Complex target pole feature extraction method improved by adopting density peak clustering
Through density peak clustering and sliding window technology, the matrix beam method is improved, solving the problem of difficult parameters to determine and false extremes in complex target pole extraction, achieving higher extraction accuracy and robustness.
Patent Information
- Application Number
- CN202411045596.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-01
AI Technical Summary
The matrix beam method has problems in the extraction of complex target poles, which are difficult to determine parameters and false poles, resulting in poor extraction results.
A complex target pole feature extraction method is adopted to improve density peak clustering. By obtaining the target time domain response, dynamically setting the number of modes, using sliding window traversal and density peak clustering, false poles are eliminated and the real poles are determined.
It improves the accuracy and robustness of pole extraction, reduces parameter dependence, and effectively solves the problem of false pole removal by matrix bundle method in complex goals.
Smart Images

Figure CN119089179A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of target pole feature extraction, and in particular to a complex target pole feature extraction method improved by density peak value clustering. Background Art
[0002] When the target feature size is comparable to the signal wavelength, the radar operates in the resonance region, and the pole is the most important feature of the target in the resonance region. The pole is insensitive to the target attitude and polarization mode, and is mainly related to the target size, shape and material, so it can be used to identify radar targets. The accurate extraction of pole features is crucial. Existing extraction methods include the prony method, the E pulse method, the Cauchy method, the matrix beam method, etc., among which the matrix beam method is the most widely used and commonly used method. The time domain transient response of the target can be divided into early response and late response. The early response is a forced response and is related to the incident pulse signal, while the late response is independent of the incident pulse signal. In the actual test environment, a signal with an extremely narrow pulse width is generally used to incident the target to obtain the scattered echo of the target. The signal with a narrow pulse width facilitates the distinction between early and late responses.
[0003] The matrix bundle method is to extract poles through the late transient response of the target in the time domain. The matrix bundle method has good noise resistance, and the algorithm is efficient and robust under a certain signal-to-noise ratio, so it is very suitable for extracting the pole features of the target. Most matrix bundle methods artificially distinguish and truncate the transient response of the target in the time domain. However, for complex targets, there are generally problems such as the difficulty in determining the aliasing interception time of early and late responses, the unknown order of complex target modes, and the difficulty in eliminating false poles. Therefore, it is difficult to apply it to the pole extraction problem of actual complex targets. If the pole features are to be used for target recognition, the pole extraction problem of complex targets is crucial. Summary of the invention
[0004] The purpose of the present invention is to solve the problem that some parameters of the matrix bundle method cannot be determined and false poles are difficult to remove, so as to obtain the pole features of complex targets, and provide a complex target pole feature extraction method improved by density peak clustering, comprising the following steps:
[0005] S1. Obtain the target time domain response based on the scattered frequency domain echo data;
[0006] S2, obtaining the theoretical late response start time according to the target size and signal duration, and obtaining the target late response;
[0007] S3, dynamically set the number of modes, use a sliding window to traverse the late response sequence, and obtain a series of possible extreme point sets;
[0008] S4. Perform density peak clustering on the possible pole set to obtain the final pole.
[0009] In a preferred solution, step S1 specifically includes the following steps:
[0010] The target frequency band [f min ,f max ]The frequency interval is δ f Scattered echo data of:
[0011]
[0012] H(ω) is the frequency domain impulse response of the target.
[0013] When the time domain incident signal is a Singh signal:
[0014] e i (t) = Sa(ω c t),
[0015] ω c is the maximum calculation frequency, then the frequency domain expression of the scattered echo signal is:
[0016]
[0017] This formula shows that if the incident signal is a Singh signal, the frequency domain scattering signal data can be obtained by truncating the frequency domain response. The actual time domain impulse signals are all real signals, and H(ω) should have a conjugate symmetric form:
[0018]
[0019] Then the time domain scattered echo signal is:
[0020]
[0021] in:
[0022]
[0023] Therefore, the time domain signal obtained by inverse Fourier transform after conjugate symmetry of the frequency domain scattered echo data calculated by the moment method is equivalent to the time domain transient response signal obtained by the Singh signal acting on the target;
[0024] The Hamming window function is applied to the frequency domain data to concentrate the signal energy in the main lobe before performing an inverse Fourier transform.
[0025] In a preferred solution, step S2 specifically includes the following steps:
[0026] When the signal transmission position and the receiving position are the same, the early response start time is T1, and the width of the pulse signal is T p , the time it takes for the signal to pass through the target is T r , the onset of the late response is usually defined as:
[0027] t1=T1+2T r +T p
[0028] Due to the delay in signal propagation, there is a time difference between the response generated by the target part that contacts the signal earliest and the target part that contacts the signal latest. The start time of the late response is defined in advance as:
[0029] t2=T1+T r +T p ;
[0030] Calculate the distance from the origin of the coordinate system to the spatial point where the wavefront of the incident wave just touches the target and the equivalent signal width T of the incident signal. ep , assuming that the projection distance from the earliest contact point between the target and the incident signal to the origin to the incident direction of the incident signal is L1, then the corresponding echo delay T1 = 2L1 / c, c is the speed of light, and the equivalent form of the incident signal used is the Singh signal, and its equivalent signal width T ep The main lobe width is calculated based on its first zero point:
[0031]
[0032] f m is the highest frequency corresponding to the incident signal.
[0033] In a preferred solution, step S2 also includes the following steps:
[0034] The target late transient response is expressed as a decaying exponential sum in the complex plane:
[0035]
[0036] Where R i =α i +jβ i ,s i =σ i +jω i ; R i is the residue, s i is the extreme point; i and ω i are the attenuation factor and attenuation angular frequency respectively;
[0037] Assume that the sampled discrete form of the late transient response y(t) in the time domain is y(n), which can be expressed as:
[0038]
[0039] Where N is the number of sampling points, M is the number of modes (number of poles);
[0040] Define the Hankel matrix:
[0041]
[0042] Where L is the beam parameter. In order to overcome noise interference, the above formula is subjected to singular value decomposition to remove smaller non-zero singular values:
[0043]
[0044] where Σ=diag{σ1,σ2,...,σ M},σ1,σ2,...,σ M are the largest M singular values of Y, Σ' is a diagonal matrix containing smaller eigenvalues of noise information;
[0045] Take the low-rank approximation Y' of Y:
[0046] Y'=UΣV H =[Y1'y0] = [y L Y2']
[0047] The matrix Y1' is obtained by removing the last column of Y', and Y2' is obtained by removing the first column of Y';
[0048] By solving the eigenvalue of Y1'+Y2', we can obtain the target pole z i , where Y1'+ is the generalized inverse of Y1';
[0049] The residues corresponding to the extreme points can be obtained by the least squares method:
[0050] x=(A T A) -1 A T b
[0051] in:
[0052]
[0053] x=[R1,R2,...,R M ] T ,
[0054] b=[y(0),y(1),...,y(N-1)] T ,
[0055] After the poles are extracted using the matrix bundle method, some false poles are removed.
[0056] In the preferred solution, the criteria for removing false poles include:
[0057] The poles are conjugate symmetric, so only the poles in the second quadrant are considered, and the attenuation factor σ iLess than zero, oscillation frequency ω i The poles greater than zero are screened out;
[0058] The target pole must satisfy the condition 2πf min <ω i <2πf max , those that do not meet the conditions are eliminated;
[0059] Calculate the energy value E i =|R i | / σ i |(R i is the residue corresponding to the extreme point), let E i =E i / max(E i ), will satisfy E i >The extreme point of p is left, and the value of p is 0.001;
[0060] The number of patterns is set dynamically, and a sliding window is used to traverse the response sequence.
[0061] In the preferred solution, in step S3, the number of patterns is dynamically set and a sliding window is used for traversal:
[0062] According to the start time, the late response is intercepted to obtain the late response sequence y(n), n=1,2,…N;
[0063] Setting mode number M range M min ~M max , minimum M min Not less than 10, maximum M max Less than N / 6;
[0064] Set the sliding window size N w , for M min ~M max For each M in the range, slide the window from the time n = 1 when the late response sequence is obtained to n = NN w +1 moment, generates W num =NN w +1 sliding window, apply the matrix bundle method to each window and remove some false poles, and finally get a set of possible poles.
[0065] In a preferred solution, step S4 further includes the following steps: using a density peak clustering algorithm:
[0066] Suppose there is a data set of size N×M, where N is the number of samples and M is the sample dimension. There are two ways to calculate the local density of sample points. The discrete value is calculated using the truncated kernel, while the continuous value is calculated using the Gaussian kernel:
[0067] Truncated core:
[0068]
[0069] Gaussian Kernel:
[0070] Where, d ij is the Euclidean distance between data point i and data point j, d c is the neighborhood cutoff distance of the data point;
[0071] The local density ρ calculated using the truncated kernel i is equal to the number of sample points distributed within the neighborhood cutoff distance of sample point i; the local density ρ calculated using Gaussian i Equal to the sum of the Gaussian distances from all sample points to sample point i;
[0072] Relative distance δ i It refers to the minimum distance between sample point i and other points with higher density. Before calculating sample point i, the local density of each data point needs to be sorted. For the sample with the highest density, the relative distance is defined as:
[0073]
[0074] For the remaining data points, the relative distance is defined as:
[0075]
[0076] Find the decision value γ and satisfy the local density ρ i Higher and relative distance δ i The density peak for the larger condition:
[0077] γ i =ρ i ×δ i .
[0078] In a preferred solution, the selection of the cutoff distance includes the following steps:
[0079] Given n sample points, the local density ρ1, ρ2, … ρ n , when the density value of each sample point is equal, the greater the uncertainty of the data distribution, the larger the Shannon entropy is, and the smallest Shannon entropy is when the uncertainty is the smallest. The density estimation entropy is introduced to measure the rationality of the local density estimation of the sample point, that is:
[0080]
[0081] Where Z is a normalization coefficient;
[0082] When the density estimation entropy takes the minimum value, the optimal value of the truncation distance is obtained.
[0083] In the preferred solution, step S4 also includes determining the cluster center according to the γ decision diagram:
[0084] γ i The larger the value, the more likely it is the cluster center. Arrange in descending order, label the sorted points, i represents the label, and obtain the sorting diagram of γ; select slope k to describe the degree of change of γ, k i It represents the slope of the line segment between the i-th point and the i+1-th point. When i is the point with the largest number among the points with a slope greater than the average value, i is the critical point.
[0085] The points in the sorting diagram whose serial numbers are before the critical point are regarded as suspected cluster centers.
[0086] In a preferred solution, step S4 further includes: removing duplicates from suspected cluster centers, including the following steps:
[0087] Determine whether the distance between the suspected cluster centers is less than the cutoff distance. When the distance between two suspected cluster centers is less than the cutoff distance, the point with a larger local density between the two is taken as the actual cluster center. When the local densities of the two are the same, the mean of the two is taken as the cluster center, and finally the actual cluster center set is obtained.
[0088] Determine whether the distance between the suspected cluster centers is less than the cutoff distance. When the distance between two suspected cluster centers is less than the cutoff distance, take the point with a larger local density between the two as the actual cluster center. When the local densities of the two are the same, take the mean of the two as the cluster center, and finally get the actual cluster center set.
[0089] In a preferred solution, step SS4 further includes: classifying the remaining points into clusters, specifically including the following steps:
[0090] After determining the actual cluster center set, you need to classify the remaining points into the corresponding clusters, determine the distance between the remaining points and each point in the actual cluster center set in turn, and find the actual cluster center closest to the point:
[0091] When the distance is less than the cutoff distance d c , it is classified into the cluster of this cluster center;
[0092] When the distance between the point and each point in the actual cluster center set is greater than the cutoff distance d c , then the point is regarded as a noise point.
[0093] In a preferred solution, step S4 also includes the following steps:
[0094] Calculate the number of points in the cluster to which each cluster center belongs. When the number is greater than the set density threshold, the cluster is a true extreme point cluster. Otherwise, it is considered a false extreme point cluster and is removed.
[0095] The cluster centers of the true pole clusters are the desired poles.
[0096] The beneficial effects of the present invention are as follows: the present invention utilizes the clustering and orientation invariance of real poles and proposes to use a density peak clustering algorithm to improve the matrix bundle method. Dynamically traversing the mode order avoids the error caused by the unknown order of the complex target mode. At the same time, the sliding window is introduced to slide backward from the theoretical earliest start time, reducing the influence of the aliasing of the front and back time on the real poles. The possible poles obtained are clustered to further determine the real pole clusters. The results show that the algorithm is robust and accurate, and effectively solves the parameter dependence problem of the matrix bundle method. BRIEF DESCRIPTION OF THE DRAWINGS
[0097] Figure 1 It is a schematic diagram of early and late response;
[0098] Figure 2 It is the algorithm flow chart of the present invention;
[0099] Figure 3 This is the effect diagram of conductor sphere extraction using the basic matrix bundle method;
[0100] Figure 4 This is the effect diagram of conductor sphere extraction using the DPC-SWMPM algorithm;
[0101] Figure 5 It is a picture of an airplane model;
[0102] Figure 6 This is the result diagram of the ten orientation extractions of the basic matrix bundle method for the aircraft target;
[0103] Figure 7 This is the result diagram of DPC-SWMPM extraction of ten directions of aircraft target. DETAILED DESCRIPTION
[0104] 1. Obtain the target time domain response based on the scattered frequency domain echo data: obtain the target frequency domain scattered echo data, make conjugate symmetry on the frequency domain scattering data, and obtain the target time domain response by inverse Fourier transform.
[0105] The target frequency band [f min ,f max ]The frequency interval is δ f Scattered echo data of:
[0106]
[0107] H(ω) is the frequency domain impulse response of the target.
[0108] If the time domain incident signal is a Singh signal:
[0109] e i (t) = Sa(ω c t),
[0110] ω c is the maximum calculation frequency, then the frequency domain expression of the scattered echo signal is:
[0111]
[0112] This formula shows that if the incident signal is a Singer signal, the frequency domain scattering signal data can be obtained by truncating the frequency domain response. Because the actual time domain impulse signals are all real signals, H(ω) should have a conjugate symmetric form:
[0113]
[0114] Then the time domain scattered echo signal is:
[0115]
[0116] in:
[0117]
[0118] Therefore, the time domain signal obtained by inverse Fourier transform after conjugate symmetry of the frequency domain scattering echo data calculated by the moment method is equivalent to the time domain transient response signal obtained by the Singh signal acting on the target.
[0119] Ideally, the frequency domain data should be infinitely long. Frequency domain truncation will cause the sidelobe energy leakage of the time domain transient response signal. When the incident signal is incident on the target, the sidelobe of the signal will also be convolved with the system response of the target. This process is cumulative, and the energy of the sidelobe cannot be ignored. In order to avoid the leakage of sidelobe energy, the Hamming window function is applied to the frequency domain data to concentrate the energy of the signal on the main lobe before performing an inverse Fourier transform, so that the obtained time domain response data is more accurate.
[0120] 2. Obtain the theoretical late response start time based on the target size and signal duration to obtain the target late response:
[0121] If the signal transmission position and reception position are the same, the early response start time is T1, and the width of the pulse signal is T p , the time it takes for the signal to pass through the target is T r , the onset of the late response is usually defined as:
[0122] t1=T1+2T r +T p
[0123] However, due to the delay in signal propagation, there is a time difference between the responses generated by the target part that contacts the signal first and the target part that contacts the signal last. This time delay will cause aliasing of early and late responses. Although the late response start time defined above can ensure that there is no early response in the late response, it will lose some information and energy of the late response. The loss of late response information will cause inaccurate pole extraction and easy loss of related poles. Therefore, the start time of the late response needs to be defined in advance as:
[0124] t2=T1+T r +T p
[0125] This definition method can make the information and energy of the late response more complete. If the theoretical time is used as the late start time for extraction, it can not only reduce the interference of the early response as much as possible, but also ensure that the time domain signal with higher energy is not lost. Figure 1 shown.
[0126] The zero time of the time domain transient response obtained by inverse Fourier transform of the scattered electric field data in the frequency domain corresponds to the time when the center of the incident signal reaches the origin of the coordinate system, so there will be 1 / 2T in the incident time. ep If you want to find the real starting time of the incident signal's interaction with the target, you need to calculate the distance from the coordinate origin to the spatial point where the wavefront of the incident wave just touches the target and the equivalent signal width T of the incident signal. ep If the projection distance from the earliest contact point between the target and the incident signal to the origin to the incident direction of the incident signal is L1, then the corresponding echo delay T1 = 2L1 / c, where c is the speed of light. The equivalent form of the incident signal used is the Singer signal, and its equivalent signal width T ep The main lobe width can be calculated based on its first zero point:
[0127]
[0128] f m is the highest frequency corresponding to the incident signal.
[0129] 3. Dynamically set the number of modes and use a sliding window to traverse the late response sequence to obtain a series of possible pole sets:
[0130] Matrix bundle extraction principle:
[0131] The target late transient response can be expressed as a decaying exponential sum in the complex plane:
[0132]
[0133] Where R i =α i+jβ i ,s i =σ i +jω i ; R i is the residue, s i is the extreme point; i and ω i are the attenuation factor and attenuation angular frequency respectively.
[0134] Assume that the sampled discrete form of the late transient response y(t) in the time domain is y(n), which can be expressed as:
[0135]
[0136] Where N is the number of sampling points and M is the number of modes (number of poles). Define the Hankel matrix:
[0137]
[0138] Where L is the beam parameter. In order to overcome noise interference, the above formula is subjected to singular value decomposition to remove smaller non-zero singular values:
[0139]
[0140] where Σ=diag{σ1,σ2,...,σ M},σ1,σ2,...,σ M are the largest M singular values of Y, and Σ' is a diagonal matrix containing smaller eigenvalues of noise information. Take the low-rank approximation Y' of Y:
[0141] Y'=UΣV H =[Y1' y0] = [y L Y2']
[0142] The matrix Y1' is obtained by removing the last column of Y', and Y2' is obtained by removing the first column of Y'.
[0143] By solving Y1' + The eigenvalue of Y2' is used to obtain the target pole z i , where Y1' + is the generalized inverse of Y1'.
[0144] The residues corresponding to the extreme points can be obtained by the least squares method:
[0145] x=(A T A) -1 A T b
[0146] in:
[0147]
[0148] x=[R1,R2,...,R M ] T ,
[0149] b=[y(0),y(1),...,y(N-1)] T ,
[0150] After the poles are extracted using the matrix bundle method, some false poles are removed.
[0151] Criteria for false pole removal:
[0152] (1) The poles are conjugate symmetric. Generally, only the poles in the second quadrant are considered. The attenuation factor σ i Less than zero, oscillation frequency ω i The poles greater than zero are screened out;
[0153] (2) The target pole must satisfy the condition 2πf min <ω i <2πf max , those that do not meet the conditions are eliminated;
[0154] (3) Calculate the energy value E i =|R i | / σ i |(R i is the residue corresponding to the extreme point), let E i =E i / max(E i ), will satisfy E i >p (the value of p is generally 0.001) is retained.
[0155] For high-Q and simple objects, the above method can obtain relatively accurate poles, but for complex objects and low-Q objects, the extraction results are not accurate enough and false poles are difficult to remove. In addition, the number of poles of complex objects is unknown. Previous methods for estimating the order include the minimum description length method (MDL), calculating the error between the reconstructed signal and the original signal when different M values are used, etc. However, the MDL method generally produces results greater than the number of true poles for simulated and measured signals; the calculation method of reconstruction error has problems such as being unable to eliminate the reconstruction error caused by the residual error and being unable to remove the false poles caused by the signal itself. Therefore, it is necessary to dynamically set the number of modes and use a sliding window to traverse the response sequence.
[0156] Dynamically set the number of patterns and use a sliding window traversal:
[0157] According to the theoretical start time mentioned above, the late response is intercepted to obtain the late response sequence y(n), n=1, 2, ...N.
[0158] Setting mode number M range Mmin ~M max , minimum M min Generally not less than 10, the maximum M max Generally less than N / 6.
[0159] Set the sliding window size N w , for M min ~M max For each M in the range, slide the window from the time n = 1 when the late response sequence is obtained to n = NN w +1 moment, generates W num =NN w +1 sliding window, apply the matrix bundle method to each window and remove some false poles, and finally get a set of possible poles.
[0160] 4. Perform density peak clustering on the possible pole set to obtain the final pole.
[0161] Density Peaks Clustering (DPC) is a density-based clustering algorithm that has the advantages of not requiring the specification of the number of clusters and being able to discover non-spherical clusters.
[0162] Two important concepts of the DPC algorithm:
[0163] (1) Local density
[0164] Suppose there is a data set with a data size of N×M, where N is the number of samples and M is the sample dimension.
[0165] For the local density of sample point i, there are two ways to calculate the local density. The discrete value is calculated using the truncated kernel, and the continuous value is calculated using the Gaussian kernel:
[0166] Truncated core:
[0167]
[0168] Gaussian Kernel:
[0169]
[0170] Where, d ij is the Euclidean distance between data point i and data point j, d c is the neighborhood cutoff distance for data points.
[0171] The local density ρ calculated using the truncated kernel i is equal to the number of sample points distributed within the neighborhood cutoff distance of sample point i; and the local density ρ calculated using Gaussian i It is equal to the sum of the Gaussian distances from all sample points to sample point i.
[0172] (2) Relative distance
[0173] Relative distance δ i Refers to the minimum distance between sample point i and other points with higher density. Before calculating sample point i, the local density of each data point needs to be sorted.
[0174] For the sample with the highest density, the relative distance is defined as:
[0175]
[0176] For the remaining data points, the relative distance is defined as:
[0177]
[0178] Since there is no point with higher density than the sample with the highest density, DPC considers this point to be the density peak (cluster center) and artificially sets its relative distance to the maximum value. The remaining density peaks need to meet two conditions at the same time: local density ρ i Higher, relative distance δ i Large. Find this type of density peak through the decision value γ:
[0179] γ i =ρ i ×δ i
[0180] Cut-off distance selection:
[0181] In information theory, Shannon entropy is used as a measure of system uncertainty. Given n sample points, the local density ρ1, ρ2, … ρ n , if the density value of each sample point is equal, the greater the uncertainty of the data distribution, the greater the Shannon entropy. In any case, the smallest uncertainty has the smallest Shannon entropy. The density estimation entropy is introduced to measure the rationality of the local density estimation of the sample point, that is:
[0182]
[0183] Where Z is a normalization coefficient.
[0184] When the density estimation entropy takes the minimum value, the optimal value of the truncation distance can be obtained.
[0185] Selection of cluster centers:
[0186] The density peak clustering algorithm (DPC) assumes that the cluster center also has a large local density ρ i and the relative distance δ iThe cluster center needs to be determined manually based on the γ decision diagram, but for some special or complex data sets, the generated decision diagram is often difficult to visually observe the cluster center. Since the DPC clustering process is not an iterative process, once the cluster center is selected incorrectly, it will directly affect the accuracy of the final clustering result.
[0187] γ i The larger the value, the more likely it is the cluster center. Arrange in descending order, label the sorted points, i represents the label, and get the sorting diagram of γ. The slope k is used to describe the degree of change of γ, k i Represents the slope of the line segment between the i-th point and the i+1-th point. When i is the point with the largest number among the points with a slope greater than the average value, i is the critical point. The points in the sorting diagram with numbers before the critical point are considered as suspected cluster centers.
[0188] Deduplication of suspected cluster centers:
[0189] Determine whether the distance between the suspected cluster centers is less than the cutoff distance. If the distance between two suspected cluster centers is less than the cutoff distance, take the point with the larger local density between the two as the actual cluster center; if the local densities of the two are the same, take the mean of the two as the cluster center. Finally, the actual cluster center set is obtained.
[0190] The remaining points are classified into clusters:
[0191] After determining the actual cluster center set, the remaining points need to be classified into the corresponding clusters. The distance between the remaining points and each point in the actual cluster center set is determined in turn, and the actual cluster center closest to the point is found. If the distance is less than the cutoff distance d c , then it is classified into the cluster of this cluster center; if the distance between this point and each point in the actual cluster center set is greater than the cutoff distance d c , then the point is regarded as a noise point.
[0192] Get clusters of true extreme points:
[0193] Calculate the number of points in each cluster center. If the number is greater than the set density threshold, the cluster is a true pole cluster. Otherwise, it is considered a false pole cluster and is removed. The cluster center of the true pole cluster is the desired pole.
[0194] The specific process of the algorithm (DPC-SWMPM) is as follows Figure 2 shown.
[0195] 5. Results comparison
[0196] (1) Simple low-Q target comparison: conducting sphere.
[0197] The scattered electric field data of a conducting sphere with a radius a of 0.075m at 200 frequency points from 10MHz to 5.5GHz are calculated, and the results extracted by the basic matrix beam method and this algorithm (DPC-SWMPM) are compared. The mode number of the basic matrix beam method is set to 50, and the mode number of this algorithm ranges from 40 to 60. The results are normalized by c / a, where a is the radius of the sphere and c is the speed of light. Figure 3 The comparison of the basic matrix bundle method for extracting the conductor sphere is shown in the figure. Figure 4 The DPC-SWMPM algorithm conductor sphere extraction effect is shown. It can be seen that the basic matrix bundle method has false poles that cannot be eliminated and the extracted values are greatly different from the theoretical values. DPC-SWMPM clusters the poles extracted from different mode numbers, further eliminates false poles, and makes the poles closer to the theoretical values. The effect is significantly better than the basic matrix bundle method.
[0198] (2) Complex target extraction comparison: aircraft model.
[0199] right Figure 5 The aircraft model is simulated with a frequency range of 50KHz-80MHz. The single-station scattered echo data at ten azimuth angles are obtained and extracted using the basic matrix beam method and this algorithm (DPC-SWMPM). Because the poles are azimuthally invariant, a group of poles mentioned in the azimuth with the largest number of poles are taken as base poles, and the window is integrated around the base poles to obtain the mean of the poles in the window. The mode number of the basic matrix beam method is set to 60, and the mode number of this algorithm ranges from 40 to 80. The results are normalized by πc / L, where L is the length of the aircraft 18.6m and c is the speed of light.
[0200] contrast Figure 7 DPC-SWMPM algorithm extraction effect and Figure 6 The extraction effect of the basic matrix bundle method shows that the poles extracted by the DPC-SWMPM algorithm are more clustered, and under the same windowing and integration conditions, the number of poles obtained by DPC-SWMPM is greater than that of the matrix bundle method. The poles with high attenuation and small residue are difficult to extract accurately. From this example, it can be clearly seen that the pole attenuation factor extracted by DPC-SWMPM is more accurate.
[0201] By comparing the extraction effects of simple and complex targets, it is proved that this algorithm can achieve good results for the pole extraction problem. The pole features extracted by this algorithm are consistent in all directions, and the attenuation factor of the pole is more accurate than that of the basic matrix bundle method.
[0202] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. The complex target pole feature extraction method improved by density peak clustering is characterized by: S1. Obtain the target time domain response based on the scattered frequency domain echo data; S2, obtaining the theoretical late response start time according to the target size and signal duration, and obtaining the target late response; S3, dynamically set the number of modes, use a sliding window to traverse the late response sequence, and obtain a series of possible extreme point sets; S4. Perform density peak clustering on the possible pole set to obtain the final pole.
2. The complex target pole feature extraction method improved by density peak clustering according to claim 1 is characterized in that: Step S1 specifically includes the following steps: The target frequency band [f min ,f max ]The frequency interval is δ f Scattered echo data of: H(ω) is the frequency domain impulse response of the target; When the time domain incident signal is a Singh signal: it i (t)=Sa(ω c t), ω c is the maximum calculation frequency, then the frequency domain expression of the scattered echo signal is: This formula shows that if the incident signal is a Singh signal, the frequency domain scattering signal data can be obtained by truncating the frequency domain response. The actual time domain impulse signals are all real signals, and H(ω) should have a conjugate symmetric form: Then the time domain scattered echo signal is: in: Therefore, the time domain signal obtained by inverse Fourier transform after conjugate symmetry of the frequency domain scattered echo data calculated by the moment method is equivalent to the time domain transient response signal obtained by the Singh signal acting on the target; The Hamming window function is applied to the frequency domain data to concentrate the signal energy in the main lobe before performing an inverse Fourier transform.
3. The complex target pole feature extraction method improved by density peak clustering according to claim 1 is characterized in that: Step S2 specifically includes the following steps: When the signal transmission position and the receiving position are the same, the early response start time is T1, and the width of the pulse signal is T p , the time it takes for the signal to pass through the target is T r , the onset of the late response is usually defined as: t1=T1+2T r +T p Due to the delay in signal propagation, there is a time difference between the response generated by the target part that contacts the signal earliest and the target part that contacts the signal latest. The start time of the late response is defined in advance as: t2=T1+T r +T p ; Calculate the distance from the origin of the coordinate system to the spatial point where the wavefront of the incident wave just touches the target and the equivalent signal width T of the incident signal. ep , assuming that the projection distance from the earliest contact point between the target and the incident signal to the origin to the incident direction of the incident signal is L1, then the corresponding echo delay T1 = 2L1 / c, c is the speed of light, and the equivalent form of the incident signal used is the Singh signal, and its equivalent signal width T ep The main lobe width is calculated based on its first zero point: f m is the highest frequency corresponding to the incident signal.
4. The complex target pole feature extraction method improved by density peak clustering according to claim 1 is characterized in that: Step S2 also includes the following steps: The target late transient response is expressed as a decaying exponential sum in the complex plane: Where R i =α i +jβ i ,s i =σ i +jω i ; R i is the residue, s i is the extreme point; i and ω i are the attenuation factor and attenuation angular frequency respectively; Assume that the sampled discrete form of the late transient response y(t) in the time domain is y(n), which can be expressed as: Where N is the number of sampling points, M is the number of modes (number of poles); Define the Hankel matrix: Where L is the beam parameter. In order to overcome noise interference, the above formula is subjected to singular value decomposition to remove smaller non-zero singular values: where Σ=diag{σ1,σ2,...,σ M },σ1,σ2,...,σ M are the largest M singular values of Y, Σ' is a diagonal matrix containing smaller eigenvalues of noise information; Take the low-rank approximation Y' of Y: Y'=UΣV H =[Y’1 y0]=[y L Y’2] The matrix Y'1 is obtained by removing the last column of Y', and Y'2 is obtained by removing the first column of Y'; By solving The eigenvalue of the target pole z is obtained i ,in is the generalized inverse of Y'1; The residues corresponding to the extreme points can be obtained by the least squares method: x=(A T A) -1 A T b in: x=[R1,R2,...,R M ] T , b=[y(0),y(1),...,y(N-1)] T , After the poles are extracted using the matrix bundle method, some false poles are removed.
5. The complex target pole feature extraction method improved by density peak clustering according to claim 4 is characterized in that: The criteria for false pole removal include: The poles are conjugate symmetric, so only the poles in the second quadrant are considered, and the attenuation factor σ i Less than zero, oscillation frequency ω i The poles greater than zero are screened out; The target pole must satisfy the condition 2πf min <ω i <2πf max , those that do not meet the conditions are eliminated; Calculate the energy value E i =|R i | / |σ i |(R i is the residue corresponding to the extreme point), let E i =E i / max(E i ), will satisfy E i >The extreme point of p is left, and the value of p is 0.001; The number of patterns is set dynamically, and a sliding window is used to traverse the response sequence.
6. The complex target extreme point feature extraction method improved by density peak clustering according to claim 1 is characterized by: In step S3, the number of patterns is dynamically set and a sliding window is used for traversal: According to the start time, the late response is intercepted to obtain the late response sequence y(n), n=1,2,...N; Setting mode number M range M min ~M max , minimum M min Not less than 10, maximum M max Less than N / 6; Set the sliding window size N w , for M min ~M max For each M in the range, slide the window from the time n = 1 when the late response sequence is obtained to n = NN w +1 moment, generates W num =NN w +1 sliding window, apply the matrix bundle method to each window and remove some false poles, and finally get a set of possible poles.
7. The method for extracting complex target pole features using density peak clustering improvement according to claim 1, characterized in that: Step S4 also includes the following steps: using a density peak clustering algorithm: Suppose there is a data set of size N×M, where N is the number of samples and M is the sample dimension. There are two ways to calculate the local density of sample points. The discrete value is calculated using the truncated kernel, while the continuous value is calculated using the Gaussian kernel: Truncated core: Gaussian Kernel: Where, d ij is the Euclidean distance between data point i and data point j, d c is the neighborhood cutoff distance of the data point; The local density ρ calculated using the truncated kernel i is equal to the number of sample points distributed within the neighborhood cutoff distance of sample point i; the local density ρ calculated using Gaussian i Equal to the sum of the Gaussian distances from all sample points to sample point i; Relative distance δ i It refers to the minimum distance between sample point i and other points with higher density. Before calculating sample point i, the local density of each data point needs to be sorted. For the sample with the highest density, the relative distance is defined as: For the remaining data points, the relative distance is defined as: Find the decision value γ and satisfy the local density ρ i Higher and relative distance δ i The density peak for the larger condition: c i =ρ i ×d i 。 8. The complex target pole feature extraction method improved by density peak clustering according to claim 7, characterized in that: The cutoff distance selection includes the following steps: Given n sample points, the local density ρ1, ρ2, … ρ n , when the density value of each sample point is equal, the greater the uncertainty of the data distribution, the larger the Shannon entropy is, and the smallest Shannon entropy is when the uncertainty is the smallest. The density estimation entropy is introduced to measure the rationality of the local density estimation of the sample point, that is: Where Z is a normalization coefficient; When the density estimation entropy takes the minimum value, the optimal value of the truncation distance is obtained.
9. The method for extracting complex target pole features using density peak clustering improvement according to claim 1, characterized in that: Step S4 also includes determining the cluster center, which needs to be done based on the γ decision diagram: γ i The larger the value, the more likely it is the cluster center. Arrange in descending order, label the sorted points, i represents the label, and obtain the sorting diagram of γ; select slope k to describe the degree of change of γ, k i It represents the slope of the line segment between the i-th point and the i+1-th point. When i is the point with the largest number among the points with a slope greater than the average value, i is the critical point. The points in the sorting diagram whose serial numbers are before the critical point are regarded as suspected cluster centers.
10. The method for extracting complex target pole features using density peak clustering improvement according to claim 1, characterized in that: Step S4 also includes: removing duplicates from suspected cluster centers, including the following steps: Determine whether the distance between the suspected cluster centers is less than the cutoff distance. When the distance between two suspected cluster centers is less than the cutoff distance, the point with a larger local density between the two is taken as the actual cluster center. When the local densities of the two are the same, the mean of the two is taken as the cluster center, and finally the actual cluster center set is obtained. Determine whether the distance between the suspected cluster centers is less than the cutoff distance. When the distance between two suspected cluster centers is less than the cutoff distance, take the point with a larger local density between the two as the actual cluster center. When the local densities of the two are the same, take the mean of the two as the cluster center, and finally get the actual cluster center set.
11. The method for extracting complex target pole features using density peak clustering improvement according to claim 1, characterized in that: Step SS4 also includes: classifying the remaining points into clusters, specifically including the following steps: After determining the actual cluster center set, you need to classify the remaining points into the corresponding clusters, determine the distance between the remaining points and each point in the actual cluster center set in turn, and find the actual cluster center closest to the point: When the distance is less than the cutoff distance d c , it is classified into the cluster of this cluster center; When the distance between the point and each point in the actual cluster center set is greater than the cutoff distance d c , then the point is regarded as a noise point.
12. The method for extracting complex target pole features using density peak clustering improvement according to claim 1, characterized in that: Step S4 also includes the following steps: Calculate the number of points in the cluster to which each cluster center belongs. When the number is greater than the set density threshold, the cluster is a true extreme point cluster. Otherwise, it is considered a false extreme point cluster and is removed. The cluster centers of the true pole clusters are the desired poles.
Citation Information
Patent Citations
Method and device for extracting pole characteristic of complicated target
CN101776748A
Pole characteristic solving method in resonance region target narrow-band mode
CN110907912A
Radar signal density peak value clustering method based on improved community merging
CN114004259A
FMCW radar moving and micro-moving target detection and recognition method and system
WO2024000610A1
