An adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion

Abstract

An adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion is disclosed, belonging to the field of radar clutter suppression. The method is as follows: Echo data is acquired through a radar receiver, and after pulse compression and fast / slow time dimension rearrangement, the slow-time signal of each range cell is extracted. A local training window is used to calculate the sample covariance matrix and perform eigenvalue decomposition to extract a five-dimensional feature vector comprising log-condition number, clutter-to-noise ratio, effective rank ratio, eigenvalue entropy, and correlation length. A hierarchical mapping mechanism is employed to calculate the basic loading factor by integrating three objectives: optimal mean square error, adaptive clutter-to-noise ratio, and condition number control. Multiplicative corrections are then performed to obtain the final adaptive diagonal loading factor. The sample covariance matrix is ​​diagonally loaded and regularized, and a whitening matrix is ​​calculated. The signal of the cell to be detected is then whitened, making the transformed clutter plus noise exhibit white noise characteristics. Moving target detection processing is then performed on the whitened data to achieve clutter suppression.

Description

Technical Field

[0001] This invention relates to a diagonal loading clutter suppression method based on multi-feature fusion, belonging to the field of radar clutter suppression technology. Background Technology

[0002] Radar acquires information such as target range, azimuth, and velocity by emitting electromagnetic waves and receiving target echo signals. In real-world environments, echo signals contain not only target information but also clutter, which can severely interfere with target signal detection. In complex environments such as on land and sea, radar echo signals often contain even stronger environmental clutter. This clutter can generate false target information, leading to incorrect radar decisions and hindering effective signal detection. Clutter suppression techniques can reduce the impact of clutter on target detection.

[0003] Traditional clutter suppression methods primarily filter clutter in the time or frequency domain, such as those based on Moving Target Indication (MTI) and Moving Target Detection (MTD) technologies. However, for slow-moving targets, the echo Doppler frequency is near zero, resulting in significant spectral overlap with both stationary and slow-moving clutter. Furthermore, the target's power is often lower than the clutter's power, leading to a significant performance degradation in traditional MTI and MTD techniques. In recent years, subspace-based clutter suppression methods, including Singular Value Decomposition (SVD) and Eigenvalue Decomposition (EVD), have been proposed. These methods utilize the low-rank characteristics of clutter signals, decomposing the signal space into a clutter subspace and a target-plus-noise subspace through matrix decomposition, and then projecting the clutter subspace to eliminate it. The key to subspace methods lies in the accurate estimation of the clutter rank. However, due to the spatial non-stationarity of clutter statistical characteristics, when clutter characteristics differ significantly across different distance regions, subspace-based clutter suppression methods can lead to insufficient clutter suppression or target signal loss in some areas.

[0004] Whitening transform is an effective clutter suppression technique. Its core idea is to decorrelate the received signal using the covariance matrix of the clutter and noise, making the transformed clutter and noise exhibit white noise characteristics. The key to the success of whitening transform lies in the accurate estimation of the covariance matrix. However, in practical radar systems, the statistical characteristics of clutter exhibit significant spatial non-stationarity, and the independent and identically distributed (ICD) assumption holds only within a finite spatial range. This leads to a severe shortage of available training samples. When training samples are insufficient, the sample covariance matrix suffers from severe ill-conditioned problems, mainly manifested as a sharp increase in the condition number, the existence of zero or minimal eigenvalues, and numerical instability in matrix inversion. Therefore, it is essential to develop a method that can fully utilize the environmental information of the covariance matrix and achieve effective clutter suppression under conditions of severe training sample shortage. Summary of the Invention

[0005] To address the problems in existing technologies, such as the decline in whitening clutter suppression performance due to ill-conditioned covariance matrix and the inability of single-feature adaptive loading methods to fully characterize the clutter environment, the present invention aims to provide an adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion. This method acquires echo data through a radar receiver, extracts the slow-time signal of each range cell after pulse compression and fast / slow time dimension rearrangement, and performs DC removal preprocessing. This invention employs a local training window to calculate the sample covariance matrix and performs eigenvalue decomposition, extracting a five-dimensional eigenvector comprising the log-condition number, clutter-to-noise ratio, effective rank ratio, eigenvalue entropy, and correlation length to comprehensively characterize the clutter environment. A hierarchical mapping mechanism is used: first, a basic loading factor is calculated by integrating three objectives—optimal mean square error, adaptive clutter-to-noise ratio, and condition number control; then, a multiplicative correction is performed using three correction factors—effective rank ratio, eigenvalue entropy, and correlation length—to obtain the final adaptive diagonal loading factor. The sample covariance matrix is ​​diagonally loaded and regularized, and a whitening matrix is ​​calculated. The signal of the unit to be detected is then whitened, making the transformed clutter plus noise exhibit white noise characteristics. Moving target detection processing is performed on the whitened data, thus achieving adaptive clutter suppression even under conditions of severely insufficient training samples. This invention features strong adaptability to clutter environments, good numerical stability, and excellent suppression performance.

[0006] To achieve the above objectives, the present invention adopts the following technical solution.

[0007] This invention discloses an adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion, comprising the following steps:

[0008] S1. Receive radar target echo signals, obtain radar echo data, and compress it to obtain pulse-compressed echo data. The radar echo data includes target velocity, range, amplitude, and range-gate echo signals.

[0009] The formula for pulse compression processing is as follows:

[0010] 𝑠(𝑘)∗ℎ(−𝑘) (1)

[0011] in, For signal number The magnitude of each data point. This represents the corresponding pulse pressure coefficient. This indicates a convolution operation. The first pulse after compression processing The signal amplitude of each distance unit.

[0012] S2. Rearrange the echo data according to pulse period and distance units to form a fast and slow time-dimensional echo matrix, and extract the first... The slow-time signal vector of each distance cell is given by the following formula:

[0013] (2)

[0014] in, For the first Unit 3D complex signal vector For the first The pulse at the ... The echo sample value of each distance cell. N is the pulse accumulation number, which is greater than or equal to 1.

[0015] S3. Perform DC removal preprocessing on the slow-time signal vector to eliminate fixed noise components and obtain the DC-removed signal vector of the unit to be detected.

[0016] The method for removing DC is as follows: For the first Slow-time signal vector of each distance unit First, calculate its mean:

[0017] (3)

[0018] Among them No. The average value of the signal in each distance cell. for A column vector of all 1s, i.e. .

[0019] Then subtract the mean from the signal, as shown in the following formula:

[0020] (4)

[0021] in This is the signal vector after DC removal. This indicates that the mean scalar is expanded to... dimensional vector;

[0022] S4. Training samples are collected using a local training window to detect the distance unit. Centered on the center, set up on both sides Each protection unit takes [a certain number of units] outside the protected area. There are training units, and the total number of training samples is [number]. The sample covariance matrix is ​​calculated using the training samples, and then its eigenvalues ​​are decomposed.

[0023] The formula for estimating the sample covariance matrix is:

[0024] (5)

[0025] in, for 3D sample covariance matrix. For the first One training sample. Superscript This indicates the conjugate transpose. This represents the total number of training samples.

[0026] For the sample covariance matrix The formula for eigenvalue decomposition is as follows:

[0027] (6)

[0028] in For the eigenvalue matrix, For the first 1 eigenvector Satisfies the properties of a unitary matrix. .

[0029] (7)

[0030] in This represents a diagonal matrix formed by using the elements within the parentheses as diagonal elements. For the first There are 10 eigenvalues, and the eigenvalues ​​are arranged in descending order:

[0031] (8)

[0032] S5. Extract the five-dimensional clutter environment feature vector from the eigenvalue decomposition results to construct the clutter environment feature vector. For logarithmic condition number, For noise ratio, For effective rank ratio, For eigenvalue entropy, For the relevant length.

[0033] S5.1 Calculate the logarithmic condition number The formula reflects the degree of ill-conditioning of the matrix as follows:

[0034] (9)

[0035] in: The largest eigenvalue, The smallest eigenvalue, The larger the value, the more ill-conditioned the matrix becomes, requiring a larger diagonal loading factor for regularization.

[0036] S5.2 Calculate the noise ratio This reflects the intensity of clutter relative to noise. First, the clutter rank is estimated using the cumulative energy ratio method, as shown in the following formula:

[0037] (10)

[0038] in, This is the clutter rank estimate. This is the energy threshold. For feature value index. The first covariance matrix is ​​the first... Each feature value. This represents the number of pulse accumulations.

[0039] The formulas for estimating clutter power and noise power are:

[0040] (11)

[0041] (12)

[0042] The noise-to-speech ratio is defined as follows:

[0043] (13)

[0044] in, This is an estimate of the clutter power. This is an estimate of the noise power.

[0045] S5.3 Calculate the effective rank ratio This reflects the degree of energy concentration in the eigenvalue space.

[0046] (14)

[0047] in, The minimum number of eigenvalues ​​required to reach 95% cumulative energy. This represents the number of pulse accumulations.

[0048] S5.4 Calculate the eigenvalue entropy The information entropy is calculated by treating the normalized eigenvalues ​​as a probability distribution:

[0049] (15)

[0050] (16)

[0051] in, For the first Normalized values ​​of each eigenvalue.

[0052] S5.5 Calculate the relevant length This reflects the time-dependent characteristics of clutter. Define the first... The correlation coefficient with lag:

[0053] (17)

[0054] in: The sample covariance matrix is ​​the first Line number Column elements, This indicates the modulo operation.

[0055] The correlation length is defined as the lag value at which the correlation coefficient first falls below a threshold:

[0056] (18)

[0057] in: The relevant threshold value ranges from [value range missing]. , The larger the value, the stronger the time correlation of the clutter.

[0058] S6, according to The basic loading factor is calculated using a weighted combination of three components. Three correction factors were introduced to Adjustments were made to obtain the final adaptive diagonal loading factor. .

[0059] S6.1 Calculate the basic loading factor using a weighted combination of three components. :

[0060] (19)

[0061] in: Let be the weighting coefficient, satisfying .

[0062] The formula for calculating the optimal component of the mean square error is:

[0063] (20)

[0064] in, This is an adjustment coefficient, and its value range is... This represents clutter power. noise power

[0065] The formula for calculating the adaptive component of the noise-to-noise ratio is:

[0066] (twenty one)

[0067] in, Represents the trace of a matrix. The adaptive coefficients related to the noise-to-speech ratio are determined according to the following rules:

[0068] (twenty two)

[0069] The formula for calculating the condition number control component is:

[0070] (twenty three)

[0071] in: Let be the target condition number, and its value range is . This component ensures that the condition number of the regularized matrix does not exceed [a certain threshold]. .

[0072] S6.2. Three correction factors are introduced to adjust the basic loading factor:

[0073] (twenty four)

[0074] The effective rank correction factor is calculated as follows:

[0075] (25)

[0076] The eigenvalue entropy correction factor is calculated as follows:

[0077] (26)

[0078] The relevant length correction factor is calculated as follows:

[0079] (27).

[0081] S7, according to The sample covariance matrix is ​​diagonally loaded with regularization to obtain the regularized covariance matrix, and the whitening matrix is ​​obtained by taking the inverse square root.

[0082] The method for performing diagonal loading regularization on the sample covariance matrix is ​​as follows:

[0083] (28)

[0084] in: This is the regularized covariance matrix. This is an adaptive diagonal loading factor. for Diagonal loading of the identity matrix is ​​equivalent to adding white noise components to the diagonal of the covariance matrix, thereby improving the numerical stability of the matrix.

[0085] The whitening matrix is ​​defined as the inverse square root of the regularized covariance matrix, expressed as:

[0086] (29)

[0087] right Perform eigenvalue decomposition:

[0088] (30)

[0089] in: It is a unitary matrix composed of eigenvectors. yes The conjugate transpose of .

[0090] (31)

[0091] The method for performing whitening transformation is as follows:

[0092] (32)

[0093] in: This is the whitened signal vector. This is the whitening matrix. This is the signal vector after DC removal.

[0094] S8. Using a whitening matrix, the signal vector of the unit to be detected after DC removal is transformed to obtain whitened data. This makes the transformed clutter and noise exhibit white noise characteristics. MTD processing is then performed on the whitened data to achieve adaptive clutter suppression.

[0095] Beneficial effects:

[0096] 1. This invention discloses an adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion. It extracts a five-dimensional clutter environment feature vector from the eigenvalue decomposition of the sample covariance matrix, including log condition number, clutter-to-noise ratio, effective rank ratio, eigenvalue entropy, and correlation length, thereby achieving a comprehensive characterization of the clutter environment. Compared with existing methods that only utilize a single feature, multi-feature fusion can capture information such as matrix stability, energy distribution, and temporal structure, thus more accurately determining the optimal loading factor.

[0097] 2. The present invention discloses an adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion. It adopts a hierarchical loading factor mapping mechanism, integrates three objectives—optimal MSE, adaptive clutter-to-noise ratio, and condition number control—to calculate the basic loading factor. It uses effective rank ratio, eigenvalue entropy, and correlation length for multiplicative correction. This mechanism ensures both the adaptability of the loading factor to the clutter environment and the robustness and stability of the adaptive diagonal loading whitening clutter suppression method.

[0098] 3. This invention discloses an adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion, which can still achieve effective clutter suppression even under conditions of severely insufficient training samples. According to the RMB criterion, traditional methods require... Only a sufficient number of independent, identically distributed training samples can guarantee performance. This invention utilizes adaptive diagonal loading regularization technology to achieve clutter suppression even with an insufficient number of training samples.

[0099] 4. The present invention discloses an adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion, which adopts local training window configuration, effectively solves the contradiction between the spatial non-stationarity of clutter and the validity of the independent and identically distributed assumption, ensures the reliability of covariance matrix estimation, has low overall computational complexity, and meets the requirements of real-time processing. Attached Figure Description

[0100] Figure 1 This is a flowchart of an adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion according to the present invention.

[0101] Figure 2 Flowchart for feature extraction in cluttered environments;

[0102] Figure 3 A schematic diagram of the adaptive diagonal loading factor hierarchical mapping;

[0103] Figure 4 Distance-Doppler 3D plot of echo data;

[0104] Figure 5 This is a three-dimensional range-Doppler image of the echo data after clutter suppression. Detailed Implementation

[0105] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0106] Example 1

[0107] The data in this embodiment was obtained from an experiment at an outdoor site. Radar equipment was used to collect ground clutter during the experiment.

[0108] like Figure 1 As shown in the figure, this embodiment discloses an adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion. The specific implementation steps are as follows:

[0109] S1. Receive radar target echo signals through a radar receiver to obtain radar echo data; the radar transmits 64 pulses within one coherent processing interval (CPI) and arranges the echo data according to the speed of the pulses; perform pulse compression processing on the radar echo data to obtain pulse-compressed echo data.

[0110] The formula for pulse compression processing is as follows:

[0111] (33)

[0112] in, For receiving the signal The amplitude of each sampling point; The corresponding pulse pressure coefficient; Indicates the convolution operation; The first pulse after compression processing The signal amplitude of each distance unit;

[0113] S2. The radar echo data after pulse compression is rearranged according to pulse period and range element to form a fast and slow time-dimensional echo matrix, and the first... A slow-time signal vector of a distance cell;

[0114] (34)

[0115] S3. Perform DC removal preprocessing on the slow-time signal vector of each distance unit to eliminate fixed clutter components and obtain the DC-removed signal vector of the unit to be detected.

[0116] The formula for removing DC is as follows:

[0117] (35)

[0118] (36)

[0119] in, For the first The mean of the signals of each distance unit; It is a 64-dimensional column vector of all ones; This is the signal vector after DC removal;

[0120] S4. Training samples are collected using a local training window to detect the distance unit. Centered on the target, skipping one protection unit on each side, four training units are selected on each side, resulting in a total training sample size of [number missing]. The training sample index is:

[0121] (37)

[0122] The formula for estimating the sample covariance matrix is:

[0123] (38)

[0124] Perform eigenvalue decomposition on the sample covariance matrix;

[0125] (39)

[0126] 64 eigenvalues ​​were obtained and the corresponding eigenvector matrix ;

[0127] S5. Extract the five-dimensional clutter environment feature vector from the eigenvalue decomposition results. ;

[0128] S5.1 Calculate the logarithmic condition number:

[0129] (40)

[0130] S5.2 Calculate the clutter rank and select the energy threshold. :

[0131] (41)

[0132] Calculate clutter power and noise power:

[0133] (42)

[0134] Calculate the noise-to-noise ratio:

[0135] (43)

[0136] S5.3 Calculate the effective rank ratio:

[0137] (44)

[0138] in, The minimum number of eigenvalues ​​required for the cumulative energy to reach 95%;

[0139] S5.4 Calculate the eigenvalue entropy:

[0140] (45)

[0141] S5.5 Calculate the correlation length and select the correlation threshold. :

[0142] (46)

[0143] S6, according to The basic loading factor is calculated by weighted combination of three components, and then three correction factors are introduced to obtain the final adaptive diagonal loading factor.

[0144] S6.1, Due to the present embodiment (Severe pathological condition), select weighting coefficients Select the adjustment coefficient Objective condition number .

[0145] Calculate the three fundamental components:

[0146] (47)

[0147] (48)

[0148] (49)

[0149] Calculate the base load factor:

[0150] (50)

[0151] S6.2 Calculate the three correction factors and perform multiplicative corrections to obtain the final adaptive diagonal loading factor:

[0152] (51)

[0153] S7, according to Diagonal loading regularization is applied to the sample covariance matrix:

[0154] (52)

[0155] Calculate the whitening matrix:

[0156] (53)

[0157] S8. The signal vector of the detected unit after DC removal is transformed using a whitening matrix to obtain whitened data; the transformed clutter and noise exhibit white noise characteristics; finally, the whitened data is processed by MTD to achieve effective adaptive clutter suppression.

[0158] (54).

[0160] Comparative Example 1

[0161] This comparative example uses a singular value decomposition clutter suppression method based on K-means clustering to compare with the adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion described in this invention, thus verifying the superiority of this invention.

[0162] The data in the comparative example is from the same source as that in Example 1, both obtained from an outdoor field experiment. In the experiment, radar equipment was used to collect clutter, and the data conditions were completely identical.

[0163] S1. Receive radar target echo signals through a radar receiver to obtain radar echo data; the radar transmits 64 pulses within one coherent processing interval (CPI) and arranges the echo data according to the speed of the pulses; perform pulse compression processing on the radar echo data to obtain pulse-compressed echo data.

[0164] The formula for pulse compression processing is as follows:

[0165] (55)

[0166] in, For receiving the signal The amplitude of each sampling point; The corresponding pulse pressure coefficient; Indicates the convolution operation; The first pulse after compression processing The signal amplitude of each distance unit.

[0167] S2. The radar echo data after pulse compression is rearranged according to pulse period and range element to form a fast and slow time-dimensional echo matrix, and the first... A slow-time signal vector of a distance cell.

[0168] (56)

[0169] in, For the first A 64-dimensional complex signal vector of distance cells; For the first The pulse at the ... Echo sample values ​​of each distance cell.

[0170] S3. After transposing the global echo matrix, perform singular value decomposition to obtain the left singular vector matrix, the singular value matrix, and the right singular vector matrix.

[0171] (57)

[0172] in, It is a left singular vector matrix. It is a right singular vector matrix;

[0173] S4. Extract three-dimensional clustering feature vectors from the singular value decomposition results to comprehensively describe the energy characteristics, spatial correlation, and Doppler spectrum characteristics of each singular component, and form a feature matrix.

[0174] The decomposition results Each singular component is used to extract the following three types of features to form the first singular component. Eigenvectors of singular components .

[0175] S4.1 Calculate normalized singular values Reflecting the first The energy percentage of each singular component relative to the strongest component:

[0176] (58)

[0177] in, For the first One singular value; It is the maximum singular value;

[0178] S4.2 Calculate the correlation coefficient between each right singular vector and the first right singular vector. Reflecting the first Spatial similarity between the singular component and the main clutter component:

[0179] (59)

[0180] in, This is the first right singular vector, corresponding to the main clutter component with the strongest energy; For the first Right singular vector, This represents the modulo operation; The L2 norm of a vector.

[0181] S4.3 Calculate the normalized Doppler frequency estimate for each left singular vector. Reflecting the first The time-domain spectral characteristics of the singular components.

[0182] Estimate the first element using the adjacent element correlation method The Doppler frequencies corresponding to the left singular vectors are first calculated by averaging the correlation values ​​of adjacent elements:

[0183] (60)

[0184] in, For the first The first left singular vector One element; for The complex conjugate of ; the total number of N-dimensional pulses.

[0185] S5. The K-means clustering algorithm is used to automatically classify each singular component and identify the set of singular components belonging to the clutter subspace.

[0186] S6. Construct a clutter subspace projection matrix from the right singular vectors corresponding to the clutter class, perform orthogonal projection on the global echo matrix, eliminate the clutter subspace components, and obtain the echo data after clutter suppression.

[0187] clutter The right singular vectors are arranged in columns to form the basis matrix of the clutter subspace:

[0188] (61)

[0189] Construct the orthogonal complementary projection matrix, i.e., the clutter suppression operator. :

[0190] (62)

[0191] in for identity matrix The signal is projected onto the orthogonal complement space of the clutter subspace, which eliminates clutter components while preserving the target component.

[0192] The transposed echo matrix Projecting onto the orthogonal complement of the clutter subspace achieves clutter component removal:

[0193] (63)

[0194] This is the global matrix after clutter suppression.

[0195] S7. Apply a Hamming window to the echo data after clutter suppression and perform MTD processing to achieve effective adaptive clutter suppression and slow-moving target detection.

[0196] Compared with the above-mentioned SVD clutter suppression method based on K-means clustering, the adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion described in this invention can improve the signal-to-clutter ratio by about 3dB. Furthermore, it outperforms the SVD clutter suppression method based on K-means clustering in the comparative example in terms of slow target protection capability, numerical stability, and real-time processing efficiency in non-stationary clutter environments.

[0197] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion, characterized in that: Includes the following steps: S1. Receive radar target echo signal, obtain radar echo data, and compress it to obtain pulse-compressed echo data; the radar echo data includes target velocity, range, amplitude, and range gate echo signal; S2, rearrange the echo data into a fast-slow time dimension echo matrix by pulse period and range cell, extract the slow time signal vector of the first range cell from the matrix; S3. Perform DC removal preprocessing on the slow-time signal vector to eliminate fixed noise components and obtain the DC-removed signal vector of the unit to be detected. S4, collect training samples by using local training window, for detecting distance unit For the center, each side is provided with A protection unit, each take The total number of training samples is Calculate the sample covariance matrix by using the training samples, and perform eigenvalue decomposition S5. Extract the five-dimensional clutter environment feature vector from the eigenvalue decomposition results to construct the clutter environment feature vector. ; For logarithmic condition number, For noise ratio, For the effective rank ratio, For eigenvalue entropy, For the relevant length; S6, according to The basic loading factor is calculated using a weighted combination of three components. Three correction factors were introduced to Adjustments were made to obtain the final adaptive diagonal loading factor. ; S7, according to Diagonal loading regularization is applied to the sample covariance matrix to obtain the regularized covariance matrix, and the whitening matrix is ​​obtained by taking the inverse square root. S8. By using the whitening matrix, the signal vector of the unit to be detected after DC removal is transformed to obtain the whitened data; the transformed clutter plus noise exhibits white noise characteristics, thereby achieving clutter suppression.

2. The method of claim 1, wherein the method is based on multi-feature fusion and adaptive diagonal loading. In step S5 is represented as: (10) where: is the largest eigenvalue, is the smallest eigenvalue, The larger the condition number, the more ill-conditioned the matrix, and the larger the diagonal scaling factor needed for regularization.

3. The method of claim 1, wherein the method is based on multi-feature fusion and adaptive diagonal loading. In step S5 was obtained by the following method: The clutter rank is estimated using the cumulative energy ratio method, as shown in the following formula: (11) in, This is the clutter rank estimate; Energy threshold; For feature value index; The first covariance matrix is ​​the first... One eigenvalue; The number of pulses accumulated; The formulas for estimating clutter power and noise power are: (12) (13) The noise-to-speech ratio is defined as follows: (14) wherein is an estimate of the clutter power; is an estimate of the noise power.

4. The adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion as described in claim 1, characterized in that, In step S5 is represented as: (15) wherein, is the minimum number of eigenvalues required for the cumulative energy to reach 95%; is the number of pulse accumulations.

5. The adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion as described in claim 1, characterized in that, In step S5 is represented as: (17) wherein is the normalized value of the th eigenvalue, denoted as (16)。 6. The adaptive diagonal loading whitening clutter suppression method based on multi-feature fusion as described in claim 1, characterized in that, The steps described in step S5 Represented as: (19) in: For the relevant threshold, The larger the value, the stronger the time correlation of the clutter; the... The correlation coefficient for each lag is: (18) in: The sample covariance matrix is ​​the first Line number Column elements, This indicates the modulo operation.

7. The method of claim 2, wherein the method is based on multi-feature fusion and adaptive diagonal loading. In step S6, the calculation method of the hierarchical mapping mechanism is as follows: S6.

1. Compute the base loading factor using a three-component weighted combination : (20) wherein: are weight coefficients, satisfying ; The optimal component of the mean square error is expressed as: (21) wherein, is a modulation coefficient; is a clutter power; is a noise power; The clutter-to-noise ratio adaptive component is expressed as: (22) wherein denotes the trace of a matrix; is an adaptive coefficient related to the noise ratio, determined according to the following rule: (23) The condition number control component is represented as: (24) wherein: is the target condition number; ensuring that the condition number of the regularized matrix does not exceed ; S6.2, Introducing three correction factors to the Adjustments made: (25) The effective rank correction factor is calculated as follows: (26) The eigenvalue entropy correction factor is calculated as follows: (27) The relevant length correction factor is calculated as follows: (28)。 8. The method according to claim 1, 2 or 3, wherein, In steps S7 and S8, The method for performing diagonal loading regularization on the sample covariance matrix is ​​as follows: (29) in: This is the regularized covariance matrix; An adaptive diagonal loading factor; for The identity matrix, when diagonally loaded, is equivalent to adding white noise components to the diagonal of the covariance matrix, thereby improving the numerical stability of the matrix. The whitening matrix is ​​defined as the inverse square root of the regularized covariance matrix, expressed as: (30) right Perform eigenvalue decomposition: (31) in: It is a unitary matrix composed of eigenvectors. yes The conjugate transpose of; (32) The method for performing whitening transformation is as follows: (33) wherein: is the whitened signal vector; is the whitening matrix; is the de-DC signal vector.

9. The method according to any one of claims 1 to 7, wherein the method is a multi-feature fusion based adaptive diagonal loading and whitening clutter suppression method. The application scenarios include a slow target detection scenario in a complex ground environment and an adaptive clutter suppression scenario under a condition of a serious lack of training samples Under the condition of a lack of a number of training samples, the adaptive diagonal loading regularization can still achieve clutter suppression.

10. The method according to any one of claims 9, wherein the method is a multi-feature fusion based adaptive diagonal loading and whitening clutter suppression method. The clutter suppression method described herein is applicable to both ground clutter and sea clutter.

Images