Single target detection and DOA tracking method and system based on Bernoulli filter

Abstract

The invention discloses a Bernoulli filter-based single target detection and DOA tracking method and system, and relates to the field of target tracking. The method solves the problems that a traditional tracking algorithm after detection cannot jointly detect the target when the underwater target repeatedly enters under the conditions of low signal-to-noise ratio and small snapshot, and the tracking effect is unstable. The method comprises the steps of constructing a Bernoulli random finite set model, defining a single target state as a dynamic description target state or a dynamic description target state, predicting the target existence probability by using a Bernoulli Markov process, and generating a prediction state distribution particle set composed of new particles and survival particles through particle filtering. A hydrophone array signal is decomposed into a noise subspace observation model and a signal subspace observation model, a generalized likelihood function after exponential weighting is used for updating a target posteriori existence probability and a particle weight, and an equal-weight particle set is generated through resampling. And judging whether the target exists or not according to the single-target posterior existence probability, thereby realizing single-target DOA joint detection and tracking.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of underwater acoustic signal processing and target tracking, and particularly relates to a single target detection and DOA tracking method and system based on Bernoulli filter. BACKGROUND

[0002] When using the traditional detection-after-tracking (TAD-Ber) method to track the azimuth of an underwater target, a report set is first generated through spatial spectrum analysis, and then input into a tracking filter to obtain the azimuth tracking result. However, the traditional detection-after-tracking method has the following problems: when the target repeatedly enters or leaves the monitoring area, or under the condition of low signal-to-noise ratio and small snapshot number, the false alarm rate increases sharply, the report set is cluttered, the miss detection rate is high, the tracking filter fails due to unreliable input, and the detection threshold is difficult to accurately set and is easily disturbed by environmental noise. Moreover, in actual situations, considering that the parameters such as sound source signal and environmental noise power are unknown, blindly using no-information priori to marginalize unknown parameters will lead to detection failure. Therefore, a robust and efficient single target joint detection and azimuth tracking method is needed to solve the problems raised in the background. SUMMARY

[0003] The present application is to solve the problem that the detection-after-tracking (TAD-Ber) algorithm in the prior art cannot stably track the existence of a target under the condition of low signal-to-noise ratio and small snapshot number when the underwater target repeatedly enters or leaves the monitoring area and the priori information is missing.

[0004] To solve the above technical problems, the present application is implemented by the following technical scheme: Scheme one, the present application proposes a single target detection and DOA tracking method based on Bernoulli filter, which comprises the following steps: Step 1, a Bernoulli random finite set model is constructed, the state of a single target is defined as existing or not existing to dynamically describe the target state, the Bernoulli Markov process is used to predict the target existence probability, and the particle filter is used to generate a predicted state distribution particle set composed of new-born particles and surviving particles, and the single target predicted state distribution particle set is constructed after the combination of the two; Step 2, the hydrophone array signal is decomposed into a noise subspace and a signal subspace observation model, the generalized likelihood function with a penalty term is calculated using the information theory criterion, which is used to compensate for the uncertainty of unknown parameters, and an exponential weighting factor is introduced to focus on the high likelihood area to enhance the sharpness of the main lobe of the generalized likelihood function under extreme conditions; Step 3, the target posterior existence probability and the single-target prediction state distribution particle set weight are updated by using the exponential weighted generalized likelihood function through the Bayesian updating equation, and an equal-weight single-target prediction state distribution particle set is generated by resampling; a number of equal-weight particles are obtained by resampling the updated single-target prediction state distribution particle set, the target existence condition is determined by using the single-target posterior existence probability, and the DOA estimation result is calculated by using the particle state, so that the single-target DOA joint detection and tracking are realized.

[0005] Further, a preferred embodiment is provided, and in step 1, the Bernoulli random finite set model is constructed, the single-target state is defined as the existence or nonexistence of a dynamic description of the target state, and the method for predicting the target existence probability by using the Bernoulli Markov process is as follows:

[0006] In the formula, represents there is no target at the moment, and only background noise is observed; represents that the observation is the superposition of the target signal and the background noise; represents that the probability of the number of targets being greater than or equal to 2 is equal to 0; represents the hydrophone array time domain output at the moment, is the predicted single-target existence probability, is the predicted single-target state distribution.

[0007] Further, a preferred embodiment is provided, and in step 2, the method for calculating the generalized likelihood function with a penalty term by using the information theory criterion is as follows: the information theory criterion includes the MDL criterion and the AIC criterion, Generalized likelihood function based on MDL criterion is

[0008] Generalized likelihood function based on AIC criterion is

[0009] In the formula, is the number of hydrophone elements.

[0010] Further, a preferred embodiment is provided, and in step 2, the method for introducing an exponential weighting factor to focus on the high likelihood region to enhance the sharpness of the main lobe of the generalized likelihood function under extreme conditions is as follows: .

[0011] Furthermore, a preferred embodiment is provided, wherein the method in step 3 for updating the target posterior existence probability and the weights of the single-target predicted state distribution particle set using the Bayesian update equation and the exponentially weighted generalized likelihood function is as follows: The update equation based on Bayes' theorem is:

[0012] In the formula This represents the likelihood function, i.e., the state in which the likelihood is expressed. The following observations The probability density, and

[0013] Due to single-objective state It can be represented by Bernoulli random finite set variables, and the single-objective posterior state particle set is updated according to the update equation. Among them, the probability of the existence of a single target posterior. for

[0014] In the formula

[0015] The weight update expression for the target posterior state distribution is as follows:

[0016] The generalized likelihood function is the exponentially weighted generalized likelihood function based on information theory criteria.

[0017] Furthermore, a preferred embodiment is provided, wherein in step 3, the presence of a target is determined by utilizing the posterior probability of a single target and the DOA estimation result is calculated by calculating the particle state, thereby realizing the method of joint detection and tracking of single target DOA: Resampling from a single-objective posterior state particle set A set of particles with equal weights Assume the posterior probability of the target exists satisfies If the target is determined to exist, it is determined to not exist. The single target's azimuth estimate is a particle state-weighted average. , .

[0018] Option 2: A single-target detection and DOA tracking system based on a Bernoulli filter, the system comprising: A single-target predicted state distribution particle set module is configured to construct a Bernoulli random finite set model, define a single-target state as presence or absence of a dynamic description of a target state, predict a target presence probability by using a Bernoulli Markov process, and generate a predicted state distribution particle set composed of newborn particles and surviving particles by particle filtering, and the predicted state distribution particle set is constructed by combining the newborn particles and the surviving particles. An edge module is configured to decompose a hydrophone array signal into a noise subspace and a signal subspace observation model, calculate a generalized likelihood function with a penalty term by using an information theory criterion, compensate for uncertainty of unknown parameters, and introduce an exponential weighting factor to focus on a high likelihood region to enhance sharpness of a main lobe of the generalized likelihood function under extreme conditions. A single-target DOA joint detection and tracking module is configured to update a target posterior existence probability and a single-target predicted state distribution particle set weight by using the exponentially weighted generalized likelihood function by a Bayesian update equation, and generate an equal-weight single-target predicted state distribution particle set by resampling; a plurality of equal-weight particles are obtained by resampling the updated single-target predicted state distribution particle set, a target presence condition is determined by using the single-target posterior existence probability, and a DOA estimation result is calculated by particle state to realize single-target DOA joint detection and tracking.

[0019] Scheme three, a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program is executed by the processor to realize the steps of the method of scheme one.

[0020] Scheme four, a computer device, comprising a memory and a processor, the memory stores a computer program, when the processor runs the computer program stored in the memory, the processor executes the method of scheme one.

[0021] The present application has the advantages that: The single-target detection and DOA tracking method and system based on a Bernoulli filter can not only solve the detection failure problem caused by edge of unknown parameters by constructing a generalized likelihood function by using an information theory criterion, significantly improve the reliability of detection, but also enhance the robustness of bearing estimation under extreme conditions by focusing on a high likelihood region of a log-likelihood function by introducing an exponential weighting method to optimize particle filtering, and reduce the operation complexity, so that the algorithm can meet the real-time processing requirements of a passive sonar system while maintaining high-precision tracking.

[0022] The present application is also applicable to the fields of target detection and target positioning under low signal-to-noise ratio and small number of snapshots. BRIEF DESCRIPTION OF DRAWINGS

[0023] Fig. 1A flow chart of a single target detection and DOA tracking method based on Bernoulli filter according to the embodiment one.

[0024] Fig. 2 A comparison diagram of tracking results of different filters in different environments according to the embodiment eleven.

[0025] Wherein, (a) is a tracking result diagram of TAD-Ber filter in SNR=-8dB, M=200 experimental environment, (b) is a tracking result diagram of TBD-Ber filter in SNR=-8dB, M=200 experimental environment, (c) is a tracking result diagram of TAD-Ber filter in SNR=-8dB, M=200 experimental environment, (d) is a tracking result diagram of TBD-Ber filter in SNR=-8dB, M=200 experimental environment.

[0026] Fig. 3 A comparison diagram of detection results of different filters in different environments according to the embodiment eleven.

[0027] Wherein, (a) is a comparison diagram of tracking results in SNR=-8dB, and the number of snapshots is respectively equal to 50, 100, and 200, (b) is a comparison diagram of tracking results in SNR=-14dB, and the number of snapshots is respectively equal to 50, 100, and 200. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme of the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all the embodiments of the present application.

[0029] Embodiment one, the embodiment proposes a single target detection and DOA tracking method based on Bernoulli filter, the method comprises the following steps: Step 1, a Bernoulli stochastic finite set model is constructed, the single target state is defined as the existence or non-existence of dynamic description target state, the Bernoulli Markov process is used to predict the target existence probability, and the particle filter is used to generate a predicted state distribution particle set composed of newborn particles and surviving particles, and the single target predicted state distribution particle set is constructed after the combination of the two; Step 2, the hydrophone array signal is decomposed into noise subspace and signal subspace observation model, the generalized likelihood function with penalty term is calculated by using information theory criterion, which is used to compensate the uncertainty of unknown parameters, and the exponential weighting factor is introduced to focus on the high likelihood area to enhance the main lobe sharpness of the generalized likelihood function under extreme conditions; Step 3, the target posterior existence probability and the single-target prediction state distribution particle set weight are updated by using the exponential weighted generalized likelihood function through the Bayesian updating equation, and an equal-weight single-target prediction state distribution particle set is generated by resampling.

[0030] Embodiment two, the embodiment is one kind based on Bernoulli filter's single-target detection and DOA tracking method of further limitation of the method for detecting and tracking single target described in embodiment one, step 1 in the Bernoulli stochastic finite set model is constructed, and the single-target state is defined as existence or nonexistence dynamic description target state, and the method for predicting target existence probability by using Bernoulli Markov process is as follows:

[0031] In the formula, Indicates When there is no target existence at time, only background noise in observation; Indicates that the observation is the superposition of target signal and background noise; Indicates that the probability of target number being greater than or equal to 2 is equal to 0; Indicates The hydrophone array time domain output at time, For predicting single-target existence probability, For predicting single-target state distribution.

[0032] Embodiment three, the embodiment is one kind based on Bernoulli filter's single-target detection and DOA tracking method of further limitation of the method for detecting and tracking single target described in embodiment one, step 1 in the Bernoulli stochastic finite set model is constructed, and the method for predicting target existence probability by using Bernoulli Markov process is as follows:

[0033]

[0034] In the formula, For target birth probability, For target survival probability, For the Markov density function of new target.

[0035] Embodiment four, the embodiment is one kind based on Bernoulli filter's single-target detection and DOA tracking method of further limitation of the method for detecting and tracking single target described in embodiment one, step 2 in the method for calculating generalized likelihood function with penalty term by using information theory criterion is as follows: the information theory criterion includes MDL criterion and AIC criterion, Generalized likelihood function based on MDL criterion For

[0036] Generalized likelihood function based on AIC criterion For

[0037] wherein, is the number of hydrophone elements.

[0038] Embodiment five, this embodiment is a further limitation of the single target detection and DOA tracking method based on Bernoulli filter according to embodiment one, the method for introducing an exponential weighting factor to focus on the high likelihood region in step 2 to enhance the sharpness of the main lobe of the generalized likelihood function under extreme conditions is: .

[0039] Embodiment six, this embodiment is a further limitation of the single target detection and DOA tracking method based on Bernoulli filter according to embodiment one, the method for updating the target posterior existence probability and the single target predicted state distribution particle set weight by using the exponential weighted generalized likelihood function through the Bayesian update equation in step 3 is: The update equation based on Bayes theorem is

[0040] wherein represents the likelihood function, that is, the probability density of observing under the state , and

[0041] Since the single target state can be represented in the form of Bernoulli stochastic finite set variable, the single target posterior state particle set is updated according to the update equation, wherein the probability of the existence of the single target posterior is .

[0042] wherein

[0043] The weight update expression of the target posterior state distribution is

[0044] wherein the generalized likelihood function is the exponential weighted generalized likelihood function based on information theory criterion.

[0045] Embodiment seven, the embodiment is one kind based on Bernoulli filter's single target detection and DOA tracking method of further limitation to the method of embodiment one, step 3 utilizes single target posteriori existence probability to determine target existence and realizes single target DOA joint detection and tracking through particle state calculation, and the method is: Resample from single target posteriori state particle set Equal weight particle set Assume that the probability of target posteriori existence satisfies Then determine that the target exists, otherwise determine that the target does not exist, and the single target azimuth estimation value is the weighted average of particle state , .

[0046] Embodiment eight, the embodiment proposes a kind of single target detection and DOA tracking system based on Bernoulli filter, the system includes: Single target prediction state distribution particle set module, for constructing Bernoulli stochastic finite set model, defining single target state as existence or nonexistence dynamic description target state, predicting target existence probability using Bernoulli Markov process, and generating prediction state distribution particle set composed of newborn particles and surviving particles by particle filtering, and the single target prediction state distribution particle set is constructed after the combination of the two; Marginalization module, for decomposing hydrophone array signal into noise subspace and signal subspace observation model, calculating generalized likelihood function with penalty term using information theory criterion, for compensating the uncertainty of unknown parameters, introducing exponential weighting factor to focus on high likelihood area to enhance the main lobe sharpness of generalized likelihood function under extreme conditions; Single target DOA joint detection and tracking module, for updating target posteriori existence probability and single target prediction state distribution particle set weight using exponential weighted generalized likelihood function through Bayesian update equation, and generating equal weight single target prediction state distribution particle set by resampling;Resample a plurality of equal weight particles from the updated single target prediction state distribution particle set, determine target existence using single target posteriori existence probability and calculate DOA estimation result through particle state, realize single target DOA joint detection and tracking.

[0047] Embodiment nine, the embodiment proposes a computer readable storage medium, the computer readable storage medium stores computer program, the computer program is executed by processor to realize the steps of the method described in any one of embodiments one to seven. Embodiment ten, the embodiment proposes a computer device, comprising a memory and a processor, the memory stores a computer program, when the processor runs the computer program stored in the memory, the processor executes the method in any one of embodiments one to seven.

[0048] Embodiment eleven, the embodiment proposes an example for explaining the above embodiments one to eight, which is specifically: Referring to Figs. 1 to 3 The embodiment is described with reference to Fig. 1 The Bernoulli filter-based single target detection and DOA tracking method according to the embodiment specifically includes the following steps: Step one, constructing Bernoulli random finite set based on Bayes theorem and completing initialization, predicting target existence probability by Bernoulli Markov process, sampling to generate surviving particles and newborn particles, and merging to construct predicted particle set.

[0049] Since the target existence problem can be represented by a binary random variable, the particle set is selected to approximate the target state distribution, wherein the prior assumption of no target is adopted, and the azimuth (DOA) is uniformly distributed in the full space monitoring range, and the azimuth change rate is subject to zero mean Gaussian distribution to reflect the uncertainty of target motion.

[0050] Assume The posterior single target state at time t is expressed in the form of Bernoulli random finite set, and the corresponding posterior parameter set is known, wherein represents the posterior existence probability of the single target, represents the posterior state distribution of the single target. The predicted single target state can also be expressed in the form of Bernoulli random finite set,

[0051] In the formula, represents There is no target at time t, and only background noise in the observation; represents that the observation is the superposition of target signal and background noise; represents that the probability of the number of targets being greater than or equal to 2 is equal to 0. represents The hydrophone array time domain output at time t, is the predicted single target existence probability, is the predicted state distribution of the single target.

[0052] According to Bayes theorem, the particle set at the initial time t0 is constructed , wherein represents the first The single-objective state and corresponding weights of each particle. The prediction equation based on Bayes' theorem can be written as...

[0053] In the formula, Let represent the single-objective Bernoulli Markov density function. (2) Assuming that the probability of the target's survival is independent of the target's state at the previous moment, the single-target state transition equation can be substituted into the prediction equation to obtain the predicted single-target existence probability and the predicted single-target state distribution when the target is assumed to exist, which can be used as the prediction parameter set. .

[0054]

[0055]

[0056] In the formula, For the target newborn probability, For the target survival probability, Let be the Markov density function for newly formed targets. The predicted probability of a single target's existence and its state distribution both contain two terms: a newly formed term and a survival term. The newly formed term consists of the newly formed density, the probability that no target existed in the previous time step, and the probability of a target being newly formed. The survival term consists of the probability that the target existed in the previous time step, the probability of the target surviving, and the standard Chapman-Kolmogorov prediction term.

[0057] (3) Sample surviving particles from the state transition model to generate There are 10 surviving particles, where the expression for the state transition model is... The target state and corresponding weight of the surviving particles at each time point are:

[0058] ,

[0059] In the formula Here is the state transition matrix. Let be the noise-driven matrix. Assume the DOA of the new target follows a function defined in the state space. Uniform distribution on, i.e. This indicates that the new target can appear at any location in the orientation space; the DOA rate of change parameter follows a real Gaussian distribution, i.e. At this point, sampling of surviving targets reflects the response of newly formed particles to the target's first appearance. The state and weight of the newly generated target at each time step can be represented as follows: ,

[0060] Merge the surviving target particle set with the new-born target particle set to obtain the single-target predicted state distribution particle set , where the total number of particles is At this time, the single-target predicted state distribution is

[0061] where is the Dirac function, which is used to discretize the particle approximation of the continuous distribution.

[0062] Step two, decompose the hydrophone array signal into signal subspace and noise subspace, calculate the generalized likelihood function with a penalty term based on the information theory criterion, and enhance the likelihood function focusing under low signal-to-noise ratio through exponential weighting technology.

[0063] The time-domain output of the hydrophone array can be separated into two independent subspace vectors, and the observation can be represented by Bernoulli random finite set variables .

[0064] wherein is the steering vector, is the narrowband sound source (target) far-field incident plane wave signal; is the additive noise received by the array. The sample covariance matrix is used as an estimate of the true covariance matrix.

[0065]

[0066] wherein is the number of snapshots. In order to marginalize the position parameters such as sound source signal power and noise power, the observation is separated into two complementary subspaces: the signal subspace determined by the steering vector, and the corresponding orthogonal space is the noise subspace. At this time, the time-domain output of the hydrophone array can be separated into two independent subspace vectors:

[0067] wherein represents the signal subspace observation, represents the noise subspace observation, represents the unit coordinate transformation matrix, wherein and represent the orthogonal vectors of the signal and noise subspace expansion, respectively.

[0068] (2) Calculate the steering vector according to the state of each particle, thereby constructing the orthogonal projection of the signal subspace and the orthogonal projection of the noise subspace , which are written as

[0069]

[0070] wherein, denotes the identity matrix of dimension .

[0071] (3) Since there is no prior information about the parameters such as the sound source signal and the noise power, the accuracy of the detection is low when the non-informative prior is selected, therefore, a penalty function is introduced by using the information theory criterion, and the logarithmic likelihood function with the penalty term is used as the generalized likelihood function of the observation, so as to compensate for the detection uncertainty caused by the arbitrary value of the unknown parameter when there is no prior information. For the single-target state at time , the generalized likelihood function based on the MDL criterion is

[0072] the generalized likelihood function based on the AIC criterion is

[0073] In the formula, is the number of hydrophone elements. (4) Since the main lobe of the logarithmic likelihood function is flat and divergent under the condition of low signal-to-noise ratio and small number of snapshots, in order to enhance the high likelihood region, the logarithmic likelihood function needs to be exponentially weighted, and the exponential weighting factor is . Thus, the main lobe of the logarithmic likelihood function is more sharp, and the robustness is stronger under the condition of low signal-to-noise ratio and small number of snapshots, and the corresponding exponentially weighted logarithmic likelihood function is

[0074] Step three, the target posterior existence probability and particle weight are updated by using the exponentially weighted generalized likelihood function, an equal-weight particle set is generated by resampling, and it is determined whether the target exists according to the existence probability and the DOA is estimated by the particle state estimation to realize tracking.

[0075] (1) The update equation based on the Bayes theorem is

[0076] In the formula, denotes the likelihood function, that is, the probability density of observing under the state , and

[0077] Since the single-target state ​The single-target posterior state particle set can be updated according to an updating equation in the form of a Bernoulli random finite set variable , wherein a probability of existence of a single-target posterior state is

[0078]

[0079] A weight updating expression of the target posterior state distribution is

[0080] The generalized likelihood function is an information criterion-based generalized likelihood function after exponential weighting.

[0081] (2) Resampling from the single-target posterior state particle set to obtain an equal-weight particle set , wherein a probability of existence of a target posterior state satisfies If the target exists, otherwise, the target does not exist, and a single-target azimuth estimation value is a particle state weighted average . .

[0082]

[0083] The embodiments have the following: A 6-element uniform linear array is set, an element spacing is set according to a half wavelength of a 500 Hz signal, a sound speed is 1500 m / s, an environmental noise field is assumed to be an isotropic noise field, noise and signal power are unknown and time-invariant, experiments are respectively performed under SNR=-8 dB and-14 dB, and a snapshot number is M=50. In the designed simulation scene, no target exists in a scene from a first time to a 15th time, a target appears at a 16th time and survives to a 40th time, the target disappears from the scene from a 41st time to a 50th time. An initial azimuth of the target is-30°, and an azimuth change rate of the target is 2° / s. Parameters of the TBD-Ber filter are set as follows: a particle number is , a new particle number is , a target survival probability is , and a target birth probability is . An exponential weighting factor is . A new target DOA obeys a uniform distribution , that is, the new target can appear at any position in an azimuth space. A DOA change rate parameter obeys a real Gaussian distribution , wherein a variance is .

[0084] The specific implementation process is as follows: ​​Initialize the Bernoulli RFS parameters, set the initial target existence probability as 0 (no target); According to the prediction equation, obtain the predicted single-target existence probability and the predicted single-target state distribution at each time. Sample and assign weights to the surviving particles, sample and assign weights to the azimuth and azimuth change rate of the new-born particles, and combine to obtain a set of single-target predicted state distribution particles; Calculate the sampling covariance matrix, calculate the generalized likelihood function according to the single-target state at the current time, and exponentially weight the corresponding likelihood function, update the target existence probability and the corresponding weight, and obtain a set of single-target posterior state distribution particles; Resample from the weighted particle set Equal weight particles, determine whether the target exists according to the target posterior existence probability, if the target posterior existence probability , it is determined that the target exists, and the target azimuth estimation result is output, otherwise it is determined that the target does not exist.

[0085] Finally, the target tracking result obtained by the method is shown in Fig. 2 , the TBD-Ber filter not only reduces the clutter, but also accurately detects the target and gives the target azimuth when the target exists. The algorithm detection performance comparison results are shown in Fig. 3 , it can be seen that the TBD-Ber filter significantly reduces the false alarm probability compared with the TAD-Ber filter, and the method can still maintain excellent detection performance under the condition of low signal-to-noise ratio and small snapshot number.

[0086] Those skilled in the art can understand that the above description is only preferred embodiments of the present application, and the features described in each embodiment of the present disclosure and / or claims can be combined or combined, even if such combination or combination is not explicitly described in the present disclosure. It is not intended to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, and those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or make equivalent replacement to part of the technical features, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0087] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the foregoing description without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims be interpreted as including all such variations and modifications as fall within the spirit and scope of the application. It is apparent that those skilled in the art can modify and adapt the application without departing from the spirit and scope of the application. It is therefore intended that the application not be limited to the disclosed embodiments, but that it can also cover modifications and variations within the scope of the present application.

Claims

1. A single-target detection and DOA tracking method based on a Bernoulli filter, characterized in that, The method includes the following steps: Step 1: Construct a Bernoulli random finite set model, define the single target state as the dynamic description of the target state as either present or absent, use the Bernoulli Markov process to predict the probability of the target's existence, and generate a predicted state distribution particle set consisting of newly generated particles and surviving particles through particle filtering. After merging the two, construct the single target predicted state distribution particle set. Step 2: Decompose the hydrophone array signal into a noise subspace and a signal subspace observation model. Calculate the generalized likelihood function with a penalty term using information theory criteria to compensate for the uncertainty of unknown parameters. Introduce an exponential weighting factor to focus on the high likelihood region to enhance the sharpness of the main lobe of the generalized likelihood function under extreme conditions. Step 3: Using the Bayesian update equation, the weights of the target posterior existence probability and the single-target predicted state distribution particle set are updated using the exponentially weighted generalized likelihood function. The particle set of the single-target predicted state distribution is then resampled to generate an equally weighted particle set. Several equally weighted particles are obtained by resampling the updated particle set of the single-target predicted state distribution. The existence of the target is determined by the single-target posterior existence probability, and the DOA estimation result is calculated by the particle state, thus realizing the joint detection and tracking of single-target DOA.

2. The single-target detection and DOA tracking method based on Bernoulli filter according to claim 1, characterized in that, In step 1, a Bernoulli stochastic finite set model is constructed, and the single-objective state is defined as a dynamic description of the objective state, indicating its existence or non-existence. The method for predicting the probability of the objective's existence using a Bernoulli Markov process is as follows: In the formula, express There is no target present at any given time, and only background noise is observed. This indicates that the observation is a superposition of the target signal and background noise; This means the probability that there are more than or equal to 2 targets is 0. express The time-domain output of the hydrophone array at any given moment. To predict the probability of the existence of a single target, To predict the state distribution of a single objective.

3. The single-target detection and DOA tracking method based on Bernoulli filter according to claim 1, characterized in that, In step 1, the method for generating a predicted state distribution particle set consisting of newly generated particles and surviving particles through particle filtering, and then merging the two to construct a single-target predicted state distribution particle set, is as follows: In the formula, For the target newborn probability, For the target survival probability, The Markov density function for the newborn objective.

4. The single-target detection and DOA tracking method based on Bernoulli filter according to claim 1, characterized in that, The method for calculating the generalized likelihood function with a penalty term using information theory criteria in step 2 is as follows: the information theory criteria include the MDL criterion and the AIC criterion. Generalized likelihood function based on MDL criterion for Generalized likelihood function based on AIC criterion for In the formula, The number of elements in the hydrophone array.

5. The single-target detection and DOA tracking method based on Bernoulli filter according to claim 1, characterized in that, The method in step 2 that introduces an exponential weighting factor to focus on the high likelihood region to enhance the main lobe sharpness of the generalized likelihood function under extreme conditions is as follows: 。 6. The single-target detection and DOA tracking method based on Bernoulli filter according to claim 1, characterized in that, Step 3 involves updating the target posterior existence probability and the weights of the single-target predicted state distribution particle set using the Bayesian update equation and the exponentially weighted generalized likelihood function. The update equation based on Bayes' theorem is: In the formula This represents the likelihood function, i.e., the state in which the likelihood is expressed. The following observations The probability density, and Due to single-objective state It can be represented by Bernoulli random finite set variables, and the single-objective posterior state particle set is updated according to the update equation. Among them, the probability of the existence of a single objective posterior. for In the formula The weight update expression for the target posterior state distribution is as follows: The generalized likelihood function is the exponentially weighted generalized likelihood function based on information theory criteria.

7. The single-target detection and DOA tracking method based on Bernoulli filter according to claim 1, characterized in that, Step 3 utilizes the posterior existence probability of a single target to determine its presence and calculates the DOA estimation result using particle states. The method for joint detection and tracking of single-target DOA is as follows: Resampling from a single-objective posterior state particle set A set of particles with equal weights Assume the posterior probability of the target exists satisfies If the target is determined to exist, it is determined to not exist. The single target's azimuth estimate is a particle state-weighted average. , 。 8. A single-target detection and DOA tracking system based on a Bernoulli filter, characterized in that, The system includes: The Single-Target Predicted State Distribution Particle Set Module is used to construct a Bernoulli random finite set model. It defines the single-target state as the dynamic description of the target state, whether it exists or not. It uses a Bernoulli Markov process to predict the probability of the target's existence and generates a predicted state distribution particle set consisting of newly generated particles and surviving particles through particle filtering. The two are then merged to construct the single-target predicted state distribution particle set. The edge-decomposition module is used to decompose the hydrophone array signal into a noise subspace and a signal subspace observation model. It uses information theory criteria to calculate the generalized likelihood function with a penalty term to compensate for the uncertainty of unknown parameters. An exponential weighting factor is introduced to focus on the high likelihood region to enhance the main lobe sharpness of the generalized likelihood function under extreme conditions. The single-target DOA joint detection and tracking module is used to update the target posterior existence probability and the weight of the single-target predicted state distribution particle set by updating the Bayesian update equation and using the exponentially weighted generalized likelihood function. It then resamples to generate an equally weighted single-target predicted state distribution particle set. The updated single-target predicted state distribution particle set is resampled to obtain several equally weighted particles. The single-target posterior existence probability is used to determine the existence of the target, and the DOA estimation result is calculated through the particle state, thus realizing single-target DOA joint detection and tracking.

9. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.

10. A computer device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the method of any one of claims 1 to 7.

Images