A radar multipath target recognition method based on Gaussian similarity

Abstract

The present invention discloses a radar multipath target recognition method based on Gaussian similarity, which belongs to the field of radar signal processing technology. First, the radar receiving signal is preprocessed to obtain the distance set and angle set of all detection targets. For a scene model containing a column of unilateral reflectors, the theoretical multipath distance and the theoretical multipath angle are calculated respectively. Then, a Gaussian similarity function based on maximum likelihood is used to calculate the similarity between the theoretical multipath distance and the detection target distance, and between the theoretical multipath angle and the detection target angle, to form a distance / angle similarity matrix. The four similarity matrices are subjected to two dimensionality reduction integrations to maximize the multipath relationship information covered by the combined similarity matrix. The combined similarity matrix is ​​solved using an optimization processing algorithm to solve the maximum sum similarity problem, and an adaptive global optimal multipath recognition result is obtained. The present invention improves the accuracy and real-time performance of multipath target recognition and improves the reliability of the radar detection system.

Description

Technical Field

[0001] The invention belongs to the technical field of radar signal processing, and in particular relates to a radar multipath target recognition method based on Gaussian similarity. Background Art

[0002] Radar target recognition, positioning, and signal extraction are prerequisites for radar system functionality. Radar is often used as a detection sensor, enabling multi-dimensional environmental detection such as ranging, speed, azimuth, and elevation in a variety of detection scenarios, including personnel monitoring and intelligent driving. However, when detecting moving or slightly moving targets, the presence of multipath targets (also known as multipath ghosting) caused by unnecessary reflections increases false alarm rates, distorts signal processing results, and significantly interferes with high-level decision-making.

[0003] Therefore, it is very necessary to deal with the system misjudgment caused by multipath targets. The identification of multipath targets can increase the effectiveness and reliability of the detection system. In order to deal with multipath targets, adaptive filters, signal correlation detection and discrimination are often used to achieve multipath target identification. Summary of the Invention

[0004] The present invention proposes a radar multipath target recognition method based on Gaussian similarity for multipath targets existing in detection scenarios containing a column of unilateral reflectors. For radars capable of measuring distance and angle, multipath target recognition is performed based on a separated Gaussian similarity function, thereby improving the recognition accuracy of multipath targets.

[0005] The radar multipath target recognition method comprises the following specific steps:

[0006] Step 1: Build a radar system for target detection. Taking the FMCW-MIMO millimeter wave radar system as an example, preprocess the radar received signal to obtain the distance set and angle set of all detected targets.

[0007] The specific preprocessing process is:

[0008] Step 101: Sample the intermediate frequency signal s(t) obtained by the FMCW-MIMO radar and perform fast-slow time segmentation to obtain a radar intermediate frequency signal sampling sequence in the fast-slow time dimension:

[0009]

[0010] Among them, n is the fast time index, p is the slow time index, A IF is the intermediate frequency signal amplitude, S is the Chirp signal slope, R0 is the radial distance of the target relative to the radar, x is the micro-motion phase signal of the target, T f is the fast time interval, T sis the slow time interval, λ c is the wavelength of the center frequency, and c is the speed of light.

[0011] Step 102: Perform a fast Fourier transform (FFT) on the fast time dimension of the radar intermediate frequency signal s[n,p] in the fast time-slow time dimension to obtain a range profile:

[0012]

[0013] Where N is the number of fast time sampling points and k is the index of the range image.

[0014] Step 103: remove the static clutter component in the range image, and obtain the range spectrum:

[0015]

[0016] Where, P is the number of slow time sampling points;

[0017] Step 104: Assume that the MIMO radar has M equivalent array elements, and each equivalent array element m corresponds to a range spectrum right Construct the autocorrelation matrix R(k):

[0018]

[0019] The range-azimuth two-dimensional spectrum is obtained by using Capon spectrum estimation based on forward and reverse autocorrelation matrices (FB-Capon).

[0020] The forward and reverse autocorrelation matrix is:

[0021]

[0022] The range-azimuth two-dimensional spectrum is:

[0023]

[0024] in, is the forward and reverse autocorrelation matrix, R * (k) is the conjugate matrix of R(k), is the permutation matrix, θ is the azimuth angle, is the steering vector, λ is the wavelength of the center frequency, and d is the spacing between the equivalent array elements.

[0025] Step 105: The reflection peaks in the range-azimuth two-dimensional spectrum are divided into weak reflection peaks and strong reflection peaks. The weak reflection peaks are suppressed using the constant false alarm rate (CFAR) algorithm, and the strong reflection peaks are classified into different clusters using the density clustering algorithm (DBSCAN) to determine the number of targets. Finally, N detection targets are obtained, and the distance set of all detection targets is ρ i∈D, the angle set is

[0026] The distances and azimuths of N detected targets are Among them, the distance of the detection target satisfies ρ1<ρ2<…<ρ N .

[0027] Step 2: For the scene model containing a column of single-sided reflectors, the theoretical multipath distance and theoretical multipath angle of each detection target are calculated using the theoretical calculation formula to obtain the theoretical multipath distance set and the theoretical multipath angle set.

[0028] In a scenario with a single-sided column of reflectors, each detected target generates three types of multipath targets: first-order direct multipath targets, first-order side-arrival multipath targets, and second-order multipath targets. These three types of multipath targets include two theoretical multipath distances and two theoretical multipath angles. The two theoretical multipath distances are the first-order multipath distance and the second-order multipath distance, and the two theoretical multipath angles are the direct multipath angle and the side-arrival multipath angle. The specific calculation is:

[0029] According to the geometric relationship, the first-order multipath distance and the second-order multipath distance between each detection target and the mth array element of the radar are calculated, that is,

[0030]

[0031]

[0032] Among them, (x i ,y i ) is the rectangular coordinate of each detection target, which is represented by the polar coordinate Perform coordinate transformation to obtain; x m is the horizontal coordinate of the mth array element of the radar; x w is the horizontal coordinate of the reflection surface. Then we get the first-order multipath distance set r i, ' m ∈D1 and the second-order multipath distance set r″ i,m ∈D2.

[0033] Similarly, calculate the direct multipath angle a of each detected target i and the Punta Multipath Angle

[0034]

[0035]

[0036] Further obtain the direct multipath angle set a i ∈A st and the Punta polypath angle set

[0037] Step 3: Use a one-dimensional Gaussian similarity function based on maximum likelihood to calculate the similarity between the theoretical multipath distance and the detection target distance, and the theoretical multipath angle and the detection target angle, to form a distance / angle similarity matrix.

[0038] First, construct the first-order multipath distance similarity matrix related to array element m

[0039] The elements in The average of each element in the pair m, that is

[0040]

[0041] for:

[0042]

[0043] sim in the above formula D (·,·) is the distance similarity function. Since the radar's measurement errors of distance and angle are independent of each other and the errors satisfy the Gaussian distribution, according to the maximum likelihood criterion, the distance similarity function is expressed as:

[0044]

[0045] Among them, u i represents the theoretical multipath parameter of target i, v j represents the parameter measurement value of target j, η D =k D (σ D +ε D ) represents the parameter related to the root mean square error (RMSE) of distance measurement, σ D is the root mean square error of distance measurement, ε D k is the protection value when the root mean square error is too small, D A correction value to unify the distance and angle dimensions numerically.

[0046] Similarly, construct the second-order multipath distance similarity matrix Direct angle similarity matrix Ponda angle similarity matrix The elements in each matrix are:

[0047]

[0048]

[0049]

[0050] Where i, j = 1, 2,…, N-1.

[0051] The angle similarity function is

[0052]

[0053] Among them, η A =k A (σ A +ε A ), the parameter meaning is the same as that of the distance similarity function.

[0054] Step 4: Perform two dimensionality reduction integrations on the four similarity matrices and use principal component projection to maximize the multipath relationship information covered by the combined similarity matrix.

[0055] First dimensionality reduction integration: Use the maximum function and second-order side constraints to integrate the similarity matrix:

[0056] Using the second-order side-reach constraint, since there is no multipath target with a distance of the second-order multipath distance and an angle of the direct multipath angle in this scenario model, in order to exclude this situation, it is set season

[0057] Considering the mutual exclusivity of multipath types, that is, a multipath target has only one type of multipath distance and multipath angle, the maximum function is used to merge and obtain the distance similarity matrix S D And the angle similarity matrix S A , whose elements are

[0058]

[0059]

[0060] Second dimensionality reduction integration: Based on the principle of maximizing the variance of each element in the merged similarity matrix, the principal components of the deformed similarity vector are extracted to obtain the integrated merged similarity matrix.

[0061] First, the distance and angle similarity matrices after the first dimensionality reduction integration are vectorized, that is, the non-zero elements of the distance or angle similarity matrix are rearranged into column vectors, that is,

[0062]

[0063]

[0064] Then, calculate The covariance matrix of and perform singular value decomposition (SVD) on it:

[0065]

[0066] Among them, λ1 and λ2 are the singular values ​​obtained after decomposition, u1 and v1 are the singular vectors corresponding to the singular value λ1, and u2 and v2 are the singular vectors corresponding to the singular value λ2.

[0067] The projection on u1 is The principal component of , which is also the vectorized form of the merged similarity matrix Right now

[0068]

[0069] Rearrange Restore it to its original form and S D and S A Same merged similarity matrix

[0070]

[0071] Step 5: Use the optimization algorithm to solve the maximum sum similarity problem on the combined similarity matrix S, and compare it with the original similarity matrix to obtain the adaptive global optimal multipath identification result.

[0072] For the element S of the merged similarity matrix S ij Solve the maximum and similarity problem and get the solution set {x ij}, where i, j = 1, 2, ... N - 1:

[0073]

[0074]

[0075] Among them, S iso is the isolation threshold, used to exclude independent targets.

[0076] Adaptive threshold calculation method is used to obtain S iso , specifically:

[0077] First, select the largest element in each column of the merged similarity matrix S column by column, sort the largest elements in each column in descending order, and add them to the set S one by one according to the element value from large to small up In. S up Every time a change occurs, calculate S up The variance of each element in .

[0078] Then, judge S up Is the variance of each element in less than the variance threshold v th If so, continue to incorporate the elements in S into S up In, until Sup The variance of each element in is not less than v th , or the largest elements of each column in S are all included in S up At this time, S up The smallest element in is S iso .

[0079] In addition, considering that there are no multipath targets in the scene, that is, all detected targets are independent targets, if S iso If it is less than a fixed similarity lower bound, then S iso Replace with this fixed value to select all independent targets.

[0080] According to the solution set {x ij}Get the identification result of multipath relationship, use vector X=[X1,X2,…X N Considering that the target closest to the radar must be the original target, let X1 = [1,0,0] T , for X2,…X N Then there is

[0081]

[0082] The multipath relationship identification result of each target includes three items. The first item is the target serial number. The second item indicates the existence and ownership of the multipath target. If the second item is 0, it means that the target is a primary target. If the second item is the serial number of another target, it means that the target is a multipath of this other target. Indicates the type of multipath target:

[0083]

[0084] represents the first-order direct multipath target, represents the first-order side-reaching multipath target, Indicates a second-order multipath target.

[0085] At this point, the existence, belonging and multipath categories of the multipath relationship are all identified.

[0086] The advantages of the present invention are:

[0087] (1) The present invention uses the principal component extraction method to process distance and angle data, which significantly improves the accuracy of multipath target recognition.

[0088] (2) The present invention uses a pure geometric method based on a scene model, which does not require long-term recording of radar received signals and has good real-time recognition performance.

[0089] (3) The present invention uses a geometric discrimination method to identify all multipath targets and their corresponding source targets in a detection scene containing a column of single-sided reflectors, and to identify more available target information such as the order and type of the multipath targets, thereby improving the reliability of the radar system. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 Flowchart of the radar multipath target recognition method based on Gaussian similarity of the present invention;

[0091] Figure 2 The figure is a schematic diagram of the principle of generating multipath targets in a scene model containing a column of single-sided reflectors according to the present invention. DETAILED DESCRIPTION

[0092] The present invention will be described in further detail below with reference to the accompanying drawings.

[0093] The present invention addresses the multipath target problem that occurs during radar detection. For each detected radar target, the theoretical multipath target position is generated based on the scene geometry relationship containing a column of unilateral reflectors. Based on a separated Gaussian similarity matrix, the similarity between the theoretical multipath target position and the actual detection position is matched in two dimensions: radial distance and azimuth. By merging and optimizing multiple similarity matrices, a multipath relationship recognition result is obtained, which has good recognition accuracy.

[0094] A radar multipath target recognition method based on Gaussian similarity, such as Figure 1 The specific steps are as follows:

[0095] Step 1: Build a radar system, pre-process the radar received signal, and obtain the distance and angle set of all detected targets.

[0096] The present invention makes discrimination based on geometric relationships, so it is necessary to obtain the geometric position parameters of all targets and perform effective processing. Here, the most widely used FMCW-MIMO millimeter wave radar in the market is used as an example for explanation.

[0097] The intermediate frequency signal received and down-converted by the FMCW-MIMO radar is pre-processed to obtain the range and angle set of all targets. The specific process is as follows:

[0098] The intermediate frequency signal s(t) obtained by the radar is discretized and sampled and divided into fast and slow time to obtain the radar intermediate frequency signal sampling sequence in the fast time-slow time dimension.

[0099]

[0100] Among them, n is the fast time index, p is the slow time index, A IFis the intermediate frequency signal amplitude, S is the Chirp signal slope, R0 is the radial distance of the target relative to the radar, x is the micro-motion phase signal of the target, T f is the fast time interval, T s is the slow time interval, λ c is the wavelength of the center frequency, and c is the speed of light.

[0101] Perform fast Fourier transform (FFT) on the fast time dimension of s[n,p] to measure the distance and obtain the range image, that is,

[0102]

[0103] Among them, N is the number of fast time sampling points, and k is the index of the range image. It can be seen that the range image S[k,p] after FFT will be The peak appears at , which corresponds to the range unit index, indicating that FFT can realize the ranging function of FMCW radar.

[0104] Then, the static clutter components that do not change with slow time in the range image are eliminated to obtain the range spectrum after clutter removal. Right now

[0105]

[0106] Where P is the number of slow time sampling points.

[0107] According to the equivalent array element theory of MIMO radar, suppose the MIMO radar with co-location of transmission and reception has N T transmit array elements and N R receiving array elements, then it can be equivalent to M=N T ×N R There are equivalent array elements with the same transmitting and receiving positions, and each equivalent array element m corresponds to a distance spectrum. right Construct the autocorrelation matrix R(k)

[0108]

[0109] In order to solve the rank deficiency of the autocorrelation matrix under the condition that the multipath signal appears coherent, the present invention uses the Capon spectrum estimation (FB-Capon) based on the forward and reverse autocorrelation matrix to obtain the range-azimuth two-dimensional spectrum, that is,

[0110]

[0111]

[0112] in, is the forward and reverse autocorrelation matrix, R * (k) is the conjugate matrix of R(k); is a permutation matrix with the characteristic that all elements except the anti-diagonal elements are 1 and all other elements are 0; under the condition that the M equivalent array elements have the same spacing d, is the steering vector, θ is the azimuth angle, λ is the wavelength of the center frequency, and d is the spacing between the equivalent array elements.

[0113] For the range-azimuth two-dimensional spectrum, the constant false alarm rate (CFAR) algorithm is used to suppress the weak reflection peaks generated by external interference or noise; the density clustering algorithm (DBSCAN) is used to classify the strong reflection peaks into different clusters to determine the number of targets, and finally N detection targets are obtained, whose polar coordinates are Among them, the distance of the detection target satisfies ρ1<ρ2<…<ρ N . We can further obtain the distance set ρ of all detected targets i ∈D and the angle set

[0114] Step 2: For a scene model containing a column of single-sided reflectors, use theoretical calculation formulas to calculate theoretical multipath distances and theoretical multipath angles, and obtain a theoretical multipath distance set and a theoretical multipath angle set.

[0115] Taking into account the attenuation of received signal energy caused by multiple reflections, it is assumed that a detected multipath target within the scene reflects no more than three times in the environment. Based on the law of reflection and established geometric relationships, the theoretical multipath positions of the detected targets and reflectors can be calculated, given the known positions of the detected targets and reflectors.

[0116] A scene model containing a column of single-sided reflectors is as follows: Figure 2 As shown in the figure, the radar detects the original target i, its generated multipath targets, other original target k, and original target j on the other side of the reflector, causing a false alarm and affecting the system's decision-making. This scenario can be generalized to more complex detection scenarios by simply adjusting the theoretical calculation formula to calculate the theoretical multipath distance and theoretical multipath angle, without affecting subsequent algorithms.

[0117] In the above scenario model, each detected target may generate three types of multipath targets: first-order direct multipath targets, first-order side-path targets, and second-order multipath targets. These three types of multipath targets include two multipath distances: first-order multipath distance and second-order multipath distance, and two multipath angles: direct multipath angle and side-path angle. Figure 2 The geometric relationship shown can be used to calculate the first-order multipath distance r′ between each detected target and the mth array element of the radar i,m and the second-order multipath distance r″ i,m ,Right now

[0118]

[0119]

[0120] Among them, (x i ,y i ) is the rectangular coordinate of each detection target, which can be obtained by polar coordinates Transform; x m is the horizontal coordinate of the mth array element of the radar; x w is the horizontal coordinate of the reflector. We can further obtain the first-order multipath distance set r′ i,m ∈D1 and the second-order multipath distance set r″ i,m ∈D2.

[0121] Similarly, the direct multipath angle a of each detected target can be calculated i and the Punta Multipath Angle

[0122]

[0123]

[0124] Similarly, we can get the direct multipath angle set a i ∈A st and the Punta polypath angle set

[0125] Step 3: Using the Gaussian similarity function based on maximum likelihood, the first-order multipath distance similarity matrix and the second-order distance similarity matrix are constructed based on the elements in the detection target distance set D and the two multipath distance sets D1 and D2. Similarly, the detection target angle set A and the two multipath angle sets A st ,A si The elements in are used to construct the direct multipath angle similarity matrix and the indirect multipath angle similarity matrix, respectively. Finally, the similarity matrix with distance and angle separation is obtained, which can effectively represent the possibility of multipath relationship and facilitate matrix operations and processing.

[0126] First-order multipath distance similarity matrix For example, the construction process is as follows:

[0127] Since there are M first-order multipath distances corresponding to each detection target, the first-order multipath distance similarity matrix related to array element m is first constructed The elements in The average of the elements in the array m, that is

[0128]

[0129] The form is

[0130]

[0131] The matrix is ​​an upper triangular matrix because the distance of the detection target relative to the radar satisfies ρ1<ρ2<…<ρ N , so the probability that close target i is the multipath of long-distance target j is very small. Therefore, the lower triangular elements of the matrix can be ignored to reduce the amount of calculation.

[0132] The element value in is determined by the distance similarity function sim D (·,·). The distance similarity function is composed of a one-dimensional Gaussian function. This is because the radar's measurement errors of distance and angle are independent of each other, and the errors satisfy the Gaussian distribution. According to the maximum likelihood criterion, the distance similarity function can be expressed as:

[0133]

[0134] Among them, u i represents the theoretical multipath parameter of target i, v j represents the parameter measurement value of target j, η D =k D (σ D +ε D ) represents the parameter related to the root mean square error (RMSE) of distance measurement, σ D is the root mean square error of distance measurement, ε D k is the protection value when the root mean square error is too small, D is a correction value that numerically unifies the distance and angle dimensions. It can be seen that the closer the theoretical multipath distance of target i is to the measured distance of target j, the higher the distance similarity, and the higher the possibility that target j is the first-order multipath of target i.

[0135] Similarly, construct the second-order multipath distance similarity matrix Direct angle similarity matrix Ponda angle similarity matrix Its elements are constructed by the corresponding detection target parameter set and the theoretical multipath target parameter set:

[0136]

[0137]

[0138]

[0139] Where i, j = 1, 2, ..., N-1. Similarly, these similarity matrices are also upper triangular matrices, and their lower triangular elements can be ignored.

[0140] Similar to the distance similarity function, the angle similarity function in the above formula is

[0141]

[0142] Among them, η A =k A (σ A +ε A ), the parameter meaning is the same as that of the distance similarity function.

[0143] So far, the first-order multipath distance similarity matrix constructed Second-order distance similarity matrix Direct angle similarity matrix Ponda angle similarity matrix The relationship between the detection target parameters and the theoretical multipath target position parameters has been fully described. Elements with the same subscripts represent the same attribution. It is worth noting that in this scenario model, there is no multipath target with a distance of the second-order multipath distance and an angle of the direct multipath angle. Therefore, in order to exclude this situation, it is set when season

[0144] Step 4: Perform matrix dimensionality reduction on the four similarity matrices, that is, merge the various similarity matrices into one similarity matrix.

[0145] First, merge the first-order and second-order distance similarity matrices, and the direct and side-reach angle similarity matrices. Considering the mutual exclusivity of multipath types, that is, a multipath target can only have one type of multipath distance and multipath angle, the merged similarity matrix only needs to retain the elements with higher similarity without affecting the attribution of the multipath relationship and the final judgment result. Therefore, the distance similarity matrix S is obtained by merging using the maximum function. D And the angle similarity matrix S A , whose elements are

[0146]

[0147]

[0148] Then, the distance similarity matrix and the angle similarity matrix are merged. This step has two purposes: one is to unify the similarities of different dimensions, and the other is to maximize the degree of differentiation of each element in the similarity matrix to facilitate the final multipath relationship judgment. Based on the principle of maximizing the variance of each element in the merged similarity matrix, principal component analysis (PCA) can be used to analyze the deformed distance similarity matrix. and angle similarity matrix Perform principal component projection to obtain the merged similarity matrix

[0149] First, the distance and angle similarity matrices are vectorized, that is, the non-zero elements of the distance or angle similarity matrix are rearranged into a column vector, that is,

[0150]

[0151]

[0152] calculate The covariance matrix of is decomposed into singular value (SVD), namely

[0153]

[0154] Among them, λ1 and λ2 are the singular values ​​obtained after decomposition, u1 and v1 are the singular vectors corresponding to the singular value λ1, and u2 and v2 are the singular vectors corresponding to the singular value λ2. The projection on u1 is The principal component of , which is also the vectorized form of the merged similarity matrix Right now

[0155]

[0156] Rearrange Restore it to its original form and S D and S A Same merged similarity matrix

[0157]

[0158] Step 5: Optimize and solve the combined similarity matrix S to obtain the multipath relationship.

[0159] According to the maximum likelihood criterion, the higher the similarity of the elements in the combined similarity matrix, the more likely the corresponding multipath relationship is to hold. Through optimization, a series of highly similar elements are obtained, which can then determine the multipath relationships in the scene and avoid inconsistencies in these relationships.

[0160] For the element S of the merged similarity matrix S ij Solve as follows and get the solution set {x ij}, where i, j = 1, 2, ..., N-1:

[0161]

[0162]

[0163] Here the objective function is ∑x ij ·S ij, that is, select the largest similarity from S as much as possible, and at the same time, meet the conditions of non-independent goals, that is, S ij >S iso .

[0164] S iso is the isolation threshold, which is used to exclude independent targets, that is, targets that do not generate multipath in the scene. Taking into account the spatial distribution of targets in the scene and the uncertainty of the radar system's distance and angle measurement errors, the present invention adopts an adaptive threshold calculation method to obtain S iso :

[0165] First, select the largest element in each column of the merged similarity matrix S column by column, sort the selected elements in descending order, and add them to the set S one by one according to the element value from large to small up In. S up Every time a change occurs, calculate S up The variance of each element in. If S up The variance of each element in is less than the variance threshold v th , then continue to incorporate the elements in S into S up Until S is not satisfied up The variance of each element in is less than v th The condition, or the largest elements of each column in S are all included in S up At this time S up The smallest element in is S iso .

[0166] In addition, considering that there are no multipath targets in the scene, that is, all detected targets are independent targets, if S iso If the similarity is less than a fixed lower bound, such as 0.5, then S iso Replace it with 0.5 to select all independent targets.

[0167] The last two constraints and The physical meanings of are: each multipath target corresponds to only one source target and each source target generates at most three different types of multipath targets. These two constraints avoid the occurrence of contradictions in the optimization results.

[0168] Finally, according to the solution set {x ij}Get the identification results of multipath relationship one by one, and use vector X=[X1,X2,…X N ] indicates that since the target closest to the radar must be the original target rather than the multipath target, let X1 = [1,0,0] T , for X2,…X N Then there is

[0169]

[0170] Where j = 1, 2,…, N-1.

[0171] Multipath relationship identification result X for each target N It includes three items. The first item is the target number. The second item indicates the existence and ownership of the multipath target. If the second item is 0, it means that the target is a source target. If the second item is the number of another target, it means that the target is a multipath of another target. Indicates the type of multipath target, which is uniquely determined by the source target i and the multipath target j, and is traced back to S ij The type of multipath target is determined based on the first-order or second-order distance similarity matrix, the direct angle similarity matrix, or the non-direct angle similarity matrix. The specific rules are as follows:

[0172]

[0173] It can be seen that represents the first-order direct multipath target, represents the first-order side-reaching multipath target, Indicates a second-order multipath target.

[0174] At this point, all multipath relationships (existence, belongingness, and multipath categories) can be identified with a high recognition accuracy.

Claims

1. A radar multipath target recognition method based on Gaussian similarity, characterized in that: The specific steps are as follows: Step 1: Build a radar system for target detection. Use an FMCW-MIMO millimeter-wave radar system to preprocess the radar received signal and obtain the distance set and angle set of all detected targets. The specific preprocessing process is: Step 101: obtain the intermediate frequency signal from the FMCW-MIMO radar Sampling is performed and fast and slow time division is performed to obtain the radar intermediate frequency signal sampling sequence in the fast time-slow time dimension: in, For fast time index, is the slow time index, is the intermediate frequency signal amplitude, is the Chirp signal slope, is the radial distance of the target relative to the radar, is the target micro-phase signal, For fast time intervals, For slow time intervals, is the wavelength of the center frequency, is the speed of light; Step 102: Radar intermediate frequency signal in fast time-slow time dimension Perform a fast Fourier transform on the fast time dimension to obtain the range image: in, is the number of fast time sampling points, is the index of the range image; Step 103: remove the static clutter component in the range image, and obtain the range spectrum: in, is the number of slow time sampling points; Step 104, assuming that the MIMO radar has equivalent array elements, then each equivalent array element Each corresponds to a distance spectrum ,right Constructing the autocorrelation matrix : The range-azimuth two-dimensional spectrum is obtained by using Capon spectrum estimation based on the forward and reverse autocorrelation matrix; The forward and reverse autocorrelation matrix is: The range-azimuth two-dimensional spectrum is: in, is the forward and reverse autocorrelation matrix, for The conjugate matrix of is the permutation matrix, is the azimuth, is the orientation vector, is the wavelength of the center frequency, is the spacing of equivalent array elements; Step 105: The reflection peaks in the range-azimuth two-dimensional spectrum are divided into weak reflection peaks and strong reflection peaks. The weak reflection peaks are suppressed using a constant false alarm detection algorithm, and the strong reflection peaks are classified into different clusters using a density clustering algorithm to determine the number of targets. Finally, detection targets, the distance set of all detection targets is , the angle set is ; The range and azimuth of each detected target are , where the distance to the target satisfies ; Step 2: For a scene model containing a column of single-sided reflectors, the theoretical multipath distance and theoretical multipath angle of each detection target are calculated using a theoretical calculation formula to obtain a theoretical multipath distance set and a theoretical multipath angle set; In a scenario with a single-sided column of reflectors, each detected target generates three types of multipath targets: first-order direct multipath targets, first-order side-arrival multipath targets, and second-order multipath targets. These three types of multipath targets include two theoretical multipath distances and two theoretical multipath angles. The two theoretical multipath distances are the first-order multipath distance and the second-order multipath distance, and the two theoretical multipath angles are the direct multipath angle and the side-arrival multipath angle. The specific calculation is: According to the geometric relationship, the distance between each detected target and the radar is calculated. The first-order multipath distance and the second-order multipath distance between array elements are in, The rectangular coordinates of each detection target are given by polar coordinates Perform coordinate transformation to obtain; For Radar The horizontal coordinate of each array element; is the horizontal coordinate of the reflection surface; and then the first-order multipath distance set is obtained and the second-order multipath distance set ; Similarly, calculate the direct multipath angle of each detected target and the Punta Multipath Angle : Further obtain the direct multipath angle set and the Punta polypath angle set ; Step 3: Use a one-dimensional Gaussian similarity function based on maximum likelihood to calculate the similarity between the theoretical multipath distance and the detected target distance, and the theoretical multipath angle and the detected target angle, to form a distance / angle similarity matrix; First, construct the array element Related first-order multipath distance similarity matrix ; The elements in The elements in the pair The average of for: In the above formula This is the distance similarity function. Since the radar's measurement errors for distance and angle are independent of each other and the errors satisfy the Gaussian distribution, according to the maximum likelihood criterion, the distance similarity function is expressed as: in, Indicates the target The theoretical multipath parameters, Indicates the target The measured values ​​of the parameters, represents the parameter related to the root mean square error of distance measurement, is the root mean square error of distance measurement, is the protection value when the root mean square error is too small, Correction values ​​to unify the distance and angle dimensions numerically; Similarly, construct the second-order multipath distance similarity matrix , direct angle similarity matrix , Pangda angle similarity matrix , the elements in each matrix are: in, ; The angle similarity function is in, , the parameter meaning is the same as the distance similarity function; Step 4: Perform two dimensionality reduction integrations on the four similarity matrices and use principal component projection to maximize the multipath relationship information covered by the combined similarity matrix; Merge similarity matrix : Step 5: Merge similarity matrix An optimization algorithm is used to solve the maximum sum similarity problem and compare it with the original similarity matrix to obtain the adaptive global optimal multipath identification result. Merge similarity matrix Elements Solve the maximum and similarity problem and get the solution set ,in, : in, is the isolation threshold, used to exclude independent targets; Solution set based on optimization Get the identification result of multipath relationship, use vector Indicates; Considering that the target closest to the radar must be the original target, let ,right Then there is The multipath relationship identification result of each target includes three items. The first item is the target serial number. The second item indicates whether the multipath target exists and its affiliation. If the second item is 0, it means that the target is a primary target. If the second item is the serial number of another target, it means that the target is a multipath of this other target. The third item is Indicates the type of multipath target: At this point, the existence, belongingness, and multipath categories of the multipath relationship are all identified.

2. A radar multipath target recognition method based on Gaussian similarity according to claim 1, characterized in that: In step 4, the first dimensionality reduction integration is to use the maximum function and the second-order side-by-side constraint to integrate the similarity matrix, specifically: Using the second-order side-reach constraint, since there is no multipath target with a distance of the second-order multipath distance and an angle of the direct multipath angle in this scenario model, in order to exclude this situation, it is set season ; Considering the mutual exclusivity of multipath types, that is, a multipath target has only one type of multipath distance and multipath angle, the maximum function is used to merge and obtain the distance similarity matrix and angle similarity matrix , whose elements are 。 3. A radar multipath target recognition method based on Gaussian similarity according to claim 2, characterized in that: In step 4, the second dimensionality reduction integration is based on the principle of maximizing the variance of each element in the combined similarity matrix, and the principal component extraction is performed on the deformed similarity vector to obtain the integrated combined similarity matrix, which is specifically: First, the distance and angle similarity matrices after the first dimensionality reduction integration are vectorized, that is, the non-zero elements of the distance or angle similarity matrix are rearranged into column vectors, that is, Then, calculate The covariance matrix of and perform singular value decomposition (SVD) on it: in, and are the singular values ​​obtained after decomposition, and Singular values The corresponding singular vectors, and Singular values The corresponding singular vectors; exist The projection on The principal component of , which is also the vectorized form of the merged similarity matrix ,Right now Rearrange , returning it to its original form and Same merged similarity matrix .

4. A radar multipath target recognition method based on Gaussian similarity according to claim 1, characterized in that: In step 5, the isolation threshold Adaptive threshold calculation is used, specifically: First, select the merged similarity matrix column by column The largest element in each column, sort the largest elements in each column in descending order, and add them to the set one by one according to the element value from large to small middle; Every time a change occurs, calculate The variance of each element in; Then, judge Is the variance of each element in less than the variance threshold? If so, continue The elements in In, until The variance of each element in is not less than ,or The largest elements in each column are all included in In, at this time The smallest element in is ; In addition, considering that there are no multipath targets in the scene, that is, all detected targets are independent targets, if If it is less than a fixed similarity lower bound, Replace it with a fixed lower bound of similarity to select all independent targets.

5. The radar multipath target recognition method based on Gaussian similarity according to claim 1, characterized in that: In step 5, the type of multipath target Expressed as: represents the first-order direct multipath target, represents the first-order side-reaching multipath target, Indicates a second-order multipath target.

Images