CN113534054A - Improved radar target detection method based on homogeneous clutter content sharing - Google Patents

Abstract

The invention belongs to the technical field of radar signal processing, and particularly discloses an improved radar target detection method based on homogeneous clutter content sharing. Meanwhile, different training sample sets are selected according to different Doppler channels, covariance matrixes of various clustered clutters are calculated in sequence, estimation of the clutter covariance matrixes is more accurate, filtering performance of the adaptive filter is improved, the problem that plateau clutters are non-uniform and non-stable is effectively solved, and the suppression effect of the clutters is more remarkable. The homogeneous clutter samples after the Doppler spectrum K-means clustering method are adopted as the reference unit of the current unit to be detected, and the criterion of selecting the reference unit by the traditional CFAR is improved.

Description

Improved radar target detection method based on homogeneous clutter content sharing

Technical Field

The invention belongs to the technical field of radar signal processing, relates to a radar target detection method, and particularly relates to an improved radar target detection method based on homogeneous clutter content sharing, which can be used for improving the target detection performance of a radar under the non-uniform clutter background.

Background

The radar target detection technology is always the core technology of radar system design, and the main challenges of modern radar target detection are complex and changeable clutter background, small echo amplitude of low-altitude observable targets, low signal-to-noise-ratio and easy flooding by clutter, which causes the problem that the actual problem is difficult to solve by the traditional detection algorithm.

The airborne radar takes the high-altitude motion platform as a carrier, has the advantages of strong maneuverability, breakthrough of the limitation of the curvature of the earth to the sight distance, capability of effectively overcoming the problem of terrain shielding and the like, and greatly improves the detection performance of low-altitude targets and ground targets. When Space-Time Adaptive Processing (STAP) is carried out on the airborne radar before target detection, theoretically, the full-dimensional STAP can obtain the optimal Processing effect. Reed et al, in "Rapid conversion in adaptive array" IEEE Transactions on Aerospace and Electronic Systems (Volume: AES-10, Issue:6, Nov.1974), analyzed the convergence of the statistical STAP method in the Gaussian clutter background, indicating that to ensure that the output signal-to-noise ratio from the sample estimate plus noise covariance data matrix does not exceed 3dB, the number of independently identically distributed training samples required is at least twice the system degree of freedom. However, in practical applications, the space-time degree of freedom of the system of the airborne radar is usually high, so that the above condition is difficult to satisfy. In order to solve the problem of the small samples, students propose a dimension reduction STAP method, such as a 3DT (Doppler three channel combined adaptive processing) algorithm proposed in the text "airborne radar space-time two-dimensional signal processing" published in 1994 01 by shines-shou et al, the algorithm combines a plurality of Doppler channels and all the airspace channels for adaptive processing on the premise of not seriously losing the system performance, and a filter notch formed in a uniform clutter environment can be well matched with clutter. However, clutter environments faced by airborne radars are often non-uniform, and a clutter-plus-noise covariance matrix estimated by using non-uniform training samples cannot reflect the real clutter statistical characteristics of a distance unit to be detected, so that the clutter suppression performance of the 3DT algorithm is seriously reduced.

At present, the existing radar target detection method such as a Cell-Averaging CFAR (CA-CFAR) detection method has better detection performance in a uniform clutter background, but in a non-uniform clutter background, clutter sampling may no longer meet the independent same distribution characteristics, so that the target detection performance is seriously reduced.

Disclosure of Invention

Aiming at the problems in the prior art, the invention aims to provide an improved radar target detection method based on homogeneous clutter content sharing, and aims to improve the detection performance of a radar on a target under a heterogeneous complex clutter background.

In order to achieve the purpose, the invention is realized by adopting the following technical scheme.

An improved radar target detection method based on homogeneous clutter content sharing comprises the following steps:

(1) obtaining echo signals and Doppler spectrums thereof, and obtaining a characteristic matrix z of all distance units by taking the kth Doppler channel as the center based on the idea of 3DT algorithm_k＝[z_k,1,...,z_k,i...,z_k,S]^T(ii) a Superscript T denotes transpose operation; s represents the number of distance units; i represents a distance unit number;

(2) feature matrix z of sampling K mean value clustering method_kEach feature vector z in (1)_k,iClustering to obtain corresponding homogeneous clutter areas;

(3) selecting a training sample set corresponding to each Doppler channel according to the clutter areas divided in the step (2); namely, K sets formed by all similar range unit data under the same Doppler channel are used as training sample sets under the Doppler channel, wherein the training sample set corresponding to the kth Doppler channel is c_k,1,...c_k,j,...c_k,K(ii) a K is the total number of categories;

(4) performing adaptive filtering on all training samples corresponding to each Doppler channel in a null-Doppler domain to obtain a filtering output result corresponding to each Doppler channel;

(5) selecting the filtering output result of the ith distance unit under the kth Doppler channel

As a unit to be detected, selecting a reference unit corresponding to the current unit to be detected;

(6) calculating the noise and clutter estimation value Z of the current unit to be detected according to the selected reference unit_k,i；

(7) Setting false alarm probability P_faCalculating the threshold product factor alpha of the current unit to be detected_k,i；

(8) Calculating a detection threshold T of a unit to be detected_k,iAccordingly, the filtering output result of the unit to be detected

And carrying out target detection to obtain a detection target.

Compared with the prior art, the invention has the beneficial effects that:

1. according to the invention, the self-adaptive filtering method combining the traditional 3DT algorithm and the Doppler spectrum-based K-means clustering method is adopted, and compared with the traditional STAP method, the degree of freedom of the system is reduced, so that the requirement on the number of independent and identically distributed training samples is reduced. Meanwhile, different training sample sets are selected according to different Doppler channels, covariance matrixes of various clustered clutters are calculated in sequence, estimation of the clutter covariance matrixes is more accurate, filtering performance of the adaptive filter is improved, the problem that plateau clutters are non-uniform and non-stable is effectively solved, and the suppression effect of the clutters is more remarkable.

2. The method improves the criterion of selecting the reference unit by the traditional CFAR, namely, the similar clutter samples clustered by the Doppler spectrum K-means clustering method are adopted as the reference unit of the current unit to be detected. Under the improvement criterion, even non-adjacent distance units can assist each other, so that the selection of the reference unit is more suitable, the number of the reference units is expanded to a certain extent, and the problem of small samples is effectively relieved. Meanwhile, the clutter statistical characteristics of the reference unit sample and the unit to be detected are kept consistent, so that the problem that the clutter background of the reference unit and the current unit to be detected can not be guaranteed to be similar when the reference unit is screened by the traditional CFAR method is solved, and the detection performance of the radar on the target is improved.

Drawings

The invention is described in further detail below with reference to the figures and specific embodiments.

FIG. 1 is a flow chart of an implementation of the present invention;

FIG. 2 is a graph of the filtering results of clutter data at a certain wave position according to the present invention and the prior art; wherein, (a) is a distance-Doppler image before filtering, (b) is a distance-Doppler image after filtering by a 3DT algorithm, and (c) is a distance-Doppler image after filtering by the method;

fig. 3 is a graph comparing the detection performance at a certain wave position of the present invention and the prior art.

Detailed Description

Embodiments of the present invention will be described in detail below with reference to examples, but it will be understood by those skilled in the art that the following examples are only illustrative of the present invention and should not be construed as limiting the scope of the present invention.

Referring to fig. 1, the present invention provides an improved radar target detection method based on homogeneous clutter content sharing, including the following steps:

(1) obtaining echo signals and Doppler spectrums thereof, and obtaining a characteristic matrix z of all distance units by taking the kth Doppler channel as the center based on the idea of 3DT algorithm_k＝[z_k,1,...,z_k,i...,z_k,S]^T(ii) a Superscript T denotes transpose operation; s represents the total number of distance units; i represents a distance unit number;

based on the idea of 3DT algorithm, selecting wave beam dimensional data of three adjacent Doppler channels of (k-1), (k), and (k + 1) th in clutter Doppler spectrum for all antenna subarrays to obtain a characteristic vector of the ith distance unit with the kth Doppler channel as the center:

z_k,i＝[z_1(k-1),i,z_1k,i,z_1(k+1),i,...,z_r(k-1),i,z_rk,i,z_r(k+1),i,...z_N(k-1),i,z_Nk,i,z_N(k+1),i]^T，

further obtain a characteristic matrix z of all the distance units with the kth Doppler channel as the center_k＝[z_k,1,z_k,2...,z_k,S]^TWherein N represents the total number of antenna array elements, and the value range of a positive integer k is 1-M to represent the total pulse number; z is a radical of_r(k-1),i,z_rk,i,z_r(k+1),iRespectively showing the wave beam dimensional data of three adjacent Doppler channels of k-1, k, k +1 under the ith distance unit of the ith antenna array element.

(2) Feature matrix z of sampling K mean value clustering method_kEach feature vector z in (1)_k,iClustering to obtain corresponding independent clutter areas with the same distribution;

for the characteristic matrix z obtained in the step (1)_kEach feature vector z in (1)_k,iPerforming K-means clustering by taking Euclidean distance as distance measure to obtain a clustering result of each distance unit beam dimension data under each Doppler channel, wherein the category index of each distance unit sample under the kth Doppler channel is l_k,1,...l_k,i,...l_k,SThe number of distance units contained in each class of samples in the K classes is n_k,1,...n_k，j，...n_k,KAnd storing the clustering result.

(3) Selecting a training sample set corresponding to each Doppler channel according to the clutter areas divided in the step (2); namely, K sets formed by all similar range unit data under the same Doppler channel are used as training sample sets under the Doppler channel, wherein the training sample set corresponding to the kth Doppler channel is c_k,1,...c_k，j,...c_k，K；

The number of samples contained in each training sample set is n_k，1，...n_k，j，...n_k，K。

(4) Performing adaptive filtering on all training samples corresponding to each Doppler channel in a null-Doppler domain to obtain a filtering output result corresponding to each Doppler channel;

step 4a) sequentially calculating covariance matrixes of various training sample sets under different Doppler channels according to the result obtained in the step 3, wherein the covariance matrix of the jth sample under the kth Doppler channel is R_k,jThe calculation formula is as follows:

wherein x is_k,jRepresenting the distance unit array element dimensional data of the jth sample under the kth Doppler channel, and the superscript H representing the conjugate transpose; n is_k,jThe number of samples contained in the jth class training sample set is shown;

step 4b) R obtained according to step 4a)_k,jCalculating the adaptive weight vector w of the jth sample under the kth Doppler channel_k,jWherein w is_k,jThe calculation formula of (2) is as follows:

in the formula, s_kRepresenting a target guide vector under a k-th Doppler channel after 3DT dimension reduction;

step 4c) calculating the filtering output result of each distance unit sample under different Doppler channels in sequence, namely vector y_j＝[y_1,j,y_2,j,...,y_M,j]^TWherein y is_k,jThe filtering output result of the jth sample under the kth doppler channel is represented, and the calculation formula is as follows:

(5) selecting the filtering output result of the ith distance unit under the kth Doppler channel

As a unit to be detected, selecting a reference unit corresponding to the current unit to be detected;

step 5a) selecting ith distance unit data under the kth Doppler channel

As to-be-detected unit data, taking a distance unit adjacent to the left side and the right side as a protection unit, and respectively calculating Euclidean distances between two distance unit data on the left side of the protection unit and two distance unit data on the right side of the protection unit and the to-be-detected unit data;

step 5b) recording the category designation l 'of the distance unit closest to the unit to be detected in step 5 a)'_k,i；

Step 5c) according to the clustering result in the step (2), marking all categories under the kth Doppler channel as l'_k,iAs a reference unit of the current unit to be detected, recording the number n 'of the reference units'_k,j。

(6) Calculating the noise and clutter estimation value Z of the current unit to be detected according to the selected reference unit_k,i；

Calculating the mean value of the data of all the reference units of the current unit to be detected according to the result obtained in the step 5c), and taking the mean value as the estimated value of the noise and the clutter of the current unit to be detected. Wherein, the estimated value of noise and clutter when the ith distance unit data under the kth Doppler channel is the unit to be detected is z_k,iThe calculation formula is as follows:

(7) setting false alarm probability P_faCalculating the threshold product factor alpha of the current unit to be detected_k,i；

Setting false alarm probability P_faIncorporating the reference unit number n 'in step 5 b)'_k,iCalculating a threshold product factor alpha of the current unit to be detected_k,iThe calculation formula is as follows:

(8) calculating a detection threshold T of a unit to be detected_k,iAccordingly, the filtering output result of the unit to be detected

And carrying out target detection to obtain a detection target.

Multiplying the threshold by a factor alpha_k,iEstimation Z of noise and clutter from the unit to be detected_k,iMultiplying to obtain a detection threshold T_k,iThe formula is as follows:

T_k,i＝α_k,iZ_k,i

carrying out target detection on the unit data to be detected:

comparing the filtered output results of the units to be detected

Is adaptively filtered to output value y_k,jAnd a detection threshold T_k,iIf the current adaptive filter output value y of the unit to be inspected_k,jGreater than or equal to detection threshold T_k,iAnd judging that the target is detected, otherwise, judging that the target is not detected.

Simulation experiment

The effects of the present invention can be further illustrated by the following specific examples:

1. simulation conditions and contents:

the measured data come from an airborne early warning radar, the distance resolution of the radar is 60m, and the pulse number in a coherent processing interval CPI time is 321. Obtaining clutter data independently and identically distributed in the batch of measured data through clustering, adding artificial targets for detection, wherein the signal-to-noise ratio range of the clutter data and the artificial targets is 0dB to 60dB, the speed range of the targets is-254.717 m/s to 253.722m/s, and the false alarm probability is P_fa＝10^-6。

The experimental contents are as follows:

experiment 1), under the experimental condition, the airborne early warning radar receives the actual measurement data from the Tibet plateau under the background at a certain wave position. The results of the 3DT algorithm plus the conventional CFAR and the clustering-based 3DT algorithm plus the conventional CFAR are respectively adopted, as shown in fig. 2, where fig. 2(a) is a range-doppler diagram before filtering, fig. 2(b) is a range-doppler diagram after filtering by the 3DT algorithm, and fig. 2(c) is a range-doppler diagram after filtering by the present invention. The abscissa and ordinate in fig. 2 are doppler and distance, respectively.

Experiment 2), under the experimental condition, the airborne early warning radar receives the actual measurement data from the Tibet plateau background at a certain wave position, and adds the artificial target into the actual measurement data. And respectively adopting a 3DT algorithm and a traditional CFAR, a 3DT algorithm based on clustering and the traditional CFAR and the method for target detection. The results are shown in FIG. 3, in which the lines with the ` O ` number correspond to the method according to the invention; the line with the '+' sign corresponds to the 3DT algorithm plus the traditional CFAR method; with

The lines of the sign correspond to the 3DT algorithm plus the traditional CFAR method based on clustering. The horizontal and vertical coordinates in FIG. 3 are the signal-to-noise ratio and the detection probability P, respectively_d。

2. And (3) analyzing an experimental result:

referring to fig. 2, it can be seen that the method can effectively filter out non-uniform strong ground clutter caused by topographic relief in the plateau region, and the traditional 3DT algorithm still has strong clutter residues after filtering processing. The experimental background of the invention is a plateau clutter background, and the plateau clutter has the characteristics of high intensity, uneven power, non-stability and the like due to the topographic characteristics of a plateau area, so that the problems of small samples of the plateau clutter are serious, the number of uniform samples is small, and the samples distributed differently are numerous and disordered. In adaptive filtering, the number and uniformity of training samples have a great influence on clutter covariance estimation, and a clutter covariance matrix is an important component of adaptive filter design. The invention utilizes the characteristic that the distribution characteristics of training samples which are independently and identically distributed in the same Doppler channel are identical in a wave beam domain, selects different training sample sets aiming at different Doppler channels and sequentially calculates the covariance matrixes of various clustered clutters, so that the estimation of the clutter covariance matrixes is more accurate, the filtering performance of the adaptive filter is improved, and the problems of few even samples of plateau clutters, numerous samples with different distributions and disorder are effectively solved. The method can effectively improve the clutter suppression performance of the airborne early warning radar system in the plateau environment, and has a better filtering effect on pure clutter data compared with the traditional 3DT algorithm in the non-uniform clutter background.

Referring to fig. 3, it can be seen that the detection performance of the present invention in the range of 15dB to 45dB of the signal-to-noise ratio is better than the 3DT algorithm plus the conventional CFAR method and the 3DT algorithm plus the conventional CFAR method based on clustering. When the detection rate is fifty percent and the false alarm probability is 10^-6In time, the signal-to-noise ratio required by the method is 5dB less than that of a 3DT algorithm and a traditional CFAR method, and the signal-to-noise ratio required by the method is 2.5dB less than that of the 3DT algorithm and the traditional CFAR method based on clustering. The detection method can fully utilize the ground clutter types independently and identically distributed with the unit to be detected, on one hand, the detection performance of the detection method is similar to that of the traditional detection method in the non-uniform clutter environment with high signal-to-noise ratio, and on the other hand, the detection performance of the detection method is superior to that of the traditional detection method in the non-uniform clutter environment with low signal-to-noise ratio.

In conclusion, the invention selects the training sample before the adaptive filtering and selects the reference unit of the unit to be detected during the CFAR detection by the clutter Doppler spectrum-based K-means clustering method, so that non-adjacent distance units can assist each other, the selection of the training sample during the adaptive filtering and the reference unit during the CFAR detection is more suitable, the number of the reference units is expanded to a certain extent, and the problem of small samples is effectively solved. The method improves the clutter suppression performance under the non-uniform background, simultaneously avoids the problem that the clutter background of the reference unit and the current unit to be detected can not be guaranteed to be the same when the reference unit is screened by the traditional CFAR method, and improves the target detection performance under the non-uniform background.

Although the present invention has been described in detail in this specification with reference to specific embodiments and illustrative embodiments, it will be apparent to those skilled in the art that modifications and improvements can be made thereto based on the present invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.

Claims (8)

1. An improved radar target detection method based on homogeneous clutter content sharing is characterized by comprising the following steps:

(1) obtaining echo signals and Doppler spectrums thereof, and obtaining a characteristic matrix z of all distance units by taking the kth Doppler channel as the center based on the idea of 3DT algorithm_k＝[z_k,1,...,z_k,i...,z_k,S]^T(ii) a Superscript T denotes transpose operation; s represents the number of distance units; i represents a distance unit number;

(2) characteristic matrix z is clustered by adopting K mean value_kEach feature vector z in (1)_k,iClustering to obtain corresponding homogeneous clutter areas;

(3) selecting a training sample set corresponding to each Doppler channel according to the clutter areas divided in the step (2); namely, K sets formed by all similar range unit data under the same Doppler channel are used as training sample sets under the Doppler channel, wherein the training sample set corresponding to the kth Doppler channel is c_k,1,...c_k,j,...c_k,K(ii) a K is the total number of categories;

(4) performing adaptive filtering on all training samples corresponding to each Doppler channel in a null-Doppler domain to obtain a filtering output result corresponding to each Doppler channel;

(5) selecting the filtering output result of the ith distance unit under the kth Doppler channel

As a unit to be detected, selecting a reference unit corresponding to the current unit to be detected;

(6) calculating the noise and clutter estimation value Z of the current unit to be detected according to the selected reference unit_k,i(ii) a (7) Setting false alarm probability P_faCalculating the threshold product factor alpha of the current unit to be detected_k,i；

(8) Calculating a detection threshold T of a unit to be detected_k,iAccordingly, the filtering output result of the unit to be detected

And carrying out target detection to obtain a detection target.

2. The method for detecting an improved radar target based on homogeneous clutter content sharing according to claim 1, wherein the obtaining of the feature matrix of all range cells centered on the kth doppler channel based on the idea of 3DT algorithm is specifically:

selecting wave beam dimensional data of three adjacent Doppler channels of (k-1), (k) and (k + 1) th in clutter Doppler spectrum aiming at all antenna array elements to obtain a characteristic vector z of the ith distance unit with the kth Doppler channel as the center_k,i＝[z_1(k-1),i,z_1k,i,z_1(k+1),i,...,z_r(k-1),i,z_rk,i,z_r(k+1),i,...z_N(k-1),i,z_Nk,i,z_N(k+1),i]^TFurther obtain the characteristic matrix z of all the distance units with the kth Doppler channel as the center_k＝[z_k,1,z_k,2...,z_k,S]^T；

Wherein N represents the total number of antenna array elements, the value range of a positive integer k is 1-M, and M represents the total pulse number; z is a radical of_r(k-1),i,z_rk,i,z_r(k+1),iRespectively showing the wave beam dimensional data of three adjacent Doppler channels of k-1, k, k +1 under the ith distance unit of the ith antenna array element.

3. The method for improved radar target detection based on homogeneous clutter content sharing according to claim 1, wherein said feature matrix z is clustered by K-means_kEach feature vector z in (1)_k,iClustering is carried out, specifically: for the characteristic matrix z obtained in the step (1)_kEach feature vector z in (1)_k,iPerforming K-means clustering by taking Euclidean distance as distance measure to obtain the beam dimension of each distance unit under each Doppler channelClustering the data, wherein the class label of each distance unit sample under the k-th Doppler channel is l_k,1,...l_k,i,...l_k，SThe number of distance units contained in each class of samples in the K classes is n_k，1,...n_k,j,...n_k,KAnd storing the clustering result.

4. The method for detecting an improved radar target based on homogeneous clutter content sharing according to claim 1, wherein the adaptive filtering is performed on all training samples corresponding to each doppler channel in a null-doppler domain by the following specific processes:

step 4a), sequentially calculating covariance matrixes of various training sample sets under different Doppler channels according to the result obtained in the step 3, wherein the covariance matrix of the jth sample under the kth Doppler channel is R_k,jThe calculation formula is as follows:

wherein x is_k,jRepresenting the distance unit array element dimensional data of the jth sample under the kth Doppler channel, and the superscript H representing the conjugate transpose; n is_k,jThe number of samples contained in the jth class training sample set is shown;

step 4b) of obtaining R according to step 4a)_k,jCalculating the adaptive weight vector w of the jth sample under the kth Doppler channel_k,jWherein w is_k,jThe calculation formula of (2) is as follows:

in the formula, s_kRepresenting a target guide vector under a k-th Doppler channel after dimension reduction by 3 DT;

step 4c), calculating the filtering output result of each distance unit sample under different Doppler channels in sequence, namely vector y_j＝[y_1,j,y_2,j,...,y_M,j]^TWherein y is_k,jThe filtering output result of the jth sample under the kth doppler channel is represented, and the calculation formula is as follows:

5. the method for improved radar target detection based on homogeneous clutter content sharing according to claim 1, wherein the step (5) further comprises the steps of:

step 5a), taking one adjacent distance unit on the left side and the right side of the unit to be detected as a protection unit, and respectively calculating Euclidean distances between two distance units on the left side of the protection unit and two distance units on the right side of the protection unit and the data of the unit to be detected;

step 5b), recording the category label l 'of the distance unit closest to the unit to be detected in step 5 a)'_k,i；

Step 5c), according to the clustering result in the step (2), marking all categories under the kth Doppler channel as l'_k,iAs a reference unit of the current unit to be detected, recording the number n 'of the reference units'_k,j。

6. The method of claim 5 wherein the noise and clutter estimates z for the units to be detected are based on the homogeneous clutter content sharing_k,iThe calculation formula of (2) is as follows:

wherein, y_k,jAnd the filtering output result of the j-th sample under the k-th Doppler channel is shown.

7. The method for improved radar target detection based on homogeneous clutter content sharing according to claim 1, wherein the calculation formula of the threshold product factor of the current unit to be detected is:

wherein, n'_k,jIs the number of reference cells.

8. The method for improved radar target detection based on homogeneous clutter content sharing according to claim 1, wherein step (8) comprises the sub-steps of:

step 8a), multiplying the threshold by a factor alpha_k,iEstimation Z of noise and clutter from the unit to be detected_k,iMultiplying to obtain a detection threshold T_k,i：

T_k,i＝α_k,iZ_k,i

Step 8b), comparing the data of the unit to be detected

Is adaptively filtered to output value y_k,jAnd a detection threshold T_k,iIf the current adaptive filter output value y of the unit to be inspected_k,jGreater than or equal to detection threshold T_k,iAnd judging that the target is detected, otherwise, judging that the target is not detected.

Images