specified antenna elevation angle is output. Applying the above steps to the echoes of all antenna elevation layers of the low spatial resolution radar, we obtain simulated values ​​of the linear domain reflectivity factor of the phased array radar that are time-aligned with the linear domain reflectivity factor of the low-resolution radar at all antenna elevation angles.

8. The method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 7, characterized in that, Based on the operational status data of the two radars, the simulated values ​​of the linear domain reflectivity factor of the phased array radar are spatially geometrically matched with the linear domain reflectivity factor of the low-resolution radar. This yields a dataset of the phased array radar linear domain reflectivity factors spatially matched with the low-resolution radar linear domain reflectivity factor, including: Take any spatial location point U as the center point from the linear domain reflectivity factor data of low-resolution radar, and extract the radial distance, azimuth angle and antenna elevation angle corresponding to the spatial location point by combining the polar coordinate spatial location of the low-resolution radar echo. Combine the collected radial distance resolution and beamwidth of the low-resolution radar to construct the low spatial resolution volume of the low spatial resolution radar centered on point U. Based on the radial range resolution and beamwidth of the phased array radar, the phased array radar echo volume contained within the low spatial resolution volume is initially screened out. Take any point in the phased array radar echo volume, and combine it with the polar coordinate spatial position information of the phased array radar echo to extract the radial distance, azimuth angle and antenna elevation angle of the arbitrary point, and calculate the distance difference, azimuth difference and elevation angle difference between the spatial position point and the arbitrary point. Based on the set range half-width threshold, azimuth half-width threshold, and elevation half-width threshold, the region half-width overlap determination conditions are set, including the absolute value of the range difference not being greater than the range half-width threshold, the absolute value of the azimuth difference not being greater than the azimuth half-width threshold, and the elevation difference not being greater than the elevation half-width threshold. When any point simultaneously satisfies the above three inequality determination conditions, and the mask matrix of the effective echo decision corresponding to any point is 1, then it is determined that any point has effectively overlapped with the low spatial resolution volume. The above steps traverse all points in the phased array radar echo volume and filter out the effective echo data of the phased array radar contained in the low spatial resolution volume. By traversing all spatial location points in the low-resolution radar linear domain reflectivity factor data, a phased array radar linear domain reflectivity factor dataset that spatially matches the low-resolution radar linear domain reflectivity factor is constructed.

9. A method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 8, characterized in that, Based on a multi-static radar detection data conversion model and a coarse beam simulation operator, a phased array radar linear domain reflectivity factor dataset was simulated. This yielded a simulated dataset of phased array radar linear domain reflectivity factors with the same spatial location and resolution as the low-resolution radar linear domain reflectivity factors, including: Based on the linear domain reflectivity factor data of low-resolution radar, any spatial point is selected from the polar coordinate spatial location of the low-resolution radar echo. Based on low-resolution radar linear domain reflectivity factor data, a data set corresponding to the spatial geometric matching of the spatial point is extracted from the phased array radar linear domain reflectivity factor dataset. Combined with the polar coordinate spatial position of the phased array radar echo, the polar coordinate spatial position information corresponding to the data set corresponding to the spatial geometric matching of the spatial point is extracted. Then, the spatial position information of the two radars, as well as the polar coordinate spatial position information corresponding to the data set that corresponds to the spatial geometric matching of the spatial point, are input into the multi-static radar detection data conversion model. The model is run to obtain the spatial position information of the data set that corresponds to the spatial geometric matching of the spatial point in the low-resolution radar polar coordinate system. Based on the spatial position information of the data set corresponding to the spatial geometric matching of the spatial point in the low-resolution radar polar coordinate system, a coarse beam simulation operator based on three dimensions—radial range, azimuth, and elevation—is constructed. Specifically, in the low spatial resolution radar polar coordinate system, the degree of deviation of each data point in the data set corresponding to the spatial geometric matching of the spatial point from the spatial point in the three dimensions of radial range, azimuth, and antenna elevation is calculated. The degree of deviation in radial range is substituted into the radial range response function to obtain the weighted value in the range direction. The degree of deviation in azimuth and antenna elevation is substituted into the antenna main lobe and side lobe radii. In the radiation energy distribution function, weighted values ​​are obtained in the azimuth and antenna elevation directions, respectively. Then, the total weighted values ​​in the radial distance, azimuth, and elevation are calculated and normalized to obtain the normalized total weighted value. The product of the total weighted value and the data set corresponding to the spatial geometric matching of the spatial point is used as the low-resolution radar linear domain reflectivity factor simulation data at the spatial point. By traversing all points of the cleaned low-resolution radar linear domain reflectivity factor data according to the above processing flow, a phased array radar linear domain reflectivity factor simulation dataset with the same spatial location and spatial resolution as the cleaned low-resolution radar linear domain reflectivity factor data can be obtained.

10. A method for refining the reflectivity factor of low-resolution radar based on phased array radar according to claim 9, characterized in that, Using low-resolution radar linear domain reflectivity factor data as training input and the phased array radar linear domain reflectivity factor dataset as training output, the residual network model based on the self-attention mechanism is trained and iteratively optimized in conjunction with the simulated phased array radar linear domain reflectivity factor dataset to obtain an optimized model, including: Low-resolution radar linear domain reflectivity factor data is used as the input dataset for model training, and phased array radar linear domain reflectivity factor data is used as the output dataset for training the residual network model based on the attention mechanism. The residual network model based on the self-attention mechanism is then trained. The total loss function is constructed based on the low-resolution radar linear domain reflectivity factor data, the phased array radar linear domain reflectivity factor dataset, the model prediction of the residual network model based on the self-attention mechanism, and the simulated dataset of the phased array radar linear domain reflectivity factor. The total loss function includes a weighted sum of fine-scale supervision loss, coarse-scale consistency loss function, physical consistency loss, and real coarse observation consistency loss. The parameters of the residual network model based on the attention mechanism are iteratively optimized and trained using the total loss function to obtain the optimized model.

Images