Single radiation source direct tracking method based on Bernoulli filter
By using a single radiation source direct detection and tracking method based on Bernoulli filter in radiation source tracking, the problem of insufficient tracking error and low signal-to-noise ratio in complex scenarios is solved, and more efficient Object detection and tracking performance.
Patent Information
- Application Number
- CN202411084995.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-12-06
AI Technical Summary
Traditional radiation source tracking methods are difficult to achieve real-time effective tracking in complex scenarios, and are poorly robust under low signal-to-noise ratio conditions, which easily leads to error accumulation and makes it difficult to accurately detect and track targets.
Using a single radiation source direct detection and tracking method based on Bernoulli filter, the likelihood function of the received signal of matching antenna array is designed to avoid intermediate errors, and the centralized implementation is used to improve robustness and adaptability, and directly from the received signal. Carry out joint detection and tracking.
This method improves the target tracking performance in complex scenarios, enhances the robustness under low signal-to-noise ratio conditions, avoids intermediate errors in traditional two-step tracking, and can effectively detect and track radiation sources with unknown signal-to-noise ratios Target.
Smart Images

Figure CN119087420A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of radiation source tracking, and in particular relates to a single radiation source direct tracking method based on a Bernoulli filter. Background Art
[0002] With the development of network, communication and computer technology, the tracking method for radiation sources has gradually evolved from the data level to the signal level. Among them, the direct tracking method of moving radiation sources based on the original signal echo has received more and more attention. The traditional radiation source tracking is a two-step method. First, the parameters of the target to be tracked are estimated from the received signal. Then, the estimated intermediate parameters are used as input, and appropriate filters are selected to process these parameters, so as to estimate the number and state of targets inside the area to be detected. This method will cause the estimation error to be unable to be corrected in the next stage, resulting in error accumulation, and it is difficult to track the target in real time and effectively in complex scenes. Summary of the invention
[0003] In view of the above problems, the present invention proposes a single radiation source direct detection and tracking method based on Bernoulli filter. In terms of detection and tracking, a likelihood function of the signal manifold received by the matching antenna array is designed on the basis that the received signal conforms to the complex wishart distribution, and a centralized implementation is adopted, and its performance is better than the distributed likelihood function. This direct tracking method avoids the intermediate error of two-step tracking and the hard decision in the traditional detection and tracking algorithm, and can improve the adaptability of the algorithm to complex scenes and the robustness under low signal-to-noise ratio, and improve the tracking performance of the target. Starting from the received data directly, the present invention realizes the joint detection and tracking problem of single radiation source targets with unknown appearance time and unknown signal-to-noise ratio. It has good performance, adaptability and robustness to the environment, and can meet the design requirements in engineering.
[0004] The technical solution adopted by the present invention is:
[0005] A single radiation source direct tracking method based on Bernoulli filter is set in the radiation source tracking scenario, with a stationary receiving station N r The position vector of the jth receiving station composed of the x-axis and the y-axis is d j ,The receiving station has M antenna units forming a linear array, and the first antenna unit on the left is set as the reference unit. The antenna unit spacing is d, r is the distance from the radiation source to the reference unit, θ is the angle between the radiation unit and the reference unit, and the radiation source state is where p x , p y represents the 2D plane position of the radiation source, and represents the speed of the radiation source in each direction. The signal received by the receiving station at time t is expressed as:
[0006] x(t)=a(r,θ)s(t)+n(t)
[0007] in,
[0008]
[0009] m=1,...,M, s(t) is the complex signal received by the reference unit, a(r,θ) is the array signal manifold, and n(t) is the propagation noise; the tracking method includes:
[0010] S1. Obtain the Bernoulli distribution parameters of the measurement data and the target at time k-1, where the measurement data is obtained by sampling x(t), and the measurement matrix at time k is represented by z k ; Get the existence probability, particle weight and particle state of the target at time k-1, indicating that the Bernoulli distribution parameter is It is used to approximate the spatial probability density function p of the Bernoulli distribution. k (x) is the weighted particle, N is the number of predicted particles;
[0011] S2. Predict the existence probability of the target at time k and obtain the spatial probability density function, which is:
[0012] The prediction method of the existence probability at time k is:
[0013] r k|k-1 =p b (1-r k-1 )+p s r k-1
[0014] Among them, r k-1 , and They represent the existence probability, particle weight and particle state at time k-1 respectively, and p b is the probability of new birth, p s is the survival probability, l=1,…,N+B, B is the number of newborn particles;
[0015] Spatial probability density function p k|k-1 The particle of (x) consists of a predicted part and a newly created part:
[0016]
[0017] in, is the predicted probability density of the state, is a Gaussian function with mean m and variance P, Q k-1 is the process noise covariance matrix, b k (x; zk ) is the known nascent density of states, and the predicted particle weights for:
[0018]
[0019] S3. State function according to the target Construct the likelihood function:
[0020]
[0021] in,
[0022] g(x)=σ 2 a(θ)a H (θ)
[0023] f(R z |x) and f(R z |f) is the likelihood function with and without a target, and σ 2 are the noise variance and signal variance respectively, I M is the M-dimensional unit matrix, ν is the number of snapshot points, represents the probability density function that conforms to the complex Wishart distribution matrix;
[0024] S4. Calculate the parameters related to the likelihood of each particle, including:
[0025] Calculate the likelihood function of the jth receiving station when there is no target and when there is a target:
[0026]
[0027] The obtained multiple likelihood functions are fused to obtain the likelihood without target and with target respectively:
[0028]
[0029] Compute the likelihood ratio:
[0030]
[0031] S5. Use the obtained likelihood ratio to sum the integral of the approximate likelihood ratio:
[0032]
[0033] S6. Update the Bernoulli parameter to obtain the Bernoulli parameter π at time k. k|k = {r k|k ,p k|k (x)}, where the probability of existence is:
[0034]
[0035] S7. For each particle, update the particle weight according to the calculated likelihood of having a target. Then normalize the particle weights:
[0036]
[0037] S8, resampling, for l = 1, ..., N and m (l) ∈{1,…,N+B}, with The probability of selecting particles is , and the probability of particles with large weights being selected is high. The state of particles after resampling is:
[0038]
[0039] S9. Reset the particle weight to l=1,…,N;
[0040] S10, obtain the existence probability, particle weight and particle state of the target at time k, and perform state extraction to obtain Contains the estimated values of the target's position, velocity, and signal power at time k; iterative processing, k = k + 1, until all moments are processed.
[0041] Beneficial effects of the present invention:
[0042] 1) The present invention uses a Bernoulli filter to solve the tracking problem using signals as measurements, and introduces typical radiation source tracking scene modeling and likelihood functions based on fuzzy functions to solve the problem of calculating likelihood functions when the transmitted signal and the transmitted power are unknown, thus providing a solution for radiation source tracking in more complex scenarios;
[0043] 2) The present invention can solve the problem of joint detection and direct tracking of radiation sources, avoiding the intermediate error caused by the traditional two-step tracking method. The method has strong robustness and good effect under low signal-to-noise ratio. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 Schematic diagram of radiation source tracking scenario.
[0045] Figure 2 Schematic diagram of the receiving station location and the target's actual trajectory.
[0046] Figure 3 It is a schematic diagram of the specific tracking curve when SNR=-15dB.
[0047] Figure 4 Schematic diagram of the curve of average OSPA changing with signal-to-noise ratio from SNR=-15dB to SNR=0dB.
[0048] Figure 5 It is a schematic diagram of the curve of detection rate changing with signal-to-noise ratio from SNR=-15dB to SNR=0dB. DETAILED DESCRIPTION
[0049] The technical solution of the present invention is described in detail below.
[0050] like Figure 1 The following is a typical radiation source tracking scenario with a stationary receiving station N. r Let the position vector composed of the x-axis and y-axis of the j-th receiving station be d j ,The receiving station has M antenna units forming a linear array, the antenna unit spacing is d, r is the distance from the radiation source to the reference unit, θ is the angle between the radiation unit and the reference unit, and the radiation source state is where p x , p y represents the 2D plane position of the radiation source, and Represents the speed of the radiation source in each direction. Assuming the observation time is K, the signal received by the receiving station at time t can be expressed as:
[0051] x(t)=a(r,θ)s(t)+n(t)
[0052] in
[0053]
[0054] The first antenna unit on the left is regarded as the reference unit, and s(t) is the complex signal received by the reference unit. a(r,θ) is the array signal manifold, which is determined by the relative position of the target and the array and the spacing between the antenna units. n(t) is the propagation noise, and the current measurement matrix z is obtained by sampling x(t) k , then z k It is an M×L dimensional matrix, where the rows represent the sampled signals received by the corresponding antenna units and satisfy the complex Gaussian distribution with a mean of zero.
[0055] The present invention takes an antenna unit as a reference unit in the antenna array, and the signal received by the entire antenna array can be expressed as the product of the corresponding signal manifold and the reference unit received signal, and the received signal is sampled as the input of the Bernoulli filter. The direct tracking method effectively reduces the intermediate error generated, and the entire system is Bernoulli, and its information transmission is in the form of probability description, so the proposed scheme has strong robustness and scalability. Because the likelihood function uses the relevant information of all antenna units of each receiving station, it can achieve target tracking at a lower signal-to-noise ratio.
[0056] The present invention specifically includes:
[0057] S1. Bernoulli filter uses a batch of weighted particles To approximate the spatial probability density function p of the Bernoulli distribution k (x), the approximate rule is Initialization: When k = 0, the Bernoulli distribution parameter is π0 = {r0, p0(x)}, r0 is the probability of existence of the target, p0(x) is the corresponding spatial probability density function, and the particle To approximate p0(x), N is the number of predicted particles, B is the number of new particles, The weight corresponding to each particle;
[0058] Define the total observation time as K, when K≥k≥1, repeat steps S2-S10;
[0059] S2, input a priori target Bernoulli distribution parameters and the measurement matrix z at the current moment k . First, make a prediction: the prediction parameter is π k|k-1 = {r k|k-1 ,p k|k-1 (x)}, where the predicted existence probability is calculated as
[0060] r k|k-1 =p b (1-r k-1 )+p s r k-1
[0061] where r k-1 , and They represent the existence probability, particle weight and particle state at time k-1 respectively, and p b is the probability of new birth, p s is the survival probability. The composition space probability density p k|k-1 The particle of (x) consists of a predicted part and a newly created part:
[0062]
[0063] in, is the predicted probability density of the state, is a Gaussian function with mean m and variance P, Q k-1 is the process noise covariance matrix, b k (x; z k ) is the known nascent density of states. The predicted particle weights It is also necessary to calculate separately
[0064]
[0065] S3, is a function of the target state. z It conforms to the complex Wishart distribution, and the likelihood function is constructed based on it. The complex Wishart distribution is a positive definite matrix random distribution in statistics. If the p×p dimensional matrix Y conforms to the complex Wishart distribution with r degrees of freedom, its probability density function can be expressed as follows:
[0066]
[0067] Where A is a p×p positive definite symmetric matrix, tr(·) represents the trace of the matrix, Represents the complex Gamma function of p variables, defined as
[0068]
[0069] and is the Gamma function.
[0070] In the present invention,
[0071]
[0072] in
[0073] g(x)=σ 2 a(θ)a H (θ)
[0074] f(R z |x) and f(R z |f) is the likelihood function with and without a target, and σ 2 are the noise variance and signal variance respectively, I M is the M-dimensional unit matrix, and ν is the number of snapshot points.
[0075] S4. For each particle (l=1, ..., N+B), repeat steps S41-S43 to calculate the parameters related to the likelihood
[0076] S41. First, calculate the likelihood function of the jth receiving station when there is no target, which is a quantity independent of snapshot and time:
[0077]
[0078] Calculating the likelihood function with a target
[0079]
[0080] S42. The fused likelihood function is the product of the likelihood functions of multiple receiving stations. The likelihood when there is no target is:
[0081]
[0082] Then the likelihood function with a target is:
[0083]
[0084] S43. Calculate likelihood ratio
[0085]
[0086] S5, use the particle likelihood ratio calculated in S43 to sum the integral of the approximate likelihood ratio,
[0087]
[0088] S6, Update: Based on the predicted Bernoulli parameters, combined with the measurement vector z obtained at the current moment k , we can get the updated Bernoulli parameter π at time k k|k = {r k|k ,p k|k (x)}. The probability of existence is
[0089]
[0090] S7, spatial probability density is: For each particle (l = 1, ..., N + B), first use the likelihood of the target calculated in S42 to update the particle weight: Then normalize the particle weights:
[0091]
[0092] S8. Resampling: for l = 1, ..., N and m (l) ∈{1,…,N+B}, with The probability of selecting particles is , and the probability of particles with large weights being selected is high. The state of the particles after resampling is
[0093]
[0094] S9. Reset the particle weight to l=1,…,N.
[0095] S10, output N particles: For iteration at the next moment and state extraction: Contains the estimated values of the target's position, speed and signal power at time k, k=k+1, and returns to step S2.
[0096] Simulation example:
[0097] For illustration purposes, consider a representative two-dimensional scene such as Figure 2 , assuming two passive receiving stations d1 ,d 2 They are located at [3000,0], [6000,0] (m) respectively. To simplify the problem, it is assumed that the attenuation of each receiver is the same, that is, the target state vector is The initial position is determined, the speed is unknown, and the total observation time is assumed to be K = 80s. The target moves in a variable speed curve. The survival probability p of each target is s =0.99. The measurement equation has been given in the previous article. The carrier frequency of the transmitter is f c =10 6 Hz, sampling frequency f s =4×10 7 Hz, number of snapshots L = 8 × 10 4 , c = 3 × 10 8 m / s. The target is born at 0s and dies at 80s. The target appears from a fixed point, and the target birth model is as follows:
[0098]
[0099] in,
[0100] m b =[x,0,y,0] T
[0101] P b =diag{[75,75,75,75]} 2
[0102] p b =0.01
[0103] For simulation of b Natural regeneration near the target (x, y is the initial position of the target). A total of 3000 particles are used, of which N = 2500 are surviving particles, B = 500 are new particles, and the number of Monte Carlo simulations is 50.
[0104] like Figure 3 As shown in Figure 2, even at a lower signal-to-noise ratio (SNR = -15dB), the tracking track of the proposed algorithm is still relatively close to the true value. Figure 4 The curve of average OSPA changing with signal-to-noise ratio is plotted. It can be seen that the estimation error of the algorithm is within 100m, which is within a reasonable range for the monitored scene. Figure 5 This shows that the detection performance of the algorithm increases with the increase of signal-to-noise ratio. Figure 2-5 From the above, we can see that the proposed algorithm is suitable for the joint detection and direct tracking of radiation sources, and has strong robustness to low signal-to-noise ratio and strong adaptability to complex environments with unknown signal-to-noise ratio.
Claims
1. A single radiation source direct tracking method based on Bernoulli filter, set in the radiation source tracking scenario, there are stationary receiving stations N r The position vector of the jth receiving station composed of the x-axis and the y-axis is d j ,The receiving station has M antenna units forming a linear array, and the first antenna unit on the left is set as the reference unit. The antenna unit spacing is d, r is the distance from the radiation source to the reference unit, θ is the angle between the radiation unit and the reference unit, and the radiation source state is where p x , p y represents the 2D plane position of the radiation source, and represents the speed of the radiation source in each direction. The signal received by the receiving station at time t is expressed as: x(t)=a(r,θ)s(t)+n(t) in, m=1,...,M, s(t) is the complex signal received by the reference unit, a(r,θ) is the array signal manifold, and n(t) is the propagation noise; characterized in that the tracking method includes: S1. Obtain the Bernoulli distribution parameters of the measurement data and the target at time k-1, where the measurement data is obtained by sampling x(t), and the measurement matrix at time k is represented by z k ; Get the existence probability, particle weight and particle state of the target at time k-1, indicating that the Bernoulli distribution parameter is It is used to approximate the spatial probability density function p of the Bernoulli distribution. k (x) is the weighted particle, N is the number of predicted particles; S2. Predict the existence probability of the target at time k and obtain the spatial probability density function, which is: The prediction method of the existence probability at time k is: r k|k-1 =p b (1-r k-1 )+p s r k-1 Among them, r k-1 , and They represent the existence probability, particle weight and particle state at time k-1 respectively, and p b is the probability of new birth, p s is the survival probability, l=1,…,N+B, B is the number of newborn particles; Spatial probability density function p k|k-1 The particle of (x) consists of a predicted part and a newly created part: in, is the predicted probability density of the state, is a Gaussian function with mean m and variance P, Q k-1 is the process noise covariance matrix, b k (x; z k ) is the known nascent density of states, and the predicted particle weights for: S3. State function according to the target Construct the likelihood function: in, g(x)=σ 2 a(i)a H (i) f(R z |x) and f(R z |f) is the likelihood function with and without a target, and σ 2 are the noise variance and signal variance respectively, I M is the M-dimensional unit matrix, ν is the number of snapshot points, represents the probability density function that conforms to the complex Wishart distribution matrix; S4. Calculate the parameters related to the likelihood of each particle, including: Calculate the likelihood function of the jth receiving station when there is no target and when there is a target: The obtained multiple likelihood functions are fused to obtain the likelihood without target and with target respectively: Compute the likelihood ratio: S5. Use the obtained likelihood ratio to sum the integral of the approximate likelihood ratio: S6. Update the Bernoulli parameter to obtain the Bernoulli parameter π at time k. k|k = {r k|k ,p k|k (x)}, where the probability of existence is: S7. For each particle, update the particle weight according to the calculated likelihood of having a target. Then normalize the particle weights: S8, resampling, for l = 1, ..., N and m (l) ∈{1,…,N+B}, with The probability of selecting particles is , and the probability of particles with large weights being selected is high. The state of particles after resampling is: S9. Reset the particle weight to l=1,…,N; S10, obtain the existence probability, particle weight and particle state of the target at time k, and perform state extraction to obtain Contains the estimated values of the target's position, velocity, and signal power at time k; iterative processing, k = k + 1, until all moments are processed.
Citation Information
Cited By
Single radiation source tracking method under far and near field change based on Bernoulli filter
CN120610236A
Track-before-detect method based on random set theory under unknown signal-to-noise ratio
CN122362363A
A random set theory based track-before-detect method under unknown signal-to-noise ratio
CN122362363B
