Target modality adaptive screening method based on multi-dimensional correlation comprehensive score

Abstract

The application discloses a target mode adaptive screening method based on multi-dimensional correlation comprehensive score, and belongs to the technical field of radar signal processing and target detection. The method comprises the following steps: performing variational mode decomposition on a to-be-detected radar signal to obtain a plurality of mode components; taking a pre-generated ideal target template signal as a reference benchmark for mode screening; calculating multi-dimensional correlation indexes between the mode components and the ideal target template signal; performing normalization processing on the multi-dimensional correlation indexes, and calculating a comprehensive score through equal-weighted average; comparing the comprehensive score with a preset threshold value, screening out mode components containing target information, and performing superposition reconstruction to obtain a reconstructed signal; extracting target features from the reconstructed signal, performing feature fusion through kernel discriminant analysis, and obtaining reconstructed signal fusion features; and comparing the reconstructed signal fusion features with a preset detection threshold value, and generating a detection result. The application realizes accurate screening of target modes, effectively suppresses clutter, and enhances target signals.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of radar signal processing and target detection, and particularly relates to a target mode adaptive screening method based on multi-dimensional correlation comprehensive scoring. BACKGROUND

[0002] In the task of radar signal processing and target detection at sea, strong clutter suppression and target signal enhancement under high sea state conditions have always been an important and extremely challenging topic. When the sea state is complex, the target component in the radar echo signal is easily covered or distorted by the background sea clutter. Therefore, how to effectively separate the clutter component in the target echo under strong sea clutter background is a key scientific problem that needs to be solved urgently.

[0003] In the prior art, as a brand-new adaptive signal decomposition technology, the variational mode decomposition VMD can adaptively decompose a signal into a series of intrinsic mode functions IMFs with minimum bandwidth, and it has attracted widespread attention in complex radar signal processing. However, mode decomposition is only the first step of signal processing. In order to realize target component enhancement and clutter suppression, it is also necessary to screen and reconstruct each mode obtained by decomposition. In the mode reconstruction stage, it is necessary to distinguish and select the mode components that truly contain target components for superposition and reconstruction. However, the core difficulty of mode reconstruction is the lack of accurate screening criteria. The traditional reconstruction method often selects a number of modes for reconstruction according to the inherent serial number or energy order of the modes. However, in a complex sea clutter environment, due to the serious aliasing of sea clutter and target signals in the frequency domain, the center frequencies of the same serial number modes obtained by decomposing different echo signals often have significant differences, and even the phenomenon of cross-mode frequency overlap occurs. This leads to the fact that the screening criteria relying only on frequency or energy cannot effectively distinguish target modes from clutter modes, and blind screening is easy to cause target information loss or clutter residue.

[0004] In order to overcome the above limitations, a new technical solution is needed, which can evaluate the mode components from multiple dimensions more accurately to realize effective screening of target modes. SUMMARY

[0005] The purpose of the present application is to solve the problem of inaccurate radar target mode reconstruction in the prior art under strong sea clutter background, and to provide a target mode adaptive screening method based on multi-dimensional correlation comprehensive scoring. The multi-dimensional correlation index is introduced as the screening basis, the matching degree of each mode signal and the ideal target template signal is quantified, the accurate screening of the target mode is realized, the clutter is effectively suppressed, the target signal is enhanced, and finally the performance of target detection is improved.

[0006] The method comprises the following steps:

[0007] S1. Perform variational mode decomposition (VMD) on the radar signal to be measured, obtaining multiple modal components fluctuating around different center frequencies;

[0008] S2. Introduce a pre-generated ideal target template signal as a reference benchmark for modal screening. The template signal is a pure target echo signal with the same sampling rate as the radar signal to be measured;

[0009] S3. Calculate the multi-dimensional correlation indices between each modal component and the ideal target template signal. The multi-dimensional correlation indices include Pearson correlation coefficient (PCC), mutual information (MI), and cross-power spectrum (CPS);

[0010] S4. Normalize the multi-dimensional correlation indices and calculate the comprehensive score through equal-weight averaging;

[0011] S5. Compare the comprehensive score with a preset threshold, screen out the modal components containing target information, and perform superposition reconstruction to obtain a reconstructed signal with a high signal-to-clutter ratio;

[0012] S6. Extract target features from the reconstructed signal and perform feature fusion using kernel discriminant analysis (KDA) to obtain the fused features of the reconstructed signal;

[0013] S7. Compare the fused features of the reconstructed signal with a preset detection threshold to generate a detection result.

[0014] Further, the specific steps of step S1 are as follows:

[0015] S11. Perform VMD processing on the signal to be measured received by the radar [[]]to construct a variational optimization problem that minimizes the frequency-domain bandwidth of each modal component while ensuring that the sum of all modes can accurately recover the original signal. The expression of the optimization problem is:

[0016] ;

[0017] ;

[0018] where, is the th modal component, is the th center frequency of the modal component;

[0019] S12. Perform iterative solution to obtain modal components fluctuating around different center frequencies.

[0020] Further, in step S2, the pre-generated ideal target template signal For the pure target echo signal, the amplitude fluctuation and Doppler variation of the target component are completely reserved, and the environmental clutter and noise are removed, and the sampling rate is consistent with the radar signal to be measured.

[0021] Further, in step S3, the calculation formula of the Pearson correlation coefficient PCC is:

[0022] ;

[0023] wherein, is the mean value of the modal component , and is the mean value of the template signal .

[0024] Further, in step S3, the calculation formula of the mutual information MI is:

[0025] ;

[0026] wherein, is the joint distribution of the modal component and the template signal , is the marginal distribution of the modal component , is the marginal distribution of the template signal .

[0027] Further, in step S3, the calculation formula of the mutual power spectrum CPS is:

[0028] ;

[0029] wherein, is the complex time-frequency spectrum of the modal component after short-time Fourier transform, is the complex time-frequency spectrum of the template signal after short-time Fourier transform.

[0030] Further, the specific steps of step S4 are:

[0031] S41. Normalize the multi-dimensional correlation indicators, and let the normalized measure of each correlation indicator of the i-th modal component k and the template signal be , , and , and the normalization formula is:

[0032] ;

[0033] wherein, To prevent division by zero for small constants;

[0034] S42. Calculate the equal-weighted average of each normalized correlation index to obtain a comprehensive score :

[0035] .

[0036] Further, the specific steps of step S5 are as follows:

[0037] S51. Set a preset threshold , and construct a set of relevant modal index

[0038] S52. If the comprehensive score of the modal component is greater than or equal to the preset threshold, it is considered that the modal component contains target information, and its index is added to the set of relevant modal indexes ; if the comprehensive score of the modal component is less than the preset threshold, it is considered that the modal component does not completely contain target information, and the modal component is excluded; the index set is as follows:

[0039] ;

[0040] S53. Superimpose all selected modal components in the set to obtain a reconstructed signal with high signal-to-clutter ratio :

[0041] .

[0042] Further, the specific steps of step S6 are as follows:

[0043] S61. Extract target features from the reconstructed signal, including but not limited to average amplitude AA, peak height PH, and Doppler peak height DPH; the calculation formula of the average amplitude AA is as follows:

[0044] ;

[0045] The calculation formula of the peak height PH is as follows:

[0046] ;

[0047] The calculation formula of the Doppler peak height DPH is as follows:

[0048] ;

[0049] Wherein, is the frequency domain representation of ;

[0050] S62. The extracted average amplitude AA, peak height PH, and Doppler peak height DPH are combined into a raw feature vector and labeled with a class to construct a sample set;

[0051] S63. The sample set is mapped to a high-dimensional space by a kernel function, and an optimal projection matrix is obtained by solving a generalized eigenvalue problem;

[0052] S64. The raw feature vector is projected to the optimal projection matrix to obtain a reconstructed signal fusion feature F .

[0053] Further, the specific steps of step S7 are:

[0054] S71. The fusion feature is compared with a preset detection threshold to generate a detection result ;

[0055] S72. If the fusion feature F is greater than the preset detection threshold T , it is determined that the signal segment contains a target; if the fusion feature F is less than or equal to the preset detection threshold T , it is determined that the signal segment is clutter:

[0056] .

[0057] From the above technical solutions, the present application has the following advantages:

[0058] In the target modality adaptive screening method based on multi-dimensional correlation comprehensive score of the present application, the radar signal is multi-modally decomposed by variational modality decomposition (VMD) to surround modality components with different center frequencies, effectively separating the target and clutter components in the signal; the multi-dimensional correlation indexes of Pearson correlation coefficient (PCC), mutual information (MI), and cross power spectrum (CPS) are introduced to quantify the correlation between the modality and the target template from linear correlation, nonlinear dependence, and frequency domain phase-frequency association, which is more comprehensive and accurate than a single index, and improves the reliability of target modality recognition; through normalization processing and equal-weight average calculation of the comprehensive score, the fusion of different types of correlation indexes is realized, the modality components are adaptively screened in combination with a preset threshold and are superimposed for reconstruction, which improves the signal-to-clutter ratio of the reconstructed signal; multi-dimensional target features such as average amplitude, peak height, and Doppler peak height are extracted, and kernel discriminant analysis (KDA) is introduced for nonlinear feature fusion to eliminate redundant information and improve the discrimination between targets and clutter. The present application introduces multi-dimensional correlation indexes as screening criteria, quantifies the matching degree between each modality component and the target template signal, realizes accurate screening of the target modality, effectively suppresses clutter, enhances target signals, and ultimately improves the performance of target detection. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and all other drawings obtained by those skilled in the art without any creative effort based on these drawings also fall within the scope of protection of the present application.

[0060] Figure 1 The flowchart of the target modality adaptive screening method based on multi-dimensional correlation comprehensive score of the present application. DETAILED DESCRIPTION

[0061] The specific steps of the target modality adaptive screening method based on multi-dimensional correlation comprehensive score will be described in detail below, and various embodiments of the present application will be described more fully. The present application can have various embodiments, and adjustments and changes can be made therein. However, it should be understood that there is no intention to limit various embodiments of the present application to the specific embodiments disclosed herein, but the present application should be understood to cover all adjustments, equivalents and / or alternatives falling within the spirit and scope of various embodiments of the present application.

[0062] It should be understood that when used in the specification of the present application, the term "comprising" indicates the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or sets thereof. The terms "comprise", "include", "have" and their variants mean "including but not limited to", unless otherwise specifically emphasized.

[0063] The phrase "one embodiment" or "some embodiments" or similar phrases as used in the present application means that a specific feature, structure or characteristic described in the embodiment is included in one or more embodiments of the present application. Therefore, the phrases "in one embodiment", "in some embodiments", "in other some embodiments", "in yet some embodiments" and the like appearing in different places in the present application do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized.

[0064] In order to make the inventive purpose, features and advantages of the present application more obvious and easy to understand, the technical solutions protected by the present application will be described clearly and completely by using specific embodiments and drawings. Obviously, the embodiments described below are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort fall within the scope of protection of the present application.

[0065] Referring to Figure 1 Fig. 1 shows a flowchart of a target modal adaptive screening method based on multi-dimensional correlation comprehensive score, which comprises the following steps:

[0066] S1. Perform a variational modal decomposition VMD on the radar signal to be tested to obtain a plurality of modal components fluctuating around different center frequencies;

[0067] It should be noted that by constructing a variational optimization problem, the frequency domain bandwidth of each modal component is minimized while ensuring that the sum of all modes accurately restores the original signal.

[0068] S2. Introduce a pre-generated ideal target template signal as a reference benchmark for modal screening, the template signal being a pure target echo signal with a sampling rate consistent with the radar signal to be tested;

[0069] It should be noted that the template signal is a pure target echo signal that retains the amplitude fluctuations and Doppler changes of the target component, while eliminating environmental clutter and noise, and has a sampling rate consistent with the signal to be tested.

[0070] S3. Calculate the multi-dimensional correlation indicators between each modal component and the ideal target template signal, including the Pearson correlation coefficient PCC, mutual information MI and cross power spectrum CPS;

[0071] It should be noted that after obtaining each modal component and the template signal, the multi-dimensional correlation indicators between them are calculated, which quantifies the matching degree of the modal from different angles, avoiding the limitations of a single indicator. PCC can eliminate amplitude differences by introducing mean and variance standardization, and is a numerical correlation indicator that focuses on linear correlation; MI measures non-linear dependence using the deviation between joint distribution and marginal distribution of signals, and is an information correlation indicator; CPS is a frequency domain correlation indicator that can capture common frequency components and phase relationships in the frequency domain based on the complex cross power spectrum of the centered short-time Fourier transform. Correlation analysis, as an effective feature discrimination method, has been widely used in target screening tasks in many fields such as mechanical fault diagnosis and computer vision, and can quantify the matching degree between the signal to be tested and the reference template signal to achieve more accurate modal selection, thereby effectively overcoming the limitations of traditional frequency domain analysis methods.

[0072] S4. Normalize the multi-dimensional correlation indicators and calculate the comprehensive score by equal weight averaging;

[0073] It should be noted that in order to integrate the above three multi-dimensional correlation indicators of different dimensions into a unified discrimination basis, they are normalized and the comprehensive score is calculated.

[0074] S5. Comparing the comprehensive score with a preset threshold value, screening out the modal component containing the target information and performing superposition reconstruction to obtain a reconstructed signal with high signal-to-clutter ratio;

[0075] It should be noted that by adaptively comparing the comprehensive score with the preset threshold value, the modal component containing the target information is screened out and superposition reconstruction is performed, effectively eliminating clutter interference and improving the signal-to-clutter ratio of the reconstructed signal, thereby providing high-purity signal input for target detection.

[0076] S6. Extracting target features from the reconstructed signal and performing feature fusion using kernel discriminant analysis KDA to obtain reconstructed signal fusion features;

[0077] It should be noted that the features extracted from the reconstructed signal can represent the characteristics of the target, effectively distinguishing the target from the clutter; in order to fully fuse the complementary advantages between different features while effectively eliminating redundant information, kernel discriminant analysis KDA is introduced at the feature level to realize nonlinear multi-feature fusion in high-dimensional space, and finally the reconstructed signal fusion features used for the final detection task are obtained.

[0078] S7. Comparing the reconstructed signal fusion features with a preset detection threshold to generate a detection result.

[0079] It should be noted that by comparing the fusion features of the high signal-to-clutter ratio reconstructed signal with the preset detection threshold, the target and the clutter can be accurately distinguished.

[0080] As a refinement and extension of the above embodiment, in order to fully describe the specific implementation process in the embodiment, a target modal adaptive screening method based on multi-dimensional correlation comprehensive score is provided, and the method comprises:

[0081] S1. Performing variational modal decomposition VMD on the radar signal to be measured to obtain a plurality of modal components fluctuating around different center frequencies; the specific steps of step S1 are:

[0082] S11. Performing VMD processing on the radar-received signal to be measured to construct a variational optimization problem, so that the frequency domain bandwidth of each modal component is minimized while ensuring that the sum of all modes can accurately restore the original signal, and the expression of the optimization problem is:

[0083] ;

[0084] ;

[0085] wherein, is the th modal component, is the center frequency of the th modal component.

[0086] S12. Iterative solution is performed to obtain modal components fluctuating around different center frequencies.

[0087] This embodiment is directed to a marine radar target detection scenario under strong sea clutter interference in a marine environment, taking the detection of a ship target on the sea surface by a ship-borne radar as an example. The signal to be measured is the ship echo signal received by the ship-borne radar, which is mixed with strong sea clutter (formed by sea waves and sea surface reflection) and ocean electromagnetic noise. The sampling rate is set to 12 kHz, the signal length is 0.15 s, and a total of 1800 sampling points are collected, denoted as ;

[0088] The VMD processing is performed on . Considering the frequency difference between the sea clutter and the target echo, 6 modal components are preset to be obtained (i.e., K = 6), and a variational optimization problem is constructed: the frequency domain bandwidth of each modal component is minimized, and the sum of the 6 modal components is constrained to accurately restore ;

[0089] The alternating direction multiplier method (ADMM) is used for iterative solution, and the number of iterations is set to 60 and the convergence threshold is set to 5 x 10 -7 . After the iteration is completed, 6 modal components u 1[ n ] to u 6[ n ] are obtained, the center frequencies of which fluctuate around 300 Hz, 600 Hz, 900 Hz, 1300 Hz, 1600 Hz, and 1900 Hz, respectively. This decomposition effectively separates the frequency components of the strong sea clutter and the ship target echo.

[0090] It should be noted that step S1 is the process of splitting the radar signal to be measured by VMD, and the core is to split the mixed target and clutter signal into fine modal components. For the radar received signal to be measured, a variational optimization problem is designed: the frequency domain bandwidth of each modal component is as small as possible (so that the frequency range of each component is more concentrated), and all components are ensured to accurately restore the original signal after superposition, so as to realize the ordered decomposition of the signal. The above optimization problem is solved by iterative operation, and finally k modal components are obtained, each of which fluctuates around a different center frequency, which is equivalent to separating the components of different frequencies (including targets, clutter, etc.) in the original signal.

[0091] S2. The pre-generated ideal target template signal is introduced as a reference benchmark for modal selection, and the template signal is a pure target echo signal with the same sampling rate as the radar signal to be measured.

[0092] In step S2, the pre-generated ideal target template signal For the pure target echo signal, the amplitude fluctuation and Doppler variation of the target component are completely retained, and the environmental clutter and noise are removed. The sampling rate is consistent with the radar signal to be measured.

[0093] It should be noted that the ideal target template signal is generated based on the radar scattering characteristics of the target ship, and is a pure echo signal: the amplitude fluctuation and Doppler variation of the echo of the ship are completely retained, and the sea clutter, noise and other interference are removed. The sampling rate is consistent with the radar signal to be measured, and is used as a reference benchmark for subsequent modal component screening.

[0094] S3. Calculate the multi-dimensional correlation index between each modal component and the ideal target template signal, the multi-dimensional correlation index including Pearson correlation coefficient PCC, mutual information MI and cross power spectrum CPS; in step S3, the calculation formula of the Pearson correlation coefficient PCC is:

[0095] ;

[0096] wherein, is the mean value of the modal component , and is the mean value of the template signal .

[0097] In step S3, the calculation formula of the mutual information MI is:

[0098] ;

[0099] wherein, is the joint distribution of the modal component and the template signal , is the marginal distribution of the modal component , and is the marginal distribution of the template signal .

[0100] In step S3, the calculation formula of the cross power spectrum CPS is:

[0101] ;

[0102] wherein, is the complex time-frequency spectrum of the modal component after short-time Fourier transform, is the complex time-frequency spectrum of the template signal after short-time Fourier transform.

[0103] For the six modal components decomposed by the ship-borne radar, the target echo dominant u 4[ nTaking a center frequency of 1300Hz as an example, calculate its multidimensional correlation index with the template signal:

[0104] PCC calculation: First obtain u 4[ n The mean of ] Substituting the mean of the two values ​​into the PCC formula yields a result of 0.82, indicating a high degree of linear correlation between them.

[0105] MI Calculation: Statistics u 4[ n ]and Substituting the joint distribution and their respective marginal distributions into the MI formula yields a result of 1.23, which reflects a strong nonlinear dependency between the two.

[0106] CPS calculation: for u 4[ n ]、 Performing short-time Fourier transforms on each yields their complex time spectra. Calculating the sum of their products gives a CPS value of 25.6, reflecting a high correlation between frequency and phase in the frequency domain.

[0107] Repeat the above operation for the remaining modal components.

[0108] It should be noted that the Pearson correlation coefficient (PCC) measures the linear correlation between the modal component and the template signal. In its calculation, the mean of both is first calculated, and then the result is obtained by comparing the sum of their deviation products with the product of the square roots of their respective deviations. The closer the value is to 1, the stronger the linear correlation. Mutual information (MI) represents a non-linear dependency. It is calculated by statistically analyzing the joint distribution and marginal distributions of both components and then calculating the logarithmic sum of their distribution ratios. This index reflects the degree of information sharing between the two components; a larger value indicates a tighter non-linear correlation. Cross-power spectrum (CPS) characterizes the matching degree in the frequency domain. It first performs a short-time Fourier transform on the modal component and the template signal to obtain the complex time-frequency spectrum, then calculates the modulus of the sum of their products to capture common frequency components and phase correlations in the frequency domain. A higher value indicates a better matching degree in the frequency domain. These three indices complement each other from linear, non-linear, and frequency domain dimensions, comprehensively measuring whether the modal component contains target information.

[0109] S4. Normalize the multidimensional correlation indicators and calculate the comprehensive score by equal weighted average;

[0110] The specific steps of step S4 are as follows:

[0111] S41. Normalize the multidimensional correlation index, let the first... k Modal components With template signal The normalized measure of each correlation index is as follows: , and , the normalization formula is:

[0112] ;

[0113] wherein, is a very small constant to prevent division by zero;

[0114] S42. Calculate the equal-weighted average of each normalized correlation index to obtain a comprehensive score :

[0115] .

[0116] It should be noted that step S4 is a unified quantitative processing of the correlation index of the modal component and the template signal. For the PCC, MI and CPS correlation indexes of the kth modal component and the template signal, the formula is used to map them to a similar numerical interval, eliminating the influence of the difference in the original value range of different indexes, wherein the very small constant ε is to avoid the error of denominator being 0 in calculation. The equal-weighted average (each occupying 1 / 3 weight) of the three normalized indexes is taken to obtain a comprehensive score S k . The multi-dimensional correlation information is fused into a single score, which not only takes into account the characteristics of different indexes, but also realizes the fair weighting between indexes.

[0117] S5. Compare the comprehensive score with a preset threshold value, select the modal component containing the target information and perform superposition reconstruction to obtain a reconstructed signal with high signal-to-clutter ratio; the specific steps of step S5 are:

[0118] S51. Set a preset threshold value to construct a related modal index set;

[0119] S52. If the comprehensive score of the modal component is greater than or equal to the preset threshold value, it is considered that the modal component contains target information, and its index is added to the related modal index set ; if the comprehensive score of the modal component is less than the preset threshold value, it is considered that the modal component does not completely contain target information, and the modal component is excluded; the index set is:

[0120] ;

[0121] S53. Superimpose all selected modal components in the set to obtain a reconstructed signal with high signal-to-clutter ratio :

[0122] .

[0123] For example, a preset threshold is set based on the accuracy requirements of maritime target detection. S th =0.7, used to distinguish between modal components containing target information and those dominated by clutter;

[0124] The comprehensive score results from step S4 of the six modal components are retrieved and compared one by one with the set threshold. S th Comparison, only u 4 (Score 1) u If a score of 5 (out of 0.8) is ≥ 0.7, then both are determined to contain the target information, and their indices are added to the set. ={4,5}, the remaining components were excluded due to insufficient scores;

[0125] For the set u 4[ n ]and u 5[ n By superimposing and reconstructing, we obtain = u 4[ n ]+ u 5[ n The test revealed that the signal-to-clutter ratio (SCR) of the original signal was only 6 dB, while the SCR of the reconstructed signal was increased to 18 dB. Sea clutter interference was significantly suppressed, and the amplitude and Doppler characteristics of the target echo were clearly highlighted, providing a high-purity signal input for subsequent target feature extraction.

[0126] S6. Extract target features from the reconstructed signal, and perform feature fusion using kernel discriminant analysis (KDA) to obtain the fused features of the reconstructed signal; the specific steps of step S6 are as follows:

[0127] S61. Extract target features from the reconstructed signal, including but not limited to average amplitude AA, peak height PH, and Doppler peak height DPH; the formula for calculating the average amplitude AA is:

[0128] ;

[0129] The formula for calculating the peak height pH is:

[0130] ;

[0131] The formula for calculating the Doppler peak height (DPH) is as follows:

[0132] ;

[0133] in, for Frequency domain representation;

[0134] S62. The extracted average amplitude AA, peak height PH, and Doppler peak height DPH are combined into a raw feature vector and labeled with a class to construct a sample set;

[0135] S63. The sample set is mapped to a high-dimensional space by a kernel function, and a generalized eigenvalue problem is solved to obtain an optimal projection matrix;

[0136] S64. The raw feature vector is projected onto the optimal projection matrix to obtain a reconstructed signal fusion feature F .

[0137] It should be noted that the average amplitude, peak height, and Doppler peak height extracted from the reconstructed signal are combined into a raw feature vector for each sample, and each vector is labeled with a corresponding class to construct a complete sample set. In this embodiment, a kernel function that adapts to the distribution of radar signal features at sea is selected, and the raw low-dimensional feature vector in the sample set is mapped to a high-dimensional feature space using the kernel function to capture the nonlinear relationships between features. In the high-dimensional space, the intra-class scatter and inter-class scatter are calculated respectively: the intra-class scatter describes the dispersion of samples in the same class, and the inter-class scatter describes the separation of samples in different classes. By solving the generalized eigenvalue problem, the projection directions that maximize the ratio of inter-class scatter to intra-class scatter are selected, and these directions are combined into an optimal projection matrix that can enhance the discrimination between targets and clutter. The high-dimensional mapping result corresponding to the raw feature vector of each sample is projected onto the optimal projection matrix, and the resulting fusion feature of the reconstructed signal is used for subsequent target detection tasks.

[0138] S7. The reconstructed signal fusion feature is compared with a preset detection threshold to generate a detection result; the specific steps of step S7 are:

[0139] S71. The fusion feature is compared with a preset detection threshold to generate a detection result ;

[0140] S72. If the fusion feature F is greater than the preset detection threshold T , it is determined that the signal segment contains a target; if the fusion feature F is less than or equal to the preset detection threshold T , it is determined that the signal segment is clutter:

[0141] .

[0142] It should be noted that step S7 is the final determination link of the offshore radar target detection process, which converts the reconstructed signal fusion feature obtained in the previous process into a clear result of whether a target exists. If the value of the fusion feature F is greater than the thresholdT This indicates that the target features in the signal are sufficiently significant, and the signal segment is determined to contain the target, corresponding to the detection result. D =1; if F Less than or equal to T If the signal segment is determined to be interference such as sea clutter, the corresponding result is... D =0. This step, through a simple threshold comparison, achieves the transformation from features to detection conclusions.

[0143] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0144] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A target modality adaptive screening method based on multidimensional correlation comprehensive scoring, characterized in that, The method includes the following steps: S1. Perform variational mode decomposition (VMD) on the radar signal under test to obtain multiple mode components that fluctuate around different center frequencies; S2. Introduce a pre-generated ideal target template signal as a reference benchmark for mode screening. The template signal is a pure target echo signal with a sampling rate consistent with the radar signal under test. S3. Calculate the multidimensional correlation index between each modal component and the ideal target template signal, wherein the multidimensional correlation index includes Pearson correlation coefficient PCC, mutual information MI and cross power spectrum CPS; S4. Normalize the multidimensional correlation indicators and calculate the comprehensive score by equal weighted average; S5. Compare the comprehensive score with the preset threshold, select the modal components containing target information, and superimpose and reconstruct them to obtain a reconstructed signal with a high signal-to-noise ratio; S6. Extract target features from the reconstructed signal, and perform feature fusion using kernel discriminant analysis (KDA) to obtain the fused features of the reconstructed signal; S7. Compare the reconstructed signal fusion features with the preset detection threshold to generate detection results.

2. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, The specific steps of step S1 are as follows: S11. The signal to be measured received by the radar VMD processing is performed, constructing a variational optimization problem that minimizes the frequency domain bandwidth of each modal component while ensuring that the sum of all modes can accurately recover the original signal. The expression for the optimization problem is: ； ； in, It is the first One modal component, It is the first The center frequency of each modal component This represents the total number of modal components obtained from the variable modal decomposition. The sampling point number of the radar signal to be tested. It is a discrete difference operator; S12. Perform iterative solutions to obtain... A modal component that oscillates around different center frequencies.

3. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, In step S2, the pre-generated ideal target template signal It provides a pure target echo signal, fully preserving the amplitude fluctuations and Doppler variations of the target components, while eliminating environmental clutter and noise, and the sampling rate is consistent with the radar signal under test.

4. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, In step S3, the formula for calculating the Pearson correlation coefficient (PCC) is as follows: ； in, Modal components The mean, Template signal The mean, N This represents the total number of sampling points for the radar signal to be tested.

5. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, In step S3, the formula for calculating the mutual information MI is: ； in, Modal components and template signal The joint distribution Modal components marginal distribution, It is a template signal The marginal distribution.

6. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, In step S3, the formula for calculating the cross-power spectrum (CPS) is: ； in, Modal components Complex time spectrum after short-time Fourier transform Template signal Complex time spectrum after short-time Fourier transform.

7. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, The specific steps of step S4 are as follows: S41. Normalize the multidimensional correlation index, let the first... k Modal components With template signal The normalized measure of each correlation index is as follows: , and The normalization formula is: ； in, To prevent division by zero of extremely small constants, For the first k The original value of a multidimensional correlation index between a modal component and the template signal. Let be the original value of the same multidimensional correlation index between the j-th modal component and the template signal. The total number of modal components obtained from variable mode decomposition; S42. Calculate the weighted average of the normalized correlation indicators to obtain the comprehensive score. : 。 8. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, The specific steps of step S5 are as follows: S51. Set a preset threshold. Construct a set of relevant modal indexes; S52. If the overall score of modal components If the modal component is greater than or equal to the preset threshold, it is considered to contain target information and is indexed. Add to relevant modal index set If the overall score of a modal component is less than the preset threshold, then the modal component is considered not to fully contain the target information and is excluded; the index set for: ； S53. For sets All selected modal components are superimposed to obtain a reconstructed signal with a high signal-to-noise ratio. : 。 9. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, The specific steps of step S6 are as follows: S61. Extract target features from the reconstructed signal, including but not limited to average amplitude AA, peak height PH, and Doppler peak height DPH; the formula for calculating the average amplitude AA is: ； The formula for calculating the peak height pH is: ； The formula for calculating the Doppler peak height (DPH) is as follows: ； in, for The frequency domain representation, N The total number of sampling points for the radar signal under test. To reconstruct the signal, For frequency variables; S62. Combine the extracted average amplitude AA, peak height PH, and Doppler peak height DPH into an original feature vector and label the categories to construct a sample set; S63. Map the sample set to a high-dimensional space using a kernel function, and solve the generalized eigenvalue problem to obtain the optimal projection matrix; S64. Project the original feature vectors onto the optimal projection matrix to obtain the reconstructed signal fusion features. F .

10. The target modality adaptive screening method based on multidimensional correlation comprehensive scoring according to claim 1, characterized in that, The specific steps of step S7 are as follows: S71. Merging Features With preset detection threshold Compare and generate detection results ; S72. If fusion features F Greater than the preset detection threshold T If the reconstructed signal segment contains the target, then the signal segment corresponding to the reconstructed signal is determined to contain the target; if the features are fused... F Less than or equal to the preset detection threshold T If so, the signal segment corresponding to the reconstructed signal is determined to be clutter. 。

Images