Multi-target heterogeneous data centralized association fusion method and system

Abstract

The invention discloses a multi-target different-source data centralized association fusion method and system, and relates to the technical field of communication, and the method comprises the steps: predicting current data according to a historical moment data point of any target I in a multi-target cluster, obtaining k moment prediction data, and putting the k moment prediction data into a determination set; for the target I, traversing all measurement data points at the moment k, carrying out heterogenous data fusion on the measurement data points and data in the determined set, calculating the overall dispersion eta final of the measurement data points after fusion, and judging whether the measurement data points are put into the determined set, the to-be-determined set or not by using double thresholds; sequentially deleting elements with the maximum overall dispersion in the undetermined set, calculating data points after all data of the determined set and the undetermined set are fused until eta final after fusion is smaller than Umin, obtaining target I fusion data at the moment k, and storing the target I fusion data into a historical data set; k is made to be equal to k + 1, S1 to S2 are repeated, and a track result of the target I is obtained. According to the invention, direct association fusion of multi-target heterogeneous data is realized.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method and system for centralized association fusion of multi-target heterogeneous data. Background Art

[0002] Military detectors can be divided into active detectors and passive detectors according to the detection mode, and can be divided into optical detectors and radars according to the detection band. According to the characteristics of the information obtained by the detector, military detectors can be divided into three categories: positioning equipment, detection equipment, and observation and aiming equipment. The common equipment, information obtained, and characteristics of the three types of detectors are summarized in Table 1.

[0003] Table 1 Classification and characteristics of military detectors

[0004]

[0005] Positioning equipment usually uses active detection methods, generally by recording the time taken for electromagnetic waves to travel back and forth to measure the distance, and then determine the specific location of the target; detection equipment usually refers to alarm devices, which use passive detection methods to detect information related to electromagnetic waves emitted by military targets; observation and aiming equipment generally also uses passive detection methods to obtain external information of the target by taking images.

[0006] The data fusion of heterogeneous detectors requires correlation operations from multiple dimensions such as time, space, and target radiation characteristics. As can be seen from the information in Table 1, all three types of detectors can obtain the target's azimuth information, but only the positioning device can obtain the target's specific location. Therefore, the spatiotemporal correlation of heterogeneous data will be the most critical step in data fusion.

[0007] In a complex battlefield environment, when facing a multi-target cluster attack, it is necessary to effectively fuse the information obtained by multi-source heterogeneous detectors. At present, in the common heterogeneous detector data fusion method, due to the different information dimensions obtained by heterogeneous detectors, the correlation between data points is poor, and a distributed correlation fusion algorithm is often used, that is, first perform point track association on the same-source detector data, and then perform track fusion on the heterogeneous detector data. This method can match two tracks together more accurately, but its disadvantage is that the algorithm has a large amount of calculation and high complexity, and it is difficult to meet the real-time requirements of the battlefield. Therefore, there is an urgent need for a solution that can simply and directly fuse and associate heterogeneous data. Summary of the invention

[0008] The purpose of the present invention is to provide a method and system for the centralized association fusion of multi-target heterogeneous data, which reduces the amount of data fusion calculation on the basis of directly realizing the association fusion of heterogeneous data points, and can also realize the real-time fusion of data acquired by heterogeneous detectors.

[0009] To achieve the above object, the present invention provides the following solutions:

[0010] A method for centralized association fusion of multi-target heterogeneous data includes the following steps:

[0011] S1. Pre-classification of measurement data:

[0012] Predict the current data based on the historical data points of any target I in the multi-target cluster, obtain the predicted data at time k, and put it into the determination set;

[0013] For any target I in the multi-target cluster, the measurement data points in the k-time measurement data set are extracted one by one, and each measurement data point is fused with all the data in the determined set to calculate the overall discreteness η of the fused measurement data point. final , using the double threshold U max with U min Determine whether the measured data point should be put into the determined set or the pending set, or not put into it;

[0014] S2. Data fusion and reclassification:

[0015] The elements with the largest overall discreteness in the undetermined set are deleted one by one, and the data points after the fusion of all the data of the determined set and the undetermined set are calculated until the fused calculated η final Less than the lower threshold U min , and obtain the final fusion data of target I at time k;

[0016] S3. Store the fused data of the target I at time k into the historical data set, set k=k+1, and repeat steps S1-S2 to obtain the track result of the target I.

[0017] Furthermore, the method of predicting the current data according to the historical data point of any target I in the multi-target cluster to obtain the predicted data at time k specifically includes:

[0018] Based on the multi-domain extended Kalman filter algorithm, the predicted data at the previous moment is used to predict the state value at the current moment k, and the measured data at moment k is used to correct the state prediction value to obtain the prediction data at moment k; among them, for the data points that do not change with the target position, their state prediction value is the final calculated value at the previous moment; for the data points that change with the target position, the state prediction value is obtained by the following equation:

[0019]

[0020] Among them, F k-1 Indicates the value of variable F at the previous moment; That is, the predicted value at time k; R k-1 , R kare the distances between the target and the detector at time k-1 and k respectively; a is a variable, which is related to the characteristics of the quantity to be predicted. When the quantity to be predicted is a constant that does not change with distance, a=0; when the quantity to be predicted is the radiation energy in the passive detection device or the target size, a=-2; when the quantity to be predicted is the radiation energy in the active detection device, a=-4.

[0021] Furthermore, the overall discreteness η of the measured data points after calculation fusion is final , specifically including:

[0022] The overall dispersion of heterogeneous data indicates the degree of dispersion of n heterogeneous data points relative to the fusion point. The fusion point when the overall dispersion η takes the minimum value is the final result of the fusion of n heterogeneous data points, and then we get:

[0023] Y fusion =argmin(η(Y))

[0024] η finaal =min(η(Y))

[0025] Among them, Y fusion is the final fusion point; η final is the final overall discreteness; Y is the information of the fusion point, which is a variable, η(Y), which is obtained by the following formula:

[0026] Among them, α ij Represents the weight of the j-th dimension information in data point i. If data point i does not contain the j-th dimension information, α ij =0, otherwise, α ij =α j , α j Indicates the importance of the j-th dimension information;

[0027] Furthermore, the use of the double threshold U max with U min Determine whether the measured data point should be included in the determined set or the pending set, or not included, including:

[0028] According to the overall dispersion η of the fused data points final Classify all the measured data points at the k moment: if η final Less than the lower threshold U min , indicating that the data fusion degree is good, and it is determined to be the same target source data, so it is put into the determination set; if η final Greater than the lower threshold U min And less than the upper threshold U max , indicating that the data fusion degree is general, and it is possible that they are the same target source data, so they are put into the pending set; if η final Greater than the upper threshold U max, indicating that the data fusion level is very poor, and data from non-same target sources are not put into any set; among them:

[0029]

[0030] The present invention also provides a multi-target heterogeneous data centralized association fusion system, which is applied to the target heterogeneous data centralized association fusion method, comprising:

[0031] Pre-classification module, used for pre-classification of measurement data:

[0032] Predict the current data based on the historical data point of any target I in the multi-target cluster, obtain the predicted data at time k, and put it into the determination set;

[0033] For any target I in the multi-target cluster, all the measured data points at time k are traversed, and heterogeneous data are fused to calculate the overall dispersion η of the fused measured data points final , using the double threshold U max with U min Determine whether the measured data point should be put into the determined set or the pending set, or not put into it;

[0034] Data fusion and reclassification module, used for data fusion and reclassification:

[0035] The elements with the largest overall discreteness in the undetermined set are deleted one by one, and the data points after the fusion of all the data of the determined set and the undetermined set are calculated until the fused calculated η final Less than the lower threshold U min , and obtain the final fusion data of target I at time k;

[0036] The updating module is used to set k=k+1, and repeatedly perform the measurement data pre-classification and data fusion re-classification steps to obtain the track result of target I.

[0037] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the centralized association fusion method and system for multi-target heterogeneous data provided by the present invention uses a centralized fusion algorithm, that is, the heterogeneous detector data points are directly associated and fused, and creatively proposes multi-domain extended filtering, multi-criteria fusion, double threshold classification and other methods, which solves the problem that heterogeneous data points are difficult to directly associate and fuse with each other, reduces the amount of data fusion calculation, realizes the real-time fusion of heterogeneous detector data, and improves the track accuracy. In a complex battlefield environment, when facing a multi-target cluster attack, the information obtained by multi-source heterogeneous detectors can be effectively fused, thereby providing guidance for target identification, situation assessment, war decision-making and other processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0039] Figure 1 It is a schematic diagram of calculating the spatiotemporal coordinate fusion error of heterogeneous detectors of the present invention;

[0040] Figure 2 This is a flow chart of pre-classification of measurement data according to an embodiment of the present invention;

[0041] Figure 3 This is a flow chart of a dual threshold classification method according to an embodiment of the present invention;

[0042] Figure 4 This is a flow chart of data fusion and reclassification according to an embodiment of the present invention;

[0043] Figure 5 A schematic diagram of air defense reconnaissance confrontation in actual combat conditions in an embodiment of the present invention;

[0044] Figure 6 This is a flow chart of heterogeneous detector data fusion in actual combat situations in an embodiment of the present invention;

[0045] FIG7 is a target track fusion result of an embodiment of the present invention, wherein (a) represents the target track obtained by using the same source data, and (b) represents the target track obtained by using the solution of the present invention;

[0046] Figure 8 The present invention is a flowchart of a method for centralized association fusion of target heterogeneous data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The purpose of the present invention is to provide a method and system for the centralized association fusion of multi-target heterogeneous data, which reduces the amount of data fusion calculation on the basis of directly realizing the association fusion of heterogeneous data points, and can also realize the real-time fusion of data acquired by heterogeneous detectors.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] like Figure 1-8 As shown, the centralized association fusion method of multi-target heterogeneous data provided by the embodiment of the present invention includes the following steps:

[0051] S1. Pre-classification of measurement data:

[0052] Predict the current data based on the historical data points of any target I in the multi-target cluster, obtain the predicted data at time k, and put it into the determination set. The predicted data can be used as the initial reference value for pre-classifying the measured data;

[0053] For any target I in the multi-target cluster, the measurement data points in the k-time measurement data set are extracted one by one, and each measurement data point is fused with all the data in the determined set to calculate the overall discreteness η of the fused measurement data point. final , using the double threshold U max with U min Determine whether the measurement data point should be put into the determined set or the pending set, or not put into the determined set. The k-time measurement data set refers to the data of the target I detected by the detectors when the target I is located in the detection area of ​​the detectors.

[0054] It can be seen from this that the determination set is dynamically changing and new data will be continuously added, that is, the determination set contains the k-time prediction data and the k-time measurement data that have been put into the determination set.

[0055] S2. Data fusion and reclassification:

[0056] The overall dispersion of the undetermined set is deleted one by one (η required in the pre-classification process of the measurement data) final ) the largest element, calculate the data point after the fusion of all the data of the determined set and the undetermined set, until the fusion calculated η final Less than the lower threshold U min , and obtain the final fusion data of target I at time k.

[0057] S3. Store the fused data of the target I at time k into the historical data set, set k=k+1, and repeat steps S1-S2 to obtain the track result of the target I.

[0058] For different targets, the above method is adopted to select the determined set and the undetermined set corresponding to each target from all the measurement data, and then fuse and calculate the data of each target at the next moment. For ease of understanding, the following example is given: for each target (target A, target B, target C...), when executing steps S1-S3, the k-time measurement data set used is all the measurement data at k time (including the data of targets A, B, and C). For example, when predicting the track result of target A, in steps S1-S3, the data of targets B and C in the k-time measurement data set will be gradually discarded, and finally the data of target A will be left and fused to generate the data information of target A at k time, and finally the track result of target A will be obtained.

[0059] About the overall dispersion η final :

[0060] In order to facilitate the fusion classification of heterogeneous data, it is necessary to introduce the concept of overall dispersion η of heterogeneous data, which represents the degree of dispersion of heterogeneous data points relative to the fusion point. Borrowing the concept of mean relative error (MRE), for n data points, the overall dispersion η is calculated by the single-point dispersion η of each data point relative to the fusion point. i (i=1,2,…,n) to find the average. Since the discreteness of data points needs to be considered from multiple dimensions such as position, energy, and frequency, the single point discreteness η i It is also determined by the basic discreteness η of data point i in m dimensions ij (j=1,2,…,m) weighted average, the heterogeneous data dispersion η is expressed as follows:

[0061]

[0062] Where: α ij represents the weight of the j-th dimension information in data point i. If data point i does not contain the j-th dimension information (for example, the radar measurement data does not contain the frequency dimension information of the target radiation source), α ij =0, otherwise, α ij =α j , α j That is, the importance of the j-th dimension information can be adjusted according to the actual application scenario.

[0063] Any parameter of heterogeneous data points has a certain error and satisfies the normal distribution: Y~N(μ,σ). Standardization is performed to obtain the standard normal distribution: This standardized normal distribution can be used to represent the basic discreteness of data point i in dimension j:

[0064]

[0065] Where: Y rawis the information of the data point before fusion; Y is the information of the fusion point, which is a variable; σ ij is the standard deviation of the j-th dimension information of data point i.

[0066] In the m-dimensional information, the target location information is different from other information and needs to be calculated separately in different coordinate systems, such as Figure 1 The direction or position information of the target obtained by the three types of detectors are all based on data in different spherical coordinate systems; the position information of the predicted point is based on the geodetic coordinate system or the geocentric coordinate system.

[0067] The overall dispersion of heterogeneous data indicates the degree of dispersion of n heterogeneous data points relative to the fusion point. The lower the overall dispersion η, the more accurate the fusion point and the better the fusion effect. Therefore, the fusion point when the overall dispersion η takes the minimum value is the final result of the fusion of n heterogeneous data points. It is expressed as follows:

[0068]

[0069] Where: Y fusion is the final fusion point; η final is the final overall discreteness, which is an important criterion for whether n heterogeneous data points can be fused.

[0070] In this embodiment, compared with distributed fusion, the main problem of centralized fusion adopted in this embodiment is that the accuracy of data association is not high enough, and individual point classification errors are prone to occur, which is particularly obvious in the fusion of heterogeneous information. In order to improve the accuracy of association, this embodiment proposes a two-step classification method of pre-classification of measurement data of heterogeneous data points and data fusion reclassification.

[0071] In this embodiment, the dual threshold U max with U min Determine whether the measured data point should be included in the determined set or the pending set, or not included, including:

[0072] like Figure 3 As shown, according to the overall discreteness η of the fused data points final Classify all the measured data points at the k moment: if η final Less than the lower threshold U min , indicating that the data fusion degree is good, and it is determined to be the same target source data, so it is put into the determination set; if η final Greater than the lower threshold U min And less than the upper threshold U max , indicating that the data fusion degree is general, and it is possible that they are the same target source data, so they are put into the pending set; if η final Greater than the upper threshold U max, indicating that the data fusion level is very poor, and data from sources other than the same target are not put into any set;

[0073] Double Threshold U max with U min The value of is related to the distribution of η(Y). ij It obeys the standard normal distribution. Combined with formula (1), mn·η(Y) can be regarded as a chi-square distribution with mn degrees of freedom:

[0074] mnη(Y)~χ 2 (mn) (4)

[0075] The double threshold values ​​are assigned with the probability of 0.75 and 0.25, and the 2 Distribution probability table, we can get:

[0076]

[0077] In this embodiment, predicting the current data according to the historical data point of any target I in the multi-target cluster to obtain the predicted data at time k specifically includes:

[0078] According to whether the information of the data points at the historical moment changes with the target position, the data points at the historical moment are divided into two categories, one is the data points that do not change with the target position (such as radiation source frequency, pulse width, etc.), and the other is the data points that change with the target position, such as radiation energy, target size, etc., and the multi-domain extended Kalman filter algorithm is used to smooth them, predict the data points at time k, and obtain the target state prediction value at time k. For information that does not change with the target position, the prediction value is the final calculated value of the previous moment; for information that changes with the target position, the prediction value is determined according to the law of the data changing with distance. The prediction equation is expressed as follows:

[0079]

[0080] Among them, F k-1 Indicates the value of variable F at the previous moment; That is, the predicted value at time k; R k-1 , R k are the distances between the target and the detector at time k-1 and k respectively; a is a variable, which is related to the characteristics of the quantity to be predicted. When the quantity to be predicted is a constant value that does not change with distance, a=0; when the quantity to be predicted is the radiation energy in the passive detection device or the target size, a=-2; when the quantity to be predicted is the radiation energy in the active detection device, a=-4. The multi-domain extended Kalman filter algorithm is as follows:

[0081] The purpose of filtering is to smooth the current and past states of the target and predict the state parameters of the future. The Kalman filter algorithm has small computational complexity, small storage volume, and is easy to implement. Therefore, it is widely used in radar tracking systems and has become the theoretical and application basis for target tracking problems.

[0082] The Kalman filter algorithm includes a state transfer model and a measurement model. The state transfer model can use the estimated information at time k-1 to predict the target state at time k. The measurement model uses the actual measurement value of the sensor to correct the state prediction value.

[0083] For a random linear system under standard assumptions, the discrete system state space equation without considering the control input is:

[0084]

[0085] Where: x k is the state vector; z k is the observation vector; Φ (k,k-1) is the state transfer matrix; Γ (k,k-1) is the process noise distribution matrix; w k-1 and v k are the process noise vector and the observation noise vector respectively, and the two are independent zero-mean Gaussian white noises; H is the matrix for converting the state vector to the observation vector.

[0086] Based on the above state transfer equation and observation equation, the main steps of the linear Kalman filter algorithm are as follows:

[0087]

[0088] Common Kalman filter algorithms usually only predict the target's next moment's point position, speed, acceleration and other spatial information. The present invention performs multi-domain expansion on the Kalman filter algorithm so that it can predict the target's next moment's radiation source information (including energy, frequency, pulse width, etc.) and image information (including target category, target size, etc.), and needs to establish a conversion matrix H for the value to be predicted.

[0089] In this embodiment, Figure 4 As shown, the S4, performing data fusion and reclassification: reclassifying by means of data fusion, and obtaining the data information of each target at time k, specifically includes the following steps:

[0090] According to the analysis of the characteristics of the three types of detectors in Table 1, different types of detectors have different combat distances, different numbers of targets that can be monitored, and different measurement accuracy. The air defense reconnaissance confrontation situation in actual combat is as follows: Figure 5As shown in the figure, the detection range of the reconnaissance equipment is the longest, but the angle measurement accuracy is the lowest, and the search range is larger; the detection range of the observation and aiming equipment is the shortest, but the angle measurement accuracy is the highest, the field of view is very small, and it cannot detect the target alone; therefore, in an actual combat environment, a long-range detector is needed to guide the short-range detector, and the short-range detector in turn fuses and corrects the target track information.

[0091] like Figure 6 As shown, k i Indicates the time when the target enters the combat range of detector i. The final target's complete track information is spliced ​​together from multiple pieces of information, and the later the time, the more detector data is integrated, and the higher the accuracy of the target track.

[0092] Based on the above multi-target heterogeneous data centralized association fusion method, this embodiment provides an application example of the heterogeneous data centralized association fusion algorithm, which is as follows:

[0093] The simulation examples are taken from the relevant data examples of the National Big Data and Decision-making Laboratory's real-time track association and track fusion competition. The competition focuses on studying the track association problem between 2D radar and ESM. 2D radar can obtain the distance and azimuth information of the observed target (ship) within the detection range, while ESM can only obtain the azimuth, carrier frequency, pulse width and repetition rate information of the electronic equipment within the detection range but cannot obtain the distance information. Therefore, there is a large uncertainty in the track association. In reality, the trajectory information collected by a single observation station cannot meet the application requirements. It is necessary to use data fusion technology to merge the trajectories collected by multiple observation stations to expand the monitoring range, reduce the information ambiguity of the system, and improve the reliability of the system decision.

[0094] The competition sets up three observation stations, each of which contains two 2D radars and one ESM sensor. The detection ranges of each detector will partially overlap but not be covered, and all have errors. The geographical location, detection range, and error of each detector are given in the data file. The optimal GOSPA estimation is used to evaluate the similarity between the generated trajectory set and the true trajectory set, while taking into account the position error and the target number error. The specific calculation formula for the optimal GOSPA estimation is as follows:

[0095]

[0096] Where c>0, p>1, X represents the fused track, Y represents the real track, x and y represent a specific track in the set, d(x i ,y j ) represents x i With y j The similarity between .

[0097] The algorithm in this paper is applied to this example. The results are shown in Figure 7. Figure (a) shows the target track obtained by using only homologous data without using heterogeneous data fusion. The value is 0.837, and the track is rough (especially the lower left corner has large fluctuations). Figure (b) shows the target track obtained by using the algorithm in this paper. It can be seen that the target track is smooth in the figure, the value is 0.938, and the track accuracy is improved by 12.1%.

[0098] In summary, the centralized association fusion algorithm proposed in the present invention can easily meet the real-time requirements of this example. In this example, the data fusion center receives a maximum of about 2,000 pieces of data information every 6 seconds, the average operation time of a single frame of the algorithm is 1.18 seconds, and the maximum time consumption is 1.53 seconds, which meets the real-time requirements.

[0099] Example 2

[0100] The embodiment of the present invention provides a multi-target heterogeneous data centralized association fusion system, which is applied to the multi-target heterogeneous data centralized association fusion method described in Example 1, including:

[0101] Pre-classification module, used to pre-classify measurement data:

[0102] Predict the current data based on the historical data point of any target I in the multi-target cluster, obtain the predicted data at time k, and put it into the determination set;

[0103] For any target I in the multi-target cluster, all the measured data points at time k are traversed, and heterogeneous data are fused to calculate the overall dispersion η of the fused measured data points final , using the double threshold U max with U min Determine whether the measured data point should be put into the determined set or the pending set, or not put into it;

[0104] The data fusion reclassification module is used to perform data fusion reclassification: gradually delete the elements with the largest overall discreteness in the undetermined set, calculate the data points after all the data of the determined set and the undetermined set are fused, until the η calculated by fusion is final Less than the lower threshold U min , and obtain the final fusion data of target I at time k;

[0105] The updating module is used to set k=k+1, repeat the pre-classification, data fusion and re-classification steps, and obtain the track result of target I.

[0106] In summary, the centralized association fusion method and system for multi-target heterogeneous data provided by the present invention, by performing multi-domain expansion on the Kalman filter algorithm, enables it to predict the radiation source information (including energy, frequency, pulse width, etc.) and image information (including target category, target size, etc.) of the target at the next moment.

[0107] The remaining technical features in this embodiment can be flexibly selected by those skilled in the art according to actual conditions to meet different specific practical needs. However, it is obvious to those skilled in the art that it is not necessary to adopt these specific details to implement the present invention. In other examples, in order to avoid confusing the present invention, the well-known components, structures or parts are not specifically described, which are all within the technical protection scope defined by the technical solution claimed for protection in the claims of the present invention.

[0108] Modifications and changes made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the scope of protection of the claims attached to the present invention. In the above description, a large number of specific details are set forth in order to provide a thorough understanding of the present invention. However, it is obvious to those of ordinary skill in the art that these specific details are not necessary to practice the present invention. In other examples, in order to avoid confusing the present invention, well-known technologies, such as specific construction details, operating conditions and other technical conditions, are not specifically described.

[0109] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for centralized association fusion of multi-target heterogeneous data, characterized in that: The following steps are involved: S1. Pre-classification of measurement data: Predict the current data based on the historical data points of any target I in the multi-target cluster, obtain the predicted data at time k, and put it into the determination set; For any target I in the multi-target cluster, the measurement data points in the k-time measurement data set are extracted one by one, and each measurement data point is fused with all the data in the determined set to calculate the overall discreteness η of the fused measurement data point. final , using the double threshold U max with U min Determine whether the measured data point should be put into the determined set or the pending set, or not put into it; S2. Data fusion and reclassification: The elements with the largest overall discreteness in the undetermined set are deleted one by one, and the data points after the fusion of all the data of the determined set and the undetermined set are calculated until the fused calculated η final Less than the lower threshold U min , and obtain the final fusion data of target I at time k; S3. Store the fused data of the target I at time k into the historical data set, set k=k+1, and repeat steps S1-S2 to obtain the track result of the target I.

2. The method for centralized association fusion of multi-target heterogeneous data according to claim 1 is characterized in that: The method of predicting current data according to a historical data point of any target I in the multi-target cluster to obtain predicted data at time k specifically includes: Based on the multi-domain extended Kalman filter algorithm, the predicted data at the previous moment is used to predict the state value at the current moment k, and the measured data at moment k is used to correct the state prediction value to obtain the prediction data at moment k; among them, for the data points that do not change with the target position, their state prediction value is the final calculated value at the previous moment; for the data points that change with the target position, the state prediction value is obtained by the following equation: Among them, F k-1 Indicates the value of variable F at the previous moment; That is, the predicted value at time k; R k-1 , R k are the distances between the target and the detector at time k-1 and k respectively; a is a variable, which is related to the characteristics of the quantity to be predicted. When the quantity to be predicted is a constant that does not change with distance, a=0; when the quantity to be predicted is the radiation energy in the passive detection device or the target size, a=-2; when the quantity to be predicted is the radiation energy in the active detection device, a=-4.

3. The method for centralized association fusion of multi-target heterogeneous data according to claim 1, characterized in that: The overall discreteness η of the measured data points after calculation fusion final , including: The overall dispersion of heterogeneous data indicates the degree of dispersion of n heterogeneous data points relative to the fusion point. The fusion point when the overall dispersion η takes the minimum value is the final result of the fusion of n heterogeneous data points, and then we get: AND fusion =argmin(η(Y)) or final =min(η(Y)) Among them, Y fusion is the final fusion point; η final is the final overall discreteness; Y is the information of the fusion point, which is a variable, η(Y), which is obtained by the following formula: Among them, α ij Represents the weight of the j-th dimension information in data point i. If data point i does not contain the j-th dimension information, α ij =0, otherwise, α ij =α j , α j Indicates the importance of the j-th dimension information.

4. The method for centralized association fusion of multi-target heterogeneous data according to claim 1, characterized in that: The use of double threshold U max with U min Determine whether the measured data point should be included in the determined set or the pending set, or not included, including: According to the overall dispersion η of the fused data points final Classify all the measured data points at the k moment: if η final Less than the lower threshold U min , indicating that the data fusion degree is good, and it is determined to be the same target source data, so it is put into the determination set; if η final Greater than the lower threshold U min And less than the upper threshold U max , indicating that the data fusion degree is general, and it is possible that they are the same target source data, so they are put into the pending set; if η final Greater than the upper threshold U max , indicating that the data fusion level is very poor, and data from non-same target sources are not put into any set; among them:

5. A multi-target heterogeneous data centralized association fusion system, applied to the target heterogeneous data centralized association fusion method according to any one of claims 1 to 3, characterized in that: include: Pre-classification module, used for pre-classification of measurement data: Predict the current data based on the historical data point of any target I in the multi-target cluster, obtain the predicted data at time k, and put it into the determination set; For any target I in the multi-target cluster, all the measured data points at time k are traversed, and heterogeneous data are fused to calculate the overall dispersion η of the fused measured data points final , using the double threshold U max with U min Determine whether the measured data point should be put into the determined set or the pending set, or not put into it; Data fusion and reclassification module, used for data fusion and reclassification: The elements with the largest overall discreteness in the undetermined set are deleted one by one, and the data points after the fusion of all the data of the determined set and the undetermined set are calculated until the fused calculated η final Less than the lower threshold U min , and obtain the final fusion data of target I at time k; The updating module is used to set k=k+1, and repeatedly perform the measurement data pre-classification and data fusion re-classification steps to obtain the track result of target I.

Images