A multi-source heterogeneous information association method based on information fusion cycle

Abstract

The present invention relates to a multi-source heterogeneous information association method based on an information fusion cycle, belongs to the field of information fusion technology, and solves the problem of association errors under bifurcation and cross-motion of existing dense targets. It includes receiving multi-source information of multiple different types of sensors; adding multi-source information containing multiple data of the same type of sensor to a to-be-clustered set, and adding other multi-source information to a preliminary clustering set; clustering the to-be-clustered set, adding the multi-source information closest to the cluster center in each class to the preliminary clustering set, and putting other multi-source information in each class into a cache area, and marking those not classified into any class as discrete points and storing them in a history library; obtaining the preliminary clustering set and the associated data pairs in the history library based on identity information and dynamic information; adding all multi-source information in the cache area that belongs to the same information fusion cycle and the same type as the multi-source information of the preliminary clustering set to the associated data pairs to obtain the association results. Accurate association of multi-source heterogeneous information in dense targets is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of information fusion, and in particular to a multi-source heterogeneous information association method based on an information fusion cycle. Background Art

[0002] As the battlefield expands in the future, the types and number of sensors continue to increase, and the types of information become richer, the requirements for real-time information fusion are becoming higher and higher. At the same time, heterogeneous multi-source information fusion has become a trend in the development of future information fusion. The multi-source information association problem is one of the key technologies of information fusion. Without information association, there is no way to track the target. Without efficient and reliable information association, the correctness of target recognition is out of the question.

[0003] Multi-source heterogeneous information association is to merge observations or traces from multiple sensors with known or confirmed events so that they belong to the event set, that is, to ensure that the observations contained in each event set come from the same entity with a high probability or a probability close to 1. The traces that are not merged, which may include new traces from the target or traces generated by noise or clutter, are retained until the next moment for further processing.

[0004] In essence, association is achieved through an m-dimensional quantitative processing, which quantifies the spatial or attribute relationship between the observed and predicted target states to determine which of the m hypotheses can best describe the observation. In the field of multi-sensor information fusion, a single target is generally used as the object of association processing, and the information association method based on target motion information is mainly adopted. The technologies that can be adopted mainly include static association methods, such as measurement-measurement association of the same or different dimensions, and dynamic information association methods, such as measurement-track association, track-track association, etc. Specific association methods include the nearest neighbor method, joint probability information association method, all-neighbor optimal filter, multi-hypothesis method, interactive multi-model method, S-dimensional allocation method, etc. Among them, the nearest neighbor method is the most typical representative. Its basic idea is to establish a mapping relationship between the two closest measurement information and the target number. It is simple to implement, has a small amount of calculation, and is widely used in practical engineering.

[0005] However, the nearest neighbor method has the following shortcomings: first, the correlation gate relies heavily on experience selection or expert formulation; second, when dense multiple targets are in a state of motion such as crossing or bifurcation, the result of "mistaking one for another" may occur; third, when the target information is sparse discrete points, the traditional practice is to delete it directly and not use it as evidence for target identification. However, sparse discrete points, such as the information of the other party's target obtained by the detection platform, play an important role in target identification. The traditional approach may miss associations, which in turn leads to a decrease in the target recognition rate. Summary of the invention

[0006] In view of the above analysis, an embodiment of the present invention aims to provide a multi-source heterogeneous information association method based on information fusion cycle to solve the problem of incorrect association under existing dense target bifurcation and cross-motion.

[0007] On the one hand, an embodiment of the present invention provides a multi-source heterogeneous information association method based on an information fusion cycle, comprising the following steps:

[0008] Receive multi-source information from multiple different types of sensors, including identity information and dynamic information;

[0009] According to the information fusion cycle, multi-source information containing multiple data from the same type of sensors is added to the set to be clustered, and other multi-source information is added to the preliminary clustering set;

[0010] Cluster the cluster set, add the multi-source information closest to the cluster center in each class to the preliminary cluster set, put the other multi-source information in each class into the cache area, and mark the multi-source information that is not classified into any class as a discrete point and store it in the history library;

[0011] Based on the identity information and dynamic information, the associated data pairs of the multi-source information in the preliminary clustering set and the multi-source information in the historical database are obtained;

[0012] According to the multi-source information of the preliminary clustering set in the associated data pair, all the multi-source information of the same type and belonging to the same information fusion cycle in the cache area are added to the associated data pair, and the same track number is set for each group of associated data to obtain the associated result.

[0013] Based on the further improvement of the above method, based on the identity information and the dynamic information, the associated data pairs of the multi-source information in the preliminary clustering set and the multi-source information in the historical library are obtained, including: firstly based on the same identity information, the associated data pairs of the multi-source information in the preliminary clustering set and the multi-source information in the historical library are identified, and then based on the dynamic information, the remaining multi-source information in the preliminary clustering set and the remaining multi-source information in the historical library are associated, wherein, based on the dynamic information, the remaining multi-source information in the preliminary clustering set and the remaining multi-source information in the historical library are associated, including:

[0014] According to the preset dynamic information association threshold, candidate historical points are obtained from the remaining multi-source information in the history library;

[0015] Based on the dynamic information, the correlation gate between the remaining multi-source information in the preliminary clustering set and the multi-source information in the candidate historical traces is calculated, and the correlation data pairs between the remaining multi-source information in the preliminary clustering set and the multi-source information in the candidate historical traces are identified according to the correlation gate.

[0016] Based on the further improvement of the above method, clustering the cluster set includes: firstly identifying the same type of multi-source information in the same information fusion cycle based on the same identity information, and taking the information fusion cycle as the cluster center, and then clustering the remaining multi-source information in the cluster set; wherein clustering the remaining multi-source information in the cluster set includes:

[0017] Traverse the remaining multi-source information in the set to be clustered, and determine multiple cluster centers and information sets in turn;

[0018] Calculate the average value of multi-source information in each type of information set;

[0019] Calculate the sum of squares of errors between each piece of multi-source information and the average value of multi-source information in each type of information set as the clustering criterion function value;

[0020] When the clustering criterion function value is less than the preset error threshold, or the difference between two consecutive clustering criterion function values ​​is less than the error threshold, the iteration is exited, otherwise it returns to determine the new cluster center and information set.

[0021] Based on the further improvement of the above method, the remaining multi-source information in the set to be clustered is traversed to determine multiple cluster centers and information sets in turn, including:

[0022] Select a piece of multi-source information from the remaining multi-source information in the set to be clustered as the initial center, and according to the preset initial distance threshold, obtain the multi-source information whose Euclidean distance from the initial center is less than the initial distance threshold as the set to be selected, select each piece of multi-source information in the set to be selected in turn, calculate the number of multi-source information whose Euclidean distance is less than the initial distance threshold, take the multi-source information with the largest number as the first cluster center, and the multi-source information whose Euclidean distance from the first cluster center is less than the initial distance threshold as the first category of information set;

[0023] The Euclidean distances between the multi-source information except the first type of information set and the first cluster center are calculated in sequence, and the multi-source information with the largest distance is selected as the second cluster center. If the maximum distance is greater than the preset minimum distance, the difference between the maximum distance and the initial distance threshold is taken as the second cluster distance, and the multi-source information with the Euclidean distance from the second cluster center less than the second cluster distance is obtained as the second type of information set;

[0024] In this way, the mth cluster center is calculated until the Euclidean distance between the m+1th cluster center and the mth cluster center is less than the minimum distance. The iteration ends and m cluster centers and m types of information sets are obtained.

[0025] Based on the further improvement of the above method, the sum of square errors between each multi-source information and the average value of the multi-source information in each type of information set is calculated as the clustering criterion function value. The formula is:

[0026]

[0027] Among them, m represents the total number of cluster centers, C i represents the i-th cluster center, i=1,…,m, SenInfo represents each multi-source information in each type of information set, Represents the average value of multi-source information in the i-th information set.

[0028] Based on the further improvement of the above method, the preset dynamic information association thresholds include: longitude association threshold, latitude association threshold and altitude association threshold;

[0029] Obtain candidate historical points from the remaining multi-source information in the history library, including:

[0030] Take out each remaining multi-source information in the preliminary clustering set in turn, and compare them with each remaining multi-source information in the historical database, and obtain the longitude difference, latitude difference and altitude difference of the dynamic information in the multi-source information. When the absolute values ​​of the three differences are all less than the corresponding association thresholds, the multi-source information in the historical database is included in the candidate historical points, and the Kalman filtering method is used to predict and extrapolate the track points related to the multi-source information in the historical database, and extrapolate to the acquisition time of the multi-source information in the preliminary clustering set compared with it.

[0031] Based on the further improvement of the above method, the correlation wave gate includes a distance correlation wave gate and an azimuth correlation wave gate;

[0032] Based on the dynamic information, the correlation gate between the remaining multi-source information in the preliminary clustering set and the multi-source information in the candidate historical traces is calculated, and the correlation data pairs in the preliminary clustering set and the candidate historical traces are identified according to the correlation gate, including:

[0033] Take out each remaining multi-source information in the preliminary clustering set in turn, and each multi-source information in the candidate historical point trace, and calculate the distance association gate and azimuth association gate of the multi-source information, as well as the distance difference and azimuth difference. When the absolute value of the distance difference is less than the distance association gate, and the absolute value of the azimuth difference is less than the azimuth association gate, the two multi-source information are successfully associated, otherwise the association fails, and the multi-source information that fails to be associated in the preliminary clustering set is marked as a discrete point and stored in the history library.

[0034] Further improvements based on the above method also include:

[0035] The multi-source information marked as discrete points in the history database is used as discrete point traces, and the other multi-source information in the history database is used as historical point traces;

[0036] Based on identity information and proprietary information, the associated data pairs of discrete traces and historical traces are obtained, and then based on time and dynamic information, multiple similarity matrices are established for the remaining discrete traces and the remaining historical traces in the historical database;

[0037] According to the preset measurement threshold of each similarity matrix, multiple similarity matrices are converted into multiple 0-1 measurement matrices; the number of element values ​​​​of the same position in multiple 0-1 measurement matrices is counted, and when the number is greater than the preset number threshold, the discrete point traces corresponding to the position are successfully associated with the historical point traces and added to the associated data pair;

[0038] The same track number is set for each set of associated data pairs to obtain the associated result.

[0039] Based on the further improvement of the above method, the rows of the similarity matrix correspond to each discrete point trace, the columns correspond to each historical point trace, and the element values ​​correspond to the comparison values ​​between each discrete point trace and each historical point trace; the similarity matrix includes: time similarity matrix and dynamic information similarity matrix; the multi-source information comparison values ​​include: time information comparison values ​​and dynamic information comparison values.

[0040] Based on the further improvement of the above method, according to the preset measurement threshold of each similarity matrix, multiple similarity matrices are converted into multiple 0-1 measurement matrices, including:

[0041] Under the condition that the measurement threshold of the preset time similarity matrix is ​​less than the preset time similarity matrix, the element values ​​of the time similarity matrix are traversed, and under the condition that the measurement threshold of the dynamic similarity matrix is ​​less than the preset dynamic similarity matrix, the element values ​​that meet the conditions are obtained respectively, and then according to the rule of taking the minimum value for each row and the maximum value for each column, the element values ​​that meet the conditions are optimized again, and the element values ​​that meet the rules are set to 1, otherwise, the element values ​​are set to 0, and the 0-1 time measurement matrix corresponding to the time similarity matrix and the 0-1 dynamic information measurement matrix corresponding to the dynamic information similarity matrix are obtained respectively.

[0042] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0043] 1. For multi-source heterogeneous information, the association process of classified and hierarchical processing is carried out through the methods of clustering multi-source information in the same period, associating information in different periods and associating discrete points, which reduces the computational complexity of multi-target and multi-source information association in dense environments and solves the problem of high complexity of information association;

[0044] 2. For multi-source information in the same period, the initial cluster center is dynamically selected based on the maximum density iterative processing method through identity information matching and improved fuzzy C-means clustering, which solves the problem that the initial cluster center is highly dependent on artificial and prior information, realizes fuzzy clustering of multiple information from the same sensor in the same period and rough selection of sparse discrete points, and solves the problems of low utilization rate of existing identity information and difficulty in determining sparse points;

[0045] 3. For heterocyclic information association, a dynamic design criterion for the association gate is designed, and a method based on identity information matching and dynamic association of multi-cycle and multi-dimensional information is implemented, which solves the problem of erroneous association that may occur under dense multi-target and complex motion conditions.

[0046] 4. Through the construction of the multi-dimensional similarity matrix of discrete points and the design of association optimization rules, the problems of difficulty in associating discrete points and missed associations are solved, providing the premise and guarantee for improving the target recognition rate.

[0047] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0049] Figure 1 This is a flow chart of a multi-source heterogeneous information association method based on an information fusion cycle in Embodiment 1 of the present invention;

[0050] Figure 2 This is a flow chart of a method for associating discrete points in multi-source heterogeneous information in Example 2 of the present invention. DETAILED DESCRIPTION

[0051] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0052] A specific embodiment of the present invention discloses a multi-source heterogeneous information association method based on information fusion cycle, such as Figure 1 As shown, the following steps are included:

[0053] S11: receiving multi-source information from multiple sensors of different types, including identity information and dynamic information;

[0054] It should be noted that this embodiment receives multi-source information radiated, emitted, and transmitted by multiple sensors of different types, different working mechanisms, and different data rates equipped on the target platform through an information association processing platform, such as acquisition time, dynamic information, identity information, and proprietary information, and stores the multi-source information into the target information library with the sensor type as the index.

[0055] Assume that the information fusion processing cycle is T, the multi-source information is MultInfo, and each sensor information is SenInfo. SenInfo consists of sensor type senClass, information acquisition time t, target identity information staticInfo, target dynamic information dynaInfo, and target-specific information specialInfo acquired by the sensor. The target dynamic information dynaInfo includes: target longitude senLon, target latitude senLat, target altitude senAlti, target distance senDis, target azimuth senAzi, and target pitch senPitch. That is:

[0056] SenInfo={senClass,t,senLon,senLat,senAlti,senDis,senAzi,senPitch,staticInfo,specialInfo}

[0057] Exemplarily, the target identity information is a 24-bit address code of the sensor, and the proprietary information is an icon of the sensor.

[0058] The i-th information fusion processing cycle T i The internal multi-source information can be recorded as:

[0059]

[0060] Where N represents the number of sensor categories in the i-th cycle, M j Indicates the nth j The number of multi-source information of sensor type, n j represents the jth sensor category, Indicates the nth j Type t of sensor k Multi-source information at all times.

[0061] The information fusion processing cycle is related to the performance of the sensor and the processing efficiency of the information association processing platform, and is exemplarily set to 40 ms.

[0062] S12: according to the information fusion cycle, multi-source information containing multiple data of the same type of sensor is added to the to-be-clustered set, and other multi-source information is added to the preliminary clustering set;

[0063] It should be noted that for the same information fusion period T i If the same type of sensor contains multiple data, the multi-source information is added to the set to be clustered, and step S13 is executed to cluster the same period information; if the same type of sensor contains only one data, it is added to the preliminary clustering set, and step S14 is executed to associate different periods.

[0064] S13: clustering the cluster set to be processed, adding the multi-source information closest to the cluster center in each class to the preliminary cluster set, putting the other multi-source information in each class into the buffer area, and storing the multi-source information that is not classified into any class into the history library as discrete information;

[0065] It should be noted that clustering the cluster set includes: firstly identifying the same type of multi-source information in the same information fusion cycle based on the same identity information, taking the information fusion cycle as the cluster center, and then clustering the remaining multi-source information in the cluster set.

[0066] Specifically, the multi-source information in the cluster set is first matched by identity information. Assume that for the i-th fusion processing cycle T i Get the nth j Class sensor information includes M j Multi-source information, M j >1, and the identity information in the multi-source information is not empty, then traverse multiple multi-source information, compare them two by two, and classify the multi-source information with the same identity information into one category, and use the information fusion cycle T i As the cluster center, select the distance T i The most recent r Multi-source information at all times As the clustering result, add the preliminary clustering set, where t r The calculation formula is as follows:

[0067]

[0068] Where r represents the number of moments, r=1,2,...,M j .

[0069] In this embodiment, considering that identity information is the unique identifier of multi-source information, the association efficiency of multi-source information is improved by preferentially matching identity information. If the identity information fails to match, it means that the identity information of the target does not always exist. Then, the dynamic information in the multi-source information is used to cluster the remaining multi-source information in the clustering set, including:

[0070] S131: traverse the remaining multi-source information in the set to be clustered, and determine multiple cluster centers and information sets in sequence;

[0071] Specifically, the process of determining cluster centers and information sets includes:

[0072] ① Select a piece of multi-source information from the remaining multi-source information in the set to be clustered as the initial center, and according to the preset initial distance threshold d1, obtain the multi-source information whose Euclidean distance with the initial center is less than the initial distance threshold as the selected set, select each multi-source information in the selected set in turn, calculate the number of multi-source information whose Euclidean distance is less than the initial distance threshold, take the multi-source information with the largest number as the first cluster center C1, and the multi-source information whose Euclidean distance with the first cluster center is less than the initial distance threshold is taken as the first category of information set; It should be noted that the initial cluster center is dynamically selected based on the maximum density iterative processing method, which solves the problem of the strong dependence of the initial cluster center on artificial and prior information.

[0073] ② Calculate the Euclidean distance between the multi-source information except the first type of information set and the first cluster center C1 in turn, and select the multi-source information with the largest distance as the second cluster center C2. If the maximum distance is greater than the preset minimum distance d min , take the difference between the maximum distance and the initial distance threshold as the second clustering distance, and obtain the multi-source information whose Euclidean distance to the second cluster center is less than the second clustering distance as the second type of information set;

[0074] ③ And so on, calculate the mth cluster center C m , until the Euclidean distance between the m+1th cluster center and the mth cluster center is less than the minimum distance d min When , the iteration ends, and m cluster centers and m types of information sets are obtained.

[0075] S132: Calculate the average value of multi-source information in each type of information set;

[0076] For each cluster center, the average value of all multi-source information in each type of information set is calculated using the following formula:

[0077]

[0078] Among them, C i represents the i-th cluster center, i=1,2,...,m, It represents the average value of multi-source information in the i-th information set, and SenInfo represents each piece of multi-source information in each information set.

[0079] Preferably, since there are multiple attribute values ​​in the multi-source information, the target distance senDis may be selected to calculate the average value.

[0080] S133: Obtain the sum of square errors between each piece of multi-source information and the average value of the multi-source information in each type of information set as the clustering criterion function value, and the formula is as follows:

[0081]

[0082] S134: When the clustering criterion function value is less than the preset error threshold, or the difference between two adjacent clustering criterion function values ​​obtained is less than the error threshold, exit the iteration; otherwise, return to step S131, select a new multi-source information from the remaining set to be clustered as the initial center, and determine the new clustering center and information set again.

[0083] After steps S131 to S134, the multi-source information closest to the cluster center in each category is added to the preliminary clustering set, the other multi-source information in each category is put into the cache area, and the multi-source information not classified into any category is marked as discrete information and stored in the history library.

[0084] Preferably, different flags are set to indicate the status of the multi-source information, and a two-dimensional data structure is used to store the different flags of the multi-source information and the information corresponding to each flag, so as to facilitate the acquisition of the corresponding data information and compress the data storage space.

[0085] S14: Based on the identity information and the dynamic information, obtaining the associated data pairs of the multi-source information in the preliminary clustering set and the multi-source information in the history library;

[0086] It should be noted that for the preliminary clustering set and the multi-source information in the historical database, the identity information is still used for matching and identification. Once the identity information is the same, it means that the association is successful and the associated data pair is added without further processing. If the identity information is not matched successfully, the remaining multi-source information in the preliminary clustering set and the remaining multi-source information in the historical database are associated based on the dynamic information, including:

[0087] S141: acquiring candidate historical points from the remaining multi-source information in the history library according to a preset dynamic information association threshold;

[0088] It should be noted that the preset dynamic information association thresholds include: longitude association threshold e Lon , latitude correlation threshold e Lat and the highly correlated threshold e Alti ;

[0089] The process of obtaining candidate historical points from the remaining multi-source information in the history library includes:

[0090] ① Take out each remaining multi-source information in the preliminary clustering set in turn, compare it with each remaining multi-source information in the historical database, and obtain the longitude difference, latitude difference and altitude difference of the dynamic information in the multi-source information. When the absolute values ​​of the three differences are all less than the corresponding association threshold, the multi-source information in the historical database will be included in the candidate historical point traces.

[0091] Specifically, assume that the remaining multi-source information in the history library is:

[0092]

[0093] Among them, TN is the number of remaining targets in the history library, S l Indicates his l The number of targets for each target, Indicates his l The tth target p The composition of multi-source information at a given moment is the same as that of SenInfo.

[0094] It should be noted that the number of cycles stored in the history library and the number of targets for each cycle, that is, the storage depth, are related to the movement and speed of the target platform itself. For example, the history library stores data of 3 cycles, and the storage depth of each cycle is 10s. The historical data is stored and updated according to the first-in-first-out rule.

[0095] If in the preliminary clustering set the i-th fusion processing cycle T i Within, nth j Sensor Type k Multi-source information at all times Central longitude Dimensions and height With History Library Central longitude Dimensions and height If the following equation is satisfied, Candidates for inclusion in the list of historical sites:

[0096]

[0097] ② For the candidate historical points, the Kalman filter method is used to predict and extrapolate the track points related to the multi-source information in the historical database, and extrapolate to the time when the multi-source information in the preliminary clustering set is obtained. Extrapolate the track points to t k At each moment, time and space calibration is achieved through filtering, and then association is achieved, which improves the accuracy of the association.

[0098] For example, the current time is 10s, and the historical track points have information from 1s to 9s. Assuming that the motion model is uniform linear motion, the historical track points can be extrapolated to the 10th second through Kalman filtering.

[0099] S142: Based on the dynamic information, calculate the correlation gate between the remaining multi-source information in the preliminary clustering set and the multi-source information in the candidate historical points, and identify the correlation data pairs between the remaining multi-source information in the preliminary clustering set and the multi-source information in the candidate historical points according to the correlation gate.

[0100] It should be noted that, according to the distance covariance and azimuth covariance of different sensors, the corresponding correlation gates are obtained, including the distance correlation gate ΔR(t i ) and the azimuth-correlated wave gate Δθ(t i ), the formula is:

[0101]

[0102] Among them, σ rS1 , σ θS1 Respectively represent the distance covariance and orientation covariance of multi-source information in the preliminary clustering set, σ rS2 , σ θS2 They represent the distance covariance and orientation covariance of the multi-source information in the candidate historical points, respectively; K represents the influence coefficient, ranging from 0 to 1; t i Indicates the time of obtaining multi-source information in the preliminary clustering set. Through the dynamic design of the correlation gate, the dependence of the correlation gate on expert experience is reduced.

[0103] According to formula (5), each remaining multi-source information in the preliminary clustering set is taken out in turn, and each multi-source information in the candidate historical point trace is compared with each other. The distance association gate and the azimuth association gate of the multi-source information, as well as the distance difference and the azimuth difference are calculated two by two. When the absolute value of the distance difference is less than the distance association gate, and the absolute value of the azimuth difference is less than the azimuth association gate, the two multi-source information are successfully associated and added to the associated data pair. Otherwise, the association fails, and the multi-source information that fails to be associated in the preliminary clustering set is marked as a discrete point and stored in the history library.

[0104] The association of hetero-periodic multi-source information realizes the association of hetero-periodic non-discrete points and the secondary determination of discrete points, avoids the misidentification of dense points and improves the determination accuracy of discrete points.

[0105] S15: According to the multi-source information of the preliminary clustering set in the associated data pair, all multi-source information of the same type and belonging to the same information fusion cycle in the cache area are added to the associated data pair, and the same track number is set for each group of associated data to obtain an associated result.

[0106] It should be noted that the associated data pairs in step S15 include two parts of associated data pairs obtained in step S14: one part is the associated data pairs of the multi-source information in the preliminary clustering set and the multi-source information in the historical library obtained based on the same identity information, and the other part is based on dynamic information, and the multi-source information remaining in the preliminary clustering set and the multi-source information remaining in the historical library are associated to obtain the associated data pairs of the multi-source information remaining in the preliminary clustering set and the multi-source information in the historical candidate points.

[0107] Based on the associated data pairs, the multi-source information in the cache area that belongs to the same type as the preliminary clustering set in the same information fusion cycle is added to the associated data pairs, that is, the multi-source information of the same type in the same information fusion cycle is associated with the multi-source information in the corresponding historical library or historical candidate point traces.

[0108] Compared with the prior art, the present embodiment provides a multi-source heterogeneous information association method based on information fusion cycle. For multi-source heterogeneous information, the association process of classified and hierarchical processing is carried out through the method of clustering multi-source information of the same period and associating information of different periods, thereby reducing the computational complexity of multi-target multi-source information association in a dense environment and solving the problem of high complexity of information association. Among them, for multi-source information of the same period, the initial clustering center is dynamically selected based on the maximum density iterative processing method through identity information matching and improved fuzzy C-means clustering, thereby solving the problem that the initial clustering center has a strong dependence on artificial and prior knowledge, realizing fuzzy clustering of multiple information of the same period and the same sensor and rough selection of sparse discrete points, solving the problems of low utilization rate of existing identity information and difficulty in determining sparse points. For information association of different periods, a dynamic design criterion of the association wave gate is designed, realizing a method based on identity information matching and dynamic association of multi-period and multi-dimensional information, thereby solving the problem of erroneous association that may occur under dense multi-target and complex motion conditions.

[0109] Embodiment 2,

[0110] This embodiment is based on the first embodiment, and further performs discrete information association on target points with a long adjacent time interval, a large position distance, and a newly appeared target point trace, including:

[0111] S21: taking the multi-source information marked as discrete points in the history library as discrete point traces, and taking other multi-source information in the history library as historical point traces;

[0112] It should be noted that, according to the acquisition time of the discrete point trace, the Kalman filter method is used to extrapolate the track point prediction of the multi-source information not marked as discrete points in the history library to the acquisition time of the discrete point trace.

[0113] The multi-source information of discrete points and historical points includes four categories: time t, dynamic information dynaInfo, identity information staticInfo and proprietary information specialInfo.

[0114] S22: based on the identity information and the proprietary information, obtaining the associated data pairs of the discrete traces and the historical traces, and then based on the time and dynamic information, establishing multiple similarity matrices for the remaining discrete traces and the remaining historical traces in the history library;

[0115] It should be noted that each discrete point trace and each historical point trace are first matched pairwise based on identity information and proprietary information. When a discrete point trace is identical to the identity information and proprietary information of only one historical point trace, the association is successful. When a discrete point trace is identical to the identity information and proprietary information of multiple historical point traces, and a discrete point trace is different from the identity information or proprietary information of each historical point trace, the association fails, and then judgment and association are performed based on time and dynamic information.

[0116] Assuming that after matching and associating identity information and proprietary information, there are P historical points that have been extrapolated to the latest acquisition time and N unassociated discrete points in the history database, then the N discrete points are compared with the P historical multi-source information in pairs to establish a similarity matrix, including: a time similarity matrix and a dynamic information similarity matrix.

[0117] The rows of the similarity matrix correspond to each discrete point trace, the columns correspond to each historical point trace, and the element value corresponds to the multi-source information comparison value of each discrete point trace and each historical point trace, which is expressed as: Among them, q represents the type of similarity matrix, q=1,2, the similarity matrix is ​​N rows and P columns, a represents the row of the similarity matrix, b represents the column of the similarity matrix, a∈[1,N], b∈[1,P].

[0118] Specifically, the temporal similarity matrix The value obtained by dividing the absolute value of the time difference between each discrete point of the element value in the matrix and the historical multi-source information by the absolute value of the maximum time difference is calculated as follows:

[0119]

[0120] It should be noted that when the time of the historical traces extrapolated to the latest acquisition time is the same as that of the discrete traces, there is no need to establish a time similarity matrix. When the historical traces in the history library cannot be extrapolated, time is used to measure similarity, expand the comparison dimension, and increase the accuracy of the association.

[0121] Dynamic Information Similarity Matrix The element value in the matrix is ​​the value obtained by dividing the absolute value of the dynamic information difference between each discrete point and the historical multi-source information by the absolute value of the maximum dynamic information difference. The calculation formula is as follows:

[0122]

[0123] It should be noted that the dynamic information includes target longitude, target latitude, target altitude, target distance, target azimuth, target pitch, target heading, and target speed. According to actual needs, multiple attributes can be selected and the corresponding similarity matrix can be established according to formula (7).

[0124] S23: according to the preset measurement threshold of each similarity matrix, multiple similarity matrices are converted into multiple 0-1 measurement matrices; the number of element values ​​1 at the same position in the multiple 0-1 measurement matrices is counted, and when the number is greater than the preset number threshold, the discrete point trace corresponding to the position is successfully associated with the historical point trace, and the associated data pair is added;

[0125] Specifically, according to the preset measurement threshold of each similarity matrix, multiple similarity matrices are converted into multiple 0-1 measurement matrices, including:

[0126] Under the condition that the measurement threshold of the preset time similarity matrix is ​​less than the preset time similarity matrix, the element values ​​of the time similarity matrix are traversed, and under the condition that the measurement threshold of the dynamic similarity matrix is ​​less than the preset dynamic similarity matrix, the element values ​​that meet the conditions are obtained respectively, and then according to the rule of taking the minimum value for each row and the maximum value for each column, the element values ​​that meet the conditions are optimized again, and the element values ​​that meet the rules are set to 1, otherwise, the element values ​​are set to 0, and the 0-1 time measurement matrix corresponding to the time similarity matrix and the 0-1 dynamic information measurement matrix corresponding to the dynamic information similarity matrix are obtained respectively.

[0127] The number of elements with the same value of 1 in multiple 0-1 metric matrices is counted. When the number of 1s is greater than the preset threshold, it means that the association is successful, and the discrete point traces and historical point traces corresponding to the position are added to the associated data pair.

[0128] S24: Setting the same track number for each group of associated data pairs to obtain an associated result.

[0129] It should be noted that the correlation relationship of multi-source heterogeneous information is used for target identification. In practice, the correlation results are sent to the target multi-attribute identification system based on information fusion, and finally the binary attributes of information are expanded to multi-attributes, the recognition of the target is improved, and more hidden relationships are excavated.

[0130] Compared with the prior art, this embodiment adds an association method of discrete points to multi-source heterogeneous information on the basis of Embodiment 1. By constructing a multi-dimensional similarity matrix of discrete points and designing association optimization rules, the problems of difficulty in associating discrete points and missed associations are solved, providing a premise and guarantee for improving the target recognition rate.

[0131] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0132] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A multi-source heterogeneous information association method based on information fusion cycle, characterized in that: The steps include: Receive multi-source information from multiple different types of sensors equipped on a target platform, including identity information and dynamic information; the identity information is a 24-bit address code of the sensor; the dynamic information includes: target longitude, target latitude, target altitude, target distance, target azimuth and target pitch; According to the information fusion cycle, multi-source information containing multiple data from the same type of sensors is added to the set to be clustered, and other multi-source information is added to the preliminary clustering set; Cluster the cluster set, add the multi-source information closest to the cluster center in each class to the preliminary cluster set, put the other multi-source information in each class into the cache area, and mark the multi-source information that is not classified into any class as a discrete point and store it in the history library; Based on the identity information and the dynamic information, the associated data pairs of the multi-source information in the preliminary clustering set and the multi-source information in the historical library are obtained, specifically including: firstly based on the same identity information, the associated data pairs of the multi-source information in the preliminary clustering set and the multi-source information in the historical library are identified, and then based on the dynamic information, the remaining multi-source information in the preliminary clustering set and the remaining multi-source information in the historical library are associated, wherein, based on the dynamic information, the remaining multi-source information in the preliminary clustering set and the remaining multi-source information in the historical library are associated, including: according to a preset dynamic information association threshold, the candidate historical point traces are obtained from the remaining multi-source information in the historical library; based on the dynamic information, the association wave gate between the remaining multi-source information in the preliminary clustering set and the multi-source information in the candidate historical point traces is calculated, and the associated data pairs of the remaining multi-source information in the preliminary clustering set and the multi-source information in the candidate historical point traces are identified according to the association wave gate; According to the multi-source information of the preliminary clustering set in the associated data pair, all the multi-source information of the same type and belonging to the same information fusion cycle in the cache area are added to the associated data pair, and the same track number is set for each group of associated data to obtain the associated result.

2. The multi-source heterogeneous information association method based on information fusion cycle according to claim 1 is characterized in that: The clustering of the clustering set includes: firstly identifying the same type of multi-source information in the same information fusion cycle based on the same identity information, and taking the information fusion cycle as the clustering center, and then clustering the remaining multi-source information in the clustering set; wherein, clustering the remaining multi-source information in the clustering set includes: Traverse the remaining multi-source information in the set to be clustered, and determine multiple cluster centers and information sets in turn; Calculate the average value of multi-source information in each type of information set; Calculate the sum of squares of errors between each piece of multi-source information and the average value of multi-source information in each type of information set as the clustering criterion function value; When the clustering criterion function value is less than the preset error threshold, or the difference between two consecutive clustering criterion function values ​​is less than the error threshold, the iteration is exited, otherwise it returns to determine the new cluster center and information set.

3. The multi-source heterogeneous information association method based on information fusion cycle according to claim 2 is characterized in that: The traversing the remaining multi-source information in the set to be clustered, and sequentially determining a plurality of cluster centers and information sets, includes: Select a piece of multi-source information from the remaining multi-source information in the set to be clustered as the initial center, and according to the preset initial distance threshold, obtain the multi-source information whose Euclidean distance from the initial center is less than the initial distance threshold as the set to be selected, select each piece of multi-source information in the set to be selected in turn, calculate the number of multi-source information whose Euclidean distance is less than the initial distance threshold, take the multi-source information with the largest number as the first cluster center, and the multi-source information whose Euclidean distance from the first cluster center is less than the initial distance threshold as the first category of information set; The Euclidean distances between the multi-source information except the first type of information set and the first cluster center are calculated in sequence, and the multi-source information with the largest distance is selected as the second cluster center. If the maximum distance is greater than the preset minimum distance, the difference between the maximum distance and the initial distance threshold is taken as the second cluster distance, and the multi-source information with the Euclidean distance from the second cluster center less than the second cluster distance is obtained as the second type of information set; In this way, the mth cluster center is calculated until the Euclidean distance between the m+1th cluster center and the mth cluster center is less than the minimum distance. The iteration ends and m cluster centers and m types of information sets are obtained.

4. The multi-source heterogeneous information association method based on information fusion cycle according to claim 2 is characterized in that: The sum of square errors between each piece of multi-source information and the average value of the multi-source information in each type of information set is calculated as the clustering criterion function value, and the formula is: Among them, m represents the total number of cluster centers, C i represents the i-th cluster center, i=1,…,m, SenInfo represents each multi-source information in each type of information set, Represents the average value of multi-source information in the i-th information set.

5. The multi-source heterogeneous information association method based on information fusion cycle according to any one of claims 2 to 4, characterized in that: The preset dynamic information association thresholds include: a longitude association threshold, a latitude association threshold and an altitude association threshold; The step of acquiring candidate historical points from the remaining multi-source information in the historical database includes: Take out each remaining multi-source information in the preliminary clustering set in turn, and compare them with each remaining multi-source information in the historical database, and obtain the longitude difference, latitude difference and altitude difference of the dynamic information in the multi-source information. When the absolute values ​​of the three differences are all less than the corresponding association thresholds, the multi-source information in the historical database is included in the candidate historical points, and the Kalman filtering method is used to predict and extrapolate the track points related to the multi-source information in the historical database, and extrapolate to the acquisition time of the multi-source information in the preliminary clustering set compared with it.

6. The multi-source heterogeneous information association method based on information fusion cycle according to claim 5 is characterized in that: The correlation wave gates include a distance correlation wave gate and an azimuth correlation wave gate; The method of calculating the correlation gate between the remaining multi-source information in the preliminary clustering set and the multi-source information in the candidate historical traces based on the dynamic information, and identifying the correlation data pairs between the preliminary clustering set and the candidate historical traces according to the correlation gate, includes: Take out each remaining multi-source information in the preliminary clustering set in turn, and each multi-source information in the candidate historical point trace, and calculate the distance association gate and azimuth association gate of the multi-source information, as well as the distance difference and azimuth difference. When the absolute value of the distance difference is less than the distance association gate, and the absolute value of the azimuth difference is less than the azimuth association gate, the two multi-source information are successfully associated, otherwise the association fails, and the multi-source information that fails to be associated in the preliminary clustering set is marked as a discrete point and stored in the history library.

7. The multi-source heterogeneous information association method based on information fusion cycle according to claim 1 is characterized in that: Also includes: The multi-source information marked as discrete points in the history database is used as discrete point traces, and the other multi-source information in the history database is used as historical point traces; The multi-source information also includes proprietary information, which is an icon of the sensor; based on the identity information and the proprietary information, a pair of associated data of discrete points and historical points is obtained, and then based on the time and dynamic information, multiple similarity matrices are established for the remaining discrete points in the history library and the remaining historical points; According to a preset measurement threshold of each similarity matrix, multiple similarity matrices are converted into multiple 0-1 measurement matrices; Count the number of elements with the same value of 1 in multiple 0-1 metric matrices. When the number is greater than the preset threshold, the discrete point traces corresponding to the position are successfully associated with the historical point traces and added to the associated data pair. The same track number is set for each set of associated data pairs to obtain the associated result.

8. The multi-source heterogeneous information association method based on information fusion cycle according to claim 7 is characterized in that: The rows of the similarity matrix correspond to each discrete point trace, the columns correspond to each historical point trace, and the element values ​​correspond to the comparison values ​​of each discrete point trace and each historical point trace; The similarity matrix includes: a time similarity matrix and a dynamic information similarity matrix; the multi-source information comparison value includes: a time information comparison value and a dynamic information comparison value.

9. The multi-source heterogeneous information association method based on information fusion cycle according to claim 8 is characterized in that: The method of converting the plurality of similarity matrices into a plurality of 0-1 measurement matrices according to the preset measurement threshold of each similarity matrix comprises: Under the condition that the measurement threshold of the preset time similarity matrix is ​​less than the preset time similarity matrix, the element values ​​of the time similarity matrix are traversed, and under the condition that the measurement threshold of the dynamic similarity matrix is ​​less than the preset dynamic similarity matrix, the element values ​​that meet the conditions are obtained respectively, and then according to the rule of taking the minimum value for each row and the maximum value for each column, the element values ​​that meet the conditions are optimized again, and the element values ​​that meet the rules are set to 1, otherwise, the element values ​​are set to 0, and the 0-1 time measurement matrix corresponding to the time similarity matrix and the 0-1 dynamic information measurement matrix corresponding to the dynamic information similarity matrix are obtained respectively.

Images