CN116662830A - Target grouping method and device based on confidence decision tree clustering - Google Patents

Abstract

The application provides a target grouping method and device based on a confidence decision tree, which acquire individual attributes, capability attributes and communication relations of targets through detection; taking all targets as root nodes; node division is carried out from the root node; traversing all nodes to be divided until all nodes do not need to be divided, and finally forming a confidence decision tree; and determining the group where the target is located according to the confidence degree of the target belonging to a certain child node in the confidence decision tree. The application determines the cutting point of the confidence decision tree node in the dividing process, and provides a confidence dividing method based on the cutting point, thereby generating child nodes and finally forming the confidence decision tree. Finally, the countermeasure population obtained by the opposite decision tree can be adjusted under the condition that the population number is known. The countermeasure group obtained by the application can describe the countermeasure group through the path of the decision tree, and meanwhile, the confidence division can realize the accurate judgment of the relationship between the countermeasure target and the countermeasure group.

Description

A method and device for target grouping based on confidence decision tree clustering

technical field

The invention belongs to the technical field of situation analysis, and in particular relates to a target grouping method and device based on confidence decision tree clustering.

Background technique

Target grouping is an important part of situation analysis, which provides the information basis for commanders to judge the intention of confrontation. The task of target grouping is the defender's understanding and judgment of the organizational structure of the confrontation system. Its purpose is to find out the relationship between various groups in the confrontation system, and on this basis, combine their confrontation capabilities to infer their confrontation tasks. In order to understand the tactical intention of the opponent in the entire confrontation environment, this process can be abstracted as a typical clustering problem.

The clustering results obtained by existing clustering models such as density-based methods and entropy-based methods often lack an intuitive description of the confrontation group, making it difficult for commanders to understand the clustering results when they are applied to target groups. In addition, in the existing clustering methods, a target can only belong to one adversarial group, so it is difficult to accurately describe some adversarial targets that serve multiple target groups at the same time in the application of target grouping, and these adversarial targets are often in the adversarial system. It has the identity of a "hub" and is an important member of the confrontation system.

The existing clustering-based target grouping methods are difficult to generate easily understood adversarial groups, and it is difficult to accurately classify some functionally complex targets that participate in multiple adversarial group tasks.

Contents of the invention

In order to solve the above-mentioned problems in the prior art, the present invention provides a method and device for grouping objects based on confidence decision tree clustering. The technical problem to be solved in the present invention is realized through the following technical solutions:

The present invention provides a kind of target grouping method based on confidence decision tree clustering comprising:

Step 1, obtain the target's individual attributes, ability attributes and communication relationships through detection;

Step 2, place all targets at the root node;

Step 3, starting from the root node to divide the child nodes, the i-th division process is:

(1) For a single-cluster node, select any attribute of the node, and select a possible cutting point in this attribute, and determine by calculating the average confidence profile measure of all possible cutting points and finding its maximum value real cut point;

(2) For the above-mentioned single-cluster node, assuming that the node needs to be split to generate three sub-nodes, the maximum evaluation value is obtained by calculating the confidence profile metrics of all targets at all possible cutting points, and judging whether the obtained maximum evaluation value is greater than the division The previous evaluation value; if so, continue to execute (3), otherwise, there is no need to divide on this node;

Wherein, if the node is the root node, there is no need to judge, and (3) can be directly executed, and if so, continue to execute (3), otherwise, there is no need to divide on the node;

(3) generating sub-nodes for the single-cluster node division, and determining the center point of each sub-node according to the real cutting point;

(4) calculate the degree of confidence of each sub-node by the distance from all targets to each central node; wherein said degree of confidence expresses the degree to which the target belongs to the population cluster represented by the sub-node;

Step 4, traverse all nodes until all nodes do not need to be divided, and finally form a confidence decision tree;

Among them, the traversal order is from top to bottom, traversing sibling nodes at the same level first, and then traversing lower-level nodes if there are no sibling nodes that can be divided;

Step 5, according to the confidence that the target belongs to a certain child node in the confidence decision tree, so as to determine the group cluster where the target is located.

The present invention provides a target grouping device based on confidence decision tree clustering comprising:

The detector is used to obtain the target attribute and related functional attributes of the target through detection;

The executor is used to place all targets on the root node; starting from the root node to divide the child nodes, the i-th division process is:

(1) For a single-cluster node, select any attribute of the node, and select a possible cutting point in this attribute, and determine by calculating the average confidence profile measure of all possible cutting points and finding its maximum value real cut point;

(2) For the above-mentioned single-cluster node, assuming that the node needs to be split, the maximum evaluation value is obtained by calculating the confidence contour metrics of all targets at all possible cutting points, and judging whether the obtained maximum evaluation value is greater than the evaluation before division value; if so, proceed to (3), otherwise, there is no need to divide on this node;

Wherein, if the node is the root node, there is no need to judge, and (3) can be directly executed, and if so, continue to execute (3), otherwise, there is no need to divide on the node;

(3) generating sub-nodes for the single-cluster node division, and determining the center point of each sub-node according to the real cutting point;

(4) calculate the degree of confidence of each sub-node by the distance from all targets to each central node; wherein said degree of confidence expresses the degree to which the target belongs to the population cluster represented by the sub-node;

Step 4, traverse all nodes until all nodes do not need to be divided, and finally form a confidence decision tree;

Among them, the traversal order is from top to bottom, traversing sibling nodes at the same level first, and then traversing lower-level nodes if there are no sibling nodes that can be divided;

The classifier is configured to determine the group cluster where the target belongs to according to the confidence that the target belongs to a certain child node in the confidence decision tree.

The present invention proposes a target grouping method and device based on a confidence decision tree, which obtains the individual attributes, capability attributes, and communication relationships of targets through detection; takes all targets as root nodes; divides nodes from the root node; traverses all The nodes to be divided, until all nodes do not need to be divided, finally form a confidence decision tree; according to the confidence that the target belongs to a certain child node in the confidence decision tree, the group where the target belongs is determined. The invention determines the cutting point when the node of the confidence decision tree is divided in the division process, and proposes a confidence division method based on the cutting point, so as to generate sub-nodes and finally form the confidence decision tree. Finally, the confrontation group obtained by the confidence decision tree can be adjusted when the number of groups is known. The confrontation group obtained in the present invention can describe the confrontation group through the path of the decision tree, and at the same time, the confidence division can realize the accurate judgment of the relationship between the confrontation target and the confrontation group.

The present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

Description of drawings

Fig. 1 is a schematic flow chart of a target grouping method based on confidence decision tree clustering provided by the present invention;

Fig. 2 is a plane display diagram of the confrontation target provided by the present invention;

Fig. 3 is a tree structure diagram when performing a target grouping task.

Detailed ways

The present invention will be described in further detail below in conjunction with specific examples, but the embodiments of the present invention are not limited thereto.

As shown in Figure 1, the present invention provides a kind of target grouping method based on confidence decision tree clustering comprising:

Step 1, obtain the target's individual attributes, ability attributes and communication relationships through detection;

Step 2, place all targets at the root node;

Step 3, starting from the root node to divide the child nodes, the i-th division process is:

(1) For a single-cluster node, select any attribute of the node, and select a possible cutting point in this attribute, calculate the average confidence profile measure of all possible cutting points, and find out the way of its maximum value , to determine the real cutting point;

It is worth noting that: if the communication relationship between targets can be obtained, the average profile measure needs to be replaced by the evaluation index in Equation 15;

(2) For the above-mentioned single-cluster node, assuming that the node needs to be split, the maximum evaluation value is obtained by calculating the confidence profile metrics of all targets at all possible cutting points, and judging whether the obtained maximum evaluation value is greater than the evaluation value before division ; If so, continue to execute (3), otherwise, there is no need to divide on this node;

Among them, in the first method, the maximum evaluation value is the maximum average value of the confidence profile metrics of the three sub-nodes, and the evaluation value before division is the maximum average value of the confidence profile metrics of the nodes at the real cutting point; in the second method Among them, using the formula of formula 15, the maximum evaluation value is calculated on the basis of the maximum average value of the confidence profile metrics of the three sub-nodes; the evaluation value before division is calculated using the formula of formula 15, and the confidence profile metrics of the nodes at the real cutting point obtained on the basis of the maximum average value. The second method is suitable for the case where the communication relationship can be obtained.

It is worth noting that: if the node is the root node, there is no need to make a judgment, and (3) can be directly executed.

(3) generating sub-nodes for the single-cluster node division, and determining the center point of each sub-node according to the real cutting point;

(4) calculate the degree of confidence of each sub-node by the distance from all targets to each central node; wherein said degree of confidence expresses the degree to which the target belongs to the population cluster represented by the sub-node;

Step 4, traverse all single-cluster leaf nodes until all nodes do not need to be divided, and finally form a confidence decision tree;

Among them, the traversal order is from top to bottom, traversing sibling nodes at the same level first, and then traversing lower-level nodes if there are no sibling nodes that can be divided.

Step 5, according to the confidence that the target belongs to a certain child node in the confidence decision tree, so as to determine the group cluster where the target is located.

The present invention proposes a target grouping method and device based on a confidence decision tree, which obtains the individual attributes, capability attributes, and communication relationships of targets through detection; takes all targets as root nodes; divides nodes from the root node; traverses all The nodes to be divided, until all nodes do not need to be divided, finally form a confidence decision tree; according to the confidence that the target belongs to a certain child node in the confidence decision tree, the group where the target belongs is determined. The invention determines the cutting point when the node of the confidence decision tree is divided in the division process, and proposes a confidence division method based on the cutting point, so as to generate sub-nodes and finally form the confidence decision tree. Finally, the confrontation group obtained by the confidence decision tree can be adjusted when the number of groups is known. The confrontation group obtained in the present invention can describe the confrontation group through the path of the decision tree, and at the same time, the confidence division can realize the accurate judgment of the relationship between the confrontation target and the confrontation group.

In a specific embodiment of the present invention, in the ith division process (1) includes:

(11) For a single cluster node, select any attribute of the node;

(12) traverse all possible cutting points in the selected target attribute, and calculate the confidence profile measure of all targets after cutting the attribute according to each possible cutting point;

(13) Calculate the average value of the confidence profile measure under each cut point, and select the cut point with the largest average value of the confidence profile measure as the real cut point.

To build a decision tree, one needs to find the best split attribute and associated cut point at each node for continuous splitting, where the tree is initialized with a root node containing the attribute values of all targets. The key issue is how to evaluate and find the best cut point for splitting. In the field of unsupervised learning, the profile metric is a popular internal criterion for assessing the quality of partitions, which takes into account both cohesion (average distance between an object and all other objects in the same adversarial population) and dissociative (object and nearest adversarial distance between all other objects in the group). The silhouette metric was originally defined for hard partitioning, it is not suitable for confident partitioning of nodes. Therefore, there is a need to extend the original contour metric in the framework of confidence partitioning.

Definition 1 (Confidence Profile Measure) The confidence profile measure (ES) of any target x _i ∈ X is defined as

where a( _xi ) denotes confidence cohesion, calculated as the weighted average distance between _xi and all other targets within the same adversarial population ^ωs :

And b( _xi ) represents the confidence separation, calculated as the weighted average distance between _xi and all other targets in the nearest adversarial population ^ωn :

Among them, BetP _j (ω ^s ) and BetP _j (ω ⁿ ) represent the pignistic probability that the target x _j belongs to the adversarial group ω ^s and the adversarial group ω ⁿ respectively. For proposition A, its pignistic probability can be obtained by the following formula

where || represents the cardinality of the set V. The confidence profile metric defined above covers the original profile metric defined for hard segmentation. When the pignistic probabilities BetP _j (ω ^s ) and BetP _j (ω ⁿ ) take 1 (that is, the confidence of the adversarial group is certain), it will degenerate into the original profile measure. Furthermore, similar to the original profile metric, the confidence profile metric defined above also takes values between -1 and 1. When the silhouette metric of _xi is close to 1, the adversarial population containing _xi is compact and _{xi is} far away from other adversarial populations, which is a good case. However, when the profile measure of _xi has a negative value (i.e. b( _xi )<a( _xi )), it means that _xi is more likely to be in another adversarial population than the target in the same adversarial population as _xi . The target is closer. In many cases, this is an unfavorable situation and should be avoided. In order to evaluate the quality of segmentation using each cut point, the average Confidence Contour (AES) value of all targets in the dataset needs to be calculated as follows

where BetP _j (ω ^s ) is the pignistic probability that target _xi belongs to its adversarial group ω ^s , which has the largest pignistic probability. For each node that needs to be split, the average confidence profile metric for all possible cut points is computed by traversing all attributes. The present invention will select the cut point with the largest average confidence profile metric value, and this cut point will be used for the confidence partition.

In a specific embodiment of the present invention, in the division process of the ith time (2) includes:

(21) For the single cluster node, assuming that the node needs to be split, calculate the confidence profile metrics of all targets at all possible cut points;

(22) For the single cluster node, calculate the average value of the confidence profile measure after splitting under each possible cut point;

(23) Select the maximum average value from the average values of the confidence profile metrics obtained in (22), obtain the maximum evaluation value according to the maximum average value, and judge whether the maximum evaluation value obtained is greater than the evaluation value before division. Otherwise, there is no need to split on this node.

Using the cut points selected in (2), the present invention needs to test in advance whether the divided contour metric value is greater than the previous value. If yes, the node to be divided will be divided using the selected cut point to generate a new child node; otherwise, this node does not need to be split.

In the developed unsupervised confidence decision tree, there are two kinds of nodes: single-cluster nodes and compound-cluster nodes (ie union of single clusters). This embodiment will describe how to divide the confidence between these two types of nodes.

In a specific embodiment of the present invention, in the i-time division process (3) includes:

(31) If the condition in (2) is met, then execute to generate a child node;

(32) If the child node is a single cluster node, then calculate the central point of the single cluster node according to the calculation expression of the single cluster node center;

(33) If the child node is a compound cluster node, then determine the real cutting point as the center point of the compound cluster node, and calculate the center points of the remaining single cluster nodes.

In a specific embodiment of the present invention, in the i-time division process (4) includes:

(41) Calculate the distance from all targets to the center point corresponding to each child node;

(42) Calculate the confidence of each child node according to the distance in (41).

In a specific embodiment of the present invention, before (3) in the ith division process, the target grouping method based on confidence decision tree clustering also includes:

a1, determine whether the node is a single cluster node or a composite cluster node;

b1, if the node is a single-cluster node, then use the real cut point in (1) as the cut point for the node to generate child nodes;

1) Confidence division of single-cluster nodes

Represents the node A of a single adversarial group to be split, and the split child nodes are denoted as L, R, and L∪R. Initially, the centers of these child nodes need to be estimated. Intuitively, the Center _m of the compound node L∪R can be represented by the selected cut point at node A. Therefore, the centers of the left child node L and the right child node R are respectively calculated as

where Left＝{ _xi | _xi ∈ A, x _ij < Center _m } and Righ＝{ _xi | x _i ∈ A, x _ij >Center _m } represent the left set and the right set separated by the cutting point, where x _ij is the target on the jth attribute chosen for splitting. m _i (A) is the confidence membership degree that target x _i belongs to node A.

In order to eliminate the influence of different adversarial group shapes on the mass function results, the normalized distance is used as

where γ > 0 is a parameter that regulates the importance of composite clusters. Larger values of γ will penalize the membership of composite clusters, and as γ approaches infinity, the aforementioned confident memberships will only reduce to fuzzy memberships. A default value of 1 is recommended to obtain proper composite clusters for characterizing overlap between adversarial populations. The mass function m _i constructed in Equation 6, A confident partition of the target in node A to its three children is provided, where the composite cluster represented by the middle child characterizes the overlap between the left and right individual adversarial populations. Furthermore, based on these mass functions, the boundaries between these generated adversarial populations can also be determined to make the confidence partition results easier for users to interpret.

Then, the confidence membership of each target _xi in node A to the three child nodes is inversely proportional to the distance to the center of the corresponding adversarial population. Therefore, the mass function representing the confidence membership of the target x _i can be designed as:

in, are the distances from x _i to Center _l , Center _r and Center _m on the selected feature, respectively.

c1, if the node is a compound cluster node, it needs to divide its two sibling nodes;

d1, obtain at least one real cutting point of sibling nodes in the process of dividing child nodes;

Among them, if neither of the two sibling nodes can be divided, then the node does not need to be divided;

e1, use the real cut obtained in d as the real cut point of the child node generated by the compound cluster node, and calculate the center and confidence of its child node at one time.

2) Confidence division of composite cluster nodes

For a confident partition of a node representing a composite cluster, there is no need to find a new cut point at this node, instead the present invention will use the bounds computed from its sibling nodes to directly form the confident partition. Therefore, before this step, it is necessary to ensure that the split of its sibling nodes has been completed. Since the composite cluster is the union of the adversarial populations represented by its sibling nodes, the boundary of splitting its sibling nodes can be inherited to split the composite cluster. If there are multiple sibling nodes generating new child nodes, the boundaries of these sibling nodes will be used one by one.

In a specific implementation of the present invention, if only one of the sibling nodes in d needs to generate a child node, then the obtained real cut point is 1; if both sibling nodes need to generate a child node, the obtained real cut point Points are 2;

In the process of dividing the compound cluster node, when there are 2 real cutting points, you can select any real cutting point to perform a division to generate two child nodes, and select another real cutting point to perform a division generation on the generated two child nodes two child nodes.

For nodes with composite clusters, using the above partitioning method, the number of leaf nodes generated by the entire decision tree is exponential in the number of different adversarial groups. For a dataset consisting of K different adversarial populations, the tree will generate at most 2 ^K leaf nodes, however, most leaf nodes representing compound clusters composed of many individual adversarial populations are meaningless, which also makes the tree structure become complicated. To address this issue, compound clusters can be directly partitioned using the cut points obtained from their sibling nodes, which prevents the generation of compound clusters consisting of more than two single adversarial populations. With this simplified belief partitioning rule, the number of generated leaf nodes is reduced to the quadratic size of the number of different adversarial groups. This simplified decision tree is preferred in applications due to its low complexity and better interpretability without sacrificing too much flexibility of the belief partition.

In a specific embodiment of the present invention, after step 4, the target grouping method based on confidence decision tree clustering also includes:

a2. In the confidence decision tree, it is determined that there is no need to divide the child node, and the child node is used as a leaf node;

b2, determine whether the number of leaf nodes is greater than or less than a given number;

b3, if it is greater than or less than a given number, adjust the leaf nodes in the confidence decision tree until it is consistent with the given number.

The procedures developed in the above two sections are initially executed at the root node and will be repeated until all leaf nodes have been evaluated and are not suitable for splitting. Up to this point, a preliminary unsupervised confidence decision tree has been constructed, with (nodes of separate clusters) leaf nodes representing the generated adversarial population. If the number of adversarial populations is unknown, or happens to be equal to the number obtained from the generated trees, the final clustering result can be obtained directly from the trees. Otherwise, an adversarial population adjustment of the tree is required to accommodate a given number of adversarial populations.

In a specific embodiment of the present invention, b3 includes:

If the number of leaf nodes is greater than the given number, the leaf node with the worst confidence is merged into the nearest leaf node to obtain the merged node;

Determine the composite cluster node of the two leaf nodes to be merged;

Merge the merge node with the compound cluster node to reduce the number of single clusters by one, and repeat the above process until the number of leaf nodes is consistent with the given number;

If the number of leaf nodes is less than the given number, try to divide the current leaf node. During the division process, there is no limit to its maximum evaluation value which needs to be greater than the current evaluation value;

Select the method with the largest evaluation value among these division methods to divide until the number of leaf nodes is consistent with the given number.

There are two possible situations in which confrontation group adjustments may be required. First, the number of adversarial groups is larger than the given adversarial groups, in which case the generated adversarial groups need to be merged. The average confidence contour (AES) value of all targets in each generated individual adversarial population ω _j is used to measure the quality of that adversarial population

Among them, BetP _i (ω _j ) is the probability that target _xi belongs to its adversarial population ω _j , ES(xi ₎ is the confidence profile metric value of target _xi defined in Equation 1. Using the above evaluation metric, the worst-quality individual adversarial population is merged with its closest individual adversarial population (measured by the Euclidean distance between the centers of the two adversarial populations). At the same time, those composite clusters containing merged adversarial populations will also be merged together. For example, if adversarial population A is merged with adversarial population B, the composite cluster A∪C should also be merged with B∪C. This step will continue until the specified number of adversarial groups is reached.

The second case is when the number of adversarial groups is smaller than a given adversarial group. In this case, the division needs to be continued. All possible splits are tested on a leaf node with a single adversarial population, and the split with the largest average confidence profile metric across all objects in the dataset is chosen. This step will continue until the specified number of adversarial groups is reached.

The present invention provides a target grouping device based on confidence decision tree clustering comprising:

The detector is used to obtain the target attribute and related functional attributes of the target through detection; the executor is used to use all targets as the root node; starting from the root node to divide the sub-nodes, the i-th division process is:

(1) For a single-cluster node, select any attribute of the node, and select a possible cutting point in this attribute, calculate the average confidence profile measure of all possible cutting points, and find out the way of its maximum value , to determine the real cutting point;

Among them, if the communication relationship between targets can be obtained, the average profile measure needs to be replaced by the evaluation index in formula 15;

(2) For the above-mentioned single-cluster node, assuming that the node needs to be split to generate three sub-nodes, the maximum evaluation value is obtained by calculating the confidence profile metrics of all targets at all possible cutting points, and judging whether the obtained maximum evaluation value is greater than the division The previous evaluation value; if so, continue to execute (3), otherwise, there is no need to divide on this node;

Wherein, if the node is the root node, there is no need to judge, and (3) can be directly executed, and if so, continue to execute (3), otherwise, there is no need to divide on the node;

(3) If the condition in (2) is met, then execute to generate child nodes, and determine the center point of each child node according to the real cutting point;

(4) calculate the degree of confidence of each sub-node by the distance from all targets to each central node; wherein said degree of confidence expresses the degree to which the target belongs to the population cluster represented by the sub-node;

Traverse all single-cluster leaf nodes until all nodes do not need to be divided, and finally form a confidence decision tree;

The traversal order is top-down, first traversing sibling nodes at the same level, if there are no sibling nodes that can be divided, then traversing lower-level nodes.

The classifier is configured to determine the group cluster where the target belongs to according to the confidence that the target belongs to a certain child node in the confidence decision tree.

The effect of the present invention is evaluated and verified on the problem of target grouping in an adversarial environment.

(1) Problem description

The main task of target grouping is to divide the confrontation target into different confrontation groups based on the attributes of the confrontation target detected by the sensor and the communication links between the confrontation targets. The attributes of the confrontation target detected by the sensor will be used as the attribute when the confidence decision tree is divided at the node, such as the three-dimensional coordinates of the target, the angle with the x-axis, the angle with the y-axis, the speed, the air defense capability, the anti-ship capability, the ground Attack capability, reconnaissance and early warning, electromagnetic suppression, communication capability, obstacle clearance capability, etc. The relevant attributes of the target's confrontation ability can be digitized according to the strength of the ability to facilitate the selection of cutting points when dividing the decision tree. The intercepted communication link between confrontation targets can be used as an item in the objective function, and the division methods that have communication links but are not classified into the same cluster are punished. In this way, the present invention adjusts the objective function for calculating a single target evaluation value in formula 1 to obtain the following formula:

Among them, Node ⁼ is the node where X is located, Node ^≠ is the target group closest to X, and Re(') is a set of targets that are associated with target X, and θ(X') is the confidence of X' to Node ^≠ Spend. γ is the proportion of the target profile measurement in its evaluation value, which can be adjusted according to the importance of the relationship in the evaluation. The value of γ is between [0,1], and the smaller the value, the greater the proportion of the relationship in the evaluation. big. The part of evaluating the object through the association relationship is composed of the second part of formula 9, where the fraction represents the ratio of the membership degree of the object that is associated with the object and located in the same cluster to the membership degree of all associated objects. If the communication relationship cannot be obtained, it can be calculated according to formula 1.

(2) Experimental results

The tree structure obtained through the confidence decision tree is shown in Figure 3 (since there is no target with the highest confidence for the node in some leaf nodes, such nodes are not shown in the figure).

The present invention respectively carries out confidence division to the anti-submarine capability (F3) of the confrontation target, the speed (F14), the X axis (F9), the Y axis (F10), the Z axis (F11), generates 6 confrontation groups and any two of them. The union between them, so as to obtain the division result and its corresponding confidence. The final grouping results of the adversarial targets under the plane are shown in Figure 2 through color distinction. Among them, cluster A is green, cluster B is pink, cluster C is black, cluster D is yellow, cluster E is red, cluster F is blue, A∪D is dark gray, A∪E is light green, A∪F is Brown, B∪D is orange, D∪E is purple, D∪F is gray, and E∪F is cyan. The confrontation targets contained in each confrontation group belong to the confrontation group with the maximum confidence. The specific confrontation targets and their confidence levels included in each confrontation group are shown in Table 1.

Table 1 Target intelligence information

In addition, the terms "first" and "second" are used for descriptive purposes only, and cannot be interpreted as indicating or implying relative importance or implicitly specifying the quantity of indicated technical features. Thus, a feature defined as "first" and "second" may explicitly or implicitly include one or more of these features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

Although the present application has been described in conjunction with various embodiments here, however, in the process of implementing the claimed application, those skilled in the art can understand and Other variations of the disclosed embodiments are implemented. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality.

The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be assumed that the specific implementation of the present invention is limited to these descriptions. For those of ordinary skill in the technical field of the present invention, without departing from the concept of the present invention, some simple deduction or replacement can be made, which should be regarded as belonging to the protection scope of the present invention.

Claims (10)

1. A method for clustering targets based on confidence decision tree clustering, comprising:

step 1, acquiring individual attributes, capability attributes and communication relations of a target through detection;

step 2, placing all targets on the root node;

step 3, dividing the child nodes from the root node, wherein the ith dividing process is as follows:

(1) For a single cluster node, selecting any attribute of the node, selecting a possible cutting point in the attribute, and determining a real cutting point by calculating the average confidence contour measurement of all the possible cutting points and finding out the maximum value of the average confidence contour measurement;

(2) Aiming at the single cluster node, assuming that the node needs to be split to generate three child nodes, obtaining a maximum evaluation value by calculating confidence contour metrics of all targets at all possible cutting points, and judging whether the obtained maximum evaluation value is larger than an evaluation value before division; if yes, continuing to execute the step (3), otherwise, not dividing on the node;

if the node is a root node, the method can be directly executed (3) without judgment, if yes, the method continues to be executed (3), otherwise, the method does not need to be divided on the node;

(3) Dividing the single cluster node into child nodes, and determining the center point of each child node according to the real cutting point;

(4) Calculating the confidence coefficient of each child node through the distances from all the targets to each center node; wherein the confidence level expresses the degree to which the target is affiliated to the cluster represented by the child node;

step 4, traversing all nodes until all nodes do not need to be divided, and finally forming a confidence decision tree;

the traversal sequence is from top to bottom, the sibling nodes of the same level are traversed preferentially, and if no partitionable sibling nodes exist, the subordinate nodes are traversed again;

and step 5, determining the group cluster where the target is located according to the confidence degree of the target belonging to a certain child node in the confidence decision tree.

2. The confidence decision tree clustering based target clustering method of claim 1, wherein in the ith partitioning process (1) comprises:

(11) For a single cluster node, selecting any target attribute of the node;

(12) Traversing all possible cutting points in the selected target attributes, and calculating confidence contour metrics of all targets after cutting the attributes according to each possible cutting point;

(13) And calculating the average value of the confidence contour measurement under each cutting point, and selecting the cutting point with the maximum average value of the confidence contour measurement as the real cutting point.

3. The confidence decision tree clustering based target clustering method of claim 1, wherein (2) in the ith partitioning process comprises:

(21) For the single cluster node, assuming that the node needs to be split, calculating confidence contour metrics of all targets at all possible cutting points;

(22) Calculating the average value of the confidence contour measurement after splitting under each possible cutting point aiming at the single cluster node;

(23) Selecting a maximum average value from the average values of the confidence profile metrics obtained in the step (22), obtaining a maximum evaluation value according to the maximum average value, judging whether the obtained maximum evaluation value is larger than the evaluation value before division, if yes, splitting can be performed, otherwise, splitting on the node is not needed.

4. The confidence decision tree clustering based target clustering method as claimed in claim 1, wherein (3) in the ith partitioning process comprises:

(31) Executing a generate child node if the condition in (2) is satisfied;

(32) If the child node is a single cluster node, calculating the central point of the single cluster node according to the single cluster node central calculation expression;

(33) If the child node is a composite cluster node, determining the real cutting point as the center point of the composite cluster node, and calculating the center points of the rest single cluster nodes.

5. The confidence decision tree clustering based target clustering method as claimed in claim 1, wherein in the i-th partitioning process (4) comprises:

(41) Calculating the distances from all the targets to the center points corresponding to each sub-node;

(42) From the distances in (41), a confidence level is calculated for each child node.

6. The confidence decision tree clustering based target clustering method of claim 1, wherein prior to (3) in the ith partitioning process, the confidence decision tree clustering based target clustering method further comprises:

a1, judging whether the node is a single cluster node or a composite cluster node;

b1, if the node is a single cluster node, taking the real cutting point in the step (1) as the cutting point of the node generation child node;

c1, if the node is a compound cluster node, two brother nodes of the compound cluster node need to be divided;

d1, obtaining at least one real cutting point of the brother node in the process of dividing the child nodes;

if two brothers are not partitionable, the node does not need to be partitionable;

and e1, taking the real cut obtained in the d1 as a real cut point of a child node generated by the composite cluster node, and calculating the center and the confidence of the child node at one time.

7. The method for clustering targets based on confidence decision tree according to claim 6, wherein in d, if only one of the sibling nodes needs to generate a child node, the obtained real cut points are 1, and if both sibling nodes need to generate child nodes, the obtained real cut points are 2;

in the process of dividing the composite cluster node, when the number of the real cutting points is 2, any one real cutting point can be selected to perform one-time division so as to generate two child nodes, and the other real cutting point is selected to perform one-time division on the generated two child nodes so as to generate two child nodes.

8. The confidence decision tree clustering based target clustering method of claim 1, further comprising, after step 4:

a2, determining child nodes which do not need to be divided in the confidence decision tree, and enabling the child nodes to be leaf nodes;

b2, judging whether the number of the leaf nodes is larger or smaller than a given number;

b3, if the number is larger or smaller than the given number, adjusting the leaf nodes in the confidence decision tree until the leaf nodes are consistent with the given number.

9. The confidence decision tree clustering based target clustering method of claim 1, wherein b3 comprises:

if the number of the leaf nodes is larger than the given number, merging the leaf node with the worst confidence into the leaf node closest to the leaf node to obtain a merged node;

determining composite cluster nodes of two leaf nodes to be combined;

combining the combined nodes with the composite cluster nodes to reduce the number of single clusters by one, and repeating the process until the number of leaf nodes is consistent with the given number;

if the number of the leaf nodes is smaller than the given number, the current leaf nodes are attempted to be divided, and the maximum evaluation value is not limited in the dividing process and is required to be larger than the evaluation value before division;

and selecting the mode with the largest evaluation value from the dividing modes for dividing until the number of the leaf nodes is consistent with the given number.

10. A confidence decision tree clustering-based target clustering device, comprising:

the detector is used for acquiring the target attribute and the related functional attribute of the target through detection;

an executor for placing all targets at the root node; the sub-node is divided from the root node, and the ith dividing process is as follows:

(1) For a single cluster node, selecting any attribute of the node, selecting a possible cutting point in the attribute, and determining a real cutting point by calculating the average confidence contour measurement of all the possible cutting points and finding out the maximum value of the average confidence contour measurement;

(2) Aiming at the single cluster node, assuming that the node needs to be split to generate three child nodes, obtaining a maximum evaluation value by calculating confidence contour metrics of all targets at all possible cutting points, and judging whether the obtained maximum evaluation value is larger than an evaluation value before division; if yes, continuing to execute the step (3), otherwise, not dividing on the node;

if the node is a root node, the method can be directly executed (3) without judgment, if yes, the method continues to be executed (3), otherwise, the method does not need to be divided on the node;

(3) Dividing the single cluster node into child nodes, and determining the center point of each child node according to the real cutting point;

(4) Calculating the confidence coefficient of each child node through the distances from all the targets to each center node; wherein the confidence level expresses the degree to which the target is affiliated to the cluster represented by the child node;

step 4, traversing all nodes until all nodes do not need to be divided, and finally forming a confidence decision tree;

the traversal sequence is from top to bottom, the sibling nodes of the same level are traversed preferentially, and if no partitionable sibling nodes exist, the subordinate nodes are traversed again;

and the classifier is used for determining the group cluster where the target is located according to the confidence that the target belongs to a certain child node in the confidence decision tree.