Rain measurement radar data quality control method and system based on multi-source data fusion and deep learning

Abstract

This invention discloses a data quality control method and system for rain-measuring radar based on multi-source data fusion and deep learning, relating to the field of radar data processing technology. The method includes: performing quality assessment on pre-processed raw radar echo data using a first deep learning model; performing quality analysis on pre-processed radar base data using a correlation analysis model; performing interference suppression on the quality assessment results, quality analysis results, and radio frequency interference monitoring data using an adaptive filtering model and a second deep learning model; and constructing a comprehensive evaluation index system based on multi-source data fusion for the interference suppression results. Compared to existing technologies, the technical solution of this invention can effectively improve data accuracy and reliability, enabling quantitative evaluation of rain-measuring radar performance.

Description

Technical Field

[0001] This invention relates to the field of rain measurement radar data processing technology, and in particular to a rain measurement radar data quality control method and system based on multi-source data fusion and deep learning. Background Technology

[0002] As a core device for meteorological and hydrological detection, rainfall radar is increasingly being deployed in networked observations to meet the growing demands for precision in precipitation monitoring for flash flood disaster prevention. However, existing rainfall radar data processing technologies still have significant limitations:

[0003] First, data quality control relies on traditional physical rules and human experience, lacking intelligent error identification and interference suppression methods. Factors such as ground clutter, radio frequency interference, and terrain obstruction lead to a large amount of false information in radar echo data, seriously affecting the accuracy of precipitation inversion. Second, the quality assessment system for raw and baseline data is incomplete, lacking standardized error analysis and quantitative evaluation methods, making it difficult to accurately pinpoint the causes of data quality problems. Third, existing quality control methods do not fully integrate statistical analysis and deep learning technologies, resulting in poor adaptability to processing severe convective precipitation data under complex weather scenarios, failing to meet the demand for high-precision and high-timeliness data in flash flood monitoring. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a data quality control method and system for rain-measuring radar based on multi-source data fusion and deep learning, thereby achieving intelligent assessment of data quality, precise suppression of interference, and comprehensive optimization of radar performance, thus enhancing the application effectiveness of rain-measuring radar in flash flood disaster prevention.

[0005] In a first aspect, the present invention provides a method for quality control of rain-measuring radar data based on multi-source data fusion and deep learning, comprising the following steps: collecting multi-source data and preprocessing the multi-source data; wherein the multi-source data includes at least radar raw echo data, radar base data, and radio frequency interference monitoring data; performing quality assessment on the preprocessed radar raw echo data using a first deep learning model; performing quality analysis on the preprocessed radar base data using a correlation analysis model; performing interference suppression on the quality assessment results, quality analysis results, and radio frequency interference monitoring data using an adaptive filtering model and a second deep learning model; and constructing a comprehensive evaluation index system based on multi-source data fusion based on the interference suppression results.

[0006] One possible implementation involves evaluating the quality of preprocessed raw radar echo data using a first deep learning model, including: optimizing the quality evaluation capability of the first deep learning model based on UAV external calibration data and raw observation data; wherein the raw observation data is the echo data collected by the radar during UAV external calibration; and evaluating the quality of the preprocessed raw radar echo data using the optimized first deep learning model.

[0007] One possible implementation involves optimizing the quality assessment capability of a first deep learning model based on UAV external calibration data and raw observation data. This includes: acquiring raw observation data, which includes radar echo intensity and radar coordinates; aligning the UAV coordinates with the radar coordinates, where the UAV external calibration data includes the UAV coordinates; calculating the theoretical value of the radar echo intensity; obtaining an estimated value of the radar echo intensity based on the radar echo intensity; calculating the deviation between the estimated value and the theoretical value of the radar echo intensity; and inputting the deviation into the first deep learning model to optimize its quality assessment capability.

[0008] One possible implementation method further includes: acquiring the spatial geometric information of the radar during UAV external calibration; inputting the spatial geometric information and the original observation data into a first deep learning model to obtain first feature information; and inputting the first feature information and the bias into the fully connected layer of the first deep learning model to optimize the first deep learning model.

[0009] One possible implementation involves performing quality assessment on the preprocessed raw radar echo data using an optimized first deep learning model. This includes: inputting the preprocessed raw radar echo data into the optimized first deep learning model and outputting a quality assessment level; where the quality assessment level includes excellent, good, and poor; and removing echo data with a poor quality assessment level to obtain the quality assessment result.

[0010] One possible implementation involves using multi-source data, including GIS topographic data and ground observation data. For the pre-processed radar base data, a quality analysis is performed using a correlation analysis model. This includes: inputting the pre-processed GIS topographic data, ground observation data, and radio frequency interference monitoring data into the correlation analysis model to obtain the radar base data error; compensating the pre-processed radar base data based on the radar base data error to obtain the quality analysis results.

[0011] One possible implementation involves using an adaptive filtering model and a second deep learning model to suppress interference in the quality assessment results, quality analysis results, and radio frequency interference monitoring data. This includes: performing adaptive filtering on the quality assessment results and radio frequency interference monitoring data using the adaptive filtering model; suppressing clutter in the quality analysis results using the second deep learning model; and optimizing the adaptive filtering results and clutter suppression results through interpolation.

[0012] One possible approach is to construct a comprehensive evaluation index system based on multi-source data fusion for the interference suppression results, including: constructing evaluation index dimensions, including data quality dimension, observation accuracy dimension, and anti-interference capability dimension; calculating the values ​​of the data optimization results under the evaluation index dimensions; comparing the values ​​under the evaluation index dimensions with preset thresholds to determine whether the multi-source data meets the standards.

[0013] One possible implementation method further includes: the data quality dimension being characterized by the mean difference and root mean square difference of the rainfall samples; the observation accuracy dimension being characterized by the correlation coefficient of the rainfall samples; and the anti-interference capability dimension being characterized by the clutter recognition rate.

[0014] Secondly, embodiments of the present invention also provide a rain measurement radar data quality control system based on multi-source data fusion and deep learning. This system includes: an acquisition and preprocessing module for acquiring multi-source data and preprocessing the multi-source data; wherein the multi-source data includes at least raw radar echo data, radar base data, and radio frequency interference monitoring data; a quality assessment module for performing quality assessment on the preprocessed raw radar echo data using a first deep learning model; a quality analysis module for performing quality analysis on the preprocessed radar base data using a correlation analysis model; an interference suppression module for performing interference suppression on the quality assessment results, quality analysis results, and radio frequency interference monitoring data using an adaptive filtering model and a second deep learning model; and a comprehensive evaluation module for constructing a comprehensive evaluation index system based on multi-source data fusion for the interference suppression results.

[0015] Thirdly, the present invention also provides an electronic device, comprising:

[0016] Memory stores computer-executable instructions non-transiently;

[0017] The processor is configured to run computer-executable instructions.

[0018] The computer-executable instructions are executed by the processor to implement the above-mentioned rain measurement radar data quality control method.

[0019] Fourthly, the present invention also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the above-described rain measurement radar data quality control method.

[0020] This invention provides a method and system for quality control of rain-measuring radar data based on multi-source data fusion and deep learning, comprising: performing quality assessment on preprocessed raw radar echo data using a first deep learning model; performing quality analysis on preprocessed radar base data using a correlation analysis model; performing interference suppression on the quality assessment results, quality analysis results, and radio frequency interference monitoring data using an adaptive filtering model and a second deep learning model; and constructing a comprehensive evaluation index system based on multi-source data fusion for the interference suppression results.

[0021] Compared with existing technologies, the technical solution of this invention improves the efficiency of quality control by constructing a quality assessment system for raw radar echo data and radar base data, thereby enabling automatic identification and quantitative analysis of data quality problems; it also improves data accuracy and reliability by effectively eliminating clutter and interference based on adaptive filtering and intelligent interference suppression methods; and it constructs a comprehensive index evaluation system for multi-source data fusion to achieve quantitative evaluation of the performance of rain-measuring radar. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a rain measurement radar data quality control method based on multi-source data fusion and deep learning, provided for an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of the structure of a rain measurement radar data quality control system based on multi-source data fusion and deep learning, provided in an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0026] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Figure 1 A data quality control method for rainfall radar based on multi-source data fusion and deep learning is presented, such as... Figure 1 As shown, the method includes the following steps.

[0028] S100. Collect multi-source data and preprocess the multi-source data; wherein, the multi-source data includes at least radar raw echo data, radar base data, and radio frequency interference monitoring data.

[0029] As one possible approach, multi-source data can also include ground observation data and Geographic Information System (GIS) topographic data.

[0030] Specifically, multi-source data acquisition can include: collecting multi-source data and constructing a basic database for quality control of rainfall radar data based on the multi-source data. Specifically, this involves collecting raw echo data and radar base data from rainfall radar, simultaneously collecting measured rainfall data from ground rain gauges and raindrop spectrometers, as well as auxiliary information such as GIS topographic data and radio frequency interference monitoring data, to construct a basic database for quality control of rainfall radar data.

[0031] For example, data covering different seasons and rainfall scenarios are collected from May 2025 to November 2025, and collected continuously from 00:00 to 24:00 every day.

[0032] The raw radar echo data is in binary format, containing I / Q signals (in-phase / quadrature signals), with a sampling rate of 10MHz.

[0033] Radar base data: NetCDF, a common network data format, including reflectivity Z (dBZ) and differential reflectivity Z. DR (dB), differential propagation phase rate K DP (° / km) and 6 other channel parameters;

[0034] Ground-based measured data: rain gauge data (CSV format), raindrop spectrometer data (HDF5 format);

[0035] Auxiliary information: GIS topographic data (resolution 30m×30m), radio frequency interference monitoring data (sampling interval 1 minute, recording interference frequency and intensity).

[0036] As a possible implementation, preprocessing operations such as pulse compression, format conversion, denoising, and coordinate correction can be performed on multi-source data; z-score and min-max normalization methods can be used to normalize radar data of different physical quantities to eliminate dimensions.

[0037] For example, the raw radar data is pulse-compressed using a linear frequency modulated (LFM) signal with a pulse width of 1 μs, resulting in a compressed pulse width of 10 ns, thus improving the range resolution to 5 m. The radar polar coordinate data is then projected onto an equidistant latitude and longitude grid (spatial resolution 0.5 km × 0.5 km) using the WGS84 coordinate system.

[0038] Among them, Z and Z DR K DP The channels were normalized using z-score (mean μ was taken as the mean of the last 3 months of historical data, and standard deviation σ was taken as the standard deviation of the corresponding channel's historical data); the raindrop number concentration (N) of the raindrop spectrometer was normalized to [0,1] using min-max normalization.

[0039] S200. The preprocessed raw radar echo data is evaluated for quality using the first deep learning model.

[0040] The first deep learning model includes a convolutional neural network (CNN) and a long short-term memory network (LSTM).

[0041] Specifically, it may include the following steps.

[0042] S210. Optimize the quality assessment capability of the first deep learning model based on the external calibration data of the UAV and the original observation data.

[0043] S211. Obtain raw observation data, which includes radar echo intensity and radar coordinates.

[0044] Specifically, the drone is controlled to hover at each preset distance point for 30 seconds. During the hovering period, the rain-measuring radar synchronously collects a continuous echo data sequence (including amplitude and phase changes of I / Q orthogonal signals) at that location and records the corresponding high-precision GPS positioning data of the drone to obtain the required raw observation data.

[0045] S212. Align the UAV coordinates with the radar coordinates, wherein the UAV external calibration data includes the UAV coordinates.

[0046] This includes the step of acquiring external calibration data for the UAV, which includes the UAV coordinates and aligns them with the radar coordinates. It can be understood that UAV external calibration refers to the process where a UAV carrying calibration equipment collects benchmark reference data in a real outdoor electromagnetic and geographical environment to calibrate the performance of radars (such as weather radars, measurement radars, synthetic aperture radars, SAR, etc.). This data is used to calibrate key indicators of the radar, such as distance, angle, and radar cross section (RCS) measurement accuracy, and to correct system and environmental errors.

[0047] S213. Calculate the theoretical value of radar echo intensity.

[0048] Specifically, the theoretical value of the rain-measuring radar echo intensity is calculated using radar meteorological equations and RCS.

[0049] S214. Obtain the estimated value of radar echo intensity based on the radar echo intensity, and calculate the deviation between the estimated value of radar echo intensity and the theoretical value of radar echo intensity.

[0050] Among them, when the deviation is less than or equal to 1dB, the radar echo intensity in the original observation data is considered to be qualified.

[0051] S215. Input the spatial geometric information and the original observation data into the first deep learning model to obtain the first feature information.

[0052] Specifically, spatial geometric information and raw observation data are converted into input feature vectors, which are then input into the first deep learning model to obtain the first feature information.

[0053] One possible implementation method includes acquiring the spatial geometric information of the radar during the external calibration of the UAV; wherein the spatial geometric information includes distance, altitude, and azimuth. For example, the UAV carries a metal spherical reflector (RCS=10m²) and sets different flight altitude layers (0.5km, 1km, 2km) within the radar detection range, with different distance points (5km, 10km, ..., 40km) set along the radial direction of the radar at each altitude layer.

[0054] As one possible implementation, the distance, azimuth, altitude, and echo intensity of the rain-measuring radar in the original observation data are converted into input feature vectors, resulting in a distance vector, an altitude vector, an azimuth vector, and an echo intensity vector.

[0055] For example, the spatial geometric information, including distance, azimuth, height, and echo intensity in the original observation data, is converted into distance vector, height vector, azimuth vector, and echo intensity vector, respectively. These vectors are then input into a CNN model to extract echo spatial texture features, and the texture features are input into an LSTM model to extract signal temporal evolution features.

[0056] S216. Input the first feature information and bias into the fully connected layer of the first deep learning model to optimize the first deep learning model.

[0057] For example, signal temporal evolution characteristics and echo intensity deviation are input into a fully connected layer to optimize the first deep learning model. It is understood that the echo intensity deviation allows the deep learning model to continuously learn, ensuring that the error between the model's output and the actual value is within an acceptable range, thereby improving the accuracy of the deep learning model.

[0058] S220. The preprocessed raw radar echo data is evaluated for quality using an optimized first deep learning model.

[0059] Specifically, the preprocessed raw radar echo data is input into the optimized first deep learning model, which outputs a quality assessment level, categorized as Excellent, Good, or Poor. Echo data with a Poor quality assessment level are then removed to obtain the final quality assessment result. In essence, the quality assessment result represents the echo data with an Excellent / Good quality assessment level from the original echo data, thus ensuring the purity of the echo data.

[0060] S300 performs quality analysis on preprocessed radar base data using a correlation analysis model.

[0061] Specifically, the preprocessed GIS topographic data, ground observation data, and radio frequency interference monitoring data are input into the correlation analysis model to obtain the radar base data error. The preprocessed radar base data is then compensated for based on the radar base data error to obtain the quality analysis results. The multi-source data includes the GIS topographic data and the ground observation data.

[0062] For example, the terrain shading angle is calculated based on GIS terrain data, the radio frequency interference intensity is obtained based on radio frequency interference monitoring data, and the rainfall intensity is obtained based on ground observation data. The terrain shading angle, radio frequency interference intensity (unit dBμV / m), and rainfall intensity (mm / h) are input into the correlation analysis model to obtain the deviation ΔZ of reflectivity Z. Based on this deviation, compensation is performed (Z'=Z-ΔZ) to obtain the compensated reflectivity Z.

[0063] In one possible implementation, the correlation analysis model can be a gradient boosting tree (XGBoost). For example, using XGBoost (learning rate 0.1, tree depth 6, iterations 100), parameters are optimized through 5-fold cross-validation, and the base data error is output. For instance, the base data error can be the deviation ΔZ of reflectivity Z, in dBZ. Specifically, when the terrain occlusion angle is >30°, ΔZ increases by an average of 2 dBZ; when the radio frequency interference intensity is >20 dBμV / m, the fluctuation range of ΔZ expands by 3 dBZ. This deviation parameter is mainly used in the basic data quality control stage to quantitatively compensate for deviations caused by terrain obstruction or signal attenuation (Z'=Z-ΔZ). The specific compensation process is as follows: First, the radar beam obstruction rate is calculated based on GIS terrain data; referring to the beam obstruction correction rules, a mapping relationship between the obstruction rate and the compensation value is established. For example, when the beam obstruction rate is in the range of 30% to 43%, the reflectivity factor correction value is +2dB, and when the beam obstruction rate is in the range of 44% to 55%, the correction value is +3dB; the correlation analysis model (XGBoost) learns the above physical laws and interference characteristics and outputs a comprehensive deviation. (here) A negative value represents the attenuation of the signal amplitude. By subtracting this deviation (i.e., This enables the recovery and correction of the original echo intensity, thereby eliminating systematic observation errors.

[0064] S400: Based on the quality assessment results, quality analysis results, and the radio frequency interference monitoring data, interference suppression is performed using an adaptive filtering model and a second deep learning model.

[0065] S410. Adaptive filtering is performed on the quality assessment results and radio frequency interference monitoring data using an adaptive filtering model.

[0066] Specifically, it may include the following steps.

[0067] S4101. Set the filtering parameters of the adaptive filtering model.

[0068] A 5-minute time window (one radar volume scan cycle) and a 3×3 grid spatial window (corresponding to a 1.5km×1.5km area) filtering window are set. Based on the frequency characteristics of radio frequency interference, a band-stop filter is set with a center frequency range of 400-450MHz. This frequency corresponds to known communication base station interference within the watershed.

[0069] S4102. Adaptive filtering is performed on the quality assessment results and radio frequency interference monitoring data based on the set filtering parameters.

[0070] As one possible implementation, adaptive filtering can also be performed on quality assessment results, radio frequency interference monitoring data, ground observation data, and GIS topographic data using an adaptive filtering model.

[0071] It is understandable that the quality assessment result is the echo data after the preprocessed raw radar echo data has been quality assessed.

[0072] S420. Clutter suppression is performed on the quality analysis results using a second deep learning model.

[0073] The second deep learning model is the U-Net model, and the quality analysis result is the radar base data after quality analysis.

[0074] Specifically, it includes the following steps.

[0075] S4021. Normalize the quality analysis results.

[0076] Specifically, reflectivity Z and differential reflectivity Z DR and the differential phase K DP The numerical values ​​are uniformly mapped to the [0,1] interval to eliminate the difference in magnitude between different physical dimensions.

[0077] S4022. Input the normalized preprocessed quality analysis results into the U-Net model.

[0078] Specifically, the normalized reflectance Z and the differential reflectance Z are... DR and the differential phase K DP The data is input into the U-Net model for inference. An output value of 1 indicates clutter, while an output value of 0 indicates valid data. Understandably, removing identified ground clutter improves data accuracy.

[0079] S430 optimizes the data by interpolation operations on the adaptive filtering results and clutter suppression results.

[0080] Specifically, interpolation calculations are performed on the quality assessment results after adaptive filtering, radio frequency interference monitoring data, ground observation data, GIS topographic data, and radar base data after clutter suppression.

[0081] For example, the terrain shading angle is optimized using neighborhood interpolation and data compensation. For areas with a terrain shading angle > 30°, the average of the three unshaded grids surrounding the area is taken, and then the deviation under the same historical rainfall pattern is superimposed to obtain the correction value.

[0082] For example, time-series smoothing is used to optimize data for periods of radio frequency interference. For data in periods where the radio frequency interference intensity is >20 dBμV / m, valid data from 10 minutes before and after the interference are taken, and linear interpolation is used to supplement the data for the interference period.

[0083] S500: Construct a comprehensive evaluation index system based on multi-source data fusion for interference suppression results.

[0084] Specifically, the evaluation index dimensions are constructed, including: data quality dimension, observation accuracy dimension, and anti-interference capability dimension; the values ​​of the data optimization results under the evaluation index dimensions are calculated; the calculated values ​​under the evaluation index dimensions are compared with the preset thresholds to determine whether the multi-source data meets the standards.

[0085] The data quality dimension is characterized by the mean deviation and root mean square error (RMSE) of the rainfall samples. For example, the bias is calculated as (estimated value - measured value) / measured value, with a bias less than or equal to 10% considered acceptable. The estimated value refers to the radar-estimated rainfall value, calculated from radar echo data, while the measured value refers to the actual rain gauge reading, obtained from ground observation data. The root mean square error (RMSE) is... An RMSE of less than or equal to 3 mm / h is considered to meet the standard.

[0086] The observation accuracy dimension is characterized by the correlation coefficient between rainfall samples, which reflects the synchronicity between radar echo data and measured rainfall. For example, the correlation coefficient CC = Cov(estimated value, measured value) / ( ), where Cov is the sample covariance, The standard deviation is the sample standard deviation; a CC value greater than or equal to 0.85 is considered acceptable.

[0087] The noise reduction capability is represented by the clutter identification rate. For example, the clutter identification rate = number of correctly identified clutter samples / total number of clutter samples; a clutter identification rate greater than or equal to 85% is considered satisfactory. The total number of clutter samples can be calculated from GIS terrain data.

[0088] Optionally, a stability dimension may also be included, using equipment availability to characterize the online stability of the system. For example, equipment availability = (total uptime - downtime) / total uptime; an equipment availability of 98% or higher is considered satisfactory.

[0089] This invention provides a data quality control method for rain-measuring radar based on multi-source data fusion and deep learning. The method includes: performing quality assessment on pre-processed raw radar echo data using a first deep learning model; performing quality analysis on pre-processed radar base data using a correlation analysis model; performing interference suppression on the quality assessment results, quality analysis results, and radio frequency interference monitoring data using an adaptive filtering model and a second deep learning model; and constructing a comprehensive evaluation index system based on multi-source data fusion for the interference suppression results. This technical solution improves quality control efficiency by constructing a quality assessment system for raw radar echo data and radar base data, enabling automatic identification and quantitative analysis of data quality problems; effectively eliminating clutter and interference based on adaptive filtering and intelligent interference suppression methods, improving data accuracy and reliability; and constructing a comprehensive index evaluation system based on multi-source data fusion to achieve quantitative evaluation of rain-measuring radar performance.

[0090] like Figure 2 As shown, an embodiment of this application provides a rain measurement radar data quality control system based on multi-source data fusion and deep learning, which includes the following modules.

[0091] The acquisition and preprocessing module is used to acquire multi-source data and preprocess the multi-source data; the multi-source data includes at least radar raw echo data, radar base data, and radio frequency interference monitoring data.

[0092] The quality assessment module is used to assess the quality of the preprocessed raw radar echo data using a first deep learning model.

[0093] The quality analysis module is used to perform quality analysis on preprocessed radar base data using a correlation analysis model.

[0094] The interference suppression module is used to suppress interference from quality assessment results, quality analysis results, and radio frequency interference monitoring data through an adaptive filtering model and a second deep learning model.

[0095] The comprehensive evaluation module is used to construct a comprehensive evaluation index system based on multi-source data fusion for interference suppression results.

[0096] Embodiments of this application also provide a method based on... Figure 2 The model training process implemented by the quality control system shown includes the following steps.

[0097] Step 1: Dataset partitioning.

[0098] 100,000 sets of samples were collected (each set of samples includes radar echo data, radar base data, ground observation data, and GIS topographic data), and divided into training set (80,000 sets), validation set (10,000 sets), and test set (10,000 sets) in a ratio of 8:1:1 to ensure that the samples cover four scenarios: clear sky, light rainfall (<10mm / h), moderate rainfall (10-25mm / h), and heavy rainfall (>25mm / h).

[0099] Step 2: Model training.

[0100] (1) Training of CNN+LSTM model.

[0101] Input data: Preprocessed raw radar echo sequences and their corresponding spatial geometric information. Specifically, this includes: continuous radar echo intensity sequences (temporal characteristics) collected during UAV hovering observations, and the distance, azimuth, and altitude (spatial characteristics) corresponding to each echo point.

[0102] Loss function: Categorical Cross-Entropy Loss. The intensity deviation calculated from externally calibrated UAV data is used as the ground truth label (e.g., deviation absolute value ≤ 1dB is labeled "Excellent", 1dB < deviation absolute value ≤ 3dB is labeled "Good", deviation absolute value > 3dB is labeled "Poor"). The model parameters are iteratively optimized to enable the model to accurately identify echo characteristics within different error ranges.

[0103] Output data: Quality assessment level of the raw radar echo (Excellent, Good, Poor). This output is directly used in the data cleaning process. By removing data at the "Poor" level, the impact of non-meteorological echoes on the subsequent inversion accuracy is reduced from the source.

[0104] (2) Training of the XGBoost model

[0105] The input data are: terrain occlusion angle (calculated from GIS terrain data), radio frequency interference intensity (calculated from radio frequency interference monitoring data), and rainfall intensity (obtained from actual rain gauge measurements based on ground observation data).

[0106] Loss function: Huber Loss (turning point parameter δ=1.5), iterative calculation to adjust model parameters and optimize interference caused by terrain occlusion.

[0107] Output data: Radar base data error (deviation of reflectivity Z ΔZ). This output directly reduces the root mean square error (RMSE) in the overall system evaluation by correcting the original echo.

[0108] (3) Training of the U-Net model

[0109] Input data: Normalized three-channel basis data (reflectivity Z, differential reflectivity Z) DR Differential propagation phase rate K DP .

[0110] Loss function: Focal Loss (α=0.25, γ=2.0), optimized to address the sparsity of clutter samples.

[0111] Output data: Clutter mask. The accuracy of this output is directly reflected in the clutter detection rate (POD) in the comprehensive evaluation index.

[0112] The optimizer used was Adam (with an initial learning rate of 0.001, which decayed to 0.9 times the previous rate every 5 rounds).

[0113] Training process and results: The model was trained for 50 rounds, and the training set loss, validation set loss, and test set accuracy were analyzed.

[0114] The quality control system of this invention can be applied to electronic devices. Figure 3 A schematic diagram of the architecture of an electronic device suitable for implementing embodiments of the present invention is shown.

[0115] It should be noted that, Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0116] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by instructions (computer programs), or by instructions (computer programs) controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor. The electronic device of this embodiment includes a storage medium and a processor, wherein the storage medium stores multiple instructions that can be loaded by the processor to execute any step of the method provided in the embodiments of the present invention.

[0117] Specifically, the storage medium and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more signal lines. The storage medium stores computer-executable instructions that implement data access control methods, including at least one software functional module that can be stored in the storage medium in the form of software or firmware. The processor executes various functional applications and data processing by running the software program and module stored in the storage medium. The storage medium can be, but is not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The storage medium stores the program, and the processor executes the program after receiving the execution instructions.

[0118] Furthermore, the software programs and modules within the aforementioned storage medium may also include an operating system, which may include various software components and / or drivers for managing system tasks (e.g., memory management, storage device control, power management, etc.) and can communicate with various hardware or software components to provide an operating environment for other software components. The processor may be an integrated circuit chip with signal processing capabilities. The aforementioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc., which can implement or execute the methods, steps, and logic flowcharts disclosed in this embodiment. The general-purpose processor may be a microprocessor or any conventional processor.

[0119] Since the instructions stored in the storage medium can execute the steps in any of the methods provided in the embodiments of the present invention, the beneficial effects of any of the methods provided in the embodiments of the present invention can be achieved, as detailed in the preceding embodiments, and will not be repeated here.

[0120] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A quality control method for rainfall radar data based on multi-source data fusion and deep learning, characterized in that, The method includes: Collect multi-source data and preprocess the multi-source data; wherein the multi-source data includes at least radar raw echo data, radar base data, and radio frequency interference monitoring data; The preprocessed raw radar echo data is then evaluated for quality using a first deep learning model. The preprocessed radar base data is then subjected to quality analysis using a correlation analysis model. Interference suppression is performed on the quality assessment results, quality analysis results, and the radio frequency interference monitoring data using an adaptive filtering model and a second deep learning model. A comprehensive evaluation index system based on multi-source data fusion was constructed for the interference suppression results.

2. The method according to claim 1, characterized in that, The preprocessed raw radar echo data is then subjected to quality assessment using a first deep learning model, including: The quality assessment capability of the first deep learning model is optimized based on the external calibration data of the UAV and the raw observation data; wherein, the raw observation data is the echo data collected by radar during the external calibration of the UAV; The preprocessed raw radar echo data is then evaluated for quality using the optimized first deep learning model.

3. The method according to claim 2, characterized in that, The quality assessment capability of the first deep learning model is optimized based on UAV external calibration data and raw observation data, including: The raw observation data is obtained, including radar echo intensity and radar coordinates; Align the UAV coordinates with the radar coordinates, wherein the UAV external calibration data includes the UAV coordinates; Calculate the theoretical value of radar echo intensity; Based on the radar echo intensity, an estimated value of radar echo intensity is obtained, and the deviation between the estimated value of radar echo intensity and the theoretical value of radar echo intensity is calculated. The deviation is input into the first deep learning model to optimize its quality assessment capability.

4. The method according to claim 3, characterized in that, The method further includes: Acquire spatial geometric information of the radar during UAV external calibration; The spatial geometric information and the original observation data are input into the first deep learning model to obtain the first feature information; The first feature information and the deviation are input into the fully connected layer of the first deep learning model to optimize the first deep learning model.

5. The method according to claim 2, characterized in that, The preprocessed raw radar echo data is then subjected to quality assessment using the optimized first deep learning model, including: The preprocessed raw radar echo data is input into the optimized first deep learning model, which outputs a quality assessment level; wherein the quality assessment level includes excellent, good, and poor. The quality assessment results are obtained by removing echo data with a poor quality assessment level.

6. The method according to claim 1, characterized in that, The multi-source data also includes geographic information system (GIS) topographic data and ground observation data; The preprocessed radar-based data is then subjected to quality analysis using a correlation analysis model, including: The preprocessed GIS terrain data, ground observation data, and radio frequency interference monitoring data are input into the correlation analysis model to obtain the radar base data error; The preprocessed radar base data is compensated for errors based on the radar base data to obtain quality analysis results.

7. The method according to claim 1, characterized in that, Interference suppression is performed on the quality assessment results, quality analysis results, and the aforementioned radio frequency interference monitoring data using an adaptive filtering model and a second deep learning model, including: The quality assessment results and the radio frequency interference monitoring data are adaptively filtered using the adaptive filtering model. The quality analysis results are then subjected to clutter suppression using a second deep learning model. The adaptive filtering results and clutter suppression results are optimized by interpolation.

8. The method according to claim 7, characterized in that, A comprehensive evaluation index system based on multi-source data fusion is constructed for the interference suppression results, including: The evaluation index dimensions are constructed, including data quality dimension, observation accuracy dimension, and anti-interference capability dimension; Calculate the value of the data optimization result under the evaluation index dimension; Compare the values ​​under the evaluation index dimensions with the preset thresholds to determine whether the multi-source data meets the standards.

9. The method according to claim 8, characterized in that, The method further includes: The data quality dimension is characterized by the mean difference and root mean square difference of the rainfall samples. The observation accuracy dimension is characterized by the correlation coefficient of the rainfall samples; The anti-interference capability dimension is characterized by the clutter recognition rate.

10. A quality control system for rainfall radar data based on multi-source data fusion and deep learning, characterized in that, The system includes: The acquisition and preprocessing module is used to acquire multi-source data and preprocess the multi-source data; wherein the multi-source data includes at least radar raw echo data, radar base data, and radio frequency interference monitoring data; The quality assessment module is used to assess the quality of the preprocessed raw radar echo data using a first deep learning model. The quality analysis module is used to perform quality analysis on the preprocessed radar base data using a correlation analysis model. The interference suppression module is used to suppress interference based on the quality assessment results, quality analysis results, and the radio frequency interference monitoring data through an adaptive filtering model and a second deep learning model. The comprehensive evaluation module is used to construct a comprehensive evaluation index system based on multi-source data fusion for interference suppression results.

Images