Self-adaptive particle filtering method for maritime maneuvering target
By combining IMM-PF and MHM-PF filters, the maneuver probability is adaptively adjusted, which solves the problem of untimely model response in tracking maneuvering targets at sea, improves tracking performance and reduces computational complexity.
Patent Information
- Application Number
- CN202511798489.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-03-03
AI Technical Summary
In existing technologies for tracking maneuvering targets at sea, the motion model cannot adapt to the target's actual motion state in a timely manner, resulting in large tracking errors. In particular, when the motion pattern of the maneuvering target changes frequently, the IMM-PF algorithm responds slowly and has large tracking errors.
By combining IMM-PF and MHM-PF filters and introducing the concept of multiple hypotheses, the state estimates of the two filters are adaptively fused. Target measurement data is used for prediction updates, and the maneuver probability is dynamically adjusted to achieve adaptive tracking of maneuvering targets at sea.
It improves the tracking performance of maneuvering targets at sea, reduces tracking errors, and reduces computational complexity during high-maneuver phases, achieving an optimal balance between tracking performance and computational efficiency.
Smart Images
Figure CN121602964A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of target tracking technology, and more specifically, relates to an adaptive particle filtering method for maneuvering targets at sea. Background Technology
[0002] Monitoring and tracking targets such as fishing boats and speedboats near ports is a crucial aspect of maintaining port security and has significant practical implications. Currently, in maritime mobile target tracking scenarios, the IMM-PF (Interactive Multiple Model Particle Filter), which combines interactive multiple models with particle filtering, is widely used as a high-performance filtering algorithm.
[0003] However, when dealing with non-cooperative targets (such as fishing boats or speedboats that are unwilling to cooperate) that are maneuvering at high speeds and whose motion patterns are frequently changing, IMM-PF fails to adapt to the target's actual motion state in a timely manner, leading to a significant increase in tracking error. Specifically, IMM-PF mainly relies on the model probabilities of the current and previous moments to adjust the particle distribution, making relatively limited use of historical information. Coordinated turning is a common motion pattern for highly maneuverable targets at sea, and these targets typically have large turning radii with small differences in their states between adjacent moments. This means that IMM-PF needs time to adjust the probability weights of each model, resulting in an insufficiently timely response to maneuvering targets and causing significant tracking errors when the target's maneuvering state changes. Summary of the Invention
[0004] The purpose of this application is to provide an adaptive particle filtering method for maneuvering targets at sea, in order to solve the technical problem in the prior art that the motion model suffers from mismatch and untimely response when facing frequent changes in the motion state of maneuvering targets at sea, resulting in large tracking errors.
[0005] To achieve the above objectives, embodiments of this application provide an adaptive particle filtering method for maneuvering targets at sea, comprising the following steps: Based on the target measurement data, the IMM-PF filter and the MHM-PF filter which introduces the idea of multiple hypotheses are predicted and updated to obtain the target state estimates of the IMM-PF filter and the MHM-PF filter, respectively. The target's three index scores are calculated based on the target state estimation using the IMM-PF filter or MHM-PF filter, and then weighted to obtain the fused original target maneuver score function. The target maneuver probability is then obtained by smoothing the function. By setting a maneuverability discrimination threshold, comparing the target maneuver probability with the maneuverability discrimination threshold, and adaptively fusing the target state estimates of the IMM-PF filter and the MHM-PF filter, the final target state estimate is obtained.
[0006] Preferably, the formula for estimating the final state of the target is as follows: ; In the formula, Let K+1 be the target maneuver probability. These are the mixed weighting coefficients, with a value range of [value range missing]. , and The target state estimates at time K+1 are obtained using the IMM-PF filter and the MHM-PF filter, respectively. For weak maneuvering state threshold, This is the threshold for a strong maneuvering state.
[0007] Preferably, the process of predicting and updating the MHM-PF filter, which incorporates multiple hypotheses, based on target measurement data includes: The hypothetical path at the current time is expanded based on the hypothetical path retained at the previous time step to obtain the hypothetical path retained at the current time step. Under the assumed path retained at the current moment, the state of each particle is updated by selecting the state transition function of the corresponding maneuvering target motion model; The particle weights are calculated based on the sampled particles at the current time, and the scores of the hypothetical paths at the current time are updated based on the particle weights. The scores are accumulated, and the hypothetical path with the highest score is retained to obtain the target state estimate of the MHM-PF filter.
[0008] Preferably, the formula for obtaining the target maneuver probability is: ; In the formula, Let k+1 be the target maneuver probability. Let be the target maneuver probability at time k. It is a smoothing constant. Let be the original target maneuver scoring function after fusion at time k+1.
[0009] Preferably, the formula for obtaining the fused original target maneuver scoring function is: ; In the formula, Let k+1 be the original target maneuver scoring function after fusion. , , All are fusion weights. The new information fraction at time k+1 The score is given for the change in velocity direction at time k+1. The entropy score of the model at time k+1.
[0010] Preferably, the process of predicting and updating the IMM-PF filter based on the target measurement data includes: The weighted sum of the particle states of each model is obtained by summing the weighted sum of the particle states of model j. This sum is then fused with the covariance matrix of the state value of the nth particle of model i to obtain the covariance matrix of model j. The probability of model j is updated based on the covariance matrix of model j to obtain the transition probability weights of model j. The transition probability weights of model j are weighted and summed with the state value of the nth particle of model j to obtain the model mixture state estimate of the particle. The average model mixture state estimates of all particles are then used to obtain the target state estimate of the IMM-PF filter.
[0011] Preferably, the formula for estimating the target state of the MHM-PF filter is as follows: ; In the formula, For target state estimation at time K using the MHM-PF filter, For particle weights, For the sampled particles at time k, This represents the number of particles in the hypothetical path.
[0012] Preferably, the formula for obtaining the target state estimate of the IMM-PF filter is: ; In the formula, This is the state estimate of the IMM-PF filter at time k, where N is the total number of particles. This is the model mixture state estimate for the nth particle at time k.
[0013] Preferably, the three index scores include the innovation score, the velocity direction change score, and the model entropy score; The formula for obtaining the new interest fraction is: ; In the formula, The score for the new information at time k+1. To innovate the threshold constant, For the new information at time k+1, For the average innovation margin, The standard deviation is denoted as .
[0014] Preferably, the formulas for obtaining the velocity direction change score and the model entropy score are as follows: ; ; In the formula, The score is given for the change in velocity direction at time k+1. The model entropy score at time k+1. Let be the normalized constant for velocity variation. The length of the sliding window. The change in a unit vector. The information entropy at time k+1, This represents the number of motion models for the maneuvering target.
[0015] The beneficial effects of this application are as follows: This application proposes an adaptive particle filtering method for maneuvering targets at sea. First, the IMM-PF filter and the MHM-PF filter, which incorporates multiple hypotheses, are updated based on target measurement data to obtain target state estimates from the IMM-PF and MHM-PF filters. Then, the target maneuver probability is dynamically determined through three index scores, and the state estimation results of the IMM-PF and MHM-PF are adaptively adjusted based on this target maneuver probability, achieving complementarity between the two filters. This design can fully utilize the tracking advantages of MHM-PF in strong maneuvering scenarios, quickly respond and achieve globally optimal matching of the motion model of the maneuvering target, effectively reducing the impact of model mismatch. At the same time, IMM-PF is dominant in weak maneuvering or stable phases, avoiding the computational redundancy of MHM-PF. While improving the robustness and accuracy of the system in tracking strong maneuvering targets, it significantly reduces the global computational complexity, achieving an optimal balance between tracking performance and computational efficiency. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 A schematic diagram of the overall process of an adaptive particle filtering method for maneuvering targets at sea, provided as an embodiment of this application; Figure 2 This is a schematic diagram illustrating the switching of the real motion model of a maneuvering target according to an embodiment of this application; Figure 3 A flowchart of the prediction update of the MHM-PF filter provided in one embodiment of this application; Figure 4 This is a schematic diagram of the actual trajectory of a maneuvering target and the algorithm tracking trajectory provided in an embodiment of this application; Figure 5 This is a comparison chart of algorithm tracking errors provided in an embodiment of this application; Figure 6A comparison chart of the computational complexity of the algorithm provided in one embodiment of this application. Detailed Implementation
[0018] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.
[0019] This application builds upon the existing IMM-PF (Interactive Multi-Model Particle Filter) framework by introducing the concept of multiple hypotheses to construct MHM-PF (Multi-Hypothesis Particle Filter). MHM-PF improves the tracking capability for highly maneuvering targets by comprehensively scoring the possible motion state models of the target at each time step and obtaining the most probable motion model at each time step.
[0020] Since MHM-PF (Multiple Hypothesis Particle Filter) is designed to adapt to the dynamic changes of highly maneuverable targets, it becomes redundant in stable scenarios with weakly maneuverable targets, increasing the system's computational complexity. Therefore, this application fuses MHM-PF and IMM-PF (Interactive Multi-Model Particle Filter) estimations, introducing a maneuverability discrimination mechanism into the tracking filtering process for maritime targets. This mechanism adaptively fuses the filtering estimation results of MHM-PF and IMM-PF based on the target's maneuverability score: IMM-PF dominates in the weak maneuvering phase, while MHM-PF dominates in the strong maneuvering phase. This improves tracking performance during the strong maneuvering phase while effectively reducing the global computational load.
[0021] Please see Figure 1 An adaptive particle filtering method for maneuvering maritime targets, provided in one embodiment of this application, includes: S1: Based on the target measurement data, the IMM-PF filter and the MHM-PF filter with the introduction of multiple hypotheses are predicted and updated to obtain the target state estimates of the IMM-PF filter and the MHM-PF filter, respectively.
[0022] Specifically, the IMM-PF filter and the MHM-PF filter, which incorporates the idea of multiple hypotheses, are initialized. Based on the input target measurement data, the prediction updates of the MHM-PF and IMM-PF filters are performed for each frame of the target motion (from frame 1 k=1 to frame T k=T).
[0023] First, the target motion and noise are modeled: Maneuvering Target Motion Model Near-uniform motion model (CV); maneuvering target motion model A coordinated turning motion model (CT1) with a coordinated turning rate ω=0.25 is adopted; a maneuvering target motion model is also used. A coordinated turning motion model (CT2) with a coordinated turning rate ω = -0.25 was adopted. The target maneuver lasted for a total of 200 seconds. The actual motion model is as follows: Figure 2 As shown in the figure, the horizontal axis represents the time step, corresponding to the time progress of the target's movement (from 0 to 200, representing a total duration of 200 seconds), and the vertical axis represents the actual model, with a value of 1 indicating the motion model of the maneuvering target. The value 2 represents the motion model of the maneuvering target. The value 3 indicates the motion model of the maneuvering target. The process noise and measurement noise in the figure both follow a zero-mean Gaussian distribution.
[0024] Next, the target is detected using radar to obtain target measurement data. The target measurement data is then input into the MHM-PF filter, and the MHM-PF filter is updated based on the input target measurement data to obtain the target state estimate output by the MHM-PF filter.
[0025] In an optional embodiment, this application uses radar to detect the target at time k to obtain target measurement data at time k. The specific steps are as follows: Please see Figure 3 The parameters in the MHM-PF filter are initialized, including the number of global hypothesis paths to be retained. Number of models in each hypothesis path Historical frame count Number of particles in each hypothetical path Model set Initial time assumed path and the initial sampling particle set .
[0026] Based on the hypothetical path retained from the previous time step (k-1 time step). The hypothetical path at the current time (time k) is expanded to obtain the hypothetical path retained at time k. The extended formula is as follows: ; In the formula, The hypothetical path retained at time K. For the first Hypothetical path, For model sets.
[0027] Next, state prediction is performed for each hypothetical path. The hypothetical paths retained at time K are... Next, select the state transition function of the corresponding maneuvering target motion model. The state of each particle is updated using the following formula: ; In the formula, For the sampled particles at time K, Let be the state transition function of the motion model of the maneuvering target. For the sampled particles from the previous frame (time k-1), The process noise follows a zero-mean Gaussian distribution.
[0028] Furthermore, based on the sampled particles at time K Calculate particle weights The formula is as follows: ; In the formula, For particle weights, Let k be the observed state. For the observation matrix, For the sampled particles at time k, To observe the noise covariance matrix.
[0029] Based on the obtained particle weights Update the hypothetical path score at time k The formula is as follows: ; In the formula, The hypothetical path score at time k. This is the assumed number of particles in the path. For particle weights.
[0030] Cumulative history The path score for a frame is calculated using the following formula: ; In the formula, For history Frame path score, For the current frame number, For frame j, The hypothetical path score is given for the j-th frame.
[0031] Scored all hypothetical paths and retained Find the highest-scoring path, remove the remaining paths, and update the hypothesis set. , among which, The number of globally hypothetical paths retained.
[0032] right Each particle is resampled to prevent degradation and its weights are reset. .
[0033] Extract the state estimate of the highest-scoring hypothetical path from the highest-scoring hypothetical path, and output the state estimate of the MHM-PF filter at time K, as shown in the following formula: ; In the formula, Here, represents the state estimate of the MHM-PF filter at time K, and represents the state estimate of the highest-scoring path. For particle weights, For the sampled particles at time k, This represents the number of particles in the hypothetical path.
[0034] Finally, the MHM-PF filter determines whether there is target measurement data input. If there is, it iteratively obtains the state estimate of the highest-scoring path; otherwise, it outputs the state estimate of the highest-scoring path and ends the prediction update of the MHM-PF filter.
[0035] The IMM-PF filter is updated based on the target measurement data at time K, and the target state estimate of the IMM-PF filter at time K is obtained. The specific steps are as follows: Randomly generated particles: based on model set (Movement model of a maneuvering target) Motion model of maneuvering target Motion model of maneuvering target Based on the mean and covariance of the state variables at time k, particles are randomly selected. The state value of the nth particle in model i is denoted as... The covariance matrix is denoted as .
[0036] The weighted sum of the particle states for each model particle is obtained by weighting and summing the values. The formula is as follows: ; In the formula, The weighted sum of the particle states of model j, Let n be the state value of the nth particle in model j. is the transition probability weight from model i to model j, m is the number of maneuvering target motion models, and n is the number of particles.
[0037] The state value of the nth particle in model i Weighted sum of particle states of model j By fusing the covariance matrices, we obtain the covariance matrix of model j, as shown in the following formula: ; In the formula, Let J be the covariance matrix of model j. The transition probability weights from model i to model j Let n be the state value of the nth particle in model i. The weighted sum of the particle states of model j.
[0038] The probabilities of model j are updated to obtain the transition probability weights of model j. The formula is as follows: ; ; In the formula, Let these be the transition probability weights for model j. The normalization coefficient is... , Let be the likelihood function of model j at time k. Let be the prior model probability of the nth particle. This is the new information at time k.
[0039] The particles of each model are resampled to generate a new particle sequence, and the weight of the new particles is initialized to 1 / N.
[0040] Based on the transition probability weights of model j The state value of the nth particle in model j By performing a weighted summation, we obtain the model mixture state estimate of the particles. The formula is as follows: ; In the formula, To estimate the mixed state of the nth particle at time k, Let be the state value of the nth particle in model j. represents the transition probability weights of model j.
[0041] The state estimate of the IMM-PF filter at time k is as follows: ; In the formula, This is the state estimate of the IMM-PF filter at time k, where N is the total number of particles. This is the model mixture state estimate for the nth particle at time k.
[0042] S2: Calculate the three index scores of the target based on the target state estimation using the IMM-PF filter or MHM-PF filter.
[0043] Specifically, in an optional embodiment, the target state is estimated based on time k. Calculate the three index scores of the target at time K+1, including the innovation score, the velocity direction change score, and the model entropy score.
[0044] The specific calculation of the new interest fraction is as follows: First, define the new information at time k+1. for: ; In the formula, For the new information at time k+1, The observed state at time k+1, For the observation matrix, This is the target state estimate at time k+1.
[0045] Among them, the target state estimate at time k+1 is obtained. The formula is: ; In the formula, Here is the state transition matrix. The target state estimate at time k can be either the target state estimate of the MHM-PF filter or the target state estimate of the IMM-PF filter at time k.
[0046] In an alternative embodiment, before introducing the sliding window length... =Using 8 frames as the sliding window length, calculate the average innovation amplitude. The length of the sliding window is not limited here and can be set according to the actual situation. Calculate the average innovation rate. The formula is as follows: ; In the formula, For the average innovation margin, The length of the sliding window. For the new information at time k+1, To find the L2 norm.
[0047] Next, the average innovation margin Calculate the standard deviation The formula is as follows: ; In the formula, Standard deviation, For the average innovation margin, The length of the sliding window. This is the new information at time k+1.
[0048] According to standard deviation Average innovation margin and new information Calculate the new interest score The formula is as follows: ; In the formula, The score for the new information at time k+1. To innovate the threshold constant, the value is selected. , For the new information at time k+1, For the average innovation margin, The standard deviation is denoted as .
[0049] The calculation of the fraction of change in velocity direction is as follows: First, estimate the target state at time k+1. Extracting velocity prediction components Normalized to: ; In the formula, For the normalized velocity prediction component, For velocity prediction components.
[0050] Next, calculate the change in the unit vector at time k+1, using the following formula: ; In the formula, Let be the change in the unit vector at time k+1. Let be the unit vector at time k+1. Let be the unit vector at time k.
[0051] Finally, based on the previous =8 frames of unit vector change The score for the change in velocity direction is calculated using the following formula: ; In the formula, The score is given for the change in velocity direction at time k+1. Let be the normalization constant for velocity change, and take the value of . , The current sliding window length =8 frames of unit vector change.
[0052] The model entropy score is calculated as follows: Let the posterior probability of the model at time k+1 be... for: ; In the formula, Let be the posterior probability of the model at time k+1. The number of motor motion models. For the first The posterior probability of a maneuvering target motion model.
[0053] Its information entropy is: ; In the formula, The information entropy at time k+1, Let be the posterior probability of the maneuvering motion model of the i-th target. To prevent division by zero, the value is taken as an extremely small positive integer. .
[0054] The model entropy score is then: ; In the formula, The model entropy score at time k+1. The information entropy at time k+1, This represents the number of motion models for the maneuvering target.
[0055] S3: Weight the three indicators to obtain the fused original target maneuver scoring function, smooth it to obtain the target maneuver probability at each time step.
[0056] Specifically, in an optional embodiment, the three metrics innovation scores at time k+1 are... Score for changes in velocity direction and model entropy score The fusion is performed to obtain the original target maneuver scoring function after fusion at time k+1. The formula is as follows: ; In the formula, Let k+1 be the original target maneuver scoring function after fusion. , , All are fusion weights, with values ranging from 1 to 2. , , .
[0057] The original target maneuver scoring function after fusion at time k+1. After smoothing, the target maneuver probability at time k+1 is obtained. The formula is as follows: ; In the formula, Let k+1 be the target maneuver probability. Let be the target maneuver probability at time k. It is a smoothing constant. , Let be the original target maneuver scoring function after fusion at time k+1.
[0058] S4: Set a maneuverability discrimination threshold, compare the target maneuver probability with the maneuverability discrimination threshold, adaptively fuse the target state estimates of the MHM-PF filter and the IMM-PF filter to obtain the final target state estimate.
[0059] Specifically, the maneuverability discrimination threshold includes a weak maneuverability state threshold and a strong maneuverability state threshold. This application defines the weak maneuverability state threshold as... Setting it to 0.15 will lower the threshold for high-mobility states. Set to 0.4. There are no restrictions on the thresholds for weak and strong maneuver states here; you can set them according to the actual situation.
[0060] In an optional embodiment, the target maneuver probability at time k+1 At that time, the target is in a weak maneuvering state, and the target state estimation is performed using the IMM-PF filter. ;when The target is in a highly maneuverable state, and the target state estimation is performed using the MHM-PF filter. ;when The target state is estimated by adaptively fusing the IMM-PF filter and the MHM-PF filter. The final target state estimate can then be expressed as: ; In the formula, These are the mixed weighting coefficients, with a value range of [value range missing]. , and These are the target state estimates at time K+1 for the IMM-PF filter and the MHM-PF filter, respectively. Specifically, the target state estimate at time K+1 is obtained. and The steps, and obtaining the target state estimate at time K. and The steps are the same, only the target measurement data of the input IMM-PF filter and the IMM-PF filter need to be changed from time K to time K+1.
[0061] Please see Figure 4 The image shows a comparison of the tracking results of the algorithms. The adaptive particle filtering method proposed in this application is denoted as SAMM-PF, and the root mean square error (RMSE) is used as the evaluation index for comparing the tracking performance of the algorithms. The results are as follows: Figure 5 As shown. Figure 6 The table below shows a comparison of the computational complexity of the algorithms, specifically the root mean square sum (RMS) computational complexity.
[0062] As shown in the table, the adaptive multi-hypothesis particle filter method for the motion characteristics of maneuvering targets at sea, applied in this embodiment, improves tracking performance by 25.7% compared to IMM-PF and 8.0% compared to MHM-PF in the experimental simulation scenario, while reducing overall algorithm complexity by 22.5% compared to MHM-PF. This demonstrates that the adaptive particle filter method used in this application effectively reduces the impact of model mismatch and improves system robustness during the target's strong maneuvering phase, and further reduces system computational complexity during the target's weak maneuvering phase by utilizing interactive multi-model particle filtering.
[0063] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0064] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. An adaptive particle filtering method for maneuvering targets at sea, characterized in that, Includes the following steps: Based on the target measurement data, the IMM-PF filter and the MHM-PF filter which introduces the idea of multiple hypotheses are predicted and updated to obtain the target state estimates of the IMM-PF filter and the MHM-PF filter, respectively. The target's three index scores are calculated based on the target state estimation of the IMM-PF filter or the MHM-PF filter, and weighted to obtain the fused original target maneuver score function. The target maneuver probability is then obtained by smoothing. A maneuverability discrimination threshold is set, and the target maneuver probability is compared with the maneuverability discrimination threshold. The target state estimates of the IMM-PF filter and the MHM-PF filter are adaptively fused to obtain the final target state estimate.
2. The adaptive particle filtering method for maneuvering maritime targets as described in claim 1, characterized in that, The formula for estimating the final state of the target is as follows: ； In the formula, Let K+1 be the target maneuver probability. These are the mixed weighting coefficients, with a value range of [value range missing]. , and The target state estimates at time K+1 are obtained using the IMM-PF filter and the MHM-PF filter, respectively. For weak maneuvering state threshold, This is the threshold for a strong maneuvering state.
3. The adaptive particle filtering method for maneuvering maritime targets as described in claim 1, characterized in that, The process of predicting and updating the MHM-PF filter with the multiple hypothesis concept based on the target measurement data includes: The hypothetical path at the current time is expanded based on the hypothetical path retained at the previous time step to obtain the hypothetical path retained at the current time step. Under the assumed path retained at the current moment, the state of each particle is updated by selecting the state transition function of the corresponding maneuvering target motion model; The particle weights are calculated based on the sampled particles at the current moment, and the scores of the hypothetical paths retained at the current moment are updated based on the particle weights. The scores are accumulated, and the hypothetical path with the highest score is retained to obtain the target state estimate of the MHM-PF filter.
4. The adaptive particle filtering method for maneuvering maritime targets as described in claim 1, characterized in that, The formula for obtaining the target's maneuver probability is: ； In the formula, Let k+1 be the target maneuver probability. Let be the target maneuver probability at time k. It is a smoothing constant. Let be the original target maneuver scoring function after fusion at time k+1.
5. The adaptive particle filtering method for maneuvering maritime targets as described in claim 1, characterized in that, The formula for obtaining the fused original target maneuver scoring function is as follows: ； In the formula, Let k+1 be the original target maneuver scoring function after fusion. , , All are fusion weights. The new information fraction at time k+1 The score is given for the change in velocity direction at time k+1. The entropy score of the model at time k+1.
6. The adaptive particle filtering method for maneuvering maritime targets as described in claim 1, characterized in that, The process of predicting and updating the IMM-PF filter based on the target measurement data includes: The weighted sum of the particle states of each model is obtained by summing the weighted sum of the particle states of model j. This sum is then fused with the covariance matrix of the state value of the nth particle of model i to obtain the covariance matrix of model j. The probability of model j is updated based on the covariance matrix of model j to obtain the transition probability weights of model j. The transition probability weights of model j are weighted and summed with the state value of the nth particle of model j to obtain the model mixture state estimate of the particle. The average of the model mixture state estimates of all the particles is used to obtain the target state estimate of the IMM-PF filter.
7. The adaptive particle filtering method for maneuvering maritime targets as described in claim 1 or 3, characterized in that, The formula for obtaining the target state estimate of the MHM-PF filter is as follows: ； In the formula, For target state estimation at time K using the MHM-PF filter, For particle weights, For the sampled particles at time k, This represents the number of particles in the hypothetical path.
8. The adaptive particle filtering method for maneuvering maritime targets as described in claim 1 or 6, characterized in that, The formula for obtaining the target state estimate of the IMM-PF filter is as follows: ； In the formula, This is the state estimate of the IMM-PF filter at time k, where N is the total number of particles. This is the model mixture state estimate for the nth particle at time k.
9. The adaptive particle filtering method for maneuvering maritime targets as described in claim 1, characterized in that, The three index scores include the innovation score, the velocity direction change score, and the model entropy score; The formula for obtaining the innovation fraction is: ； In the formula, The score for the new information at time k+1. To innovate the threshold constant, For the new information at time k+1, For the average innovation margin, The standard deviation is denoted as .
10. The adaptive particle filtering method for maneuvering maritime targets as described in claim 9, characterized in that, The formulas for obtaining the velocity direction change score and the model entropy score are as follows: ； ； In the formula, The score is given for the change in velocity direction at time k+1. The model entropy score at time k+1. Let be the normalized constant for velocity variation. The length of the sliding window. The change in a unit vector. Let k+1 be the information entropy. This represents the number of motion models for the maneuvering target.