Methods and Systems for Modeling and Adaptive Suppression of Clutter Background in Radar Systems

Abstract

This invention discloses a method and system for clutter background modeling and adaptive suppression in radar systems, relating to the field of radar signal target detection technology. The method includes the following steps: covariance matrix construction, clutter model construction, adaptive clutter suppression, and adaptive suppression processing optimization. This invention performs a covariance matrix stability assessment based on the constructed covariance matrix and determines whether covariance matrix correction is needed. Then, an initial clutter model is established based on the covariance matrix and a preset probability distribution model. Next, the adaptability of the clutter model is assessed, and it is determined whether clutter model optimization is needed. Adaptive suppression processing is then applied to the radar signal. Finally, the result of the adaptive suppression processing optimization is obtained based on the signal-to-noise ratio of the radar signal after adaptive suppression processing. This improves the clutter suppression effect in radar signals and solves the problem of poor clutter suppression in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of radar signal target detection technology, and in particular to a method and system for radar system clutter background modeling and adaptive suppression. Background Technology

[0002] Radar systems detect targets by emitting electromagnetic waves and receiving echo signals. For example, border security radar is often used in conjunction with other security systems (such as cameras) to significantly compensate for missed detections by cameras and improve the reliability of the security system. The echo signals can be transmitted to satellite communication systems after certain packet processing, especially in scenarios where radar signal data needs to be transmitted to remote monitoring stations or for further processing. However, in real-world scenarios, such as areas around railways, radar signals contain not only target echoes but also a large amount of clutter interference. This clutter can come from various sources, including but not limited to natural environmental clutter, artificial interference clutter, and internal system noise. Background modeling primarily utilizes the characteristics of different clutter sources (such as trains, weather, trees, etc.) to select mathematical models to describe the statistical properties of clutter. Finally, a mathematical model that accurately describes the clutter distribution is constructed by analyzing the spatial and temporal distribution of the acquired radar signals to establish the spatial-temporal covariance matrix of the signal. This matrix reflects the statistical characteristics of clutter, providing support for subsequent clutter suppression and target detection.

[0003] Adaptive suppression techniques aim to dynamically adjust radar signal processing algorithms based on clutter characteristics to weaken or suppress clutter interference, thereby improving the reliability of target detection. For example, Space-Time Adaptive Processing (STAP) is a commonly used adaptive filtering technique in radar systems. It aims to suppress clutter by analyzing the spatial and temporal characteristics of the signal and designing optimized filters. For instance, when a train passes, the filter removes interference from train reflections by real-time estimation of clutter characteristics (such as power and correlation) while preserving the characteristics of the target signal.

[0004] For example, a detection before tracking method based on G0 clutter background and constant target amplitude disclosed in the Chinese invention patent announcement with the announcement number: CN104062651B includes: Step 1, initialize system parameters: the shape parameter α of G0 clutter, the scale parameter β of G0 clutter, and the target echo amplitude A, the number of frames K processed by the dynamic programming detection before tracking algorithm, and initialize the variable k = 1; Step 2, read the measurement value Zk of the k-th frame echo data from the radar receiver; Step 3, use the shape parameter α and scale parameter β of G0 clutter and the target echo amplitude A to calculate the weighted incomplete gamma function and sij for each measurement unit (i,j); Step 4, calculate the measurement value amplitude likelihood ratio for each measurement unit (i,j) using the weighted incomplete gamma function and sij; Step 5, accumulate the value function; Step 6, if k < K, update k = k + 1 and return to Step 2; if k = K, execute Step 7; Step 7, store the states corresponding to the value function exceeding the threshold VT in the K-th frame in the target candidate state set D; Step 8, track recovery; Step 9, output the detection result and the target track after deleting the false tracks.

[0005] For example, the distance extended target detection method based on Wald under interference plus clutter background disclosed in the Chinese patent application with the publication number: CN119148078A includes: assuming that the interference subspace is linearly independent of the signal subspace and the coordinates of the two subspaces are unknown, the clutter component is modeled as a complex Gaussian vector with zero mean, and using the dimensionality reduction subspace method and the skew-symmetry transformation matrix structure to solve the maximum likelihood estimates of the unknown clutter covariance matrix, the coordinate vector of the interference, and the coordinate vector of the signal, so as to establish an RPD2Wald detector for target detection.

[0006] The above technologies at least have the following technical problems:

[0007] In the prior art, extreme environments may lead to very complex clutter signals received by the radar. For example, the clutter sources in railway scenarios are diverse and dynamically changing (such as train movement, weather changes), and the existing clutter models (such as Rayleigh distribution, Kwasny-Gaussian distribution, Laplace distribution) are designed based on idealized assumptions and are difficult to fully cover the complexity of extreme environments. The inapplicability of the clutter model will lead to poor clutter suppression effects and even misjudging targets as clutter. And the covariance matrix is a key parameter describing the statistical characteristics of clutter, but the clutter characteristics in railway scenarios change rapidly over time (such as sudden changes in wind direction), resulting in unstable covariance matrix estimation. The unstable covariance matrix causes the filter parameters to deviate from the actual clutter characteristics, and there is a problem of poor clutter suppression in radar signals. Summary of the Invention

[0008] To address the problem of poor clutter suppression in existing radar signals, embodiments of the present invention provide a method and system for clutter background modeling and adaptive suppression in radar systems. The technical solution is as follows:

[0009] On the one hand, a method for radar system clutter background modeling and adaptive suppression is provided. This method is implemented by a radar system clutter background modeling and adaptive suppression system. The method includes: S1, constructing a covariance matrix based on the collected environmental detection data reflecting the characteristics of the target environment and the radar signal reflecting the interaction between the monitored target and the environment; performing a covariance matrix stability evaluation based on the constructed covariance matrix to reflect the difference between the reference covariance matrix and the covariance matrix, and determining whether to perform covariance matrix correction for suppressing environmental noise; S2, if covariance matrix correction is performed, establishing an initial clutter model reflecting the target environment and its interference characteristics based on the corrected covariance matrix and a preset probability distribution model; otherwise, based on the covariance matrix and a preset probability distribution model... S3: An initial clutter model is established based on the probability distribution model, which is selected by the statistical characteristics of the covariance matrix. S4: The clutter model is evaluated for adaptability based on the adaptability evaluation parameters of the initial clutter model, and it is determined whether to optimize the clutter model to ensure its accuracy. If the clutter model is optimized, the radar signal is adaptively suppressed based on the optimized initial clutter model and the space-time filter. Otherwise, the radar signal is directly adaptively suppressed based on the initial clutter model and the space-time filter. S5: The adaptive suppression processing optimization result is obtained based on the signal-to-noise ratio of the radar signal after adaptive suppression processing. If the result is yes, the adaptive suppression processing is unqualified; otherwise, the adaptive suppression processing is qualified.

[0010] On the other hand, a radar system clutter background modeling and adaptive suppression system is provided. This system is applied to radar system clutter background modeling and adaptive suppression methods. The system includes: a covariance matrix construction module, a clutter model construction module, a clutter adaptive suppression module, and an adaptive suppression processing optimization module. The covariance matrix construction module constructs a covariance matrix based on collected environmental detection data reflecting target environmental characteristics and radar signals reflecting the interaction between the monitored target and the environment. Based on the constructed covariance matrix, a covariance matrix stability assessment is performed to reflect the difference between the reference covariance matrix and the actual covariance matrix, and to determine whether covariance matrix correction for suppressing environmental noise is required. The monitored target represents the object that needs to be identified in the radar system. The clutter model construction module, if covariance matrix correction is required, constructs a model based on the corrected covariance matrix and a preset probability distribution model. An initial clutter model reflecting the target environment and its interference characteristics is established; otherwise, an initial clutter model is established based on the covariance matrix and a preset probability distribution model, which is selected based on the statistical characteristics of the covariance matrix. The clutter adaptive suppression module is used to evaluate the adaptability of the clutter model based on the adaptability evaluation parameters of the initial clutter model and to determine whether to optimize the clutter model to ensure its accuracy. If clutter model optimization is performed, adaptive suppression processing of the radar signal is performed based on the optimized initial clutter model and the space-time filter; otherwise, adaptive suppression processing of the radar signal is performed directly based on the initial clutter model and the space-time filter. The adaptive suppression processing optimization module is used to obtain the judgment result of the adaptive suppression processing optimization based on the signal-to-noise ratio of the radar signal after adaptive suppression processing. If the judgment result is yes, it means that the adaptive suppression processing is unqualified; otherwise, it means that the adaptive suppression processing is qualified.

[0011] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0012] 1. By combining environmental monitoring data and radar signals, the interaction characteristics between environmental noise and target information can be effectively captured, thereby establishing a covariance matrix. The stability of the covariance matrix is ​​then evaluated and corrected, effectively reducing noise interference with radar signal processing and improving its accuracy. Subsequently, an initial clutter model is established based on the covariance matrix and a preset probability distribution model. The probability distribution model is selected based on the statistical characteristics of the covariance matrix, ensuring the clutter model accurately reflects environmental noise characteristics and effectively simulates environmental noise. This provides a basis for subsequent adaptive suppression processing, further improving the accuracy of the clutter model. Next, by evaluating the adaptability of the initial clutter model, it can be determined whether it accurately reflects actual environmental noise. Poor model adaptability may lead to poor filtering effects, necessitating clutter model optimization. The optimized model improves filtering performance, enhancing radar signal quality. Finally, by calculating the signal-to-noise ratio of the radar signal after adaptive suppression processing, it can be determined whether further optimization of the adaptive suppression processing is needed to ensure monitoring is unaffected by signal noise, providing more comprehensive support for security monitoring.

[0013] 2. A noise power selection correction algorithm is used to update the covariance matrix. Kalman filtering can effectively adjust the model and filter noise, avoiding unnecessary covariance matrix updates and ensuring stable operation of the radar system even in high-noise environments. Recursive least squares has lower computational complexity and can update the covariance matrix more quickly and effectively, thus ensuring the stability of the covariance matrix correction process. However, in complex weather conditions, the stability of the covariance matrix cannot be accurately assessed. Therefore, alternative solutions are needed. By combining time window length mapping with adaptive adjustment, various complex environmental changes can be addressed. If the signal frequency change rate is higher, database mapping can more efficiently handle complex situations. Otherwise, adaptive adjustment can maintain the simplicity and real-time performance of data processing, thereby enhancing the radar system's adaptability to different signal frequency change rates during clutter background modeling. Then, adjusting the time window length according to the frequency change rate of the radar signal allows for more flexible signal processing in dynamic environments. Next, adaptively adjusting the time window length based on the signal frequency reduces computational overhead and enables more efficient adjustment of the time window length, improving the flexibility of covariance matrix correction in clutter background modeling. Finally, within the time window length, the environmental monitoring data is converted into a frequency domain signal through fast Fourier transform, thereby correcting the covariance matrix and facilitating efficient calculation of the covariance matrix.

[0014] 3. By setting the clutter model corresponding to the covariance matrix template of the maximum matrix distribution similarity as the clutter model of the current environmental monitoring data, the clutter model can be updated in a short time, ensuring the real-time performance of the radar system clutter background modeling method in complex environments. However, in high-precision radar detection tasks, the clutter model needs to accurately describe the clutter characteristics in the environment to reduce clutter interference with target signals. Furthermore, frequent switching between different environments may be required, which may prevent the timely provision of a sufficiently accurate clutter model, making it difficult to meet the demands of real-time high-precision detection. By introducing the adaptive exponential rate of change as an adjustment criterion, the optimization strategy of the clutter model can be adaptively and dynamically adjusted. In unstable environments, this method accelerates the optimization process and avoids error accumulation; in stable environments, it avoids over-adjustment and reduces computational burden. Then, by introducing a learning rate adjustment factor and correcting the parameters of the gradient descent method, the parameters of the clutter model are adjusted more precisely, improving the stability and rapid response capability of the optimization process. This allows the clutter model to quickly reach its optimal state, enabling it to adapt to changes in different environments and accelerate convergence, avoiding premature stopping or over-adjustment, thereby achieving better optimization results and improving the adaptability and accuracy of the clutter model. Finally, the clutter model is adjusted according to the preset learning rate to maintain the efficiency of the optimization process and ensure that radar signal processing can be completed quickly in stable environments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0016] Figure 1 A flowchart illustrating radar system clutter background modeling and adaptive suppression provided in this application embodiment;

[0017] Figure 2 A flowchart illustrating the radar system clutter background modeling and adaptive suppression method provided in this application embodiment;

[0018] Figure 3 A flowchart for adjusting clutter model parameters provided in this application embodiment; Detailed Implementation

[0019] This application provides a method and system for clutter background modeling and adaptive suppression in radar systems, which solves the problem of poor clutter suppression in radar signals in the prior art. It evaluates the stability of the constructed covariance matrix and determines whether to correct it. Then, it establishes an initial clutter model based on the covariance matrix and a preset probability distribution model. Next, it evaluates the adaptability of the clutter model and determines whether to optimize it. Adaptive suppression processing is then applied to the radar signal. Finally, the result of the adaptive suppression processing optimization is obtained based on the signal-to-noise ratio of the radar signal after adaptive suppression, thus improving the clutter suppression effect in radar signals.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] like Figure 1 The flowchart shown illustrates the clutter background modeling and adaptive suppression process for a radar system. The specific logic is as follows: First, environmental detection data and radar signals are collected. Then, a covariance matrix is ​​constructed. Next, the stability of the covariance matrix is ​​evaluated. If the stability index of the covariance matrix is ​​not greater than a preset stability index, the covariance matrix is ​​corrected; otherwise, no correction is performed. Simultaneously, an initial clutter model is established based on the covariance matrix and a preset probability distribution model. Then, the adaptability of the clutter model is evaluated. If the adaptability index of the clutter model is not greater than a preset adaptability index, the clutter model is optimized; otherwise, no optimization is performed. Finally, adaptive suppression processing of the radar signal is performed based on the clutter model and a space-time filter. Finally, if the radar signal-to-noise ratio (SNR) is not greater than a preset SNR, adaptive suppression processing optimization is performed; otherwise, no optimization is performed.

[0022] Example 1: This embodiment of the invention provides a method for radar system clutter background modeling and adaptive suppression, applicable to security detection in specific areas, such as railway areas. Figure 2The flowchart shown is for a radar system clutter background modeling and adaptive suppression method. The processing flow of this method may include the following steps: S1, constructing a covariance matrix based on collected environmental detection data reflecting target environmental characteristics and radar signals reflecting the interaction between the target and the environment; performing a covariance matrix stability assessment based on the constructed covariance matrix to reflect the difference between the reference covariance matrix and the covariance matrix; and determining whether to perform covariance matrix correction for suppressing environmental noise. The target to be monitored represents the object that needs to be identified in the radar system; environmental detection data includes, but is not limited to, temperature, humidity, precipitation, and wind speed; radar signals... The signal is acquired through radar, such as a security radar, a type of radar used to monitor and detect intruders or other potential security threats. It employs non-contact remote sensing technology, using radio waves to detect the presence and movement of objects. The covariance matrix can capture the relationship between radar signals and environmental noise, thereby helping to identify the impact of noise on target monitoring. Through stability assessment, if the covariance matrix shows instability, correction can be made to the covariance matrix to effectively reduce the impact of environmental noise and improve the quality of radar signals. The stability of the covariance matrix directly affects the effectiveness of subsequent adaptive suppression processing, ensuring that subsequent processing steps are more robust.

[0023] S2, if covariance matrix correction is performed, an initial clutter model reflecting the target environment and its interference characteristics is established based on the corrected covariance matrix and a preset probability distribution model. Otherwise, an initial clutter model is established based on the covariance matrix and the preset probability distribution model, which is selected based on the statistical characteristics of the covariance matrix. Specifically, the statistical characteristics extracted from the covariance matrix (such as mean, variance, correlation, eigenvalues, etc.) are used to evaluate the fitness of different distribution models (such as Gaussian distribution, Rayleigh distribution, etc.) using maximum likelihood estimation, and the distribution model with the maximum likelihood value is selected as the final clutter model. By selecting the probability distribution model, the characteristics of the interference source can be captured more accurately, thus providing a basis for subsequent clutter suppression and target identification. This ensures that the clutter model accurately reflects the statistical characteristics of clutter and interference, enabling subsequent steps to effectively distinguish between target signals and environmental noise.

[0024] S3. Based on the adaptability evaluation parameters of the initial clutter model, perform clutter model adaptability evaluation, and determine whether to optimize the clutter model to ensure the accuracy of the clutter model. If the clutter model is optimized, perform adaptive suppression processing on the radar signal based on the optimized initial clutter model and the space-time filter. Otherwise, directly perform adaptive suppression processing on the radar signal based on the initial clutter model and the space-time filter. The space-time filter is used to perform weighted processing on the received space-time signal through the optimal weight vector to obtain the optimal output signal. The space-time signal is obtained by signal acquisition at a preset number (set by a preset person) of antenna arrays and time points. The optimal weight vector is obtained by processing the space-time covariance matrix and the modified space-time covariance matrix using the minimum mean square error. The modified space-time covariance matrix is obtained by modifying the space-time covariance matrix through the Kwasny-Gaussian distribution. The space-time covariance matrix represents the result calculated from the space-time data matrix constructed by the space-time signal; by optimizing the clutter model, it can respond to different noise interference sources in real time and maintain the best radar signal quality; the space-time filter weights the radar signal by using the optimal weight vector, so that the radar system can filter clutter more precisely. [[ID=`1]]

[0025] It should be added that the specific process of adaptive suppression processing is as follows: First, select a clutter model, such as the Weibull distribution; estimate clutter parameters (such as shape parameters, scale parameters) from the training data in the target-free area, train a neural network using historical data to predict the clutter distribution, and confirm the matching degree between the clutter model and the actual clutter through a goodness-of-fit test (such as the chi-square test); then calculate the covariance matrix of clutter plus noise from the training data, perform an inverse operation on the covariance matrix, and multiply it by the target steering vector containing the spatial position and Doppler information of the target to obtain the weight vector; multiply the conjugate transpose of the weight vector by the training data to obtain the filtered signal; finally, to adapt to the non-stationary characteristics of the clutter, update the covariance matrix in real time through the recursive least squares algorithm, and recalculate the optimal weight vector according to the updated covariance matrix to ensure that the filter always matches the current clutter environment and achieve adaptive tracking.

[0026] S4. Obtain the determination result of the optimization of the adaptive suppression processing based on the signal-to-noise ratio of the radar signal after the adaptive suppression processing. If the determination result is yes, it means that the adaptive suppression processing is unqualified; otherwise, it means that the adaptive suppression processing is qualified; the judgment of the signal-to-noise ratio can effectively improve the quality of the radar signal and avoid false alarms or missed alarms due to poor signal quality.

[0027] In this embodiment, by correcting the covariance matrix, optimizing the clutter model, and weighting the space-time filter, the target radar signal can be extracted more accurately from the clutter background, reducing noise interference and ensuring that the target identification and monitoring capabilities can still be maintained under different environmental and noise conditions. Video data can supplement the target information that the radar signal cannot directly capture, providing more comprehensive support for security monitoring and improving the clutter suppression effect in the radar signal.

[0028] Furthermore, a stability assessment of the covariance matrix is ​​performed based on the constructed covariance matrix. The specific method is as follows: Each element in the reference covariance matrix corresponding to the maximum matrix distribution similarity is compared one-to-one with each element in the covariance matrix constructed within a preset time period using mean absolute error (MAE). The reference covariance matrix is ​​obtained from a preset database and represents a matrix constructed based on historical environmental monitoring data and radar signals for comparison with the covariance matrix, used to detect deviations in the covariance matrix. The preset time period is set by preset personnel. The MAE processing in this application refers to averaging the absolute values ​​of the difference calculation results. By calculating the MAE, the degree of deviation between the current covariance matrix and the reference covariance matrix can be intuitively reflected, providing basic data for assessing the stability of the covariance matrix. A larger MAE may lead to a higher rate of change in the covariance matrix, potentially resulting in a larger deviation in the stability assessment; simultaneously, the trace change coefficient may also increase, especially when the error is mainly concentrated in the diagonal elements.

[0029] The maximum matrix distribution similarity is obtained by sorting the matrix distribution similarities between the covariance matrix obtained using Jensen-Shannon divergence and each reference covariance matrix in the preset database in descending order. Jensen-Shannon divergence is used to more evenly measure the difference between two covariance matrices. The maximum matrix distribution similarity reflects the closest distribution between the current covariance matrix and a certain reference covariance matrix in the preset database.

[0030] The covariance matrix is ​​constructed by mapping each element (covariance) of the covariance matrix within a preset time period to each element of the covariance matrix constructed in the previous preset time period, and then performing a comprehensive relative deviation processing to obtain the covariance matrix change rate. In this application, the comprehensive relative deviation processing means performing relative deviation processing first, and then averaging; the relative deviation processing means first performing a difference calculation, taking the absolute value of the difference calculation result, and then performing a ratio calculation, where the denominator in the ratio calculation is each element of the covariance matrix constructed in the previous preset time period. By calculating the covariance matrix change rate, the changes in the covariance matrix between adjacent preset time periods are reflected, which helps to understand the trend of the covariance matrix over time and to promptly detect potential abnormal changes. A higher covariance change rate may lead to an increase in the trace change coefficient, especially when the matrix size is large, the trace change coefficient will be more sensitive to changes.

[0031] The trace of the covariance matrix constructed within a preset time period and the trace of the reference covariance matrix corresponding to the maximum matrix distribution similarity are deviated to obtain the trace change coefficient. The trace of the covariance matrix represents the sum of the diagonal elements of the matrix; the deviation processing in this application refers to performing a difference operation. By calculating the trace change coefficient, the overall change of the covariance matrix relative to the reference covariance matrix is ​​reflected, providing information for evaluating the stability of the covariance matrix from another perspective.

[0032] A matrix stability assessment balance factor is introduced. After weighting and coupling the matrix stability assessment coefficients, an inverse proportional operation is performed to obtain the covariance matrix stability index. The covariance matrix stability index is used to quantitatively assess the stability of the covariance matrix. The matrix stability assessment coefficients include the unit-free mean absolute error of the matrix, the rate of change of the covariance matrix, and the unit-free trace coefficient. The specific constrained expression for the covariance matrix stability index is as follows: In the formula, S represents the covariance matrix stability index, s1 represents the matrix mean absolute error between the reference covariance matrix and the covariance matrix, s2 represents the rate of change of the covariance matrix between two adjacent preset time periods, s3 represents the trace change coefficient between the reference covariance matrix and the covariance matrix, α1 represents the first matrix stability assessment balance factor, α2 represents the second matrix stability assessment balance factor, and α3 represents the third matrix stability assessment balance factor.

[0033] The matrix stability assessment balance factors involved are obtained from a preset database and set by preset personnel. Specifically, they include a first matrix stability assessment balance factor, a second matrix stability assessment balance factor, and a third matrix stability assessment balance factor, the sum of which is 1. For example, the matrix mean absolute error and the first matrix stability assessment balance factor form a corresponding first matrix stability assessment balance mapping set. The real-time matrix mean absolute error is input into the first matrix stability assessment balance mapping set to obtain the first matrix stability assessment balance factor. The first matrix stability assessment balance factor represents the degree of influence of the matrix mean absolute error on the covariance matrix stability index; the second matrix stability assessment balance factor represents the degree of influence of the covariance matrix change rate on the covariance matrix stability index; and the third matrix stability assessment balance factor represents the degree of influence of the trace change coefficient on the covariance matrix stability index.

[0034] Specifically, the process for determining whether to perform covariance matrix correction for suppressing environmental noise is as follows: If the stability index of the covariance matrix is ​​not greater than the preset stability index of the covariance matrix obtained from the preset database (set by preset personnel), then covariance matrix correction is performed; otherwise, covariance matrix correction is not performed. Covariance matrix correction refers to the selection of a covariance matrix update method based on noise power to ensure the effectiveness of the corrected covariance matrix in suppressing environmental noise.

[0035] In this embodiment, the above steps enable a comprehensive and accurate quantitative assessment of the stability of the covariance matrix, allowing for correction of the covariance matrix to adapt to environmental changes and suppress environmental noise. This helps improve the reliability of the covariance matrix construction, ensuring stable and accurate operation in different environments and reducing the impact of environmental noise on performance.

[0036] Furthermore, the specific process for covariance matrix correction is as follows: If the noise power is greater than the preset noise power, it indicates that the noise has a greater impact on the covariance matrix. Kalman filtering is then used to adaptively update the covariance matrix. Kalman filtering can better handle noise interference, making it more accurately reflect the true state of the radar system. Otherwise, it indicates that the noise has a smaller impact on the covariance matrix. Recursive Least Squares (RLS) is then used to adaptively update the covariance matrix, which can more efficiently utilize environmental monitoring data, gradually adjust the estimation of the covariance matrix, and reduce computational complexity. The preset noise power is set by preset personnel. Noise power refers to the energy of the noise signal per unit time, usually expressed as the integral of the power spectral density within a specific frequency band (set by preset personnel). In radar systems, noise power reflects the strength of noise in the received signal and is one of the important factors affecting radar performance. An adaptive filtering method is used to select a correction strategy that can dynamically adjust the covariance matrix according to the actual situation, improving the accuracy of covariance matrix correction.

[0037] It should be added that the specific process of Kalman filtering for adaptively updating the covariance matrix is ​​as follows: First, before receiving new environmental monitoring data, the covariance matrix for the current preset time period is predicted based on the covariance matrix constructed in the previous preset time period and combined with the system dynamic model (e.g., assuming the covariance matrix changes slowly over time). Then, the Kalman filtering algorithm obtains an observation estimate of the covariance matrix based on the received new environmental monitoring data. This observation estimate reflects the covariance information contained in the current data. Next, the predicted covariance matrix for the current preset time period and the observation estimate of the covariance matrix are compared. If the two are very close, it indicates that the prediction is accurate; otherwise, it indicates that the prediction has a bias and needs to be corrected based on the observation results. Then, the Kalman filter calculates the weight coefficients, which depend on the uncertainty of the prediction and the magnitude of the observation noise, determining the respective proportions of the predicted value and the observation estimate. Based on the weight coefficients, the predicted covariance matrix for the current preset time period and the observation estimate are weighted and averaged to obtain the updated covariance matrix. The specific process of adaptively updating the covariance matrix using the recursive least squares method is as follows: First, a covariance matrix estimate is set (based on historical data). Then, the error between the new environmental monitoring data and the current estimate is analyzed, that is, the difference between the performance of the new data under the current covariance matrix and the actual observation. At the same time, a gain value is obtained, which reflects the importance of the new data to correcting the covariance matrix. Next, the covariance matrix is ​​updated based on the obtained gain and error. The degree to which the information from the new environmental monitoring data is used to correct the current covariance matrix estimate is determined according to the magnitude of the gain. The larger the gain, the greater the correction.

[0038] If the stability index of the covariance matrix after correction is not greater than the preset stability index of the covariance matrix obtained from the preset database, feedback is sent to the preset personnel; otherwise, the covariance matrix correction ends, and the clutter model is directly constructed based on the covariance matrix. By monitoring the stability index of the covariance matrix, the stability of the covariance matrix changes can be intelligently identified, avoiding the computational burden caused by overly frequent correction processes. This improves the stability and efficiency of radar system clutter background modeling. When the effect of covariance matrix correction does not meet expectations, the model parameters can be further adjusted through manual feedback to avoid performance degradation caused by incorrect covariance matrix correction. This ensures a balance between automatic correction and manual intervention, enhancing the adaptive capability of covariance matrix correction in radar system clutter background modeling.

[0039] In this embodiment, by combining noise power judgment, adaptive filtering method selection, and covariance matrix stability index monitoring and feedback mechanism, intelligent and efficient correction of the covariance matrix in radar system clutter background modeling is achieved. By monitoring and feedback the covariance matrix stability index, the unstable situation of covariance matrix changes can be identified and handled in a timely manner, avoiding performance degradation caused by incorrect correction of the covariance matrix and enhancing the stability of radar system clutter background modeling.

[0040] Furthermore, the adaptability evaluation of the clutter model is based on the adaptability evaluation parameters of the initial clutter model. The specific method is as follows: The radar signal includes the target signal and the clutter signal. The initial clutter model describes the statistical characteristics of the clutter component in the radar signal. The radar signal and the initial clutter model can be compared to evaluate the effectiveness of the clutter model construction. The power spectral density of the initial clutter model and the power spectral density of the radar signal are processed to obtain the power spectral density deviation coefficient. The specific method for obtaining the power spectral density is as follows: Apply a Fourier transform to the radar signal (or the initial clutter model) to convert it into a frequency domain signal, i.e., X(f), and then obtain the amplitude spectrum, i.e., |X(f)|, based on the frequency domain signal. 2 Finally, the power spectral density is obtained by ratioing the square of the amplitude spectrum to the time window length of the radar signal (or the initial clutter model). The power spectral density deviation coefficient is calculated to quantify the degree of matching between the initial clutter model and the radar signal. A larger deviation in power spectral density means that the clutter model cannot effectively match the spectrum of the actual radar signal, resulting in greater errors during adaptive evaluation and performance degradation in practical use. A larger mean absolute error in the covariance matrix indicates that the radar system's modeling of the signal or noise is not accurate enough, leading to a larger deviation in the power spectral density, as the covariance matrix directly affects the spectral characteristics of noise and signal. Conversely, a deviation in the power spectral density may indicate a larger error in the covariance matrix, because the frequency domain characteristics of the signal and the correlation with noise are not accurately described.

[0041] The initial clutter model and radar signal are subjected to a chi-square test to obtain the chi-square statistic. The chi-square test involves taking the square of the difference between each term in the radar signal and the initial clutter model, and then proportionally dividing it by the corresponding term in the initial clutter model. The results of these ratios are then summed. The chi-square statistic is an important indicator for evaluating the fitness of a clutter model; a smaller chi-square value indicates a better fit and stronger fitness. A larger power spectral density deviation leads to a greater gap between the radar signal and the initial clutter model, thus increasing the chi-square statistic. A significantly increased chi-square statistic may indicate a larger mean absolute error in the covariance matrix, suggesting a significant difference between the radar signal and the initial clutter model.

[0042] A model fitness evaluation balance factor is introduced. After weighting and coupling the fitness evaluation parameters, an inverse proportional operation is performed to obtain the clutter model fitness index. The clutter model fitness index is used to quantitatively evaluate the effectiveness of the constructed clutter model. The fitness evaluation parameters include the unit-free mean absolute error of the matrix, the power spectral density deviation coefficient, and the chi-square statistic. The mean absolute error directly affects the correlation between the covariance matrix and the radar signal, thus affecting the accuracy of the clutter model. A larger mean absolute error indicates a model mismatch and poorer fitness. The specific limiting expression for the clutter model fitness index is as follows: In the formula, A represents the clutter model fitness index, s1 represents the matrix mean absolute error between the reference covariance matrix and the covariance matrix, a2 represents the power spectral density deviation coefficient between the initial clutter model and the radar signal, a3 represents the chi-square statistic between the initial clutter model and the radar signal, β1 represents the first model fitness evaluation balance factor, β2 represents the second model fitness evaluation balance factor, and β3 represents the third model fitness evaluation balance factor.

[0043] The model fitness evaluation balance factors involved are obtained from a preset database and set by preset personnel. Specifically, they include a first model fitness evaluation balance factor, a second model fitness evaluation balance factor, and a third model fitness evaluation balance factor, the sum of which is 1. For example, the matrix mean absolute error and the first model fitness evaluation balance factor form a corresponding first model fitness evaluation balance mapping set. The real-time matrix mean absolute error is input into the first model fitness evaluation balance mapping set to obtain the first model fitness evaluation balance factor. The first model fitness evaluation balance factor represents the degree of influence of the matrix mean absolute error on the clutter model fitness index; the second model fitness evaluation balance factor represents the degree of influence of the power spectral density deviation coefficient on the clutter model fitness index; and the third model fitness evaluation balance factor represents the degree of influence of the chi-square statistic on the clutter model fitness index.

[0044] Specifically, the process of determining whether to optimize the clutter model is as follows: If the clutter model adaptability index is not greater than the preset clutter model adaptability index, the clutter model is optimized; otherwise, the clutter model is not optimized, and the radar signal is directly adaptively suppressed based on the clutter model and the space-time filter. The preset clutter model adaptability index is set by the preset personnel.

[0045] In this embodiment, through the evaluation and weighting of multiple indicators in the above steps, the adaptability of the clutter model can be accurately quantified. The introduction of the adaptability evaluation balance factor enables flexible adjustment of the clutter model adaptability evaluation strategy according to different application scenarios or environmental characteristics. When the clutter model adaptability index is low, the optimized clutter model will be able to better filter out clutter and improve the signal-to-noise ratio of the radar signal.

[0046] Furthermore, the clutter model optimization means setting the clutter model corresponding to the covariance matrix template with the maximum matrix distribution similarity as the clutter model for the current environmental monitoring data, which can automatically select the most suitable template to reflect the changes in the current environment, making the clutter model optimization process more intelligent, ensuring the accuracy of the clutter model, and providing a more reliable basis for subsequent radar signal processing.

[0047] If the clutter model adaptability index of the optimized clutter model is not greater than the preset clutter model adaptability index, feedback is sent to the preset personnel; otherwise, the clutter model optimization is ended, and the radar signal is directly adaptively suppressed based on the clutter model and the space-time filter. When the adaptability of the optimized clutter model is not ideal, information is promptly fed back to the preset personnel and the existing problems are understood to ensure the stable and reliable performance of the entire radar signal processing. When the adaptability of the optimized clutter model is ideal, the optimization process is promptly ended and the radar signal adaptive suppression processing stage is entered, which can improve the processing efficiency of the radar signal.

[0048] In this embodiment, the clutter model corresponding to the most suitable covariance matrix template is automatically selected based on the maximum matrix distribution similarity, which can dynamically adjust the clutter model according to the changes in the actual environment, making it more accurately reflect the clutter characteristics of the current environment, reducing the signal processing error caused by inaccurate clutter models, improving the accuracy of clutter model selection, and ensuring that only clutter models with good adaptability are used for subsequent radar signal adaptive suppression processing to ensure effective removal of clutter interference.

[0049] To further clarify, the determination result for adaptive suppression processing optimization is obtained based on the radar signal-to-noise ratio (SNR) after adaptive suppression processing. The specific process is as follows: If the radar signal-to-noise ratio after adaptive suppression processing is not greater than the preset radar signal-to-noise ratio, adaptive suppression processing optimization is performed; otherwise, adaptive suppression processing optimization is not performed, and security monitoring is directly performed in conjunction with video data. The radar signal-to-noise ratio represents the ratio of the radar signal's signal power to its noise power; the preset radar signal-to-noise ratio is set by preset personnel. The comparison result of the radar signal-to-noise ratio determines whether further optimization of adaptive suppression processing is needed, avoiding unnecessary complex processing and improving radar signal processing efficiency. If the SNR already meets the requirements, security monitoring (e.g., railway perimeter security monitoring) can be directly performed in conjunction with video data, allowing for rapid transition to subsequent application stages.

[0050] Specifically, adaptive suppression processing optimization means introducing multi-level filters to adaptively suppress radar signals. The specific process is as follows: the radar signal is divided into various frequency bands by a bandpass filter. Each frequency band includes high-frequency signals and low-frequency signals. Signals with frequencies higher than a preset frequency (set by preset personnel) are recorded as high-frequency signals, and vice versa as low-frequency signals. This allows for different processing strategies to be used on different frequency bands to optimize the quality of the radar signal. Frequency band processing can more effectively optimize the quality of radar signals.

[0051] The signals of each frequency band are input into the respective frequency band space-time filters, and the spatial covariance matrix and temporal covariance matrix of each frequency band are calculated. Each frequency band space-time filter includes a high-frequency space-time filter and a low-frequency space-time filter. The spatial covariance matrix of each frequency band includes a high-frequency spatial covariance matrix and a low-frequency spatial covariance matrix. The temporal covariance matrix of each frequency band includes a high-frequency time covariance matrix and a low-frequency time covariance matrix. The spatial covariance matrix represents the spatial correlation of radar signals between different antennas, which helps to adjust the weight of the antenna array to optimize signal reception. The temporal covariance matrix represents the correlation of the same antenna signal in the time domain to suppress high-frequency clutter and help improve the purity of radar signals in the time domain. The construction methods of the spatial covariance matrix and the temporal covariance matrix are the same as those of the covariance matrix below. The space-time filter represents a device or algorithm that combines spatial and temporal information to adaptively suppress signals, and can simultaneously consider the characteristics of radar signals at different spatial locations and different time points.

[0052] Based on the spatial covariance matrix and the clutter model fitness index deviation coefficient, the optimal antenna weight is obtained by maximizing the radar signal-to-noise ratio (SNR). The clutter model fitness index deviation coefficient is obtained by performing relative deviation processing between the clutter model fitness index and a preset clutter model fitness index to maximize the distinction between the received target signal and clutter, thereby improving the SNR of the radar system. The specific formula for the antenna weight is as follows: In the formula, w represents the antenna weight, R represents the spatial covariance matrix, and R0 represents the spatial covariance matrix. -1 Let b(θ) be the inverse of the spatial covariance matrix, and b(θ) be the spatial direction of the radar signal (determined by the azimuth angle θ). H (θ) represents the conjugate transpose of b(θ). By adjusting the weights of the antenna array, the distinction between the received radar signal and clutter is improved, enhancing the radar system's ability to receive radar signals, further improving the signal-to-noise ratio, and making the radar signal clearer.

[0053] Based on the time covariance matrix and the clutter model fitness index deviation coefficient, the output error is minimized to obtain the optimal filter coefficients for each frequency band's space-time filter, thus achieving the best filter coefficients to effectively suppress high-frequency clutter in the time domain. The specific formula for the filter coefficients is as follows: In the formula, k(t+1) represents the filter coefficient at time t+1, k(t) represents the filter coefficient at time t, μ represents the step size factor (controlling the update speed), x(t) represents the radar signal at time t, and e(t) represents the error signal (the difference between the radar signal and the filtered output). By obtaining the optimal filter coefficients, interference from clutter on the radar signal is reduced, and signal quality is improved.

[0054] The algorithm uses the least mean square (LMS) method to dynamically adjust the weights of the space-time filters in each frequency band based on the received space-time signal. This maximizes the distinction between the radar signal and clutter, making the received radar signal increasingly closer to the target radar signal while suppressing clutter. The specific adjustment process is as follows: First, an initial weight is set for each weight of the space-time filter (set by a pre-defined team). Then, the filter multiplies the received current input signal with each weight and sums the results to obtain the output value, which represents the filter's response to the current input signal, i.e., the estimate of the target signal. Next, the filter's output is compared with the desired signal (a reference signal derived from historical data or a known training sequence) to obtain the error. For each weight, the algorithm checks its contribution to the error, i.e., analyzes how the error changes if the weight increases or decreases, and fine-tunes the weight in the direction that reduces the error. The adjustment magnitude is determined by two factors: the magnitude of the error (the larger the error, the larger the adjustment magnitude) and the strength of the input signal in the channel corresponding to that weight (the stronger the signal, the larger the adjustment magnitude).

[0055] In this embodiment, through the above steps, it is possible to intelligently determine whether adaptive suppression processing needs to be optimized based on the actual situation of the radar signal. When optimization is required, the radar signal is comprehensively optimized from both spatial and temporal dimensions, which can effectively improve the signal-to-noise ratio of the radar signal, enhance the distinction between the radar signal and clutter, suppress high-frequency clutter, and make the received radar signal closer to the target signal. This improves the performance and reliability of the radar system in applications such as security monitoring, and provides high-quality radar signal support for accurate security monitoring in conjunction with video data.

[0056] The specific application scenario of this application is target identification in security monitoring. For example, security monitoring is carried out by combining radar signals and video data of the visualized area reflecting the radar signals. Radar target detection is performed on the radar signals to obtain the target position of the target to be monitored. Radar target detection refers to the process of analyzing and processing radar echo signals based on the principle of radar transmitting and receiving electromagnetic waves, extracting target information (such as target distance, azimuth, speed, etc.) from complex background clutter, and then determining the position of the target in the radar monitoring space through specific target detection algorithms (such as constant false alarm rate detection algorithm, matched filtering algorithm, etc.). It is the basic function of radar system to realize target detection and tracking. It can quickly and accurately determine the position of the target to be monitored in the radar monitoring space, providing basic information for subsequent security decision-making.

[0057] If the target location is within a preset safety zone (defined by preset personnel), video target recognition is not performed; otherwise, it is. Video target recognition involves using target recognition and tracking methods to analyze the behavioral patterns of the target. The specific process is as follows: First, the target to be tracked is automatically selected in the first frame of the video. Then, in subsequent frames, target recognition technology is used to re-detect the target's position. Next, through feature matching or motion prediction, the correspondence between the target in the current frame and the target in the previous frame is determined. Finally, based on the matching results, the target's trajectory is updated. By defining a preset safety zone, unnecessary video target recognition processing is avoided, improving the operational efficiency of security monitoring. Furthermore, when the target exceeds the safety zone, video target recognition is initiated promptly, fully utilizing the intuitiveness and richness of video data for a more in-depth analysis of target behavior.

[0058] By using statistical learning, normal behavior patterns are constructed by analyzing target behavior in historical video data. These patterns can be based on factors such as target speed and dwell time. The results of video target recognition are compared with these normal behavior patterns to determine whether the target to be monitored exhibits abnormal behavior. For example, if the target's dwell time exceeds a preset threshold (set by preset personnel), it is identified as loitering behavior. If this is found, an alarm is sent to the preset personnel; otherwise, security monitoring of the target continues. This enables accurate identification of environmental threats and abnormal behavior around the clock, effectively improving security protection in complex scenarios and ensuring the safety of critical facilities and personnel.

[0059] In this embodiment, radar signals and video data are combined to fully leverage the advantages of radar's long-range detection and video's intuitive and detailed analysis. Radar can quickly detect targets and determine their locations, while video data allows for in-depth analysis of target behavior. The two complement each other, reducing the possibility of false positives and false negatives, and improving the accuracy and reliability of security monitoring.

[0060] Example 2: In complex meteorological environments, the covariance matrix is ​​influenced by multiple meteorological factors. For example, in high-altitude areas, there may be strong winds and heavy rain. These factors are interconnected and interact with each other, causing the covariance matrix to exhibit complex nonlinear characteristics. The covariance matrix stability index cannot comprehensively consider the influence of these factors on the covariance matrix and cannot accurately assess its stability. Therefore, Example 1 may not be applicable, and Example 2 needs to be selected to replace Example 1 for covariance matrix correction.

[0061] Specifically, if the radar signal frequency change rate is greater than the preset signal frequency change rate obtained from the preset database, the radar signal frequency and covariance matrix stability index are input into the time window length mapping set stored in the database to obtain the time window length. This allows for more efficient processing of the frequency domain signals of environmental monitoring data. The time window length mapping set is a collection obtained from the preset database representing the mapping relationship between the radar signal frequency, covariance matrix stability index, and time window length. First, a short-time Fourier transform is performed on the radar signal to obtain the spectrum information for different time periods. Then, the dominant frequency (the frequency corresponding to the maximum amplitude) at each time point is extracted. Finally, the rate of change of the dominant frequency over time is recorded as the frequency change rate. The preset signal frequency change rate and the preset covariance matrix change amount are set by preset personnel. By adjusting the time window length through the radar signal frequency change rate, signal processing can be performed more flexibly in dynamic environments. When the radar signal frequency change rate is larger, a shorter time window can be used for faster response to changes. When the radar signal frequency change rate is smaller, a longer time window can be used to reduce the amount of computation and improve efficiency.

[0062] If the rate of change of the radar signal frequency is not greater than the preset rate of change of the signal frequency obtained from the preset database, the time window length is adaptively adjusted based on the signal frequency. Specifically, the length of the time window is dynamically adjusted according to the frequency characteristics of the signal. The higher the signal frequency, the shorter the time window is needed to capture rapidly changing features. Commonly used methods include wavelet transform, which can reduce dependence on external databases, reduce the burden of database access, thereby reducing computational overhead and enabling more efficient adjustment of the time window length. This effectively improves the flexibility of covariance matrix correction in radar system clutter background modeling.

[0063] By combining time window length mapping set mapping with adaptive adjustment, it can cope with various complex environmental changes. When the signal frequency change rate is larger, database mapping can deal with complex situations more efficiently. When the signal frequency change rate is smaller, adaptive adjustment can maintain the simplicity and real-time performance of data processing, thereby enhancing the adaptability of radar system to different signal frequency change rates in the clutter background modeling process.

[0064] Specifically, the method for constructing the covariance matrix is ​​as follows: Environmental monitoring data is converted to a frequency domain signal using a Fast Fourier Transform within a time window to facilitate efficient calculation of the covariance matrix. The frequency domain signal is then processed using a pre-defined covariance matrix estimation method to obtain the covariance matrix, while simultaneously acquiring the changes in the covariance matrix. For example, if there are three signals D1, D2, and D3, the covariance matrix is ​​a 3×3 matrix, including the following covariances: C11 represents the covariance between X1 and X2, C12 represents the covariance between X1 and X2, C13 represents the covariance between X1 and X3, C22 represents the covariance between X2 and X2, C23 represents the covariance between X2 and X3, and C33 represents the covariance between X3 and X3. The structure of the covariance matrix is ​​as follows: .

[0065] If the change in the covariance matrix is ​​greater than the change in the preset covariance matrix obtained from the preset database, it indicates that the covariance matrix has changed and the noise characteristics have changed. The weighted average method is used to correct the covariance matrix of the current preset time period and the covariance matrix of the previous preset time period. Otherwise, it indicates that the covariance matrix has not changed and the noise characteristics have not changed, and no correction is made to the covariance matrix.

[0066] In this embodiment, by flexibly adjusting the time window length through the above steps, accurate correction of the covariance matrix in the clutter background modeling of the radar system is achieved in complex dynamic environments. On the one hand, it can quickly respond to changes in signal frequency and improve the processing capability of dynamic signals; on the other hand, it reduces the dependence on external databases, reduces computational overhead, and enhances the adaptability and real-time performance of the radar system in clutter background covariance matrix correction, thereby effectively improving the performance and reliability of the radar system in clutter background.

[0067] Example 3: In some high-precision radar detection tasks, such as the detection and identification of weak targets (e.g., small drones), the clutter model needs to accurately describe the clutter characteristics in the environment to reduce clutter interference with the target signal. These tasks may also require frequent switching between different environments, such as from clear weather to inclement weather. The template selection method for the clutter model in Example 1 may not meet the high-precision requirements. Because the number of pre-set templates is limited, and each template can only represent a specific clutter characteristic, it may not provide a sufficiently accurate clutter model in a timely manner when facing frequently changing environments. Furthermore, template updates and maintenance require time and cost, making it difficult to meet the needs of real-time high-precision detection. Therefore, the clutter model parameter adjustment method in Example 3 is needed for clutter model optimization.

[0068] like Figure 3 The flowchart for adjusting clutter model parameters shown is as follows: The clutter model parameters are adjusted according to the initial learning rate during the gradient descent process to obtain the clutter model fitness index for the first iteration. Then, the rate of change of the iterative fitness index is obtained. If the rate of change of the iterative fitness index is greater than the preset rate of change of the fitness index obtained from the preset database, a learning rate adjustment factor is introduced to correct the learning rate before adjusting the clutter model parameters. Otherwise, the clutter model parameters are adjusted according to the preset learning rate.

[0069] Specifically, the clutter model parameters are adjusted according to the initial learning rate (set by pre-defined personnel) during the gradient descent process, resulting in the clutter model fitness index for the first iteration. The clutter model parameters include the covariance matrix and power spectrum. The clutter model fitness index for the first iteration represents the clutter model fitness index re-obtained after adjusting the clutter model parameters based on the initial learning rate. This provides an initial reference point for the entire optimization process. The clutter model fitness index for the first iteration reflects the degree of matching between the clutter model and the actual environment after the initial parameter adjustment, providing basic data for subsequent judgments on whether the learning rate needs to be adjusted and the progress of the optimization process.

[0070] If the rate of change of the iterative fitness index is greater than the rate of change of the preset fitness index obtained from the preset database, a learning rate adjustment factor is introduced to correct the learning rate (the correction here means multiplying the learning rate adjustment factor and the learning rate). Then, the clutter model parameters are adjusted according to the corrected learning rate, allowing the gradient descent method to adjust the clutter model parameters more accurately, improving the stability and rapid response capability of the clutter model optimization process, enabling the clutter model to quickly reach its optimal state, and adapting to changes in different environments. This improves the robustness of radar system clutter background modeling in complex environments. By optimizing the learning rate, the convergence process can be accelerated, avoiding premature stopping or over-adjustment, thereby achieving better optimization results for the clutter model and improving its adaptability and accuracy. The rate of change of the iterative fitness index is obtained by comparing the clutter model fitness index of the first iteration with the clutter model fitness index. In this application, the deviation comparison means performing a ratio calculation after deviation processing, where the denominator in the ratio calculation is the clutter model fitness index.

[0071] The learning rate adjustment factor is obtained by mapping the clutter model fitness index and the rate of change of the iterative fitness index into a learning rate adjustment mapping set stored in the database. The learning rate adjustment mapping set is a collection obtained from a preset database that represents the mapping relationship between the clutter model fitness index, the rate of change of the iterative fitness index and the learning rate adjustment factor.

[0072] If the rate of change of the iterative fitness index is not greater than the rate of change of the preset fitness index obtained from the preset database, the clutter model parameters are adjusted according to the preset learning rate to reduce computational overhead, maintain the simplicity and efficiency of the clutter model optimization process, and ensure that radar signal processing can be performed more quickly in a more stable environment.

[0073] It should be added that the specific process of adjusting the clutter model parameters according to the learning rate is as follows: When adjusting the clutter model parameters, firstly, a loss function is defined to measure the error between the model prediction and the real data. The loss function may include the data fitting error, regularization term and other parameter constraints. Then, the gradient of the loss function with respect to the clutter model parameters is obtained, which represents the direction and magnitude of the parameter update. Finally, the clutter model parameters are updated according to the gradient descent formula, using the learning rate and gradient.

[0074] In this embodiment, by introducing the rate of change of the adaptive index as the adjustment basis, the optimization strategy of the clutter model can be dynamically adjusted. In unstable environments, the optimization process of the clutter model can be accelerated and error accumulation can be avoided. In stable environments, over-adjustment can be avoided and the computational burden can be reduced.

[0075] The radar system clutter background modeling and adaptive suppression system provided in this application includes: a covariance matrix construction module, a clutter model construction module, a clutter adaptive suppression module, and an adaptive suppression processing optimization module. The covariance matrix construction module constructs a covariance matrix based on collected environmental detection data reflecting target environmental characteristics and radar signals reflecting the interaction between the monitored target and the environment. Based on the constructed covariance matrix, it performs a covariance matrix stability evaluation to reflect the difference between the reference covariance matrix and the actual covariance matrix, and determines whether to perform covariance matrix correction for suppressing environmental noise. The monitored target represents the object that needs to be identified in the radar system. Through the construction, evaluation, and correction of the covariance matrix, a comprehensive data foundation is provided for subsequent clutter modeling, enabling the clutter model to better reflect the actual situation.

[0076] The clutter model building module is used to establish an initial clutter model reflecting the target environment and its interference characteristics based on the corrected covariance matrix and a preset probability distribution model if covariance matrix correction is performed. Otherwise, an initial clutter model is established based on the covariance matrix and the preset probability distribution model, which is selected from the statistical characteristics of the covariance matrix. By establishing an initial clutter model, the target environment and its interference characteristics can be preliminarily described, providing a model basis for subsequent adaptive clutter suppression.

[0077] The clutter adaptive suppression module is used to evaluate the adaptability of the clutter model based on the adaptive evaluation parameters of the initial clutter model, and to determine whether clutter model optimization is necessary to ensure its accuracy. If clutter model optimization is performed, adaptive suppression processing of the radar signal is applied based on the optimized initial clutter model and the space-time filter; otherwise, adaptive suppression processing of the radar signal is directly applied based on the initial clutter model and the space-time filter. Through clutter model adaptive evaluation and optimization, clutter can be effectively suppressed, radar signals can be enhanced, and the signal-to-noise ratio of the radar signal can be improved.

[0078] The adaptive suppression processing optimization module determines the optimization result of the adaptive suppression processing based on the signal-to-noise ratio of the radar signal after adaptive suppression processing. If the result is positive, the adaptive suppression processing is considered unqualified; otherwise, it is considered qualified. Through adaptive suppression processing optimization and security monitoring, clutter can be effectively suppressed, improving the detection performance of the radar system.

[0079] In summary, this embodiment of the application, by combining environmental detection data and radar signals, can effectively capture the interaction characteristics between environmental noise and target information, thereby establishing a covariance matrix. Then, the stability of the covariance matrix is ​​evaluated and corrected, effectively reducing noise interference with radar signal processing and improving signal processing accuracy. Subsequently, an initial clutter model is established based on the covariance matrix and a preset probability distribution model. The probability distribution model is selected based on the statistical characteristics of the covariance matrix, ensuring the clutter model accurately reflects the noise characteristics in the environment and effectively simulates environmental noise. This provides a basis for subsequent adaptive suppression processing, thereby improving the accuracy of the clutter model. Next, by evaluating the adaptability of the initial clutter model, it can be determined whether it accurately reflects the actual environmental noise. If the model has poor adaptability, it may lead to poor filtering effects; therefore, clutter model optimization is necessary. The optimized model can improve the filtering effect, thereby enhancing radar signal quality. Finally, by calculating the signal-to-noise ratio of the radar signal after adaptive suppression processing, it can be determined whether further optimization of the adaptive suppression processing is needed to ensure that monitoring is not affected by signal noise.

[0080] Those skilled in the art will understand that this invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0081] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for modeling and adaptively suppressing clutter background in radar systems, characterized in that, The method includes: S1. Based on the collected environmental monitoring data reflecting the characteristics of the target environment and the radar signals reflecting the interaction between the target and the environment, a covariance matrix is ​​constructed. Based on the constructed covariance matrix, a covariance matrix stability assessment is performed to reflect the difference between the reference covariance matrix and the covariance matrix, and it is determined whether to perform covariance matrix correction for suppressing environmental noise. The environmental monitoring data includes temperature, humidity, precipitation and wind speed. The reference covariance matrix represents a matrix constructed based on historical environmental monitoring data and radar signals for comparison with the covariance matrix. S2, if covariance matrix correction is performed, an initial clutter model reflecting the target environment and its interference characteristics is established based on the corrected covariance matrix and the preset probability distribution model; otherwise, an initial clutter model is established based on the covariance matrix and the preset probability distribution model, wherein the preset probability distribution model is selected based on the statistical characteristics of the covariance matrix. S3. Based on the adaptive evaluation parameters of the initial clutter model, the clutter model is evaluated for adaptability, and it is determined whether to optimize the clutter model to ensure its accuracy. If the clutter model is optimized, the radar signal is adaptively suppressed based on the optimized initial clutter model and the space-time filter. Otherwise, the radar signal is directly adaptively suppressed based on the initial clutter model and the space-time filter. S4. Based on the signal-to-noise ratio of the radar signal after adaptive suppression processing, the judgment result of the adaptive suppression processing optimization is obtained. If the judgment result is yes, it means that the adaptive suppression processing is unqualified; otherwise, it means that the adaptive suppression processing is qualified.

2. The radar system clutter background modeling and adaptive suppression method according to claim 1, characterized in that: The method for evaluating the stability of the covariance matrix based on the constructed covariance matrix is ​​as follows: The elements in the reference covariance matrix corresponding to the maximum matrix distribution similarity are matched one-to-one with the elements in the covariance matrix constructed within the preset time period, and the mean absolute error is processed to obtain the mean absolute error of the matrix within the preset time period. The maximum matrix distribution similarity is obtained by sorting the matrix distribution similarity between the acquired covariance matrix and each reference covariance matrix in the preset database in descending order; The elements in the covariance matrix constructed within the preset time period are matched one-to-one with the elements in the covariance matrix constructed in the previous preset time period, and the relative deviation is processed to obtain the rate of change of the covariance matrix. The trace of the covariance matrix constructed within a preset time period and the trace of the reference covariance matrix corresponding to the maximum matrix distribution similarity are subjected to deviation processing to obtain the trace change coefficient. The trace of the covariance matrix represents the sum of the diagonal elements of the matrix. A matrix stability evaluation balance factor is introduced, and after weighting and coupling the matrix stability evaluation coefficients, an inverse proportional operation is performed to obtain the covariance matrix stability index. The covariance matrix stability index is used to quantitatively evaluate the stability of the covariance matrix. The matrix stability evaluation coefficients include the unit-free mean absolute error of the matrix, the rate of change of the covariance matrix, and the unit-free trace coefficient. The specific process for determining whether to perform covariance matrix correction for suppressing environmental noise is as follows: If the stability index of the covariance matrix is ​​not greater than the preset stability index of the covariance matrix obtained from the preset database, then the covariance matrix is ​​corrected; otherwise, the covariance matrix is ​​not corrected. The covariance matrix correction refers to the selection of a covariance matrix update method based on noise power to ensure the effectiveness of the corrected covariance matrix in suppressing environmental noise.

3. The radar system clutter background modeling and adaptive suppression method according to claim 2, characterized in that: The specific process for correcting the covariance matrix is ​​as follows: If the noise power is greater than the preset noise power, Kalman filtering is selected to adaptively update the covariance matrix; otherwise, recursive least squares method is selected to adaptively update the covariance matrix. If the stability index of the covariance matrix after correction is not greater than the preset stability index of the covariance matrix obtained from the preset database, feedback is sent to the preset personnel; otherwise, the covariance matrix correction is terminated, and a clutter model is directly constructed based on the covariance matrix.

4. The radar system clutter background modeling and adaptive suppression method according to claim 2, characterized in that: The covariance matrix correction also includes: If the rate of change of the radar signal frequency is greater than the preset rate of change of the radar signal frequency obtained from the preset database, then the radar signal frequency and the covariance matrix stability index are input into the time window length mapping set stored in the database and mapped to obtain the time window length. The time window length mapping set is a set obtained from the preset database that represents the mapping relationship between the radar signal frequency, the covariance matrix stability index and the time window length. If the rate of change of the radar signal frequency is not greater than the preset rate of change of the signal frequency obtained from the preset database, the length of the time window is adaptively adjusted based on the signal frequency. Within the time window, environmental monitoring data is converted to frequency domain signals using fast Fourier transform, which facilitates efficient calculation of the covariance matrix. The frequency domain signals are processed using a preset covariance matrix estimation method to obtain the covariance matrix, and the change in the covariance matrix is ​​also obtained. If the change in the covariance matrix is ​​greater than the change in the preset covariance matrix obtained from the preset database, the weighted average method is used to correct the covariance matrix of the current preset time period and the covariance matrix of the previous preset time period; otherwise, the covariance matrix is ​​not corrected.

5. The radar system clutter background modeling and adaptive suppression method according to claim 2, characterized in that: The adaptive evaluation of the clutter model based on the adaptive evaluation parameters of the initial clutter model is performed using the following method: The power spectral density of the initial clutter model and the power spectral density of the radar signal are subjected to relative deviation processing to obtain the power spectral density deviation coefficient. The initial clutter model and radar signal were subjected to a chi-square test to obtain the chi-square statistic. A model fitness evaluation balance factor is introduced, and after weighting and coupling the fitness evaluation parameters, an inverse proportional operation is performed to obtain the clutter model fitness index. The clutter model fitness index is used to quantitatively evaluate the effectiveness of the constructed clutter model. The adaptive evaluation parameters include the unit-free matrix mean absolute error, power spectral density deviation coefficient, and chi-square statistic. The specific process for determining whether to perform clutter model optimization is as follows: if the clutter model fitness index is not greater than the preset clutter model fitness index, then clutter model optimization is performed; otherwise, clutter model optimization is not performed, and the radar signal is directly subjected to adaptive suppression processing based on the clutter model and space-time filter.

6. The radar system clutter background modeling and adaptive suppression method according to claim 5, characterized in that: The clutter model optimization means setting the clutter model corresponding to the covariance matrix template with the maximum matrix distribution similarity as the clutter model of the current environmental monitoring data. If the clutter model adaptability index after optimization is not greater than the preset clutter model adaptability index, feedback is sent to the preset personnel; otherwise, the clutter model optimization ends, and the radar signal is directly subjected to adaptive suppression processing based on the clutter model and space-time filter.

7. The radar system clutter background modeling and adaptive suppression method according to claim 5, characterized in that: The clutter model optimization also includes: The clutter model parameters are adjusted according to the initial learning rate in the gradient descent process, and the clutter model fitness index of the first iteration is obtained. The clutter model parameters include the covariance matrix and the power spectrum. The clutter model fitness index of the first iteration represents the clutter model fitness index re-obtained after adjusting the clutter model parameters based on the initial learning rate. If the rate of change of the iterative adaptive index is greater than the rate of change of the preset adaptive index obtained from the preset database, then a learning rate adjustment factor is introduced to correct the learning rate, and the clutter model parameters are adjusted according to the corrected learning rate. The rate of change of the iterative adaptive index is obtained by comparing the deviation between the clutter model adaptive index of the first iteration and the clutter model adaptive index. The learning rate adjustment factor is obtained by mapping the clutter model fitness index and the rate of change of the iterative fitness index into a learning rate adjustment mapping set stored in the database. The learning rate adjustment mapping set is a set obtained from a preset database that represents the mapping relationship between the clutter model fitness index, the rate of change of the iterative fitness index and the learning rate adjustment factor. If the rate of change of the iterative fitness index is not greater than the rate of change of the preset fitness index obtained from the preset database, the clutter model parameters are adjusted according to the preset learning rate.

8. The radar system clutter background modeling and adaptive suppression method according to claim 5, characterized in that: The determination result of adaptive suppression processing optimization is obtained based on the signal-to-noise ratio of the radar signal after adaptive suppression processing. The specific process is as follows: If the signal-to-noise ratio of the radar signal after adaptive suppression processing is not greater than the preset signal-to-noise ratio of the radar signal, then adaptive suppression processing optimization is performed; otherwise, adaptive suppression processing optimization is not performed. The adaptive suppression processing optimization involves introducing multi-stage filters to adaptively suppress radar signals. The specific process is as follows: The radar signal is divided into frequency bands by a bandpass filter, and each frequency band includes high-frequency signals and low-frequency signals. The signals of each frequency band are input into the space-time filters of each frequency band respectively, and the spatial covariance matrix and the time covariance matrix of each frequency band are calculated. The space-time filters of each frequency band include high-frequency space-time filters and low-frequency space-time filters. The spatial covariance matrix of each frequency band includes high-frequency spatial covariance matrix and low-frequency spatial covariance matrix. The time covariance matrix of each frequency band includes high-frequency time covariance matrix and low-frequency time covariance matrix. Based on the spatial covariance matrix and the clutter model adaptability index deviation coefficient, the radar signal-to-noise ratio is maximized to obtain the optimal antenna weight. The clutter model adaptability index deviation coefficient is obtained by performing relative deviation processing on the clutter model adaptability index and the preset clutter model adaptability index. Based on the time covariance matrix and the clutter model fitness index deviation coefficient, the output error is minimized to obtain the optimal filter coefficients corresponding to the space-time filter for each frequency band. The least mean square algorithm is used to dynamically adjust the weights of the space-time filters for each frequency band based on the received space-time signal.

9. A radar system clutter background modeling and adaptive suppression system, employing the radar system clutter background modeling and adaptive suppression method as described in any one of claims 1-8, characterized in that, include: The module includes a covariance matrix construction module, a clutter model construction module, a clutter adaptive suppression module, and an adaptive suppression processing optimization module. The covariance matrix construction module is used to construct a covariance matrix based on the collected environmental detection data reflecting the characteristics of the target environment and the radar signals reflecting the interaction between the target to be monitored and the environment. Based on the constructed covariance matrix, a covariance matrix stability assessment is performed to reflect the difference between the reference covariance matrix and the covariance matrix, and it is determined whether to perform covariance matrix correction for suppressing environmental noise. The target to be monitored refers to the object that needs to be identified in the radar system. The clutter model construction module is used to establish an initial clutter model reflecting the target environment and its interference characteristics based on the corrected covariance matrix and a preset probability distribution model if covariance matrix correction is performed; otherwise, it establishes an initial clutter model based on the covariance matrix and a preset probability distribution model, wherein the preset probability distribution model is selected based on the statistical characteristics of the covariance matrix. The clutter adaptive suppression module is used to evaluate the adaptability of the clutter model based on the adaptive evaluation parameters of the initial clutter model, and to determine whether to optimize the clutter model to ensure its accuracy. If the clutter model is optimized, the radar signal is adaptively suppressed based on the optimized initial clutter model and the space-time filter; otherwise, the radar signal is directly adaptively suppressed based on the initial clutter model and the space-time filter. The adaptive suppression processing optimization module is used to obtain the judgment result of the adaptive suppression processing optimization based on the signal-to-noise ratio of the radar signal after adaptive suppression processing. If the judgment result is yes, it means that the adaptive suppression processing is unqualified; otherwise, it means that the adaptive suppression processing is qualified.

Images