SVM-based RFI source estimation method and system
Through the SVM-based RFI source estimation method, the multi-classification model and eigenvalue difference operation are used to solve the problems of inaccurate source estimation and complex calculation in the prior art, and efficient and accurate RFI source estimation is achieved, which is suitable for dynamic interference scenarios.
Patent Information
- Application Number
- CN202510383147.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-28
- Publication Date
- 2025-08-12
AI Technical Summary
The existing RFI signal processing methods rely on manual experience to set thresholds, making it difficult to adapt to dynamic interference scenarios, lack reliable source estimation methods, resulting in missed or misdetection. The existing algorithms lack accuracy and efficiency under low signal-to-noise ratio conditions, making it difficult to process in real time.
The multi-classification support vector machine (SVM) model is used to generate correlation matrix through simulated antenna array imaging, matrix decomposition and feature value difference operations are performed, statistical feature data are generated, and RFI source estimation models suitable for different scenarios are trained, and adaptive estimation is performed based on actual measured bright temperature data.
It significantly improves the classification accuracy and model generalization capabilities of RFI source estimation, reduces the computational complexity, is suitable for real-time processing, adapts to different interference scenarios, and has physical interpretability and data-driven classification capabilities.
Smart Images
Figure CN120468787A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of radio frequency interference processing, and in particular to a method and system for estimating an RFI source based on SVM. Background Art
[0002] In passive microwave remote sensing, radio frequency interference (RFI) signals are considered unwanted electromagnetic radiation. Due to the high sensitivity of microwave radiometers, they are susceptible to interference from spatial electromagnetic frequencies during measurement, such as radar interference, communication interference, and electromagnetic waves emitted by man-made electronic and electromagnetic equipment. Even small amounts of RFI can cause significant errors in measurement results. Strong interference signals can cause radiometers to malfunction. For example, in the presence of RFI, soil moisture measurements will appear drier, sea surface salinity will be higher, sea surface temperature will be higher, and atmospheric water content will be lower. One approach to RFI mitigation is to process visibility data. A typical example is the RFI source location and mitigation method based on a direction of arrival estimation model proposed by Park et al. This RFI mitigation method requires prior knowledge of the number of RFI sources, but existing methods lack a scientific and reasonable method for estimating RFI sources. Therefore, RFI detection and suppression are one of the core challenges in microwave radiometer data processing.
[0003] Existing RFI processing methods generally have the following problems: abnormal signals are identified by setting power or statistical thresholds (such as kurtosis and skewness), and they are judged as RFI and eliminated. The thresholds of this method rely on manual experience to set, which is difficult to adapt to dynamically changing interference scenarios. It is not sensitive to low signal-to-noise ratio (SNR) interference and is prone to missed detection or false detection; existing methods lack reliable means of estimating the number of interference sources, resulting in limited practicality of the model; classification of interference signals based on algorithms such as decision trees and random forests relies on expert experience to extract time domain, frequency domain or statistical features, and the feature expression ability is limited; efficiency and accuracy are contradictory, and high-precision methods (such as subspace decomposition) are complex to calculate and difficult to process in real time; lightweight methods (such as threshold detection) have low accuracy and limited practicality. Summary of the Invention
[0004] The present invention aims to address the shortcomings of the prior art by providing a method and system for estimating RFI sources based on Support Vector Machines (SVMs). This method uses a multi-classification SVM model to predict the number of RFI interferences. Due to its simplicity and operability, this method is easy to implement in engineering.
[0005] On the one hand, a method for estimating RFI sources based on SVM is provided, comprising the following steps: S1: Simulate the antenna array imaging of the synthetic aperture microwave radiometer to obtain the correlation matrix without RF interference and with different numbers of RF interference sources; S2: performing matrix decomposition on the correlation matrix to extract its eigenvalues, and performing difference operations and second-order difference operations on the eigenvalue sequence to generate statistical feature data; S3: training a multi-classification support vector machine (SVM) model based on the statistical feature data to obtain an RFI source estimation model suitable for the target antenna array and scenario; S4: Inversely transforming the measured brightness temperature data back into the correlation matrix to extract the eigenvalue difference and the second-order difference as the input of the RFI source estimation model to achieve adaptive estimation of the number of radio frequency interference sources.
[0006] Furthermore, in step S1, obtaining the correlation matrix of no radio frequency interference and different numbers of radio frequency interference sources further includes: S11: Through simulation The antenna array of the synthetic aperture microwave radiometer receives the A signal, get The antenna signal is expressed as: ; S12: There is noise when the antenna receives the spatial signal. Antenna signal and spatial signal The following mathematical relationship exists: in, , is the spatial noise, ,in is the antenna’s steering vector; S13: The signals are uncorrelated with each other, the noises received by any two antennas are not correlated, and the spatial signal and noise are also not correlated. The antenna signals are correlated and the correlation matrix is obtained. It is expressed as follows: in, represents the noise power, ,in, Indicates the power of the signal.
[0007] Preferably, step S1 further comprises: Adapting to different synthetic aperture microwave radiometers by changing different antenna array formations, including L-type, Y-type, T-type and even distributed formations; Radio frequency interference simulation in different scenarios is achieved by modifying the correlation matrix, and the scenarios include a strong interference scenario, multiple weak interference scenarios, and a land-water boundary scenario.
[0008] Furthermore, in step S2, performing matrix decomposition on the correlation matrix to extract its eigenvalues further includes: For the correlation matrix Perform eigenvalue decomposition, because is a Hermite matrix, mathematically there exists: in, represents the eigenvector matrix of the signal subspace, Represents the eigenvector matrix of the noise subspace. Due to the characteristics of the radiometer, it is usually believed that there is no correlation between the detection signals, that is, it is usually believed that the eigenvalue matrix of the signal subspace is a small value, and the eigenvalue matrix of the noise subspace There is a large value, and the signal subspace and the noise subspace are superimposed to obtain the received signal. Therefore, the statistical characteristics of the eigenvalue are used as the classification basis of SVM.
[0009] Preferably, in step S2, performing a difference operation and a second-order difference operation on the eigenvalue sequence to generate statistical feature data further includes: The difference between adjacent eigenvalues is calculated to generate a first-order difference sequence, which reflects the decay rate of adjacent eigenvalues and highlights the mutation point between the signal subspace and the noise subspace. The first-order difference sequence is differentiated again to generate the second-order difference sequence, which further amplifies the curvature characteristics of the mutation point and enhances the distinguishability of the number of interference sources; Extract statistics from the first-order and second-order difference sequences as feature vectors, including the maximum value, minimum value, mean value, and standard deviation of the sequence: For the first-order difference sequence, the statistics are used to characterize the maximum decay amplitude, the minimum decay amplitude, the overall decay amplitude and the decay volatility; For the second-order difference sequence, the statistics are used to characterize the maximum curvature change, the minimum curvature change, the overall curvature trend and the volatility of the curvature.
[0010] Furthermore, in step S3, training a multi-classification support vector machine (SVM) model based on the statistical feature data further includes: S31: Build a multi-classification SVM model, train a binary SVM for each class, and select the class with the highest confidence; S32: Inputting an eight-dimensional statistical feature vector extracted from the difference and second sister difference sequences of the eigenvalues to train the multi-classification SVM model, outputting a label of the number of simulated interference sources, and setting a training optimization goal to maximize the classification accuracy and minimize the hinge loss; S33: The kernel function of the model is tuned by selecting the best kernel function and parameters through grid search. At the same time, the regularization parameters are adjusted to balance the model complexity and generalization ability, and finally an RFI source estimation model suitable for the target antenna array and scenario is obtained.
[0011] Furthermore, step S4 includes: S41: extracting original visibility data from the measured brightness temperature image, performing a de-interpolation operation on the extracted original visibility data, and then reconstructing a correlation matrix from the measured data after the de-interpolation operation based on a Hermite matrix; S42: performing matrix decomposition on the reconstructed correlation matrix to extract its eigenvalues, and performing difference operation and second-order difference operation on the eigenvalue sequence to generate statistical characteristic data of the measured data; S43: Inputting the statistical characteristic data of the measured data into the RFI source estimation model to obtain the predicted number of RFI interference sources in the actual brightness temperature map.
[0012] In another aspect, a SVM-based RFI source estimation system is provided, comprising: The simulation data generation module is used to simulate the antenna array imaging of the synthetic aperture microwave radiometer and obtain the correlation matrix without radio frequency interference and with different numbers of radio frequency interference sources; A feature extraction and processing module is used to perform matrix decomposition on the correlation matrix to extract its eigenvalues, and perform difference operations and second-order difference operations on the eigenvalue sequence to generate statistical feature data; A model training and optimization module is used to train a multi-classification support vector machine (SVM) model based on the statistical feature data to obtain an RFI source estimation model suitable for the target antenna array and scene; The prediction and application module is used to inversely transform the measured brightness temperature data into the correlation matrix to extract the difference and second-order difference of the eigenvalues as the input of the RFI source estimation model to achieve adaptive estimation of the number of radio frequency interference sources.
[0013] At the same time, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that the program includes a boot program and an application program, and when executed by a processor, implements any of the above-mentioned SVM-based RFI source estimation methods.
[0014] In addition, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement any of the SVM-based RFI source estimation methods described above.
[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention extracts statistical features through eigenvalue difference and second-order difference operations, and combines them with a multi-classification SVM model to significantly improve classification accuracy. By modifying the correlation matrix parameters (such as interference source power and incident angle), the present invention can quickly generate training data adapted to different scenarios, improve the model's generalization ability, and support flexible configuration of antenna arrays (L-type, Y-type, T-type, distributed) and interference scenarios (strong interference, multiple weak interference, and water-land interface); The present invention is based on the signal subspace and noise subspace separation characteristics of the Park Baud rate estimation model, uses the statistical characteristics of eigenvalues as the classification basis, and the feature extraction process has physical interpretability (for example, the eigenvalue mutation corresponds to the number of interference sources). At the same time, data-driven classification is achieved through SVM, taking into account both theoretical rigor and algorithmic flexibility. The present invention uses lightweight statistical features (8-dimensional feature vectors) instead of original high-dimensional data, significantly reducing the model's computational burden. Feature extraction only requires matrix decomposition and simple differential operations. The SVM inference speed is fast, making it suitable for real-time processing in embedded systems. The present invention inverts the correlation matrix from the measured brightness temperature image, generates input features through de-interpolation and matrix reconstruction, and adapts to the actual data flow. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of RFI source estimation based on SVM in the present invention; Figure 2 A schematic diagram of a T-shaped antenna array according to the present invention; Figure 3 This is a schematic diagram of visibility sampling of a T-type antenna according to the present invention; Figure 4 This is a schematic diagram comparing the estimation results of a 7-element signal source according to the present invention; Figure 5 This is a schematic diagram of a confusion matrix result of a 7-element source estimation according to the present invention; Figure 6 This is a schematic diagram of a strong interference brightness temperature of the present invention; Figure 7This is a schematic diagram of multiple weak interference brightness temperatures at the land-water interface of the present invention; Figure 8 This is a data flow diagram of RFI source estimation based on SVM in the present invention. DETAILED DESCRIPTION
[0017] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0018] The present invention first simulates an antenna array to obtain a correlation matrix. Radio frequency interference (RFI) is randomly added to the correlation matrix (including random number and position). The correlation matrix can also be modified based on scenario requirements (e.g., strong interference, multiple weak interferences, and water-land interface scenarios). Matrix decomposition is used to obtain eigenvalues, and the statistical characteristics of the eigenvalues are used as training input data for a multi-classification SVM model. The number of RFIs used is used as the training output data for the multi-classification SVM model, resulting in an RFI source estimation SVM model suitable for the antenna array. Matrix transformation is performed on the actual brightness temperature map to obtain a correlation matrix. The differences and second-order differences of the eigenvalues are used as prediction inputs for the SVM model, and the output is an estimate of the number of RFIs.
[0019] The specific implementation of the present invention is described below with reference to the accompanying drawings and embodiments.
[0020] Example 1 See also Figure 1 , which is a technical solution of a RFI source estimation method based on SVM provided in this embodiment, includes the following steps: S1: Simulate the antenna array imaging of the synthetic aperture microwave radiometer to obtain the correlation matrix without RF interference and with different numbers of RF interference sources; S2: performing matrix decomposition on the correlation matrix to extract its eigenvalues, and performing difference operations and second-order difference operations on the eigenvalue sequence to generate statistical feature data; S3: training a multi-classification support vector machine (SVM) model based on the statistical feature data to obtain an RFI source estimation model suitable for the target antenna array and scenario; S4: Inversely transforming the measured brightness temperature data back into the correlation matrix to extract the eigenvalue difference and the second-order difference as the input of the RFI source estimation model to achieve adaptive estimation of the number of radio frequency interference sources.
[0021] Wherein, in step S1, obtaining the correlation matrix of no radio frequency interference and different numbers of radio frequency interference sources further includes: S11: Through simulation The antenna array of the synthetic aperture microwave radiometer receives the A signal, get The antenna signal is expressed as: ; S12: There is noise when the antenna receives the spatial signal. Antenna signal and spatial signal The following mathematical relationship exists: in, , is the spatial noise, ,in is the antenna’s steering vector; S13: The signals are uncorrelated with each other, the noises received by any two antennas are not correlated, and the spatial signal and noise are also not correlated. The antenna signals are correlated and the correlation matrix is obtained. It is expressed as follows: in, represents the noise power, ,in, Indicates the power of the signal.
[0022] Specifically, we use Park's bode estimation model, where there are M signals in the space and in the direction cosine coordinate system In the following expression: in, Indicates The spatial signals at the specified position are represented by M, which represents the number of spatial signals. An antenna array consisting of N antennas receives these M spatial signals. These M spatial signals generate signals for each antenna under the action of "electric field superposition." The following formula represents these N antenna signals: in, , is the spatial noise, Where is the antenna's steering vector, and the specific relationship is shown in the following formula: In the above formula represents the coordinates of the i-th antenna; Indicates the wavelength of the spatial signal.
[0023] The M spatial signals are uncorrelated with each other. The noise received by any two antennas is not correlated, and the spatial signal and the noise are also uncorrelated. This can be expressed as follows: in, ; , represents the signal power; Represents the noise power. Combined with the above formula, the N antenna signals are correlated and the correlation matrix is It is expressed as follows: .
[0024] On this basis, different antenna array formations are changed to adapt to different synthetic aperture microwave radiometers. The antenna array formations include L-type, Y-type, T-type and even distributed formations. Radio frequency interference simulation in different scenarios is achieved by modifying the correlation matrix, and the scenarios include a strong interference scenario, multiple weak interference scenarios, and a land-water boundary scenario.
[0025] In step S2, performing matrix decomposition on the correlation matrix to extract its eigenvalues further includes: For the correlation matrix Perform eigenvalue decomposition, because is a Hermite matrix, mathematically there exists: in, represents the eigenvector matrix of the signal subspace, Represents the eigenvector matrix of the noise subspace. Due to the characteristics of the radiometer, it is usually believed that there is no correlation between the detection signals, that is, it is usually believed that the eigenvalue matrix of the signal subspace is a small value, and the eigenvalue matrix of the noise subspace There is a large value, and the signal subspace and the noise subspace are superimposed to obtain the received signal. Therefore, the statistical characteristics of the eigenvalue are used as the classification basis of SVM.
[0026] Therefore, the correlation matrix can be decomposed into eigenvalues, so as to determine the number of interference sources by the number of larger eigenvalues.
[0027] Then, we perform difference operations and second-order difference operations on the eigenvalue sequence to generate statistical feature data, further including: The difference between adjacent eigenvalues is calculated to generate a first-order difference sequence, which reflects the decay rate of adjacent eigenvalues and highlights the mutation point between the signal subspace and the noise subspace. The first-order difference sequence is differentiated again to generate the second-order difference sequence, which further amplifies the curvature characteristics of the mutation point and enhances the distinguishability of the number of interference sources; Extract statistics from the first-order and second-order difference sequences as feature vectors, including the maximum value, minimum value, mean value, and standard deviation of the sequence: For the first-order difference sequence, the statistics are used to characterize the maximum decay amplitude, the minimum decay amplitude, the overall decay amplitude and the decay volatility; For the second-order difference sequence, the statistics are used to characterize the maximum curvature change, the minimum curvature change, the overall curvature trend and the volatility of the curvature.
[0028] Next, training a multi-classification support vector machine (SVM) model based on the statistical feature data further includes: S31: Build a multi-classification SVM model, train a binary SVM for each class, and select the class with the highest confidence; S32: Inputting an eight-dimensional statistical feature vector extracted from the difference and second sister difference sequences of the eigenvalues to train the multi-classification SVM model, outputting a label of the number of simulated interference sources, and setting a training optimization goal to maximize the classification accuracy and minimize the hinge loss; S33: The kernel function of the model is tuned by selecting the best kernel function and parameters through grid search. At the same time, the regularization parameters are adjusted to balance the model complexity and generalization ability, and finally an RFI source estimation model suitable for the target antenna array and scenario is obtained.
[0029] Using the RFI source estimation model to perform RFI prediction further includes: S41: extracting original visibility data from the measured brightness temperature image, performing a de-interpolation operation on the extracted original visibility data, and then reconstructing a correlation matrix from the measured data after the de-interpolation operation based on a Hermite matrix; S42: performing matrix decomposition on the reconstructed correlation matrix to extract its eigenvalues, and performing difference operation and second-order difference operation on the eigenvalue sequence to generate statistical characteristic data of the measured data; S43: Inputting the statistical characteristic data of the measured data into the RFI source estimation model to obtain the predicted number of RFI interference sources in the actual brightness temperature map.
[0030] Specifically, in this embodiment, since the antennas of the synthetic aperture microwave radiometer are arranged in an array, different antenna arrangement arrays will bring different visibility sampling diagrams, such as Figure 2 and Figure 3The 7-element T-shaped antenna array and its visibility sampling diagram are presented. Because the correlation imaging between two antennas in a synthetic aperture radiometer has information redundancy (i.e., the correlation matrix is a Hermite matrix, and the diagonals all image the same area), the visibility sampling diagram is not 77. In theory, the visibility sampling values at the same coordinate are equal, but in practice, the same values are averaged.
[0031] A 7-element synthetic aperture radiometer was simulated with an integration length of 10,000 and a maximum number of RFI sources of 3. The simulation was repeated 2,000 times, with 80% of the data used as the training set for the SVM and 20% as the test set. Since the input data for the SVM cannot be raw data, it is necessary to extract the data features from the eigenvalues obtained after matrix decomposition. Here, the maximum value, minimum value, mean value, and standard deviation are used. The simulation of the 7-element array achieved an accuracy of 96.75% in 400 test sets. The comparison between the actual interference number and the predicted interference number is shown in the figure below. Figure 4 , the confusion matrix of the SVM model is as follows Figure 5 shown.
[0032] The multi-frequency and multi-angle passive detector integrates multi-band detection of P-band, L-band, C-band, X-band, K-band and Ka-band, and simultaneously performs multi-angle detection at 20°, 30° and 40° respectively. It is equipped with a 7-element T-type synthetic aperture radiometer in the C-band.
[0033] Multi-frequency and multi-angle passive detectors are mainly used for soil moisture detection at low frequencies (such as P band and L band), and for snow thickness detection at high frequencies. The brightness temperature image of the C band synthetic aperture radiometer near the point source after interpolation processing is as follows: Figure 6 As shown in the figure, it can be considered as the brightness temperature map under strong interference conditions. The brightness temperature map under the influence of land-water boundary and multiple weak interferences is shown in the figure. Figure 7 It should be noted that the image has not been calibrated, so the vertical axis can only represent the proportional relationship, not the actual brightness temperature.
[0034] By performing numerical extraction, deinterpolation, and matrix reconstruction on the image, the original 7×7 correlation matrix can be obtained. The eigenvalues of the matrix can then be extracted and used as the input of the SVM model. Next, the SVM-based RFI source estimation method is used to verify the number of interference sources. The matrix is decomposed and the eigenvalues are obtained as shown in Table 1 below.
[0035] Table 1 Eigenvalue table The identification results of the trained SVM model are that the number of RFIs is 1 in the strong interference scenario and 3 in the multiple weak interference scenarios at the land-water interface.
[0036] The specific data flow in this embodiment is shown in FIG. Figure 8 shown.
[0037] In addition, this embodiment also provides an SVM-based RFI source estimation system, including: The simulation data generation module is used to simulate the antenna array imaging of the synthetic aperture microwave radiometer and obtain the correlation matrix without radio frequency interference and with different numbers of radio frequency interference sources; A feature extraction and processing module is used to perform matrix decomposition on the correlation matrix to extract its eigenvalues, and perform difference operations and second-order difference operations on the eigenvalue sequence to generate statistical feature data; A model training and optimization module is used to train a multi-classification support vector machine (SVM) model based on the statistical feature data to obtain an RFI source estimation model suitable for the target antenna array and scene; The prediction and application module is used to inversely transform the measured brightness temperature data into the correlation matrix to extract the difference and second-order difference of the eigenvalues as the input of the RFI source estimation model to achieve adaptive estimation of the number of radio frequency interference sources.
[0038] It should be noted that the steps in the SVM-based RFI source estimation method provided in this embodiment can be implemented using the corresponding modules, devices, units, etc. in the SVM-based RFI source estimation system. Those skilled in the art can refer to the technical solution of the system to implement the step flow of the method, that is, the embodiments in the system can be understood as preferred examples for implementing the method, which will not be elaborated here.
[0039] In addition to implementing the system and its various devices provided by the present invention in purely computer-readable program code, it is entirely possible to implement the same functions of the system and its various devices provided by the present invention in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. by logically programming the method steps. Therefore, the system and its various devices provided by the present invention can be considered a hardware component, and the devices included therein for implementing the various functions can also be considered as structures within the hardware component; the devices for implementing the various functions can also be considered as both software modules implementing the method and structures within the hardware component.
[0040] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that improvements and modifications that do not depart from the principles of the present invention, which are apparent to those skilled in the art, should also be considered within the scope of protection of the present invention.
[0041] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
Claims
1. A method for estimating RFI sources based on SVM, characterized in that: The steps include: S1: Simulate the antenna array imaging of the synthetic aperture microwave radiometer to obtain the correlation matrix without RF interference and with different numbers of RF interference sources; S2: performing matrix decomposition on the correlation matrix to extract its eigenvalues, and performing difference operations and second-order difference operations on the eigenvalue sequence to generate statistical feature data; S3: training a multi-classification support vector machine (SVM) model based on the statistical feature data to obtain an RFI source estimation model suitable for the target antenna array and scenario; S4: Inversely transforming the measured brightness temperature data back into the correlation matrix to extract the eigenvalue difference and the second-order difference as the input of the RFI source estimation model to achieve adaptive estimation of the number of radio frequency interference sources.
2. The SVM-based RFI source estimation method according to claim 1, wherein: In step S1, obtaining a correlation matrix with no radio frequency interference and with different numbers of radio frequency interference sources further includes: S11: Through simulation The antenna array of the synthetic aperture microwave radiometer receives the A signal, get The antenna signal is expressed as: ; S12: There is noise when the antenna receives the spatial signal. Antenna signal and spatial signal The following mathematical relationship exists: in, , is the spatial noise, ,in is the antenna’s steering vector; S13: The signals are uncorrelated with each other, the noises received by any two antennas are not correlated, and the spatial signal and noise are also not correlated. The antenna signals are correlated and the correlation matrix is obtained. It is expressed as follows: in, represents the noise power, ,in, Indicates the power of the signal.
3. The RFI source estimation method based on SVM according to claim 1, characterized in that: Step S1 further comprises: By changing different antenna array formations to adapt to different synthetic aperture microwave radiometers, the antenna array formations include L-type, Y-type, T-type and even distributed formations; Radio frequency interference simulation in different scenarios is achieved by modifying the correlation matrix, and the scenarios include a strong interference scenario, multiple weak interference scenarios, and a land-water boundary scenario.
4. The SVM-based RFI source estimation method according to claim 2, wherein: In step S2, performing matrix decomposition on the correlation matrix to extract its eigenvalues further includes: For the correlation matrix Perform eigenvalue decomposition, because is a Hermite matrix, mathematically there exists: in, represents the eigenvector matrix of the signal subspace, Represents the eigenvector matrix of the noise subspace. Due to the characteristics of the radiometer, it is usually believed that there is no correlation between the detection signals, that is, it is usually believed that the eigenvalue matrix of the signal subspace is a small value, and the eigenvalue matrix of the noise subspace There is a large value, and the signal subspace and the noise subspace are superimposed to obtain the received signal. Therefore, the statistical characteristics of the eigenvalue are used as the classification basis of SVM.
5. The RFI source estimation method based on SVM according to claim 1, characterized in that: In step S2, performing a difference operation and a second-order difference operation on the eigenvalue sequence to generate statistical feature data further includes: The difference between adjacent eigenvalues is calculated to generate a first-order difference sequence, which reflects the decay rate of adjacent eigenvalues and highlights the mutation point between the signal subspace and the noise subspace. The first-order difference sequence is differentiated again to generate the second-order difference sequence, which further amplifies the curvature characteristics of the mutation point and enhances the distinguishability of the number of interference sources; Extract statistics from the first-order and second-order difference sequences as feature vectors, including the maximum value, minimum value, mean value, and standard deviation of the sequence: For the first-order difference sequence, the statistics are used to characterize the maximum decay amplitude, the minimum decay amplitude, the overall decay amplitude and the decay volatility; For the second-order difference sequence, the statistics are used to characterize the maximum curvature change, the minimum curvature change, the overall curvature trend and the volatility of the curvature.
6. The SVM-based RFI source estimation method according to claim 5, characterized in that: In step S3, training a multi-classification support vector machine (SVM) model based on the statistical feature data further includes: S31: Build a multi-classification SVM model, train a binary SVM for each class, and select the class with the highest confidence; S32: Inputting an eight-dimensional statistical feature vector extracted from the difference and second sister difference sequences of the eigenvalues to train the multi-classification SVM model, outputting a label of the number of simulated interference sources, and setting a training optimization goal to maximize the classification accuracy and minimize the hinge loss; S33: The kernel function of the model is tuned by selecting the best kernel function and parameters through grid search. At the same time, the regularization parameters are adjusted to balance the model complexity and generalization ability, and finally an RFI source estimation model suitable for the target antenna array and scenario is obtained.
7. The RFI source estimation method based on SVM according to claim 5, characterized in that: Step S4 further comprises: S41: extracting original visibility data from the measured brightness temperature image, performing a de-interpolation operation on the extracted original visibility data, and then reconstructing a correlation matrix from the measured data after the de-interpolation operation based on a Hermite matrix; S42: performing matrix decomposition on the reconstructed correlation matrix to extract its eigenvalues, and performing difference operation and second-order difference operation on the eigenvalue sequence to generate statistical characteristic data of the measured data; S43: Inputting the statistical characteristic data of the measured data into the RFI source estimation model to obtain the predicted number of RFI interference sources in the actual brightness temperature map.
8. A RFI source estimation system based on SVM, characterized in that: include: The simulation data generation module is used to simulate the antenna array imaging of the synthetic aperture microwave radiometer and obtain the correlation matrix without radio frequency interference and with different numbers of radio frequency interference sources; A feature extraction and processing module is used to perform matrix decomposition on the correlation matrix to extract its eigenvalues, and perform difference operations and second-order difference operations on the eigenvalue sequence to generate statistical feature data; A model training and optimization module is used to train a multi-classification support vector machine (SVM) model based on the statistical feature data to obtain an RFI source estimation model suitable for the target antenna array and scene; The prediction and application module is used to inversely transform the measured brightness temperature data into the correlation matrix to extract the difference and second-order difference of the eigenvalues as the input of the RFI source estimation model to achieve adaptive estimation of the number of radio frequency interference sources.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program includes a boot program and an application program, and when executed by a processor, implements the SVM-based RFI source estimation method according to any one of claims 1 to 6.
10. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement the SVM-based RFI source estimation method according to any one of claims 1 to 6.