Feature-assisted external radiation source radar target tracking method and system in non-uniform clutter environment

Abstract

The invention discloses a feature-assisted external radiation source radar target tracking method in a non-uniform clutter environment. The method comprises the steps of obtaining position measurement, Doppler frequency measurement and RCS measurement in a Cartesian coordinate system; constructing an RCS and Doppler frequency mathematical model of the target; predicting a target by using multi-model PHD to obtain target Gaussian components predicted by a plurality of models; using the converted measurement to update all the predicted Gaussian components; missing alarm detection is carried out on the Gaussian component obtained after prediction updating, and if it is judged that missing alarm occurs, missing alarm correction is carried out; and cutting and combining all Gaussian components, extracting a multi-target state for track association, and inputting the multi-target state into a filter at the next moment. According to the method, the multi-target tracking robustness of the external radiation radar in the non-uniform clutter environment is improved, and the detection tracking performance of the external radiation source radar in the complex electromagnetic environment is improved.

Description

Technical Field

[0001] The present invention belongs to the field of radar tracking, relates to an exo-radiation source radar target tracking technology, and specifically relates to a feature-assisted exo-radiation source radar target tracking method and system in a non-uniform clutter environment. Background Art

[0002] In the current field of radar tracking, multi-target tracking faces numerous challenges. The presence of clutter in the detection environment poses numerous challenges to establishing the corresponding relationship between measurements and targets. Existing multi-target tracking methods are primarily categorized into two types: those based on data association and those based on random finite sets (RFS). Data association multi-target algorithms decompose the multi-target tracking problem into multiple single-target problems. When the number of measurements is small and the tracking environment is simple, they can effectively ensure real-time performance and accuracy. However, in complex scenarios with high clutter and large numbers of targets, the algorithm complexity increases exponentially. RFS-based multi-target tracking algorithms, such as Probability Hypothesis Density (PHD) and Cardinalized Probability Hypothesis Density (CPHD), are well-suited to multi-clutter environments. These algorithms, such as those based on Probability Hypothesis Density (PHD) and Cardinalized Probability Hypothesis Density (CPHD), do not require data association and can directly estimate the state and number of multiple targets.

[0003] The above algorithm has good performance in a uniform clutter environment, but it is easy to misjudge clutter as a real target in a non-uniform clutter environment, resulting in an increase in false targets. To address the problem of non-uniform clutter, the existing technology proposes a clutter space density estimation method based on a finite mixture model (FMM). Although this method can adaptively estimate the clutter intensity and distribution in space, it is only applicable to time-invariant clutter, and the calculation process is complicated. In the "Low-altitude Multi-target Tracking Algorithm in Non-uniform Clutter Environment" published in 2025, the DBSCAN clustering algorithm was used to homogenize non-uniform clutter. However, the algorithm is complex to calculate and requires manual setting of parameters such as the neighborhood radius and the minimum number of points. Improper settings will seriously affect performance. It can be seen that the existing technology still has shortcomings in dealing with non-uniform clutter problems, and a more effective solution needs to be proposed.

[0004] In a non-uniform clutter environment, radar interference can lead to missed detections. The lack of true measurements from the target further complicates the tracking task. On the one hand, the algorithm may mistakenly identify clutter as target measurements, leading to misidentification of the target or biased state estimation. On the other hand, if true measurements are missing for a long time, established target tracks may be interrupted, lost, or require reinitialization due to a lack of necessary measurement data support. This severely impacts the continuity and accuracy of multi-target tracking, posing a significant challenge to the stable tracking of high-speed maneuvering targets. Summary of the Invention

[0005] Purpose of the invention: In order to overcome the shortcomings of the existing technology, a feature-assisted external radiation source radar target tracking method and system are provided in a non-uniform clutter environment, which solves the problems that multiple targets are difficult to track, easy to miss, and may maneuver at any time in a non-uniform clutter environment.

[0006] Technical Solution: To achieve the above objectives, the present invention provides a feature-assisted external emitter radar target tracking method in a non-uniform clutter environment, comprising the following steps:

[0007] S1: Acquire raw measurement data at multiple moments, convert the raw measurement data at each moment in turn using the dual-base target tracking method to obtain position measurement, Doppler frequency measurement, and RCS measurement in the Cartesian coordinate system;

[0008] S2: Construct a mathematical model of the target's RCS and Doppler frequency;

[0009] S3: Use multi-model PHD to predict the target and obtain the target Gaussian components predicted by multiple models; use the converted measurement to update all the predicted Gaussian components;

[0010] S4: Perform missed alarm detection on the Gaussian component obtained after the prediction update, and perform missed alarm correction if it is determined that a missed alarm occurs;

[0011] S5: Cut and merge all Gaussian components, extract multi-target states for track association, and input the multi-target states into the filter at the next moment.

[0012] Furthermore, the position measurement, Doppler frequency measurement, and RCS measurement in the Cartesian coordinate system in step S1 are expressed as follows:

[0013]

[0014] Among them, z k is the position measurement of the target at time k, z D,k and z RCS,k They are the Doppler measurement and RCS measurement of the target respectively.

[0015] Furthermore, the construction of the target RCS and Doppler frequency mathematical model in step S2 includes:

[0016] A1: Assuming that the target's Doppler information, RCS, and position are independent of each other, the likelihood function of the target and the likelihood function of the clutter are expressed as follows:

[0017]

[0018] Among them, g z (z|x), g D (D) and g RCS (RCS) are the target’s position, Doppler frequency and RCS likelihood, respectively, c z (z|x), c D (D) and c RCS (RCS) are the likelihood functions of clutter with respect to position, Doppler frequency, and RCS, respectively;

[0019] A2: The likelihood function of Doppler frequency is expressed as follows:

[0020] g D (D) = N(z D,k ,h k (x k ),R D,k )

[0021]

[0022] Among them, R D,k is the noise variance;

[0023] A3: RCS is modeled using the chi-square distribution, and its likelihood function is expressed as follows:

[0024]

[0025] Where Γ(m) is the GAMMA function, m is the degree of freedom, and RCS av is the average value of the target RCS.

[0026] Furthermore, the step S3 specifically includes:

[0027] B1: For the filter implemented by GM, the posterior strength at a given moment is:

[0028]

[0029] Prediction strength v k|k-1 (x k ,r k ) is represented as follows:

[0030] v k|k-1 (xk ,r k )=v f,k|k-1 (x k ,r k )+v γ,k (x k ,r k ),

[0031] Among them, v f,k|k-1 (x k ,r k ) represents the survival intensity and is expressed as follows:

[0032]

[0033] in:

[0034]

[0035] Among them, p s,k|k-1 (r k-1 )=p s,k-1 (r k-1 ) indicates that in model r k-1 The probability of target survival, t k|k-1 (r k |r k-1 ) represents the target motion model from r k-1 Switch to r k The probability of In the model r k-1 The weight of the next j-th component;

[0036] B2: Given a GM form, predict the strength v k|k-1 (x,r k ) and defined as follows:

[0037]

[0038] Given a measurement set Z k , then the posterior strength is expressed as:

[0039]

[0040] in:

[0041]

[0042] Among them, λ c c(z) represents the clutter intensity, c RCS (RCS) and c D (D) represents the RCS and Doppler frequency likelihood function of the clutter, Indicates that k The measured likelihood function of the jth component under the model, and Respectively expressed in r k Measurement under model The mean and covariance of the j-th Gaussian component after the Kalman filter update.

[0043] Furthermore, the step S4 specifically includes:

[0044] C1: Perform missed detection on the updated Gaussian component:

[0045] In order to distinguish different targets, the label τ is introduced. is the set of all labels of the target output at time k-1, and all components updated at time k are expressed as For all Find the Gaussian component with the same label as the lth track in all motion models at time k:

[0046]

[0047] in, is the number of Gaussian components in all models with the same label, is the weight of the Gaussian component with the same labels at time k;

[0048] If at the current moment there is And the tag track has output at least N at time k-1 min points, it means that the label at time k-1 is The target may be lost after the measurement update at time k, where ωTH is the missed alarm threshold;

[0049] C2: Perform missed alarm correction for detected missed targets:

[0050] After detecting the occurrence of missed alarms, the following formula is used to correct the missed alarms:

[0051]

[0052] in, η<1 is the weight penalty factor; and are the Cartesian velocities of the missed target at time k-1 and k-2, respectively, and |Δv| is the absolute value of the velocity change at the first two moments;

[0053] The target missed alarm is corrected by the target prediction component, and the target prediction component is added to all models. In the subsequent Gaussian component pruning, merging and state extraction, when the component is larger than the extraction threshold, it can be correctly extracted, thereby reducing the impact of target false alarm on tracking.

[0054] Furthermore, the shearing and merging of all Gaussian components in step S5 specifically includes:

[0055] Assume that at time k, there are two Gaussian components Ω(x Ω ,r),Φ(x Φ ,r), which is defined as follows:

[0056]

[0057] Prune the mixed components; prune and remove the components whose weights are less than the pruning threshold in the mixed components to avoid the explosive growth of subsequent components and the sudden increase in computational complexity;

[0058] Merge the trimmed mixture components.

[0059] Furthermore, in step S5, the mixed components are merged using a merging criterion, which is defined as follows:

[0060] DisGM(Ω,Φ)<ThGM

[0061] in, is the weight of the smaller component being merged, σ s To measure noise.

[0062] Based on the method of the present invention, the present invention also provides a feature-assisted external emitter radar target tracking system in a non-uniform clutter environment, comprising:

[0063] The measurement acquisition module is used to obtain position measurement, Doppler frequency measurement and RCS measurement in the Cartesian coordinate system;

[0064] Mathematical model building module, used to build the RCS and Doppler frequency mathematical model of the target;

[0065] Gaussian component acquisition module, used to predict the target using multi-model PHD, and obtain the target Gaussian components after multiple models predict;

[0066] A Gaussian component update module for updating all predicted Gaussian components using the transformed measurements;

[0067] The missed alarm detection and correction module is used to perform missed alarm detection on the Gaussian component obtained after the prediction update, and perform missed alarm correction if it is determined that a missed alarm has occurred;

[0068] The shearing and merging module is used to shear and merge all Gaussian components;

[0069] Track association module is used to extract multi-target states for track association.

[0070] This invention addresses the difficulty of tracking multiple targets in non-uniform clutter environments using traditional multi-target tracking methods for exo-radiation radars, which can easily lead to false targets. Instead, it proposes a feature-matching multiple-model PHD filter (FM-MM-GM-PHD) that assists in tracking targets by matching the measured radar cross-section (RCS) and Doppler frequency characteristics of the target. This invention uses RCS and Doppler frequency characteristics to suppress the effects of non-uniform clutter, introduces multiple models to address target maneuvers, and designs a missed alarm correction mechanism to mitigate the negative impact of radar missed alarms.

[0071] Beneficial Effects: Compared with the prior art, the present invention improves the robustness of external radiation radar in tracking multiple targets in a non-uniform clutter environment and improves the detection and tracking performance of external radiation radar in a complex electromagnetic environment. Specifically, the present invention has the following advantages:

[0072] 1. Using RCS and Doppler feature assistance combined with a multi-model approach can help improve tracking accuracy through feature matching in non-uniform strong clutter environments, suppress the algorithm performance degradation caused by non-uniform clutter, and improve the ability to cope with multi-target maneuvers.

[0073] 2. For non-uniform clutter environments, the Gaussian component pruning and merging process has been improved to reduce the possibility of incorrect merging.

[0074] 3. To address the problem of missed detection of targets, a missed detection correction mechanism is designed. According to the changes in the target's motion state in the previous two frames, the estimated state of the target at the current moment is adaptively corrected, reducing the impact of missed detection on tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 : Flowchart of the method of the present invention;

[0076] Figure 2 : Target trajectory diagram in Example 3;

[0077] Figure 3 : Comparison diagram of target number estimation for multi-target tracking under the first detection probability parameter in Example 3;

[0078] Figure 4 : OSPA comparison diagram of multi-target tracking under the first detection probability parameter in Example 3;

[0079] Figure 5 : Comparison diagram of target number estimation for multi-target tracking under the second detection probability parameter in Example 3.

[0080] Figure 6 : OSPA comparison diagram of multi-target tracking under the second detection probability parameter in Example 3. DETAILED DESCRIPTION

[0081] The present invention is further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, modifications of various equivalent forms of the present invention made by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0082] Example 1:

[0083] like Figure 1 As shown, this embodiment provides a feature-assisted external emitter radar target tracking method in a non-uniform clutter environment, comprising the following steps:

[0084] S1: Acquire raw measurement data at multiple moments, convert the raw measurement data at each moment in turn using the dual-base target tracking method to obtain position measurement, Doppler frequency measurement, and RCS measurement in the Cartesian coordinate system;

[0085] The expressions of position measurement, Doppler frequency measurement, and RCS measurement in the Cartesian coordinate system are as follows:

[0086]

[0087] Among them, z k is the position measurement of the target at time k, z D,k and z RCS,k They are the Doppler measurement and RCS measurement of the target respectively.

[0088] S2: Constructing a mathematical model of the target's RCS and Doppler frequency, including the following steps:

[0089] A1: Assuming that the target's Doppler information, RCS, and position are independent of each other, the likelihood function of the target and the likelihood function of the clutter are expressed as follows:

[0090]

[0091]

[0092] Among them, g z (z|x), g D (D) and g RCS (RCS) are the target’s position, Doppler frequency and RCS likelihood, respectively, c z (z|x), c D (D) and c RCS(RCS) are the likelihood functions of clutter with respect to position, Doppler frequency, and RCS, respectively;

[0093] A2: The likelihood function of Doppler frequency is expressed as follows:

[0094] g D (D) = N(z D,k ,h k (x k ),R D,k )

[0095]

[0096] Among them, R D,k is the noise variance;

[0097] A3: RCS is modeled using the chi-square distribution, and its likelihood function is expressed as follows:

[0098]

[0099] Where Γ(m) is the GAMMA function, m is the degree of freedom, and RCS av is the average value of the target RCS.

[0100] S3: Use multi-model PHD to predict the target and obtain multiple models (multiple motion models, such as turning, uniform speed, uniform acceleration, etc., r in the formula k The target Gaussian component after prediction by the model is updated using the transformed measurement.

[0101] Step S3 specifically includes:

[0102] B1: For the filter implemented by GM, the posterior strength at a given moment is:

[0103]

[0104] Prediction strength v k|k-1 (x k ,r k ) is represented as follows:

[0105] v k|k-1 (x k ,r k )=v f,k|k-1 (x k ,r k )+v γ,k (x k ,r k ),

[0106] Among them, v f,k|k-1 (x k ,rk ) represents the survival intensity and is expressed as follows:

[0107]

[0108] in:

[0109]

[0110]

[0111] Among them, p s,k|k-1 (r k-1 )=p s,k-1 (r k-1 ) indicates that in model r k-1 The probability of target survival, t k|k-1 (r k |r k-1 ) represents the target motion model from r k-1 Switch to r k The probability of In the model r k-1 The weight of the next j-th component;

[0112] B2: Given a GM form, predict the strength v k|k-1 (x,r k ) and defined as follows:

[0113]

[0114] Given a measurement set Z k , then the posterior strength is expressed as:

[0115]

[0116] in:

[0117]

[0118] Among them, λ c c(z) represents the clutter intensity, c RCS (RCS) and c D (D) represents the RCS and Doppler frequency likelihood function of the clutter, Indicates that k The measured likelihood function of the jth component under the model, and Respectively expressed in r k Measurement under model The mean and covariance of the j-th Gaussian component after the Kalman filter update.

[0119] The negative impact of non-uniform clutter primarily stems from the fact that dense clutter may appear around actual targets, misidentifying them as targets and creating false targets. This invention incorporates RCS and Doppler frequency characteristics into the measurements to distinguish between clutter and target measurements. Building on the model established in step S2, the effects of non-uniform clutter are suppressed using RCS and Doppler frequency characteristics, specifically as described above.

[0120] S4: Perform missed alarm detection on the Gaussian component obtained after the prediction update, and perform missed alarm correction if it is determined that a missed alarm occurs;

[0121] Step S4 specifically includes:

[0122] C1: Perform missed detection on the updated Gaussian component:

[0123] In order to distinguish different targets, the label τ is introduced. is the set of all labels of the target output at time k-1, and all components updated at time k are expressed as For all Find the Gaussian component with the same label as the lth track in all motion models at time k:

[0124]

[0125] in, is the number of Gaussian components in all models with the same label, is the weight of the Gaussian component with the same labels at time k;

[0126] If at the current moment there is And the tag track has output at least N at time k-1 min points, it means that the label at time k-1 is The target may be lost after the measurement update at time k, where ωTH is the missed alarm threshold;

[0127] C2: Perform missed alarm correction for detected missed targets:

[0128] After detecting the occurrence of missed alarms, the following formula is used to correct the missed alarms:

[0129]

[0130] in, η<1 is the weight penalty factor; and are the Cartesian velocities of the missed target at time k-1 and k-2, respectively, and |Δv| is the absolute value of the velocity change at the first two moments;

[0131] The target missed alarm is corrected by the target prediction component, and the target prediction component is added to all models. In the subsequent Gaussian component pruning, merging and state extraction, when the component is larger than the extraction threshold, it can be correctly extracted, thereby reducing the impact of target false alarm on tracking.

[0132] S5: Cut and merge all Gaussian components, extract multi-target states for track association, and input the multi-target states into the filter at the next moment.

[0133] The shearing and merging of all Gaussian components in step S5 specifically includes:

[0134] Assume that at time k, there are two Gaussian components Ω(x Ω ,r),Φ(x Φ ,r), which is defined as follows:

[0135]

[0136] Prune the mixed components; prune and remove the components whose weights are less than the pruning threshold in the mixed components to avoid the explosive growth of subsequent components and the sudden increase in computational complexity;

[0137] If these two components can be merged, then the merging criteria must be met; the merging of mixed components is performed through the merging criteria, which is defined as follows:

[0138] DisGM(Ω,Φ) <ThGM

[0139] in, is the weight of the smaller component being merged, σ s To measure noise.

[0140] Example 2:

[0141] Based on the method of Example 1, this embodiment provides a feature-assisted external emitter radar target tracking system in a non-uniform clutter environment, including:

[0142] The measurement acquisition module is used to obtain position measurement, Doppler frequency measurement and RCS measurement in the Cartesian coordinate system;

[0143] Mathematical model building module, used to build the RCS and Doppler frequency mathematical model of the target;

[0144] Gaussian component acquisition module, used to predict the target using multi-model PHD, and obtain the target Gaussian components after multiple models predict;

[0145] A Gaussian component update module for updating all predicted Gaussian components using the transformed measurements;

[0146] The missed alarm detection and correction module is used to perform missed alarm detection on the Gaussian component obtained after the prediction update, and perform missed alarm correction if it is determined that a missed alarm has occurred;

[0147] The shearing and merging module is used to shear and merge all Gaussian components;

[0148] Track association module is used to extract multi-target states for track association.

[0149] Example 3:

[0150] In order to verify the effect of the present invention, this embodiment is verified by the following simulation experiments, which are as follows:

[0151] In the simulation, the present invention establishes a target positioning database by analyzing the characteristics of the simulated external radiation source bistatic radar system and sets the sampling interval T = 1s. The specific simulation parameters are shown in Table 1.

[0152] Table 1 Simulation parameters of the embodiment

[0153]

[0154] In the experiment, four trajectories were simulated at the same time, and the appearance and disappearance time of the four targets were different. Figure 2 As shown in Table 2, this example uses the Optimal Subpattern Assignment (OSPA) and Cardinality Estimation metrics to measure the accuracy and robustness of the proposed method. The simulation also generates target tracks under different detection probabilities. The target track parameters are shown in Table 2.

[0155] Table 2 Target parameter settings

[0156]

[0157] This example compares GM-PHD, MM-GM-PHD, and the proposed FM-MM-GM-PHD in a non-uniform clutter environment with different detection probabilities. The simulation uses the Monte Carlo method with 50 repetitions. The target motion model is shown below:

[0158]

[0159] The following two simulation experiments are used to illustrate:

[0160] Simulation experiment 1: The number of uniform clutter per frame is set to 100 and the detection probability is 0.88; Simulation experiment 2: The number of uniform clutter per frame is still 100 and the detection probability is 0.78. In both simulation experiments, there is a Gaussian distributed dense clutter on each target track. The OSPA of simulation experiment 1 is Figure 4 Show, the cardinality is estimated by Figure 3 Presentation; OSPA of simulation experiment 2 passed Figure 6 Reflection, the base estimate corresponds to Figure 5 The simulation results of simulation experiments 1 and 2 are shown in Table 3 and Table 4 respectively.

[0161] Table 3 Simulation results of different methods in simulation experiment 1

[0162]

[0163] Table 4 Simulation results of different methods in simulation experiment 2

[0164]

[0165] pass Figures 3 to 6 As can be seen from the data in Tables 1 and 2:

[0166] In simulation experiment 1, the proposed FM-MM-GM-PHD algorithm performed well in both target state estimation and population estimation. Its OSPA distance was significantly smaller than that of conventional GM-PHD and MM-GM-PHD, and its population estimation results were closer to the true value. However, due to its multi-model approach, the MM-GM-PHD algorithm was more susceptible to non-uniform clutter, resulting in a significant overestimation of population estimates. The average OSPA distance was comparable to that of conventional GM-PHD.

[0167] In the low detection probability scenario of Simulation Experiment 2, the target estimation accuracy of the conventional GM-PHD and MM-GM-PHD algorithms was significantly affected, and performance declined significantly. In sharp contrast, the FM-MM-GM-PHD algorithm of our invention reduces the impact of low detection probability on tracking performance through a missed detection mechanism, maintaining good performance in target state estimation and number estimation. The OSPA distance is much lower than that of other methods, and the number estimation is also the closest to the true value.

[0168] In summary, the proposed FM-MM-GM-PHD algorithm can effectively assist in tracking multiple maneuvering targets in non-uniform clutter environments by leveraging measurement features, and can effectively address missed detections. Experiments have fully demonstrated the algorithm's excellent robustness.

Claims

1. A feature-assisted external emitter radar target tracking method in a non-uniform clutter environment, characterized in that: The steps include: S1: Acquire raw measurement data at multiple moments, convert the raw measurement data at each moment in turn using the dual-base target tracking method to obtain position measurement, Doppler frequency measurement, and RCS measurement in the Cartesian coordinate system; S2: Construct a mathematical model of the target's RCS and Doppler frequency; S3: Use multi-model PHD to predict the target and obtain the target Gaussian components predicted by multiple models; use the converted measurement to update all the predicted Gaussian components; S4: Perform missed alarm detection on the Gaussian component obtained after the prediction update, and make corrections if missed alarms are detected; S5: Cut and merge all Gaussian components, extract multi-target states for track association, and input the multi-target states into the filter at the next moment.

2. The feature-assisted external emitter radar target tracking method in a non-uniform clutter environment according to claim 1, characterized in that: The position measurement, Doppler frequency measurement and RCS measurement in the Cartesian coordinate system in step S1 are expressed as follows: Among them, z k is the position measurement of the target at time k, z D,k and z RCS,k They are the Doppler measurement and RCS measurement of the target respectively.

3. The feature-assisted external emitter radar target tracking method in a non-uniform clutter environment according to claim 2, characterized in that: The construction of the target RCS and Doppler frequency mathematical model in step S2 includes: A1: Assuming that the target's Doppler information, RCS, and position are independent of each other, the likelihood function of the target and the likelihood function of the clutter are expressed as follows: Among them, g z (z|x), g D (D) and g RCS (RCS) are the target’s position, Doppler frequency and RCS likelihood, respectively, c z (z|x), c D (D) and c RCS (RCS) are the likelihood functions of clutter with respect to position, Doppler frequency, and RCS, respectively; A2: The likelihood function of Doppler frequency is expressed as follows: g D (D)=N(z D,k ,h k (x k ),R D,k ) Among them, R D,k is the noise variance; A3: RCS is modeled using the chi-square distribution, and its likelihood function is expressed as follows: Where Γ(m) is the GAMMA function, m is the degree of freedom, and RCS av is the average value of the target RCS.

4. The feature-assisted external emitter radar target tracking method in a non-uniform clutter environment according to claim 3, characterized in that: The step S3 specifically includes: B1: For the filter implemented by GM, the posterior strength at a given moment is: Prediction strength v k|k-1 (x k ,r k ) is represented as follows: v k|k-1 (x k ,r k )=v f,k|k-1 (x k ,r k )+v γ,k (x k ,r k ), Among them, v f,k|k-1 (x k ,r k ) represents the survival intensity and is expressed as follows: in: Among them, p s,k|k-1 (r k-1 )=p s,k-1 (r k-1 ) indicates that in model r k-1 The probability of target survival, t k|k-1 (r k |r k-1 ) represents the target motion model from r k-1 Switch to r k The probability of In the model r k-1 The weight of the next j-th component; B2: Given a GM form, predict the strength v k|k-1 (x,r k ) and defined as follows: Given a measurement set Z k , then the posterior strength is expressed as: in: Among them, λ c c(z) represents the clutter intensity, c RCS (RCS) and c D (D) represents the RCS and Doppler frequency likelihood function of the clutter, Indicates that k The measured likelihood function of the jth component under the model, and Respectively expressed in r k Measurement under model The mean and covariance of the j-th Gaussian component after the Kalman filter update.

5. The feature-assisted external emitter radar target tracking method in a non-uniform clutter environment according to claim 4, characterized in that: The step S4 specifically includes: C1: Perform missed detection on the updated Gaussian component: In order to distinguish different targets, the label τ is introduced. is the set of all labels of the target output at time k-1, and all components updated at time k are expressed as For all Find the Gaussian component with the same label as the lth track in all motion models at time k: in, is the number of Gaussian components in all models with the same label, is the weight of the Gaussian component with the same labels at time k; If at the current moment there is And the tag track has output at least N at time k-1 min points, it means that the label at time k-1 is The target may be lost after the measurement update at time k, where ωTH is the missed alarm threshold; C2: Perform missed alarm correction for detected missed targets: After detecting the occurrence of missed alarms, the following formula is used to correct the missed alarms: in, η<1 is the weight penalty factor; and are the Cartesian velocities of the missed target at moments k-1 and k-2, respectively, and |Δv| is the absolute value of the velocity change at the first two moments.

6. The feature-assisted external emitter radar target tracking method in a non-uniform clutter environment according to claim 5, characterized in that: The shearing and merging of all Gaussian components in step S5 specifically includes: Assume that at time k, there are two Gaussian components Ω(x Ω ,r),Φ(x Φ ,r), which is defined as follows: Prune the mixed components; prune and remove the components whose weights are less than the pruning threshold in the mixed components; Merge the trimmed mixture components.

7. The feature-assisted external emitter radar target tracking method in a non-uniform clutter environment according to claim 6, characterized in that: In step S5, the mixed components are merged using a merging criterion, which is defined as follows: DisGM(Ω,Φ) <ThGM in, is the weight of the smaller component being merged, σ s To measure noise.

8. A feature-assisted external emitter radar target tracking system in a non-uniform clutter environment, characterized in that: include: The measurement acquisition module is used to obtain position measurement, Doppler frequency measurement and RCS measurement in the Cartesian coordinate system; Mathematical model building module, used to build the RCS and Doppler frequency mathematical model of the target; Gaussian component acquisition module, used to predict the target using multi-model PHD, and obtain the target Gaussian components after multiple models predict; A Gaussian component update module for updating all predicted Gaussian components using the transformed measurements; The missed alarm detection and correction module is used to perform missed alarm detection on the Gaussian component obtained after the prediction update, and perform missed alarm correction if it is determined that a missed alarm has occurred; The shearing and merging module is used to shear and merge all Gaussian components; Track association module is used to extract multi-target states for track association.

Images