L-band radar precipitation particle signal identification and extraction method

Abstract

The invention relates to an L-band radar precipitation particle signal identification and extraction method, which mainly comprises the following steps of: firstly, preprocessing acquired L-band radar data, and sequentially finishing direct current removal, three-point smoothing, noise level estimation and spectral moment calculation so as to suppress non-target signal interference and provide reliable initial reference; then, based on multi-beam radial velocity analysis, whether a rainfall signal exists or not is automatically judged, and subsequent processing is started only when it is confirmed that the rainfall signal exists; performing double-Gaussian or single-Gaussian model fitting on the power spectrum data with precipitation signals, and extracting parameters such as amplitude, expected value and standard deviation of the signals; and finally, through signal reconstruction and characteristic parameter analysis, signal attributes are judged in combination with the radial velocity mean value and the standard deviation, precipitation signals are extracted, and parameters are recorded. According to the invention, precipitation signals can be accurately extracted under the condition of multiple signals and superposition, redundancy calculation is reduced, and processing efficiency and signal identification accuracy are improved.

Description

Technical Field

[0001] The present invention belongs to the field of radar signal processing, and in particular relates to an L-band radar precipitation particle signal recognition and extraction method based on power spectrum data, which is used for accurately recognizing and extracting precipitation particle signals in the case where multiple echo signals and precipitation signals are superimposed. Background Art

[0002] Heavy precipitation is an extremely important type of meteorological disaster. The observation system of vertical change information of precipitation particles provides crucial data support for the detection, early warning and forecasting of meteorological disasters and other business and scientific research. In the meteorological radar system, L-band radar has a strong detection capability and can provide rich echo information for the monitoring of precipitation particles. In particular, it has potential advantages in detecting parameters such as the falling speed and echo intensity of precipitation particles. However, although L-band radar has these technical characteristics in precipitation particle detection, its actual application currently focuses more on the extraction of clear sky atmospheric turbulence signals, and the processing of precipitation signals has not received enough attention.

[0003] Existing L-band radar processing systems usually only perform simple classification by distinguishing the characteristics of clear sky signals and precipitation signals. However, in actual applications, precipitation signals often overlap with multi-echo signals, and the characteristics of the signals are not obviously different. This simple classification method is difficult to meet the actual needs under complex weather conditions. For example, in heavy precipitation processes or complex airflow environments, the overlap between precipitation particle signals and other interference signals is significantly enhanced, resulting in the inability of existing methods to accurately extract precipitation particle information, thereby reducing the reliability of radar data inversion results. In addition, most of the current processing flows lack a mechanism to automatically determine whether precipitation signal identification is required, which increases the complexity of operation and the risk of misidentification, limiting the application value of L-band radar in precipitation monitoring.

[0004] Compared with other precipitation signal extraction algorithms, the present invention proposes an optimized precipitation signal identification and extraction method. In the preprocessing stage, the present method can automatically determine whether there is a precipitation signal to avoid invalid processing. By adopting a double Gaussian and single Gaussian mixed model, it can effectively deal with the situation where precipitation signals are superimposed on multi-echo signals. Before fitting the precipitation signal, the present method calculates the initial values ​​of the relevant parameters, improves the fitting efficiency and effectively avoids overfitting. In addition, by introducing FFT resampling and signal reconstruction technology, while accurately extracting precipitation signals, the amount of information processed is significantly reduced and the calculation speed is improved. This comprehensive signal processing method significantly enhances the practicality and reliability of L-band radar in precipitation signal extraction. Summary of the invention

[0005] In view of the above-mentioned deficiencies of the prior art, the present invention proposes a method for identifying and extracting precipitation particle signals of an L-band radar, comprising the following steps:

[0006] Step 1: Preprocess the collected L-band radar data, including DC removal, three-point smoothing, noise level estimation, and spectral moment calculation operations;

[0007] Step 2: Based on the preprocessed data, determine whether there is a precipitation signal through a comprehensive judgment method based on multi-beam radial velocity analysis;

[0008] Step 3: Perform power spectrum fitting on the data with precipitation signals to obtain relevant signal parameters;

[0009] Step 4: According to the signal parameters obtained by fitting, perform the following processing:

[0010] Step 4.1: Reconstruct the power spectrum data of the precipitation signal echo1 and the non-precipitation signal echo2 according to the fitted signal parameters and noise level values;

[0011] Step 4.2: Analyze the reconstructed power spectrum data and calculate the key parameters of each signal, including the mean and standard deviation of the radial velocity;

[0012] Step 4.3: Extract precipitation signal;

[0013] Step 4.4: Repeat step 4.3 to process all beams and all range library data in turn: if the data of all beams and range libraries have been processed, save all precipitation signal parameters; otherwise, continue to process the next beam or range library until all calculations are completed.

[0014] Furthermore, in the step 1, the collected L-band radar data is preprocessed, including DC removal, three-point smoothing, noise level estimation, and spectral moment calculation operations, specifically: first, the collected L-band radar data is subjected to DC removal processing to suppress the influence of non-target signals such as zero-frequency interference; then, three-point smoothing processing is performed to further reduce signal noise; then, the noise level is estimated to provide an initial value reference for subsequent precipitation signal extraction; finally, spectral moment calculation is performed to calculate the Doppler velocity for subsequent precipitation signal judgment.

[0015] Furthermore, in step 2, based on the pre-processed data, a comprehensive judgment method based on multi-beam radial velocity analysis is used to determine whether there is a precipitation signal, specifically:

[0016] Step 2.1: Obtain the power spectrum data of each beam of the preprocessed radar data;

[0017] Step 2.2: Estimate the radial velocity of all range bins for the power spectrum data of each beam;

[0018] Step 2.3: In each beam, calculate the mean and variance of the radial velocity for all range bins in the range of 2-4 km;

[0019] Step 2.4: Determine the calculation results of each beam;

[0020] Step 2.5: Determine the signal containing multiple beams.

[0021] Furthermore, in step 3, power spectrum fitting is performed on the data containing precipitation signals to obtain relevant signal parameters, specifically, a double Gaussian or single Gaussian model is used to perform signal fitting;

[0022] Furthermore, in step 3, power spectrum fitting is performed on the data containing precipitation signals to obtain relevant signal parameters, and the specific steps are as follows:

[0023] Step 3.1: For each distance library power spectrum data judged to have precipitation signals, subtract the noise level value of the distance library to obtain the denoised power spectrum data;

[0024] Step 3.2: Resample the denoised power spectrum data with equal spacing, and set the FFT points of the resampled power spectrum data to 256;

[0025] Step 3.3: Use the double Gaussian model to fit the power spectrum data after equal spacing resampling;

[0026] Step 3.4: Classify the fitted signal parameters;

[0027] Step 3.5: Save the fitted parameters of precipitation signal echo1 and non-precipitation signal echo2.

[0028] Furthermore, the step 2.4: judging the calculation result of each beam is specifically as follows:

[0029] If v<v t or v >σ t , where v t and σ t If the threshold is determined based on historical radar data, it is determined that there is no precipitation signal in the beam, the information is recorded and saved in the database, and the next beam is processed;

[0030] If v ≥ v t And σ v ≤σ t , it is determined that there is a precipitation signal in the beam and the beam is marked as a precipitation signal beam.

[0031] Furthermore, the step 2.5: judging the signal including multiple beams is specifically as follows:

[0032] If at least one beam is marked as a precipitation signal beam, it is determined that there is a precipitation signal as a whole, and the relevant information is recorded; if all beams are not marked as precipitation signal beams, it is determined that there is no precipitation signal, and the information is recorded and saved in the database.

[0033] Furthermore, the step 3.3: using a double Gaussian model to fit the power spectrum data after equal-interval resampling, specifically:

[0034] If the double Gaussian model converges after fitting, record the amplitudes a1, a2, expected values ​​b1, b2 and standard deviations c1, c2 of the two Gaussian signals g1 and g2;

[0035] If the double Gaussian model fitting does not converge, use the single Gaussian model to fit the power spectrum data and record the amplitude a, expected value b, and standard deviation c of the single Gaussian signal.

[0036] Furthermore, the step 3.4: classifying the signal parameters obtained by fitting is specifically as follows:

[0037] If b1≥b2 and c1≥c2, the parameters of Gaussian signals g1 and g2 are stored in precipitation signal echo1 and non-precipitation signal echo2 respectively;

[0038] If only a single Gaussian model is used for fitting, the fitting parameters are stored in the precipitation signal echo1 or the non-precipitation signal echo2.

[0039] The further step 4.3: extracting precipitation signal, specifically comprises:

[0040] Step 4.3.1: Read the data of the reconstructed signals echo1 and echo2;

[0041] Step 4.3.2: For each beam i signal echo1, calculate its radial velocity mean v within the range below 8 km i and standard deviation d;

[0042] Step 4.3.3: For each range bin j of each beam, calculate the radial velocity v1 of the range bin signal echo1 ij and the radial velocity mean v i The difference is dv1,

[0043] Step 4.3.4: If dv1≤1.5×d, mark the distance library signal as a precipitation signal and record the precipitation signal parameters;

[0044] Step 4.3.5: If dv1>1.5×d, calculate the radial velocity v2 of the distance library signal echo2 ij The mean radial velocity The difference dv2,

[0045] Step 4.3.6: If dv2≤1.5×d, mark the distance library signal as a precipitation signal and record the precipitation signal parameters;

[0046] Step 4.3.7: If dv2>1.5×d, the distance library signal is marked as missing.

[0047] Compared with the traditional method, the present invention has the following advantages, thereby solving the corresponding technical problems:

[0048] 1. The present invention can determine whether there is a precipitation signal by itself, avoiding the drawbacks of traditional methods that require human intervention or unnecessary processing. By judging the mean and variance of multi-beam radial velocity and combining historical thresholds to automatically judge precipitation signals, a data screening mechanism for each beam and each distance library is formed, and the overall signal status is judged by the comprehensive results of multiple beams. This method avoids the limitations of traditional single beam or single threshold judgment, improves the accuracy and robustness of precipitation signal judgment, effectively reduces the processing time of invalid data, and improves real-time processing efficiency and reliability.

[0049] 2. The present invention proposes a precipitation signal extraction method based on reconstructed power spectrum. First, the power spectrum data is fitted by a double Gaussian model to accurately separate the characteristic parameters of precipitation signals and non-precipitation signals; when the double Gaussian model fitting does not converge or the signal complexity is low, it is automatically switched to a single Gaussian model, which improves the stability and calculation efficiency of the fitting. Based on the signal parameters and noise level values ​​obtained by fitting, the power spectrum data of the precipitation signal and the non-precipitation signal are reconstructed to ensure the accuracy and completeness of the signal separation; at the same time, focusing on the range below 8 kilometers, the radial velocity mean and standard deviation of the signal are calculated, and the difference between the radial velocity and the mean is used as the judgment condition. Through the cyclic processing of beam-by-beam and distance library-by-distance library, the dynamic classification and accurate extraction of precipitation signals are realized. This method effectively copes with the complex situation of multiple signals and superimposed signals, solves the problems of signal fitting failure, calculation redundancy and misjudgment in traditional methods, and ensures the comprehensiveness, accuracy and efficiency of precipitation signal extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flowchart for automatic judgment of precipitation signals;

[0051] Figure 2 This is a schematic diagram of precipitation signal extraction results; DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solution in the embodiment of the present invention in conjunction with the accompanying drawings in the embodiment of the present invention. The method includes the following steps:

[0053] Step 1: Preprocess the collected L-band radar data, including DC removal, three-point smoothing, noise level estimation, and spectral moment calculation operations;

[0054] Step 2: Based on the preprocessed data, determine whether there is a precipitation signal through a comprehensive judgment method based on multi-beam radial velocity analysis;

[0055] Step 3: Perform power spectrum fitting on the data with precipitation signals to obtain relevant signal parameters;

[0056] Step 4: According to the signal parameters obtained by fitting, perform the following processing:

[0057] Step 4.1: Reconstruct the power spectrum data of the precipitation signal echo1 and the non-precipitation signal echo2 according to the fitted signal parameters and noise level values;

[0058] Step 4.2: Analyze the reconstructed power spectrum data and calculate the key parameters of each signal, including the mean and standard deviation of the radial velocity;

[0059] Step 4.3: Extract precipitation signal;

[0060] Step 4.4: Repeat step 4.3 to process all beams and all range library data in turn: if the data of all beams and range libraries have been processed, save all precipitation signal parameters; otherwise, continue to process the next beam or range library until all calculations are completed.

[0061] Furthermore, in the step 1, the collected L-band radar data is preprocessed, including DC removal, three-point smoothing, noise level estimation, and spectral moment calculation operations, specifically: first, the collected L-band radar data is subjected to DC removal processing to suppress the influence of non-target signals such as zero-frequency interference; then, three-point smoothing processing is performed to further reduce signal noise; then, the noise level is estimated to provide an initial value reference for subsequent precipitation signal extraction; finally, spectral moment calculation is performed to calculate the Doppler velocity for subsequent precipitation signal judgment.

[0062] Furthermore, in step 2, based on the pre-processed data, a comprehensive judgment method based on multi-beam radial velocity analysis is used to determine whether there is a precipitation signal, specifically:

[0063] Step 2.1: Obtain the power spectrum data of each beam of the preprocessed radar data;

[0064] Step 2.2: Estimate the radial velocity of all range bins for the power spectrum data of each beam;

[0065] Step 2.3: In each beam, calculate the mean and variance of the radial velocity for all range bins in the range of 2-4 km;

[0066] Step 2.4: Determine the calculation results of each beam;

[0067] Step 2.5: Determine the signal containing multiple beams.

[0068] Furthermore, in step 3, power spectrum fitting is performed on the data containing precipitation signals to obtain relevant signal parameters, specifically, a double Gaussian or single Gaussian model is used to perform signal fitting;

[0069] Furthermore, in step 3, power spectrum fitting is performed on the data containing precipitation signals to obtain relevant signal parameters, and the specific steps are as follows:

[0070] Step 3.1: For each distance library power spectrum data judged to have precipitation signals, subtract the noise level value of the distance library to obtain the denoised power spectrum data;

[0071] Step 3.2: Resample the denoised power spectrum data with equal spacing, and set the FFT points of the resampled power spectrum data to 256;

[0072] Step 3.3: Use the double Gaussian model to fit the power spectrum data after equal spacing resampling;

[0073] Step 3.4: Classify the fitted signal parameters;

[0074] Step 3.5: Save the fitted parameters of precipitation signal echo1 and non-precipitation signal echo2.

[0075] Furthermore, the step 2.4: judging the calculation result of each beam is specifically as follows:

[0076] like or v >σ t , where v t and σ t If the threshold is determined based on historical radar data, it is determined that there is no precipitation signal in the beam, the information is recorded and saved in the database, and the next beam is processed;

[0077] like And σ v ≤σ t , it is determined that there is a precipitation signal in the beam and the beam is marked as a precipitation signal beam.

[0078] Furthermore, the step 2.5: judging the signal including multiple beams is specifically as follows:

[0079] If at least one beam is marked as a precipitation signal beam, it is determined that there is a precipitation signal as a whole, and the relevant information is recorded; if all beams are not marked as precipitation signal beams, it is determined that there is no precipitation signal, and the information is recorded and saved in the database.

[0080] Furthermore, the step 3.3: using a double Gaussian model to fit the power spectrum data after equal-interval resampling, specifically:

[0081] If the double Gaussian model converges after fitting, record the amplitudes a1, a2, expected values ​​b1, b2 and standard deviations c1, c2 of the two Gaussian signals g1 and g2;

[0082] If the double Gaussian model fitting does not converge, use the single Gaussian model to fit the power spectrum data and record the amplitude a, expected value b, and standard deviation c of the single Gaussian signal.

[0083] Furthermore, the step 3.4: classifying the signal parameters obtained by fitting is specifically as follows:

[0084] If b1≥b2 and c1≥c2, the parameters of Gaussian signals g1 and g2 are stored in precipitation signal echo1 and non-precipitation signal echo2 respectively;

[0085] If only a single Gaussian model is used for fitting, the fitting parameters are stored in the precipitation signal echo1 or the non-precipitation signal echo2.

[0086] The further step 4.3: extracting precipitation signal, specifically comprises:

[0087] Step 4.3.1: Read the data of the reconstructed signals echo1 and echo2;

[0088] Step 4.3.2: For each beam i signal echo1, calculate its radial velocity mean v within the range below 8 km i and standard deviation d;

[0089] Step 4.3.3: For each range bin j of each beam, calculate the radial velocity v1 of the range bin signal echo1 ij and the radial velocity mean v i The difference is dv1,

[0090] Step 4.3.4: If dv1≤1.5×d, mark the distance library signal as a precipitation signal and record the precipitation signal parameters;

[0091] Step 4.3.5: If dv1>1.5×d, calculate the radial velocity v2 of the distance library signal echo2 ij The mean radial velocity The difference dv2,

[0092] Step 4.3.6: If dv2≤1.5×d, mark the distance library signal as a precipitation signal and record the precipitation signal parameters;

[0093] Step 4.3.7: If dv2>1.5×d, the distance library signal is marked as missing.

[0094] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

Claims

1. A method for identifying and extracting precipitation particle signals from an L-band radar, comprising the following steps: Step 1: Preprocess the collected L-band radar data, including DC removal, three-point smoothing, noise level estimation, and spectral moment calculation operations; Step 2: Based on the preprocessed data, determine whether there is a precipitation signal through a comprehensive judgment method based on multi-beam radial velocity analysis; Step 3: Perform power spectrum fitting on the data with precipitation signals to obtain relevant signal parameters; Step 4: According to the signal parameters obtained by fitting, perform the following processing: Step 4.1: Reconstruct the power spectrum data of the precipitation signal echo1 and the non-precipitation signal echo2 according to the fitted signal parameters and noise level values; Step 4.2: Analyze the reconstructed power spectrum data and calculate the key parameters of each signal, including the mean and standard deviation of the radial velocity; Step 4.3: Extract precipitation signal; Step 4.4: Repeat step 4.3 to process all beams and all range library data in turn: if the data of all beams and range libraries have been processed, save all precipitation signal parameters; otherwise, continue to process the next beam or range library until all calculations are completed.

2. The method for identifying and extracting L-band radar precipitation particle signals as claimed in claim 1, characterized in that: In the step 1, the collected L-band radar data is preprocessed, including DC removal, three-point smoothing, noise level estimation, and spectral moment calculation operations, specifically: first, the collected L-band radar data is subjected to DC removal processing to suppress the influence of non-target signals such as zero-frequency interference; then, three-point smoothing processing is performed to further reduce signal noise; then, the noise level is estimated to provide an initial value reference for subsequent precipitation signal extraction; finally, spectral moment calculation is performed to calculate the Doppler velocity for subsequent precipitation signal judgment.

3. The method for identifying and extracting L-band radar precipitation particle signals as claimed in claim 1, characterized in that: In step 2, based on the pre-processed data, a comprehensive judgment method based on multi-beam radial velocity analysis is used to determine whether there is a precipitation signal, specifically: Step 2.1: Obtain the power spectrum data of each beam of the preprocessed radar data; Step 2.2: For the power spectrum data of each beam, estimate the radial velocity v of all range bins; Step 2.3: In each beam, calculate the average radial velocity of all range bins in the range 2-4 km and variance σ v ; Step 2.4: Determine the calculation results of each beam; Step 2.5: Determine the signal containing multiple beams.

4. The method for identifying and extracting L-band radar precipitation particle signals as claimed in claim 1, characterized in that: In the step 3, power spectrum fitting is performed on the data containing precipitation signals to obtain relevant signal parameters, specifically, a double Gaussian or single Gaussian model is used to perform signal fitting.

5. The method for identifying and extracting L-band radar precipitation particle signals as claimed in claim 4, characterized in that: In step 3, power spectrum fitting is performed on the data containing precipitation signals to obtain relevant signal parameters. The specific steps are as follows: Step 3.1: For each distance library power spectrum data judged to have precipitation signals, subtract the noise level value of the distance library to obtain the denoised power spectrum data; Step 3.2: Resample the denoised power spectrum data with equal spacing, and set the FFT point number of the resampled power spectrum data to 256; Step 3.3: Use the double Gaussian model to fit the power spectrum data after equal spacing resampling; Step 3.4: Classify the fitted signal parameters; Step 3.5: Save the fitted parameters of precipitation signal echo1 and non-precipitation signal echo2.

6. The method for identifying and extracting L-band radar precipitation particle signals as claimed in claim 2, characterized in that: The step 2.4: judging the calculation result of each beam, specifically: like or v >σ t , where v t and σ t If the threshold is determined based on historical radar data, it is determined that there is no precipitation signal in the beam, the information is recorded and saved in the database, and the next beam is processed; like And σ v ≤σ t , it is determined that there is a precipitation signal in the beam and the beam is marked as a precipitation signal beam.

7. The method for identifying and extracting L-band radar precipitation particle signals as claimed in claim 2, characterized in that: The step 2.5: judging the signal containing multiple beams is specifically as follows: If at least one beam is marked as a precipitation signal beam, it is determined that there is a precipitation signal as a whole, and the relevant information is recorded; if all beams are not marked as precipitation signal beams, it is determined that there is no precipitation signal, and the information is recorded and saved in the database.

8. The method for identifying and extracting L-band radar precipitation particle signals as claimed in claim 4, characterized in that: The step 3.3: using the double Gaussian model to fit the power spectrum data after equal spacing resampling, specifically: If the double Gaussian model converges after fitting, record the amplitudes a1, a2, expected values ​​b1, b2 and standard deviations c1, c2 of the two Gaussian signals g1 and g2; If the double Gaussian model fitting does not converge, use the single Gaussian model to fit the power spectrum data and record the amplitude a, expected value b, and standard deviation c of the single Gaussian signal.

9. The method for identifying and extracting L-band radar precipitation particle signals as claimed in claim 7, characterized in that: The step 3.4: classifying the signal parameters obtained by fitting, specifically: If b1≥b2 and c1≥c2, the parameters of Gaussian signals g1 and g2 are stored in precipitation signal echo1 and non-precipitation signal echo2 respectively; If only a single Gaussian model is used for fitting, the fitting parameters are stored in the precipitation signal echo1 or the non-precipitation signal echo2.

10. The method for identifying and extracting L-band radar precipitation particle signals according to claim 1, characterized in that: The step 4.3: extracting precipitation signals, specifically comprises: Step 4.3.1: Read the data of the reconstructed signals echo1 and echo2; Step 4.3.2: For each beam i signal echo1, calculate its radial velocity mean within the range below 8 km and standard deviation d; Step 4.3.3: For each range bin j of each beam, calculate the radial velocity v1 of the range bin signal echo1 ij The mean radial velocity The difference dv1, Step 4.3.4: If dv1≤1.5×d, mark the distance library signal as a precipitation signal and record the precipitation signal parameters; Step 4.3.5: If dv1>1.5×d, calculate the radial velocity v2 of the distance library signal echo2 ij The mean radial velocity The difference dv2, Step 4.3.6: If dv2≤1.5×d, mark the distance library signal as a precipitation signal and record the precipitation signal parameters; Step 4.3.7: If dv2>1.5×d, the distance library signal is marked as missing.

Images