A group target intention recognition method and system based on a dynamic Bayesian network

Abstract

The application discloses a group target intention recognition method and system based on a dynamic Bayesian network, and the method comprises the following steps: acquiring attribute and state information of n targets contained in radar detection information, and extracting target characteristic attributes associated with target intention to form a state set; taking the target intention as a root node of a Bayesian network, taking the target characteristic attributes associated with the target intention as child nodes under the root node of the Bayesian network, and obtaining a Bayesian network topological structure based on the dependency relationship between the child nodes; determining each intention prior probability, conditional probability and state transition probability based on historical data and in combination with expert knowledge, and determining a dynamic Bayesian network model; acquiring observation data corresponding to each characteristic attribute and performing preprocessing, determining a state corresponding to a node variable, and performing intention recognition based on the dynamic Bayesian network model. The method can realize group division of target swarms at a space group level and intention reasoning at a functional group level.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of situation assessment, more particularly to a group target intention recognition method and system based on dynamic Bayesian network. BACKGROUND

[0002] Situation assessment is a means of perceiving, understanding and predicting various information on the battlefield, which can provide auxiliary information for decision makers, and thus improve the correctness of decision-making. Target intention recognition is one of the main functions of situation assessment, which can provide evidence support for situation prediction. Group target intention recognition is an automatic assessment and prediction of the purpose or combat plan of group targets in the combat area, which is an important content of situation analysis and belongs to the high-level processing part of the battlefield data fusion system.

[0003] The process of target intention recognition can be generally summarized as follows: according to the information obtained from various information sources, a certain method is used to filter and extract the features of the obtained information, and a target intention result is obtained through a certain recognition reasoning mechanism. The existing intention recognition methods include template matching, Bayesian reasoning, decision tree, fuzzy reasoning and neural network, etc. The existing group target intention recognition method in the air is to first group, and then analyze the group intention based on the grouping result. The current target grouping methods mainly include three types: grouping method based on similarity matrix, grouping method based on clustering and grouping method based on Bayesian network. Most algorithms are improved on the basis of these three types of algorithms. Grouping targets is to use the extracted situation elements to merge targets with similar types, similar motion states and same combat purposes in spatial and functional levels. Then, based on the grouping result, the intention is recognized by using reasoning or learning mechanism according to the behavior characteristics of the group target. However, the group target intention recognition based on the grouping result often ignores the changes of the feature information of the target in the grouping process, and taking the group target as the research object also does not consider the intention change of the single target. SUMMARY

[0004] The present application provides a group target intention recognition method based on dynamic Bayesian network, which can divide the spatial group of the air target in real time, and obtain the group intention through the functional group judgment.

[0005] Another object of the present application is to provide a group target intention recognition system based on dynamic Bayesian network.

[0006] Technical scheme: In a first aspect, a group target intention recognition method based on dynamic Bayesian network comprises the following steps:

[0007] Obtain the attribute and state information of n targets contained in the radar detection information, and extract the target feature attributes associated with the target intention to form a state set;

[0008] the target intention as a root node of a Bayesian network, target feature attributes associated with the target intention as child nodes under the root node of the Bayesian network, and obtaining a Bayesian network topology structure based on dependency relationships between the child nodes;

[0009] determining each intention prior probability, conditional probability and state transition probability based on historical data and in combination with expert knowledge, and determining a dynamic Bayesian network model;

[0010] obtaining observation data corresponding to each feature attribute and performing preprocessing, determining a state corresponding to a node variable, performing intention recognition based on the constructed dynamic Bayesian network model, and obtaining an intention recognition result.

[0011] Preferably, the target feature attributes associated with the target intention include a target distance D, a target radar scattering cross-sectional area R, a target trajectory type M, a radiation source type P and a target type T.

[0012] Preferably, the Bayesian network topology structure is as follows: when there is no dependency relationship between the child nodes, the target intention is the root node, and the target feature attributes associated with the target intention are first-level child nodes under the root node; when there is a dependency relationship between the child nodes, an upper-lower structure between the child nodes is determined according to the dependency relationship, wherein a dependent child node is in the upper layer, a child node dependent on other child nodes is in the lower layer, the target intention is the root node, and a dependent child node not dependent on other child nodes is a first-level child node under the root node.

[0013] Preferably, the state corresponding to the node variable is determined according to a state set of the node variable, and the state set of the node variable is:

[0014] the target distance D = {first defense zone (D1), second defense zone (D2), third defense zone (D3)}

[0015] the target radar scattering cross-sectional area R = {large (R1), small (R2)}

[0016] the target trajectory type M = {straight line (M1), circle (M2), S-bend (M3), turning (M4)}

[0017] the radiation source type P = {no radiation source (P1), radar radiation source (P2), jamming radiation source (P3)}

[0018] the target type T = {fixed-wing aircraft (T1), multi-rotor aircraft (T2), missile (T3)}.

[0019] Preferably, the method represents the prior probability of the i-th intention as P(TIi), i = 1, 2, …, n, n is the number of intention types, and A jrepresents the attribute, j = 1, 2, …, m, m is the number of attributes, and the conditional probability of the intention i is:

[0020]

[0021] P (TI | A) represents the probability of the attribute being A when the intention i; i

[0022] The dynamic Bayesian network model multiplies P (TI i | A1, A2, …, A m ) by the prior probability P (TI i ) of the corresponding intention to obtain the posterior probability, and the obtained posterior probability is taken as the prior probability of the intention i at the next moment.

[0023] In a second aspect, a group target intention recognition system based on a dynamic Bayesian network is provided, and the system comprises:

[0024] A data extraction module is configured to acquire attribute and state information of n targets contained in radar detection information, and extract target feature attributes associated with target intention to form a state set.

[0025] A Bayesian network construction module is configured to take the target intention as a root node of a Bayesian network, take target feature attributes associated with the target intention as child nodes under the root node of the Bayesian network, and obtain a Bayesian network topology structure based on a dependency relationship between the child nodes.

[0026] A dynamic Bayesian network model determination module is configured to determine each intention prior probability, conditional probability, and state transition probability based on historical data and in combination with expert knowledge, and determine a dynamic Bayesian network model.

[0027] A target intention recognition module is configured to acquire observation data corresponding to each feature attribute and perform preprocessing, determine a state corresponding to a node variable, perform intention recognition based on the constructed dynamic Bayesian network model, and obtain an intention recognition result.

[0028] In a third aspect, a computer device is provided, and the computer device comprises:

[0029] One or more processors;

[0030] Memory; and

[0031] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs are implemented when executed by the processor to implement the group target intention recognition method based on the dynamic Bayesian network according to the first aspect of the present application.

[0032] ​In a fourth aspect, a computer readable storage medium is provided, and a computer program is stored on the computer readable storage medium, and the computer program is executed by a processor to implement the steps of the group target intention recognition method based on a dynamic Bayesian network according to the first aspect of the present application.

[0033] Beneficial effects: The air group target intention recognition method based on a dynamic Bayesian network according to the present application selects multiple target feature information as nodes, constructs a dynamic Bayesian network model, and uses the probability calculation result of the previous moment to learn and update the prior probability. With the passage of time, more and more effective information is obtained, and the accuracy of the reasoning result is also higher and higher. The results show that the method of the present application can group the target swarm in the space group level, and simultaneously perform intention reasoning in the functional group level. Compared with the method of grouping targets and then performing group target intention recognition, the method of the present application does not separate the grouping of targets and the recognition of intentions, and is more suitable for the demand of dynamic change of the battlefield environment. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a flowchart of the recognition method of the present application;

[0035] Figure 2 is a topological structure of the static Bayesian network of the present application;

[0036] Figure 3 is a topological structure of the dynamic Bayesian network of the present application;

[0037] Figure 4 is a flight trajectory plan view of all targets in the simulation example of the present application;

[0038] Figure 5 is a Bayesian reasoning flowchart in the simulation example of the present application;

[0039] Figure 6 is a dynamic Bayesian network model in the simulation example of the present application;

[0040] Figure 7 is the intention probability of each target in the simulation example of the present application. DETAILED DESCRIPTION

[0041] The technical solutions of the present application will be further described below with reference to the accompanying drawings.

[0042] The technical solutions adopted by the present application are to use a dynamic Bayesian network model to perform real-time analysis on the grouping process of the air unmanned aerial vehicle swarm target, obtain the grouping result, and also infer the group intention of each group. Specifically, the following steps are included:

[0043] Step 1: According to the first-level data fusion information detected by the sensor, a suitable characteristic variable is selected as a node of the dynamic Bayesian network.

[0044] The focus of the target grouping is how to extract and summarize the output information of the first-level data fusion. Each attribute set of the target is not limited to the spatial coordinates, but also includes the speed, the heading, the batch number, the radar cross section (RCS), the distance, the radiation source type, the radiation source direction, etc. In the scenario of the UAV swarm, the UAV target characteristic parameters in each functional group are similar, and when the target grouping is performed, the characteristics that are greatly different between the groups should be considered for observation. According to the embodiment of the present application, it is assumed that at time t, the radar information of the first-level data fusion includes the attributes and state information of n targets, and the state set formed is:

[0045] A=D, R, M, P, T

[0046] Wherein, D is the distance of the target, R is the radar cross section of the target, M is the trajectory type of the target, such as straight line, circle, ellipse, S-bend, parabola, etc., P is the radiation source type, such as radar radiation source, jamming radiation source, navigation radiation source, communication radiation source, etc., and T is the target type, such as fixed-wing aircraft, multi-rotor aircraft, early warning aircraft, missile, etc.

[0047] Step 2: The dependency relationship between the nodes is determined, and the network topology structure is obtained based on the dependency relationship.

[0048] The network structure of the Bayesian network is a directed acyclic graph, which is composed of nodes and arcs, and the arcs between the nodes represent the probability dependency relationship between the attributes. The combat intention of the air group target mainly includes reconnaissance, decoy, jamming, counter-radiation, and attack. The purpose of the present application is to identify the combat intention according to the characteristics extracted from the first-level data fusion information. The five characteristic variables obtained in step 1 are observable nodes, while the intention is an unobservable node, which is the target of the Bayesian network model, so the intention node is the root node, and the characteristic node is the child node. According to the causal relationship, this is a bottom-up reasoning. In theory, there are many network topology structures composed of the five characteristic variables, and the calculation amount and time are large for the calculation of each structure, so expert knowledge can be used for determination. Since the five characteristic variables described in step 1 are independent of each other and are independent, the basic static Bayesian network topology structure is as follows: Figure 2In other embodiments, there can be dependency relationship between each sub-node, then only need to connect them with arc line with arrow according to the dependency relationship. The arc line connection determines the upper and lower structure between the sub-nodes, wherein the depended sub-node is in the upper layer, and the sub-node depending on other sub-nodes (non-independent sub-node) is in the lower layer, so that in the network topology, the target intention is the root node, the depended sub-node not depending on other sub-nodes is the first level sub-node under the root node, and the other non-independent sub-nodes connected to the depended sub-node are the next level sub-nodes, and if there are multiple layers of dependency relationship, the above can be used by analogy.

[0049] The dynamic Bayesian network is an extension of the static Bayesian network in time, wherein the dynamic refers to the change of time, rather than the change of network structure. The static Bayesian network can only make inference according to the characteristic parameters at a specific time, and the grouping process often needs a period of time to complete, the dynamic Bayesian network can capture the dynamic change of each target parameter with time in the grouping process, so that the grouping result is more accurate. The topology structure of the dynamic Bayesian network is as shown in Figure 3

[0050] Step 3: determining the intention prior probability, conditional probability and state transition probability based on expert knowledge.

[0051] The prior probability represents the probability of each intention of a single target, which is generally determined by the domain experts combined with experience and knowledge. The present application learns the conditional probability by using a large amount of historical data, and adjusts it combined with expert knowledge. The following are some behavior characteristics of each sub-group summarized by experts after many times of combat experience.

[0052] The reconnaissance group is usually far away, and the RCS is small. In order to detect the position and information of the enemy, the radar state must be turned on, the radar signal can be detected, and the motion trajectory is commonly a circle. The decoy group is close, and the RCS is large. The motion trajectory is commonly a straight line. The jamming group is in the middle distance, and the RCS is small. The interference signal can be detected, and the motion trajectory is commonly a straight line. The anti-radiation group is in the middle distance, and the RCS is small. The motion trajectory is commonly a straight line. The attack group is close, and the RCS is small. In order to avoid being found by the enemy, the radar state must be turned off, and the motion trajectory is commonly a straight line.

[0053] The learning sample data is composed of the above historical experience data, and the maximum likelihood estimation method is used for parameter learning. The maximum likelihood estimation is to calculate the frequency of different values of a given node set, and this frequency is generally used as the prior probability for calculating the conditional probability.

[0054] ​The state transition probability is an empirical probability, which is generally determined by domain experts combined with experience and knowledge, and represents the state probability of each intention at the next time under the condition that the intention probability of each intention at the last time is known. Compared with static Bayesian, dynamic Bayesian needs to set the state transition probability between two time slices, i.e. P[(T+1) / T].

[0055] Assuming that P(TIi) is the prior probability of the ith intention, i = 1, 2, …, n, n is the number of intention types, and A represents the characteristic attribute of the target, according to the Bayesian probability formula, the conditional probability of intention i is,

[0056]

[0057] wherein P(TI i ) is the prior probability of intention i, and P(A|TI i ) is the probability that the attribute is A when the intention is i.

[0058] In general cases, by A j , j = 1, 2, …, m, representing the attribute, m is the number of attributes, when m characteristic attributes are observed, the conditional probability of intention i can be expressed as:

[0059]

[0060] P(TI i |A1,A2,…A m ) is multiplied by the prior probability P(TI i ) of the corresponding intention to obtain the posterior probability, and the posterior probability calculated in this way is the prior probability of intention i at the next time.

[0061] Step 4: Preprocessing the observed data, determining the state of the node variable, and performing intention recognition based on the constructed dynamic Bayesian network model to obtain the intention recognition result.

[0062] For the selected continuous variable of the distance of the air target, it refers to the straight-line distance of the target from our side. According to the task requirements, the key defense area of our side can be regionally divided: when the distance is far, it belongs to the first defense area; when the distance is medium, it is in the second defense area; and when the distance is close, it is in the third defense area. The specific distance threshold can be set according to the defense requirements.

[0063] For the RCS value of the target, it can be distinguished as large and small. If the RCS value of a target is greater than the average value of all targets, it is considered that the RCS state value of the target is large, otherwise the state value is small.

[0064] For the discrete variable of target trajectory type, the common ones are straight line, circle, S-bend and turning, and the four trajectory styles of straight line, circle, S-bend and turning are selected in the software template library to form the trajectory state set.

[0065] For the radiation source type, the types that can be output by considering the first level fusion data are no radiation source, radar radiation source and jamming radiation source.

[0066] For the target type, the common air combat target types include fixed-wing aircraft, multi-rotor aircraft and missile.

[0067] For the combat intention of the air group target, the main ones are reconnaissance, decoy, jamming, anti-radiation and attack.

[0068] Therefore, the state sets of the target variables are represented as follows:

[0069] ① Target distance D = {first defense zone (D1), second defense zone (D2), third defense zone (D3)}

[0070] ② Target RCS R = {large (R1), small (R2)}

[0071] ③ Target trajectory type M = {straight line (M1), circle (M2), S-bend (M3), turning (M4)}

[0072] ④ Radiation source type P = {no radiation source (P1), radar radiation source (P2), jamming radiation source (P3)}

[0073] ⑤ Target type T = {fixed-wing aircraft (T1), multi-rotor aircraft (T2), missile (T3)}

[0074] ⑥ Target intention TI = {reconnaissance (TI1), decoy (TI2), jamming (TI3), anti-radiation (TI4), attack (TI5)}

[0075] The states of the five characteristic variables of the target obtained in real time are determined by using the above preprocessing method, and the posterior probability of each intention can be obtained by inferring the states of the child nodes of the Bayesian network from bottom to top.

[0076] The implementation steps of the intention recognition method proposed in the application are described above, in order to verify the effectiveness of the method, the following simulation test is used for illustration.

[0077] The simulation scenario is as follows: an enemy F35 aircraft flies to our observation point, and drops 53 unmanned aerial vehicles outside our defense zone (300 Km). The unmanned aerial vehicles enter radially at a speed of 100 m / s and a height of 3000 m, in an approximate linear formation, and the F35 aircraft returns. The enemy unmanned aerial vehicle swarm target continues to enter and advances towards our side. When it reaches 100 Km, it changes formation. The enemy unmanned aerial vehicle formation is divided into 6 subgroups: detection group, reconnaissance group, jamming group, decoy group, attack group, and anti-radiation group. The entire simulation process takes 3395 seconds. Figure 4 is a flight trajectory plan view of all targets, and 53 targets are analyzed at 5 typical time points.

[0078] When flying to 500 seconds, all 53 targets are in the first defense zone and are far away from our side, and the RCS values are all 0.05 m 2 .

[0079] When flying to 1000 seconds, the distance between the unmanned aerial vehicle targets increases significantly, and the flight direction changes. It can be determined that all 53 targets are fixed-wing aircrafts.

[0080] When flying to 1600 seconds, batch numbers 1-6 unmanned aerial vehicle targets are in the first defense zone, and the other unmanned aerial vehicles are in the second defense zone.

[0081] When flying to 2100 seconds, batch numbers 7-13 and 34-53 unmanned aerial vehicle targets enter the third defense zone and are close to our side, and the other unmanned aerial vehicle targets fly in the previous defense zone. The RCS value of batch numbers 7-13 unmanned aerial vehicle targets is 0.5 m 2 , and radar signals and jamming signals are detected.

[0082] When flying to 2800 seconds, it is identified that the motion trajectory of batch numbers 1-6 targets is a circle, and the trajectories of the other unmanned aerial vehicles are straight lines.

[0083] According to the above information, the specific method process is shown in Figure 5 . Five characteristic variables such as distance, RCS, radiation source type, trajectory type, and target type are selected as nodes of the dynamic Bayesian network, and a dynamic Bayesian network model is constructed using GeNIE2.0 software, as shown in Figure 6 . The probability distribution of each intention is relatively average at the initial time, and the prior probability is set as (0.2, 0.2, 0.2, 0.2, 0.2). The conditional probability table is obtained by combining historical data and expert knowledge, as shown in Table 1.

[0084] Table 1 Conditional probability table

[0085]

[0086] The battlefield environment changes unpredictably, and each intention is independent of each other, so the state transition probability is shown in Table 2.

[0087] Table 2 State transition probability table

[0088]

[0089] The target’s feature information is preprocessed to determine the state of each feature and input into the dynamic Bayesian network to obtain the intention probability of each target, such as Figure 7 As shown, Figure 7 (a) shows the probability of each intention for targets in batches 1-6, (b) shows the probability of each intention for targets in batches 7-13, (c) shows the probability of each intention for targets in batches 14-23, (d) shows the probability of each intention for targets in batches 24-33, and (e) shows the probability of each intention for targets in batches 34-53. After 2800 seconds, the movement states of the targets remain essentially unchanged. Targets in batches 1-6 have the highest probability of reconnaissance intention, reaching 0.99, forming the reconnaissance group. Targets in batches 7-13 have the highest probability of decoy intention, reaching 0.81, forming the decoy group. Targets in batches 14-23 have the highest probability of jamming intention, reaching 0.78, forming the jamming group. Targets in batches 24-33 have the highest probability of anti-radiation intention, reaching 0.72, forming the anti-radiation group. Targets in batches 34-53 have the highest probability of reconnaissance intention, reaching 0.54, forming the attack group. The results of clustering and intention recognition are consistent with the simulation. It can be seen that the method of the present invention can simultaneously achieve clustering and intention recognition of swarm targets.

[0090] The present invention also provides a group target intention recognition system based on a dynamic Bayesian network, comprising:

[0091] The data extraction module is used to obtain the attributes and state information of n targets contained in the radar detection information, and extract the target feature attributes associated with the target intention to form a state set;

[0092] A Bayesian network construction module is used to take the target intent as the root node of the Bayesian network, take the target feature attributes associated with the target intent as the child nodes under the Bayesian network root node, and obtain the Bayesian network topology structure based on the dependency relationship between the child nodes;

[0093] Dynamic Bayesian network model determination module, used to determine the prior probability, conditional probability and state transition probability of each intention based on historical data and combined with expert knowledge, and determine the dynamic Bayesian network model;

[0094] The target intention recognition module is used to obtain the observation data corresponding to each feature attribute and perform preprocessing, determine the state corresponding to the node variable, perform intention recognition based on the constructed dynamic Bayesian network model, and obtain the intention recognition result.

[0095] It should be understood that the dynamic Bayesian network-based group target intention recognition system provided in the embodiments can implement all the technical solutions in the method embodiments, and the functions of the various functional modules can be implemented according to the methods in the method embodiments, and the specific implementation process can be referred to the related description in the above embodiments, which will not be described here.

[0096] Those skilled in the art will appreciate that embodiments of the application can be devised for use with either software, hardware, or a combination of software and hardware. Moreover, the software can be embodied on a computer readable storage medium. The present application can be realized in a centralized fashion in one computer system or it can be realized in a distributed fashion where different elements are spread across multiple computer systems. Any kind of computer system or other apparatus adapted for carrying out the methods described herein is suited. A typical combination of hardware and software could be a general purpose computer system with a computer program that, when being loaded and executed, carries out the methods described herein. As will be realized, the application is capable of great adaptability and can be implemented in numerous types of software constructs.

[0097] The present application also can be embodied in a computer program product adapted to perform any one of the methods described herein. A computer program can be written in any form suitable for use in the intended computer or data processing environment, including compiled or interpreted languages, and it can be deployed in a number of ways, such as for example, as a stand-alone program or as a utility, tool, service, function or routine running on or across one or more computers. Figure 1 one or more processes and / or blocks Figure 1 means for functionally implementing the one or more processes and / or blocks

[0098] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 one or more processes and / or blocks Figure 1 means for functionally implementing the one or more processes and / or blocks

[0099] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 one or more processes and / or blocks Figure 1 means for functionally implementing the one or more processes and / or blocks

Claims

1. A group target intention recognition method based on a dynamic Bayesian network, characterized in that, The method comprises the following steps: Obtaining attribute and state information of n targets contained in radar detection information, extracting target feature attributes associated with target intention to form a state set, wherein the target feature attributes associated with target intention include target distance D, target radar scattering cross-sectional area R, target trajectory type M, radiation source type P and target type T; Taking the target intention as a root node of a Bayesian network, taking the target feature attributes associated with the target intention as child nodes under the root node, obtaining a Bayesian network topology based on the dependency relationship between the child nodes, wherein when there is no dependency relationship between the child nodes, the target intention is the root node and the target feature attributes associated with the target intention are first-level child nodes under the root node; when there is a dependency relationship between the child nodes, determining the upper and lower layer structures between the child nodes according to the dependency relationship, wherein the dependent child nodes are in the upper layer and the child nodes dependent on other child nodes are in the lower layer, the target intention is the root node and the dependent child nodes not dependent on other child nodes are first-level child nodes under the root node; Determining the intention prior probability, the conditional probability and the state transition probability based on historical data and in combination with expert knowledge, determining a dynamic Bayesian network model, wherein the dynamic Bayesian network is a time extension of a static Bayesian network, and the dynamic refers to the change of time; Obtaining observation data corresponding to each feature attribute and performing preprocessing, determining the state corresponding to the node variable, performing intention recognition based on the constructed dynamic Bayesian network model, and obtaining an intention recognition result.

2. The method of claim 1, wherein, The state corresponding to the node variable is determined according to the state set of the node variable, and the state set of the node variable is: Target distance D={first defense zone (D1), second defense zone (D2), third defense zone (D3)} Target radar scattering area R={large (R1), small (R2)} Target trajectory type M={straight line (M1), circle (M2), S bend (M3), turning (M4)} Radiation source type P={no radiation source (P1), radar radiation source (P2), jamming radiation source (P3)} Target type T={fixed-wing aircraft (T1), multi-rotor aircraft (T2), missile (T3)}.

3. The method of claim 1, wherein, Let P(Ti) denote the prior probability of the ith intention, i = 1, 2, …, n, n is the number of intention types, which is calculated by A j represents the attribute, j = 1, 2, …, m, m is the number of attributes, and the conditional probability of intention i is: P(A | T1 i ) denotes the probability that the attribute is A given that the intent i was intended. The dynamic Bayesian network model multiplies P(TI i |A1,A2,…A m ) by the prior probability P(TI i ) of the corresponding intention to obtain a posterior probability, and the obtained posterior probability is taken as the prior probability of the intention i at the next time.

4. A dynamic Bayesian network based group target intent recognition system, characterized in that, The method comprises the following steps: A data extraction module is configured to obtain attribute and state information of n targets contained in radar detection information, and extract target feature attributes associated with target intention to form a state set, wherein the target feature attributes associated with target intention include target distance D, target radar scattering cross-sectional area R, target trajectory type M, radiation source type P and target type T; The Bayesian network construction module is configured to take the target intention as a root node of a Bayesian network, take the target feature attributes associated with the target intention as child nodes under the root node of the Bayesian network, and obtain a Bayesian network topology based on the dependency relationships between the child nodes. The Bayesian network topology is as follows: when there is no dependency relationship between the child nodes, the target intention is the root node, and the target feature attributes associated with the target intention are first-level child nodes under the root node; when there is a dependency relationship between the child nodes, the upper and lower layer structures between the child nodes are determined according to the dependency relationship, wherein the dependent child nodes are in the upper layer, the child nodes that depend on other child nodes are in the lower layer, the target intention is the root node, and the dependent child nodes that do not depend on other child nodes are first-level child nodes under the root node. The dynamic Bayesian network model determination module is configured to determine the prior probabilities, the conditional probabilities and the state transition probabilities of each intention based on historical data and in combination with expert knowledge, and determine a dynamic Bayesian network model. The dynamic Bayesian network is an extension of a static Bayesian network in time, and the dynamic refers to the change in time. The target intention recognition module is configured to acquire observation data corresponding to each feature attribute, pre-process the observation data, determine states corresponding to node variables, perform intention recognition based on the constructed dynamic Bayesian network model, and obtain an intention recognition result.

5. A computer device, comprising: The method comprises: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs, when executed by the processors, implement the group target intention recognition method based on a dynamic Bayesian network according to any one of claims 1-3.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer programs, when executed by the processors, implement the steps of the group target intention recognition method based on a dynamic Bayesian network according to any one of claims 1-3.

Images