A method for suppressing sea clutter based on autocorrelation features

Abstract

A sea clutter suppression method based on autocorrelation features, belonging to the field of image processing of measurement signals. First, obtain the autocorrelation function of the RT range-time image of the sea surface and construct an autocorrelation matrix. Then, construct the covariance matrix of the speckle component based on the autocorrelation matrix, and perform eigenvalue decomposition on the covariance matrix of the speckle component to obtain the eigenvector matrix of the speckle component. After that, obtain the subspace of the sea clutter according to the number of eigenvectors of the covariance matrix of the sea surface RT range-time image. Finally, substitute the subspace of the sea clutter into the orthogonal projection suppression algorithm to obtain an orthogonal projection operator, and use this orthogonal projection operator to effectively suppress the sea clutter in the sea surface RT range-time image. The present invention can directly construct the eigen-subspace of the sea clutter using the received data without relying on the reference data of adjacent units, avoiding the inaccuracy in the covariance matrix estimation.

Description

Technical Field

[0001] The present invention relates to the field of image processing of measurement signals, and more specifically to a sea clutter suppression method based on autocorrelation features. Background Art

[0002] With the increasing demand for real-time monitoring and precise detection of marine targets, it has become increasingly important to quickly and reliably detect small mobile targets on the sea surface (such as small boats, frogmen, low-altitude aircraft, etc.). However, affected by the movement of the sea surface, sea clutter exhibits non-uniform, non-Gaussian, and non-stationary characteristics, resulting in interference from sea clutter to sea surface targets in both the time domain and the frequency domain. Currently, the research on the detection of small sea surface targets mainly focuses on two aspects. One is the research on sea clutter suppression methods, and the other is the research on sea surface target detection algorithms. Since sea clutter suppression is a prerequisite for effective target detection, it is even more important.

[0003] Currently, the methods for suppressing sea clutter include time-domain cancellation method, wavelet transform method, deep learning method, space-time adaptive processing, and subspace-based methods. The time-domain cancellation method generates a cancellation signal based on the estimated parameters of the measured sea clutter and subtracts this signal from the measured data to achieve sea clutter suppression. However, this method depends on the accurate estimation of sea clutter characteristics, and any deviation may weaken the characteristics of the target echo, resulting in insignificant sea clutter suppression effect. The wavelet transform method separates sea clutter and target signals in the wavelet domain and uses threshold segmentation to remove clutter. However, this method depends on appropriate basis functions and will destroy the phase information of the received signal. With the development and application of deep learning, researchers have proposed new sea clutter suppression methods from the perspective of image processing. Although these methods perform well in some cases, there are still problems such as complex models and poor generalization ability, which limit their wide application. Space-time adaptive processing (STAP) can make full use of multi-domain information such as space, Doppler, and distance to suppress sea clutter. Although this method provides performance improvement to varying degrees, space-time adaptive processing is mainly applicable to airborne conditions and has a large amount of computation. In contrast, subspace-based sea clutter suppression algorithms have been widely used in the field of sea surface target detection due to their strong target feature extraction ability, low computational complexity, and strong engineering realizability.

[0004] Subspace methods usually decompose the echo signal into a sea clutter subspace and a signal subspace, set the singular values in the sea clutter subspace to zero, and then reconstruct the signal to achieve sea clutter suppression. Subspace methods usually rely on adjacent reference cells to estimate the covariance matrix of sea clutter. This strategy assumes that the sea clutter is uniform, that is, the backscattering within each range gate has the same statistical distribution. However, in practical applications, the training samples in adjacent range gates often do not satisfy the assumption of independent and identically distributed. Therefore, the estimation of the sea clutter covariance matrix may be inaccurate, thus affecting the effect of sea clutter suppression. Specifically, subspace-based suppression methods face the following problems: (1) The estimation accuracy of the covariance matrix has a great impact on performance. For example, the non-uniformity of sea clutter results in fewer samples of uniform sea clutter for estimation and larger estimation errors; (2) In actual detection, the target Doppler is unknown, and Doppler mismatch is likely to occur between the echo signal and the model, thus affecting the detection performance; (3) The non-stationarity of the signal within the coherent processing interval (CPI) also greatly limits the performance of this type of detector. Summary of the Invention

[0005] To solve the above problems existing in subspace-based suppression methods in practical applications, the present invention provides a sea clutter suppression method based on autocorrelation features. This method can directly construct the covariance matrix of sea clutter using the sea surface RT range-time image, without relying on the reference data of adjacent cells, avoiding inaccuracies in covariance matrix estimation.

[0006] The technical solution adopted by the present invention to solve the above technical problems is: A sea clutter suppression method based on autocorrelation features, including the following steps:

[0007] 1) Calculate the autocorrelation function for each range cell in the sea surface RT range-time image along the time dimension, and take the average of the results of all range cells to obtain the autocorrelation function of the sea surface RT range-time image, and then construct an autocorrelation matrix based on the autocorrelation function;

[0008] 2) Construct the covariance matrix of the speckle component according to the autocorrelation matrix;

[0009] 3) Perform eigenvalue decomposition on the covariance matrix of the speckle component to obtain the eigenvector matrix of the speckle component;

[0010] 4) Obtain the subspace of sea clutter according to the number of eigenvectors of the sea surface RT range-time image;

[0011] 5) Substitute the subspace of sea clutter into the orthogonal projection suppression algorithm to obtain an orthogonal projection operator, and use this orthogonal projection operator to effectively suppress the sea clutter in the sea surface RT range-time image.

[0012] The sea surface RT range-time image in the present invention is a range and pulse dimension image obtained after the pulse Doppler radar receives and demodulates the signal.

[0013] As an optimized scheme of the above sea clutter suppression method based on autocorrelation features, in step 1), the autocorrelation function of the sea surface RT range-time image is:

[0014] ;

[0015] In the formula, M is the number of range cells of the sea surface RT range-time image; N represents the number of pulses of the sea surface RT range-time image; represents the echo pixel value of the range cell m in the nth pulse; p represents the number of pulse intervals substituted into the formula for calculation;

[0016] The constructed autocorrelation matrix is:

[0017] ;

[0018] In the formula, K represents the number of time intervals when the autocorrelation function converges to 0.

[0019] As another optimized scheme of the above sea clutter suppression method based on autocorrelation features, the operation of constructing the autocorrelation matrix from the autocorrelation function is:

[0020] Due to the signal characteristics of sea clutter, the autocorrelation function converges to 0, and the number of time intervals K when the autocorrelation function converges to 0 is obtained through the following formula:

[0021] ;

[0022] In the formula, represents a threshold value close to 0, taking 0.01;

[0023] Set the autocorrelation function values with the number of time intervals greater than K to zero, then the autocorrelation matrix formed by the autocorrelation function of sea clutter data is a Toeplitz matrix, which is the autocorrelation matrix.

[0024] As another optimized scheme of the above sea clutter suppression method based on autocorrelation features, in step 2), the method of constructing the covariance matrix of the speckle component is:

[0025] 2.1) Take the Gaussian distribution as the starting sample for generating sample clutter, and then form the complex Gaussian white noise matrix W;

[0026] 2.2) Convert the complex Gaussian white noise matrix into a colored noise matrix with given correlation characteristics, so as to generate the speckle component reflecting the correlation information of sea clutter;

[0027] 2.3) The covariance matrix of the speckle component is represented by the autocorrelation matrix of the sea clutter.

[0028] As another optimization scheme of the above sea clutter suppression method based on autocorrelation features, in the above 2.1), the complex Gaussian white noise matrix W is:

[0029] ;

[0030] where, and are M×N order Gaussian white noise matrices.

[0031] As another optimization scheme of the above sea clutter suppression method based on autocorrelation features, the colored noise matrix in the above 2.2) is:

[0032] ;

[0033] where, W is the complex Gaussian white noise matrix, and L is the lower triangular matrix obtained by performing Cholesky decomposition on the autocorrelation matrix C.

[0034] As another optimization scheme of the above sea clutter suppression method based on autocorrelation features, the covariance matrix of the speckle component in the above 2.3) is:

[0035] ;

[0036] where, R is the covariance matrix of the speckle component, represents the colored noise matrix, C is the autocorrelation matrix of the sea clutter, , L is the lower triangular matrix obtained by performing Cholesky decomposition on the correlation matrix C, L H is the transpose of L, W is the complex Gaussian white noise matrix, and W H is the transpose of W.

[0037] As another optimization scheme of the above sea clutter suppression method based on autocorrelation features, in the above step 3), the specific operation to obtain the eigenvector matrix of the speckle component is:

[0038] Perform eigenvalue decomposition on the covariance matrix of the speckle component, then the covariance matrix, eigenvalues, and eigenvectors of the speckle component satisfy the following relationship:

[0039] ;

[0040] where, R is the covariance matrix of the speckle component, is the diagonal matrix composed of the eigenvalues of the covariance matrix of the speckle component; is the matrix composed of the eigenvectors of the covariance matrix of the speckle component.

[0041] As another optimization scheme of the above sea clutter suppression method based on autocorrelation features, in step 4), the specific operation of obtaining the subspace of sea clutter is as follows:

[0042] 4.1) Perform a unitary transformation on the covariance matrix of the sea surface RT range-time image to separate the sea clutter and the noise disk, and block the separated data covariance matrix to obtain a covariance matrix of dimension (N - 1)×(N - 1) ;

[0043] Among them, when blocking the separated data covariance matrix, its expression is:

[0044] ;

[0045] In the formula, is an (N - 1)-dimensional column vector; is the conjugate transpose of, and is also an (N - 1)-dimensional row vector; is a single element;

[0046] 4.2) Obtain the eigenmatrix of the (N - 1)-dimensional covariance matrix , and the unitary transformation matrix is

[0047] ;

[0048] In the formula: a is an (N - 1)-dimensional column vector with all elements being 0; b is an (N - 1)-dimensional row vector with all elements being 0;

[0049] At this time, the covariance matrix after unitary transformation is expressed as

[0050] ;

[0051] 4.3) According to the GDE theorem, is the Gerschgorin circle radius corresponding to , and according to the distribution characteristics of the Gerschgorin circle radius, the empirical formula for estimating the number of eigenvectors in the sea surface RT range-time image is

[0052] ;

[0053] In the formula, represents the Gerschgorin circle radius corresponding to , k represents the kth Gerschgorin circle, represents an adjustment factor related to the matrix T, and its purpose is to adjust the mean value of the radius in combination with the size of the matrix T;

[0054] 4.4) Obtain the subspace expression of sea clutter:

[0055] ;

[0056] where, is the first vectors to form a matrix.

[0057] As another optimization scheme of the above sea clutter suppression method based on autocorrelation features, in step 5), the specific operation of suppressing sea clutter in the sea surface RT range-time image according to the orthogonal projection operator is as follows:

[0058] 5.1) According to the projection theorem, the projection operator for orthogonally projecting the sea surface RT range-time image into the subspace of sea clutter is:

[0059] ;

[0060] where, represents the subspace of sea clutter;

[0061] 5.2) Let x be the original sea surface RT range-time image, be the projected sea surface RT range-time image, then the sea surface RT range-time image after orthogonal projection suppression is expressed as:

[0062] ;

[0063] In the formula: is the orthogonal projection operator, and the suppression of sea clutter in the sea surface RT range-time image is achieved through this orthogonal projection operator.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1) The method of the present invention does not need to rely on a reference unit to construct a covariance matrix, but generates the speckle component of sea clutter through the autocorrelation features in the sea surface RT range-time image, and uses the speckle component to suppress sea clutter. This method effectively avoids the influence of the texture component of sea clutter and the target in the sea surface RT range-time image, providing a new idea for suppressing sea clutter in the sea surface RT range-time image. This innovation not only simplifies the sea clutter suppression process, but also provides new ideas and methods for related fields;

[0066] 2) The present invention obtains the covariance matrix of the speckle component from the autocorrelation function of the sea surface RT range-time image, determines the number of eigenvectors using the Gerschgorin circle method, and selects eigenvectors from the covariance matrix of the speckle component, thereby constructing the eigen-subspace of sea clutter. Since small maneuvering targets have little influence on the number of eigenvectors of sea clutter, the number of eigenvectors obtained from the sea surface RT range-time image is close to the true number of eigenvectors of pure sea clutter. And because the target has little influence on the autocorrelation function of sea clutter, the eigen-subspace of sea clutter obtained by this method will be more accurate;

[0067] 3) The present invention suppresses sea clutter through the orthogonal projection method according to the eigen-subspace characteristics of sea clutter, so it can match the Doppler information of sea clutter to effectively suppress sea clutter. When the target has the same moving speed as the sea surface, certain suppression will also be obtained. Therefore, the method proposed by the present invention can also adaptively suppress floating targets in the sea surface RT range-time image, thereby improving the recognition ability of maneuvering targets. Description of the Drawings

[0068] Figure 1 It is a diagram showing the change of the normalized intensity of sea clutter in the frequency dimension in the sea surface RT range-time image before and after using the method of the present invention;

[0069] Figure 2 It is a schematic diagram showing the amplitude distribution of pixel values in the sea surface RT range-time image and the amplitude distribution of pixel values after suppression by applying the method of the present invention;

[0070] Figure 3 It is the sea surface RT range-time image without processing and the sea surface RT range-time image processed by the method of the present invention after adding a moving target to the sea clutter sample;

[0071] Figure 4 It is an image of the range dimension fluctuation obtained by non-coherently accumulating 300 pulses and performing non-correlated accumulation in the time dimension on the sea surface range-time image before and after suppressing by using the method of the present invention when the target is at 4 m / s and the SIR is -10;

[0072] Figure 5 It is an image of the range dimension fluctuation obtained by non-coherently accumulating 300 pulses and performing non-correlated accumulation in the time dimension on the sea surface range-time image before and after suppressing by using the method of the present invention when the target is at 4 m / s and the SIR is -5;

[0073] Figure 6 It is an image of the range dimension fluctuation obtained by non-coherently accumulating 300 pulses and performing non-correlated accumulation in the time dimension on the sea surface range-time image before and after suppressing by using the method of the present invention when the target is at 4 m / s and the SIR is 0;

[0074] Figure 7To perform non-coherent integration on 300 pulses, when the target speed is 4 m / s and the SIR is 5, the method of the present invention is used to suppress the range dimension fluctuations in the range-time images of the front and rear sea surfaces obtained by non-coherent integration in the time dimension. Detailed implementation mode

[0075] The method of the present invention will be further elaborated in detail below in combination with specific embodiments. For parts not clarified in the following embodiments of the present invention, they are all regarded as the prior art known or should be known to those skilled in the art.

[0076] Embodiment 1

[0077] A sea clutter suppression method based on autocorrelation features. This method is based on the sea surface RT range-time image. First, calculate the autocorrelation function of the sea surface RT range-time image, and construct an autocorrelation matrix based on this function. Then, construct the covariance matrix of the speckle component according to the autocorrelation matrix. Subsequently, perform eigenvalue decomposition on the covariance matrix of the speckle component to extract its eigenvector matrix. Next, determine the subspace of sea clutter according to the number of eigenvectors of the sea surface RT range-time image. Finally, substitute the obtained sea clutter subspace into the orthogonal projection suppression algorithm to construct an orthogonal projection operator, and use this operator to effectively suppress the sea clutter in the sea surface RT range-time image;

[0078] The specific operations are as follows:

[0079] 1) Obtain the autocorrelation function of the sea surface RT range-time image and construct an autocorrelation matrix;

[0080] The sea clutter signal z satisfies the generalized stationary property. Calculate the autocorrelation function for each range cell in the sea surface RT range-time image along the time dimension, and take the average of the results of all range cells to obtain the autocorrelation function of the sea surface RT range-time image. The expression of the autocorrelation function is:

[0081] ;

[0082] In the formula: M is the number of range cells of the sea surface RT range-time image; N represents the number of pulses of the sea surface RT range-time image; represents the echo pixel value of the range cell m in the nth pulse; p represents the number of pulse intervals substituted into the formula for calculation;

[0083] For the autocorrelation function of sea clutter data, as the number of time intervals increases, the autocorrelation function of sea clutter will slowly converge to 0. This is because the signal characteristics of sea clutter usually have strong time variability. The autocorrelation function describes the similarity between signals at different time points. As the time interval increases, the similarity weakens and finally approaches zero. Therefore, K, which represents the number of time intervals when the autocorrelation function converges to 0, is obtained through the following formula:

[0084] ;

[0085] where, represents a threshold close to 0, taking 0.01;

[0086] Set the autocorrelation function values of time intervals greater than K to zero, then the autocorrelation matrix formed by the autocorrelation function of sea clutter data is a Toeplitz matrix, that is, the autocorrelation matrix, expressed as:

[0087] ;

[0088] In the formula, K represents the number of time intervals when the autocorrelation function converges to 0, C is an N×N symmetric and positive definite matrix. Since the autocorrelation function of sea clutter data converges to 0, there are many 0 elements in C.

[0089] 2) Construct the covariance matrix of the speckle component according to the autocorrelation matrix;

[0090] The autocorrelation matrix C can describe the linear relationship between these random variables. Since the Cholesky decomposition can introduce the autocorrelation characteristics into the white noise of independent and identically distributed, the Cholesky decomposition can be used to generate a Gaussian white noise sequence with a specific autocorrelation structure. The Cholesky decomposition can decompose the positive definite matrix C into a lower triangular matrix and its transpose:

[0091] ;

[0092] where, L is the lower triangular matrix obtained by performing the Cholesky decomposition on C, and L H is the transpose of L;

[0093] According to the SIRP theory, the speckle component comes from the echo signals of a large number of basic scatterers, so its characteristics are not affected by the distribution. Since the Gaussian distribution has the excellent property of still maintaining the Gaussian characteristics after linear transformation, the Gaussian distribution is selected as the starting sample for generating correlated sample clutter. Let and be M×N order Gaussian white noise matrices, and constitute a complex Gaussian white noise matrix:

[0094] ;

[0095] Multiply the complex white noise matrix by the matrix L obtained from the Cholesky decomposition to obtain a colored noise matrix with the desired correlation:

[0096] ;

[0097] where W is the complex Gaussian white noise matrix and L is the lower triangular matrix obtained by performing the Cholesky decomposition on the autocorrelation matrix C;

[0098] Calculate the covariance matrix of the generated speckle component:

[0099] ;

[0100] where R is the covariance matrix of the speckle component, represents the colored noise matrix, C is the autocorrelation matrix of the sea clutter, , L is the lower triangular matrix obtained by performing the Cholesky decomposition on the correlation matrix C, L H is the transpose of L, W is the complex Gaussian white noise matrix, and W H is the transpose of W;

[0101] At this time, the covariance matrix of the speckle component can be expressed by the autocorrelation matrix of the sea clutter;

[0102] 3) Obtain the eigenvector matrix of the speckle component;

[0103] Perform eigenvalue decomposition on the covariance matrix of the speckle component. Then, the covariance matrix, eigenvalues, and eigenvectors of the speckle component satisfy the following relationship:

[0104] ;

[0105] where R is the covariance matrix of the speckle component, is the diagonal matrix composed of the eigenvalues of the covariance matrix of the speckle component; is the matrix composed of the eigenvectors of the covariance matrix of the speckle component;

[0106] 4) Obtain the subspace of the sea clutter according to the number of eigenvectors of the sea surface RT range-time image;

[0107] The sea clutter covariance matrix describes the correlation and variance of the sea clutter data in different dimensions. Among them, U can be used as the principal component direction of the sea clutter covariance matrix, which directly reflects the correlation characteristics of the data in these directions, while represents the variance magnitude in the corresponding direction. The amplitude information of the sea clutter is included in , while U only has the correlation information of the sea clutter.

[0108] Since the speckle component has the same eigenvectors as the covariance matrix of sea clutter, the eigenvectors corresponding to the large eigenvalues in the covariance matrix of the speckle component can also form the subspace of sea clutter. However, the eigenvalues of the covariance matrix of the speckle component and sea clutter are not consistent, and the number of eigenvectors determined only by the speckle component is inaccurate.

[0109] To solve this problem, the present invention obtains the covariance matrix of sea clutter by means of the sea surface RT range-time image, and uses the Gerschgorin disk estimation (GDE) method to determine the number of eigenvalue vectors of the sea surface echo. Usually, there is no obvious difference between the Gerschgorin disks of the sea surface echo and the thermal noise in the sea clutter covariance matrix. It is necessary to perform a certain transformation on the covariance matrix according to the Gerschgorin disk theorem so that the radius of the Gerschgorin disk of the sea surface echo in the transformed covariance matrix is significantly larger than the radius of the thermal noise Gerschgorin disk, and then the number of eigenvectors of the sea clutter can be reliably estimated based on the Gerschgorin disk radius.

[0110] The covariance matrix of the sea surface RT range-time image is unitarily transformed to separate the sea clutter and noise disks. Usually, the disk radius corresponding to the sea clutter is larger, while the disk radius corresponding to the noise is smaller. The covariance matrix of the separated data is partitioned to obtain a covariance matrix of dimension (N - 1)×(N - 1) , and the expression of the partitioning is:

[0111] ;

[0112] where, is obtained by calculating the covariance of the sea surface RT range-time image, which is basic knowledge for those skilled in the art;

[0113] After partitioning , the dimension of is (N - 1)×(N - 1); is an (N - 1)-dimensional column vector; is the conjugate transpose of , which is an (N - 1)-dimensional row vector; is a single element.

[0114] Obtain the eigenmatrix of the (N - 1)-dimensional covariance matrix , and the unitary transformation matrix is formed as:

[0115] ;

[0116] In the formula: a is an (N - 1)-dimensional column vector with all elements being 0; b is an (N - 1)-dimensional row vector with all elements being 0.

[0117] Therefore, the covariance matrix after unitary transformation can be expressed as:

[0118] ;

[0119] According to the GDE theorem, is the Gerschgorin circle radius corresponding to According to the distribution characteristics of the Gerschgorin circle radius, the empirical formula for estimating the number of eigenvectors in the sea surface RT range-time image is:

[0120] ;

[0121] In the formula, represents the Gerschgorin circle radius corresponding to k represents the k-th Gerschgorin circle, represents the adjustment factor related to matrix T, and its main function is to adjust the mean value of the radius in combination with the size of matrix T.

[0122] Therefore, the subspace representation of sea clutter is:

[0123] ;

[0124] Among them, is the matrix formed by the first vectors of

[0125] 5) Orthogonal projection to suppress sea clutter;

[0126] The orthogonal projection method can effectively suppress sea clutter. In the orthogonal projection operation, the data will be projected in a specific direction. In the present invention, the covariance matrix of the speckle component is constructed to obtain the subspace of sea clutter, and according to the projection theorem, the projection operator is:

[0127] ;

[0128] Among them, represents the subspace of sea clutter;

[0129] Let x represent the original sea surface RT range-time image, be the sea surface RT range-time image after projection. Then, the sea surface RT range-time image after orthogonal projection suppression is expressed as:

[0130] ;

[0131] In the formula: is an orthogonal projection operator (i.e., an operator obtained after orthogonal projection operation), and through this orthogonal projection operator, the suppression of sea clutter in the sea surface RT range-time image can be achieved.

[0132] Example verification:

[0133] To verify the performance of the sea clutter suppression method proposed in the present invention, the data used comes from the data file "20210106155330_01_staring" in the first issue of 2020 in the "Radar Sea Detection Data Sharing Program (SDRDSP)". This set of data is sea surface observation data without targets, with a range resolution of 6 meters and using the HH polarization mode. The change of sea clutter before and after using the method of the present invention is compared.

[0134] 1. Analysis of sea clutter suppression effect

[0135] Since the sea clutter suppression algorithm of the present invention obtains an orthogonal projection operator based on the autocorrelation matrix, and there is a corresponding relationship of Fourier transform between the autocorrelation function and the power spectrum, the present invention can more precisely suppress sea clutter in the frequency dimension. To show the effect of sea clutter suppression of the present invention, the sea surface observation data in the data file "20210106155330_01_staring" is continued to be used for the sea clutter suppression experiment. Figure 1 shows the change of the normalized intensity of sea clutter in the frequency dimension before and after applying the method of the present invention.

[0136] Since the sea surface moves with the sea breeze, which causes a frequency shift in the echo signal of the sea surface, the peak position of the sea clutter power spectrum reflects the frequency shift caused by the sea surface movement. From Figure 1 it can be seen that the peak value of the sea clutter power spectrum is near 1, and the peak of the sea clutter power spectrum after being processed by the present invention is significantly weakened. This is because the autocorrelation function can effectively describe the complex signal form of sea clutter, and the present invention can adaptively match the autocorrelation characteristics of sea clutter, filtering a large number of echoes from the moving sea surface, thereby reducing the intensity of sea clutter.

[0137] Figure 2 shows the pixel value amplitude distribution of the original sea surface RT range-time image and the pixel value amplitude distribution after applying the method of the present invention.

[0138] From Figure 2 it can be seen that the trailing feature of the pixel value amplitude distribution image after being processed by the method of the present invention is reduced, which indicates that the echo of sea spikes in the moving state is effectively suppressed. Since the echo of sea spikes has a great influence on the target detection performance, suppressing these echoes can significantly improve the effect of target detection.

[0139] 2. Improvement of the target detection performance by the suppression method of the present invention

[0140] To ensure the reliability of the experimental results, public sea surface observation data was used for comparison. Specifically, 300 consecutive pulses were randomly selected from the 2000 to 2200 range cells in the file "20210106155330_01_staring" as sea clutter samples. A moving target was added to the sea clutter samples, with the signal-to-clutter ratio of the target set to -5 and the target moving towards the observation device at a speed of 4 m / s. Figure 3 The sea surface RT range-time images without processing and the sea surface RT range-time images of the data processed by the method of the present invention are given in

[0141] In Figure 3 There is a target in the 101st range sample in . It can be observed that the sea clutter completely covers the target and it is difficult to identify the target with the naked eye. However, after being processed by the method of the present invention, the amplitude distribution of the sea clutter is significantly improved and becomes more uniform, and the target can also be clearly observed. This is because the present invention effectively filters a large number of echoes from the moving sea surface, thereby improving the recognition rate of targets with different moving speeds from the sea surface.

[0142] Non-coherent integration was performed on these 300 pulses to verify the improvement of the target detection performance by the present invention. Figures 4 - 7 The images of the power ratio of the signal to the sea clutter (SIR) of the target at 4 m / s with values of -10, -5, 0, and 5 are given in .

[0143] From Figures 4 - 7 It is not difficult to find that after being processed by the method of the present invention, the target can be clearly observed in the range dimension at SIR = -5, while the pulse accumulation image without clutter suppression can only clearly observe the target at SIR = 0. It can be seen that the method of the present invention can effectively suppress sea clutter data.

Claims

1. A method for suppressing sea clutter based on autocorrelation features, characterized in that It includes the following steps: 1) Calculate the autocorrelation function for each range cell in the sea surface RT range-time image along the time dimension, and take the average of the results for all range cells to obtain the autocorrelation function of the sea surface RT range-time image. Then, construct an autocorrelation matrix based on the autocorrelation function of the sea surface RT range-time image. 2) Construct the covariance matrix of the speckle component according to the autocorrelation matrix. 3) Perform eigenvalue decomposition on the covariance matrix of the speckle component to obtain the eigenvector matrix of the speckle component. 4) Obtain the subspace of the sea clutter according to the number of eigenvectors of the sea surface RT range-time image. The specific operation is as follows: 4.1) Perform a unitary transformation on the covariance matrix of the sea surface RT range-time image to separate the sea clutter and noise disks, and block the covariance matrix of the separated data to obtain a covariance matrix of dimension (N - 1) × (N - 1) Among them, the data covariance matrix after separation is partitioned, and its expression is: wherein, is an N-1 dimensional column vector; is 's conjugate transpose; is a single element; 4.2) Obtain the (N - 1)-dimensional covariance matrix of the eigenmatrix to form the unitary transformation matrix as In the formula: a is an (N - 1)-dimensional column vector with all elements being 0; b is an (N - 1)-dimensional row vector with all elements being 0. At this time, the covariance matrix after unitary transformation is expressed as 4.3) According to the GDE theorem, |r i | is the Gerschgorin circle radius corresponding to λ i , and according to the distribution characteristics of the Gerschgorin circle radius, the empirical formula for estimating the number of eigenvectors K' in the sea surface RT distance-time image is where, |r k | represents the Gerschgorin circle radius corresponding to λ k , k represents the k-th Gerschgorin circle, and ε1 represents an adjustment factor related to matrix T, whose purpose is to adjust the mean value of the radius in combination with the size of matrix T; 4.4) Obtain the subspace expression of the sea clutter: Q = [u1, u2, …, u K′ ; where \(Q = [u_1, u_2, \ldots, u K′ \) is the matrix composed of the first \(K'\) vectors of \(U = [u_1, u_2, \ldots, u N \); 5) Substitute the subspace of the sea clutter into the orthogonal projection suppression algorithm to obtain an orthogonal projection operator, and use this orthogonal projection operator to effectively suppress the sea clutter in the sea surface RT range-time image.

2. The method for suppressing sea clutter based on autocorrelation features according to claim 1, characterized in that, In step 1), the autocorrelation function of the sea surface RT range-time image is: Where, M is the number of range cells of the sea surface RT range-time image; N represents the number of pulses of the sea surface RT range-time image; c m,n represents the echo pixel value of the range cell m in the nth pulse; p represents the number of pulse intervals substituted into the formula for calculation; The constructed autocorrelation matrix is: In the formula, K represents the number of time intervals when the autocorrelation function converges to 0.

3. A method for suppressing sea clutter based on autocorrelation features according to claim 2, wherein, The operation of constructing the autocorrelation matrix from the autocorrelation function is: Due to the signal characteristics of the sea clutter, the autocorrelation function converges to 0, and the number of time intervals K when the autocorrelation function converges to 0 is obtained through the following formula: In the formula, ∈ represents a threshold, taking 0.

01. Set the autocorrelation function values with time intervals greater than K to zero, then the autocorrelation matrix formed by the autocorrelation function of the sea clutter data is a Toeplitz matrix.

4. A method for suppressing sea clutter based on autocorrelation features according to claim 1, characterized in that, In step 2), the method of constructing the covariance matrix of the speckle component is: 2.1) Use the Gaussian distribution as the starting sample for generating sample clutter, and then form a complex Gaussian white noise matrix W. 2.2) Convert the complex Gaussian white noise matrix into a colored noise matrix with given correlation characteristics, so as to generate a speckle component reflecting the correlation information of the sea clutter. 2.3) Represent the covariance matrix of the speckle component using the autocorrelation matrix of the sea clutter.

5. A method for suppressing sea clutter based on autocorrelation features according to claim 4, characterized in that: In 2.1), the complex Gaussian white noise matrix W is: W = W I + j·W Q ; Where, W I and W Q are M×N order Gaussian white noise matrices.

6. A method for suppressing sea clutter based on autocorrelation features according to claim 4, wherein The colored noise matrix in 2.2) is: η = WL; 7. A method for suppressing sea clutter based on autocorrelation features according to claim 4, characterized in that, In the formula, L is the lower triangular matrix obtained by performing Cholesky decomposition on the correlation matrix C. R = ηη H = LWW H L H = C; Wherein, R is the covariance matrix of the speckle component, η represents the colored noise matrix, C is the autocorrelation matrix of the sea clutter, C = LL H , L is the lower triangular matrix obtained by performing Cholesky decomposition on the correlation matrix C, L H is the transpose of L, W is the complex Gaussian white noise matrix, W H is the transpose of W.

8. A method for suppressing sea clutter based on autocorrelation features according to claim 1, characterized in that, The covariance matrix of the speckle component in 2.3) is: In step 3), the specific operation of obtaining the eigenvector matrix of the speckle component is: Perform eigenvalue decomposition on the covariance matrix of the speckle component, then the covariance matrix, eigenvalues, and eigenvectors of the speckle component satisfy the following relationship: where R is the covariance matrix of the speckle components, Σ = diag(λ1, λ2, …, λ N ) is the diagonal matrix composed of the eigenvalues of the covariance matrix of the speckle components; U = [u1, u2, …, u N is the matrix composed of the eigenvectors of the covariance matrix of the speckle components.

9. A method for suppressing sea clutter based on autocorrelation features according to claim 1, characterized in that, RU = UΣ; In step 5), the specific operation of suppressing the sea clutter in the sea surface RT range-time image according to the orthogonal projection operator is: φ x = Q(Q H Q) -1 Q H ; 5.1) According to the projection theorem, the projection operator for orthogonally projecting the sea surface RT range-time image into the subspace of the sea clutter is: Among them, Q represents the subspace of the sea clutter. 5.2) Let \(x\) be the original sea surface RT range-time image, and \(y\) be the projected sea surface RT range-time image. Then, the sea surface RT range-time image after orthogonal projection suppression is expressed as: Where: (I - Q(Q H Q) -1 Q H ) is an orthogonal projection operator, and the suppression of sea clutter in the sea surface RT range-time image is achieved through this orthogonal projection operator.

Images