A method for associating and identifying aerial targets

By dynamically setting density thresholds and pre-grouping, the memory consumption and time consumption problems of the DBSCAN algorithm under massive data are solved, the efficiency of air target association recognition is improved, and a faster and more efficient clustering process is achieved.

CN114282601BActive Publication Date: 2024-11-19ZHONGKE XINGTU SPACE TECH CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202111479716.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-06
Publication Date
2024-11-19
Estimated Expiration
2041-12-06

AI Technical Summary

Technical Problem

The existing DBSCAN algorithms are difficult to select appropriate density thresholds ε and MinPts during the target clustering process, resulting in excessive clustering time and memory consumption and inability to effectively process massive data.

Method used

Dynamic threshold settings are adopted, and the non-spatial attributes of the target object (motor time threshold t and maneuver speed v) are adjusted ε, pre-grouped and selected a list of core objects for local clustering, reducing the repeated consumption of global judgments.

Benefits of technology

The dynamics of domain thresholds are achieved, clustering time and memory consumption are reduced, the extreme situations of isolated points and class merging are avoided, and clustering speed and efficiency are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114282601B_ABST
    Figure CN114282601B_ABST
Patent Text Reader

Abstract

The present invention relates to an aerial target association recognition method, which specifically includes the following steps: step 1, first, screening the original spatial data of the target object and standardizing the format; step 2, grouping the standardized data according to the specified number of groups and each group; step 3, processing each group after grouping according to the set dynamic threshold, selecting the core object, and obtaining the core object list of each group; step 4, then, performing local clustering according to the core object list; step 5, selecting the representative points of the local clustering; step 6, finally performing global confirmation to obtain the clustering. The method of the present invention not only retains the advantage of the DBSCAN clustering method that can discover clusters of any shape and size through the number of objects in a unit area, but also overcomes the disadvantages of the DBSCAN method that the density threshold ε and MinPts are globally unique, and reduces the clustering time and memory consumption.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of target recognition, in particular to an aerial target association recognition method. Background Art

[0002] Target clustering, also known as force aggregation, is the process of forming a target group and an important function that the decision-making support system must achieve in situation estimation. The target clustering process is a forward reasoning process. It divides and abstracts each unit based on the information of each target unit output by the first-level information fusion, and aggregates the target entities layer by layer into a higher-level abstract combat group. By analyzing the various static attributes and dynamic behaviors of the group, a military system unit hypothesis at the level of the enemy's force organizational structure relationship is finally formed, providing battlefield information at the force level for military decision-making.

[0003] The ultimate goal of target segmentation is to be able to explain:

[0004] (1) What targets exist on the battlefield and their characteristics;

[0005] (2) What combat groups are there on the battlefield, what are their composition and distribution structures, and what are their conditions?

[0006] (3) What kind of relationships exist between adjacent groups;

[0007] (4) Possible combat intentions and objectives based on the characteristics of battlefield forces.

[0008] Therefore, the central issue of target clustering is how to integrate data from different sources to analyze the characteristics and status of battlefield forces, such as organizational structure, combat capability, and combat intent. The input data of target clustering is mainly the first-level fused target attributes and status information.

[0009] Let S and P i Respectively represent the situation estimation and the state information set of the i-th target entity at a certain moment, then:

[0010] S = {P1, P2, ..., P i , …P n}

[0011] P i ={T i , V i , Y i , L i , S i , E i , W i}

[0012] i=1,2…n

[0013] Where T i , V i , Y i , L i , S i , E i and W i are the time, speed, type, position, state, radiation source and weapon load of the i-th target unit respectively.

[0014] Based on the target information of each entity, using information such as tactics / campaign regulations, spatial position relationships, and functional combination relationships, the target entities are divided and abstracted layer by layer from the bottom up to form force-level relationship units. After group division, the battlefield situation can be expressed as:

[0015] S'={G1,G2,…,G N}

[0016] Where N is the number of groups formed after grouping.

[0017] These groups are divided into space groups, interaction groups, and enemy / friend / neutral groups in order from low to high. The space group divides the target entities according to the principle of close spatial position or similar behavior; the interaction group divides the generated space groups again according to the principle of similar tactical purposes; the enemy / friend / neutral group divides the interaction groups according to the confrontational relationship. At this point, the set of mutual relationships formed by the groups can be used to deduce and identify all possible conflicts.

[0018] The representative method currently used to aggregate target objects into spatial groups is the DBSCAN clustering method. This density-based clustering algorithm uses local data features as the judgment criteria for clustering. Its main idea is: as long as the density of the adjacent area exceeds a certain threshold, clustering will continue. A class is regarded as a data area where objects are dense, and areas with sparse objects separate the classes. Density-based clustering algorithms do not require the number of clusters to be set in advance. Instead, they use data density, that is, the number of objects in a unit area, to discover clusters of any shape and size, thereby overcoming the disadvantage that distance-based algorithms can only discover "quasi-circular" clusters.

[0019] Gaining decision-making advantage and gaining battlefield control is one of the key conditions for winning an informationized war. However, in an informationized combat environment, there are many targets, coordinated relationships, and frequent maneuvers. This makes it difficult for commanders to make timely and effective decisions when faced with massive amounts of information and rapidly changing battlefield situations.

[0020] In order to assist commanders in implementing combat command and gain decision-making advantages, it is necessary to use the command decision-making system to carry out information fusion, that is, to quickly and efficiently process and abstract multi-source information for subsequent situation assessment.

[0021] Target grouping can greatly reduce the number of targets that commanders are concerned about and reduce the cognitive pressure they face. In addition, target grouping can form a military system unit hypothesis at the relationship level by abstracting and dividing the information of threat units. It can reflect the collaborative relationship between entities through group division, and then reveal the mutual relationship between situation elements. One of the key steps of target grouping is target association identification, that is, grouping according to the position of the target object in space.

[0022] The DBSCAN algorithm divides areas with sufficiently high density into clusters, defining a cluster as the largest set of density-connected objects. It controls the growth of clusters based on a threshold value. It is a typical density clustering algorithm based on high-density connected areas. DBSCAN represents density by the number of objects in the ε neighborhood. The DBSCAN algorithm finds clusters and isolated points in a data set D. It first checks the ε neighborhood of each object in D. If object p is a core object, it generates a cluster containing N ε (p) is a cluster C of objects; then N of all unprocessed objects q in C are detected ε (q), if q is a core object, then N that is not yet included in C ε The objects in (q) are added to cluster C, and the ε neighborhoods of these objects are detected in the next step. This process is repeated until no new objects are added to the current cluster C.

[0023] However, this method still has the following disadvantages:

[0024] (1) The density threshold ε and MinPts are globally unique and difficult to select. If the density threshold is too large, a cluster will be split into several clusters, resulting in a large number of isolated points; if the density threshold is too small, several clusters that are far away will be merged.

[0025] (2) The algorithm operates on the entire data set, which requires large memory support and I / O consumption when the data volume is large. Summary of the invention

[0026] The problem solved by this method is to achieve the association and identification of aerial targets through spatial clustering of target objects, and to reduce the clustering time and memory consumption.

[0027] The present invention provides a clustering method for aggregating spatial target objects into spatial groups. The method not only retains the advantage of the DBSCAN clustering method that it can discover clusters of any shape and size by the number of objects in a unit area, but also overcomes the disadvantages of the DBSCAN method that the density threshold ε and MinPts are globally unique, thereby reducing the clustering time and memory consumption.

[0028] The technical solution of the present invention is as follows: A method for associating and identifying aerial targets, specifically comprising the following steps:

[0029] Step 1: First, filter the original spatial data of the target object and standardize its format;

[0030] Step 2, group the standardized data according to the specified number of groups and for each group;

[0031] Step 3: Process each group according to the set dynamic threshold, select the core object, and obtain the core object list of each group;

[0032] Step 4: Subsequently, local clustering is performed based on the core object list;

[0033] Step 5: Select representative points of local clustering;

[0034] Step 6: Finally, perform global confirmation to obtain clustering.

[0035] Furthermore, in step 3, processing each group after grouping according to the set dynamic threshold specifically includes:

[0036] Given a data object set D and a data object p, p∈D, and define the ε neighborhood N of p ε (p) is N ε (p)={q∈D|dist(p,q)≤ε}, where dist(p,q) represents the distance between two data objects p and q in D. The parameter ε is dynamically set using the non-spatial properties of the data, and ε=tv, where t is the set maneuvering time threshold and v is the maneuvering speed.

[0037] Furthermore, in step 3, selecting core objects to obtain a core object list for each group specifically includes:

[0038] Scan the data set and select any data object p. If p has been classified as a cluster or marked as noise, scan the data set again to select the data object. Otherwise, determine whether the number of points in the area is less than MinPts. If so, mark point p as a boundary point or a noise point; otherwise, mark point p as a core point and obtain a list of core point objects.

[0039] Furthermore, in step 4, for the core point p, a new cluster C is established and all points within the area of ​​point p are added to C.

[0040] Furthermore, in step (5), for the core object p, it is necessary to determine whether each point in its neighborhood is a core object. If its neighborhood is a high-density area, it is necessary to select a representative object to reduce clustering time and memory consumption; check all unmarked q points in the neighborhood ε, and determine whether the points in the area are less than MinPts. If so, continue to check the unmarked q points in the neighborhood. Otherwise, add the points in the neighborhood that are not classified into other clusters to C.

[0041] Beneficial effects:

[0042] (1) The dynamic selection of domain threshold is realized by adopting dynamic parameter setting, where ε=tv, where t is the set maneuvering time threshold and v is the maneuvering speed. This avoids the extreme situation where a large number of isolated points or class merging occur arbitrarily when the threshold is unique.

[0043] (2) The method of selecting cluster object representatives reduces the time and memory consumption of clustering and avoids the time and space overhead caused by repeated judgments due to neighborhood overlap.

[0044] (3) The target objects are grouped in advance according to static and dynamic data by pre-grouping and then clustering. The core objects are selected for each group to obtain a list of core objects for each group. Local clustering is then performed based on this list, and finally a global judgment is performed. This optimization further reduces the computational time and space overhead and improves the clustering speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 :The target objects are aggregated into the process of forming space groups;

[0046] Figure 2 : Improved DBSCAN algorithm clustering process. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all 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 protection scope of the present invention.

[0048] According to an embodiment of the present invention, the overall process of the improved DBSCAN spatial clustering method of the present invention is as follows: Figure 1 As shown, the specific steps include:

[0049] Step 1: First, filter the original spatial data of the target object and standardize its format;

[0050] Step 2: Group the standardized data according to the specified number of groups and for each group;

[0051] Step 3: Process each group after grouping according to the set dynamic threshold, select the core objects, and obtain the core object list for each group;

[0052] Step 4: Subsequently, perform local clustering based on the core object list;

[0053] Step 5: Select the representative points of the local clustering;

[0054] Step 6: Finally, perform global confirmation to obtain the clustering.

[0055] In the present invention, the number of representative objects selected is set within a predetermined range; it cannot be too many or too few. If it is too many, it is difficult to exert the efficiency of the algorithm and the significance of selection is lost; on the contrary, if it is too few, it is difficult to compare the neighborhoods of the selected representative objects and completely cover the neighborhoods of other objects, resulting in the loss of objects and splitting a clustering into several, making it difficult to exert the efficiency of the algorithm.

[0056] First, the following definitions are given

[0057] Definition 1. ε-neighborhood. Given a data object set D and a data object p, p ∈ D, the ε-neighborhood N ε (p) of p is defined as Nε ( p) = {q ∈ D | dist(p, q) ≤ ε}, where dist(p, q) represents the distance between two data objects p and q in D, and the parameter ε is dynamically set using the non-spatial attributes of the data, and let ε = tv, where t is the set maneuvering time threshold and v is the maneuvering speed.

[0058] Definition 2. Core object and boundary object. For a data object p ∈ D, given an integer MinPts, if the number of objects in the ε-neighborhood of p, |N ε (p)| ≥ MinPts, then p is called a core object; if |N ε (p)| < MinPts, but q is a core object and p ∈ N ε (q), then p is called a boundary object.

[0059] Definition 3. Direct density reachability. If p, p ∈ D, for the given ε and MinPts, there is p ∈ N ε (q), |N ε (p)| ≥ MinPts, then it is said that the object p is directly density reachable from the object q with respect to ε and MinPts.

[0060] Definition 4. Density reachability. If p1, p2,..., p n ∈ D, p1 = q, p n=p, for a given ε and MinPts, there is p i+1 From p i If it is directly density reachable, then object p is said to be density reachable from object q with respect to ε and MinPts.

[0061] Definition 5. Density connected, if p, q∈D, for a given ε and MinPts, If both object p and object q are density-reachable from o, then object p and object q are said to be density-connected with respect to ε and MinPts.

[0062] Definition 6. Clusters and Outliers. For a given ε and MinPts, a cluster C is a non-empty subset of D that satisfies the following conditions:

[0063] (1) q∈D, if p∈C and q is density-reachable from p with respect to ε and MinPts, then q∈C.

[0064] (2) q∈C, q and p are density-connected with respect to ε and MinPts. Objects not included in any cluster are called outliers.

[0065] After the present invention improves the DBSCAN algorithm, the clustering steps are as follows: Figure 2 As shown:

[0066] (1) Given a data object set D and a data object p, p∈D, and define the ε neighborhood N of p ε (p) is N ε (p) = {q∈D|dist(p,q)≤ε}, where dist(p,q) represents the distance between two data objects p and q in D, and the parameter ε is dynamically set by using the non-spatial attribute of the data, and ε = tv, where t is the set maneuvering time threshold, and v is the maneuvering speed. According to one embodiment, for example, t is the time threshold, which is a set fixed value, and the size of v is the average value of the speed of all points within the time threshold in the data point set. In the next time period t, v has a new change. The present invention (1) utilizes the non-spatial attribute of the target object when spatial clustering, that is, a dynamic parameter setting is adopted to set ε = tv, where t is the set maneuvering time threshold, and v is the maneuvering speed. In this way, the disadvantages that when the density threshold ε and MinPts are globally unique, the density threshold is too large, a cluster will be divided into several, and a large number of isolated points will appear; and when the density threshold is too small, several clusters with a long distance will be merged.

[0067] (2) Scan the data set and select any data object p. If p has been classified as a cluster or marked as noise, scan the data set again to select the data object. Otherwise, determine whether the points in the area are less than MinPts. If so, mark point p as a boundary point or a noise point. Otherwise, mark point p as a core point, establish a new cluster C, and add all points in the area of ​​point p to C.

[0068] (3) Regarding the issue of only core objects being eligible for expansion, once a core object is found, it is necessary to determine whether each point in its neighborhood is a core object in order to further expand outward. If its neighborhood is a high-density area, it means that there will be many core objects, and their neighborhoods overlap with each other. The neighborhoods of some points are even completely covered by the neighborhoods of other points, which will cause repeated judgments and lead to unnecessary consumption of time and memory. Therefore, it is necessary to select representative objects. Preferably, objects with less overlapping coverage can be selected as representative objects to reduce clustering time and memory consumption, and avoid repeated judgments caused by neighborhood overlap;

[0069] (4) Check all unmarked q points in the neighborhood ε and determine whether the number of points in the area is less than MinPts. If yes, continue to check the unmarked q points in the neighborhood. Otherwise, add the points in the neighborhood that are not classified into other clusters to C.

[0070] The main purpose of spatial clustering is to reduce the number of spatial target objects that are overly concerned, so there is no need to pay too much attention to how many classes can be clustered. The target objects should be divided and grouped according to the static and dynamic data of the location, and then each group should be locally clustered, and then judged from the global perspective to further reduce memory.

[0071] The present invention performs pre-grouping during clustering, groups the target objects according to static and dynamic data, processes each group separately to select core objects, and obtains a list of core objects for each group. Then, local clustering is performed based on this list. Finally, a global judgment is performed, which can further reduce computing consumption and improve clustering speed.

[0072] Although the above describes the illustrative specific embodiments of the present invention to facilitate those skilled in the art to understand the present invention, and it should be clear that the present invention is not limited to the scope of the specific embodiments, for those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A method for associating and identifying aerial targets, characterized in that: The specific steps include: Step 1: First, filter the original spatial data of the target object and standardize its format; Step 2: Group the standardized data according to the specified number of groups; Step 3: Process each group according to the set dynamic threshold, select the core object, and obtain the core object list of each group; Step 4: Subsequently, local clustering is performed based on the core object list; Step 5: Select representative points of local clustering; Step 6: Finally, perform global confirmation to obtain clustering; In step 3, processing each group after grouping according to the set dynamic threshold specifically includes: Given a data object set D and data objects p, p D, and define p Neighborhood (p) is (p)={ q D |dist(p,q) }, where dist(p,q) represents the distance between two data objects p and q in D, and the parameters are dynamically set using the non-spatial properties of the data ,make , where t is the set maneuvering time threshold, and v is the maneuvering speed; In step 3, selecting core objects to obtain a core object list for each group specifically includes: Scan the data set and select any data object p. If p has been classified as a cluster or marked as noise, scan the data set again to select the data object. Otherwise, determine whether the number of points in the domain is less than MinPts. If so, mark point p as a boundary point or a noise point; otherwise, mark point p as a core point and obtain a list of core point objects.

2. The method for associating and identifying aerial targets according to claim 1, characterized in that: In step 4, for the core point p, a new cluster C is established and the p point area is Add all to C.

3. The method for associating and identifying aerial targets according to claim 2, characterized in that: In step 5, for the core object p, it is necessary to determine whether each point in its neighborhood is a core object. If its neighborhood is a high-density area, it is necessary to select a representative object to reduce clustering time and memory consumption; check the neighborhood All unmarked q points in the judgment area Is the point in the neighborhood less than MinPts? If so, continue to check the unmarked q points in the neighborhood. Otherwise, add the points in the neighborhood that are not classified into other clusters to C.

Citation Information

Patent Citations

  • Medical examination report clustering method based on density core and dynamic scanning radius

    CN108960298A

  • A Web services clustering method based on improved DBSCAN algorithm

    CN109255125A

  • Moving object track clustering method based on multi-dimensional distance measurement

    CN110991475A

  • Density-based data clustering method

    US20110055212A1