CN116029379A - Method for constructing air target intention recognition model - Google Patents
Method for constructing air target intention recognition model Download PDFInfo
- Publication number
- CN116029379A CN116029379A CN202211737280.5A CN202211737280A CN116029379A CN 116029379 A CN116029379 A CN 116029379A CN 202211737280 A CN202211737280 A CN 202211737280A CN 116029379 A CN116029379 A CN 116029379A
- Authority
- CN
- China
- Prior art keywords
- data
- intention
- decision tree
- network
- recognition model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000003066 decision tree Methods 0.000 claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 29
- 230000003068 static effect Effects 0.000 claims abstract description 24
- 238000012545 processing Methods 0.000 claims abstract description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 19
- 238000010276 construction Methods 0.000 claims description 8
- 230000002068 genetic effect Effects 0.000 claims description 8
- 238000012546 transfer Methods 0.000 claims description 8
- 238000013138 pruning Methods 0.000 claims description 7
- 230000035772 mutation Effects 0.000 claims description 5
- 238000007476 Maximum Likelihood Methods 0.000 claims description 3
- 210000000349 chromosome Anatomy 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 description 11
- 238000001514 detection method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 230000002123 temporal effect Effects 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 125000002015 acyclic group Chemical group 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000000540 analysis of variance Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000009960 carding Methods 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000000546 chi-square test Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The embodiment of the invention provides a method for constructing an air target intention recognition model, which comprises the following steps: acquiring type data of all different types of aircrafts in the air and intention data of all different types of aircrafts under different intentions, and taking the type data and the intention data as a training set; establishing an airplane type recognition model based on feature selection and a decision tree, training the decision tree by using type data, and performing parameter learning by using intention data to obtain a trained airplane type recognition model; processing different types of data predicted and output by the aircraft category recognition model; establishing a plurality of static Bayesian network models; each static Bayesian network model corresponds to one type of data, the processed data of different types are input into the corresponding static Bayesian network model to be trained, and finally a trained air target intention recognition model based on a decision tree and the static Bayesian network model is obtained. Different intentions of different types of airplanes can be accurately predicted.
Description
Technical Field
The invention belongs to the technical field of air target intention recognition, and particularly relates to a construction method of an air target intention recognition model.
Background
Different kinds of aircraft in the air target intention recognition process execute different tasks, possible intentions are different, and the characteristics of each intention are different, so that the characteristics of each intention cannot be completely distinguished by using one network.
Disclosure of Invention
The invention aims to at least solve one of the technical problems in the prior art and provides an air target intention recognition model construction method.
The invention provides a method for constructing an air target intention recognition model, which is characterized by comprising the following steps:
acquiring type data of all different types of aircrafts in the air and intention data of all different types of aircrafts under different intentions, and taking the type data and the intention data as a training set;
establishing an aircraft type recognition model based on feature selection and a decision tree, training the decision tree by using the type data, and performing parameter learning by using the intention data to obtain a trained aircraft type recognition model;
processing different types of data predicted and output by the aircraft category recognition model;
establishing a plurality of static Bayesian network models; each static Bayesian network model corresponds to one type of data, the processed data of different types are input into the corresponding static Bayesian network model to be trained, and finally a trained air target intention recognition model based on a decision tree and the static Bayesian network model is obtained.
Optionally, the establishing of the decision tree comprises the following steps:
and (3) feature selection: selecting features capable of classifying training sets, wherein key criteria for feature selection comprise information gain, information gain ratio and Gini index;
generating a decision tree: firstly, constructing a root node, placing original situation training data of all aerial target airplanes in the root node, selecting optimal characteristics in the original situation training data, and dividing a training data set into subsets according to the characteristics so that each subset has optimal classification under the current condition; if the subsets can be correctly classified, constructing leaf nodes and dividing the subsets into corresponding leaf nodes; if subsets which cannot be classified correctly exist, selecting new optimal features for the subsets, continuing partitioning the subsets, and constructing corresponding nodes; if recursion is performed until all subsets of training data are substantially correctly classified, or there are no suitable features, each subset is partitioned into leaf nodes, i.e., there are explicit classes, generating a decision tree;
pruning a decision tree: the pruning of the decision tree is to prevent the overfitting of the tree and enhance the generalization capability of the tree.
Optionally, the processing the data includes: data discretization, data noise adding, data preprocessing and feature selection.
Optionally, the establishing a plurality of static bayesian network models includes:
and (3) structure learning: is responsible for extracting corresponding rules from the data and learning the structure of the network from the data;
parameter learning: learning the probability of each node value using maximum likelihood estimation;
reasoning: and (3) reasoning by using the learned network with parameters and the junction tree algorithm, and finally outputting the intention by comparing the posterior probability of each intention value.
Optionally, the step of structure learning specifically includes:
constructing an initial transfer network by using time mutual information, then performing BIC scoring, encoding the transfer network into a chromosome capable of being used for a genetic algorithm, performing optimal searching by using a self-adaptive genetic algorithm, defining a crossover operator and a mutation operator, and obtaining the optimal network by adopting a preset learning algorithm.
Optionally, the maximum evolution times, population scale and variation probability parameters are preconfigured, and the obtaining the optimal network by adopting a preset learning algorithm includes:
s1, initializing a population, and coding;
s2, calculating fitness, namely BIC scores, of all individuals in the population, arranging the individuals in descending order according to the fitness, and reserving optimal individuals based on elite strategies;
s3, judging whether the difference value between the optimal fitness obtained in the iteration and the optimal fitness obtained in the previous iteration is smaller than a set threshold value; if yes, ending; otherwise, turning to step S4;
s4, selecting parent individuals from the better individuals, generating new individuals through cross mutation, and placing the new individuals into the population to replace the worse individuals to obtain a new population;
s5, judging whether the iteration times reach the maximum evolution times or not; if yes, ending; otherwise go to step S2.
Optionally, the method further comprises:
knowing the label of each Bayesian network, calculating the maximum posterior probability of each network for reasoning out a tactical intention;
and reasoning threat levels of tactical intentions according to each network to carry out comprehensive evaluation.
The method uses a plurality of static Bayesian networks to build the intention recognition model, uses the decision tree model as an auxiliary decision to use which static Bayesian network to carry out reasoning, and can accurately predict different intentions of different types of airplanes.
Drawings
FIG. 1 is a schematic diagram illustrating a non-timing edge situation according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method of constructing an airborne target intention recognition model in accordance with another embodiment of the invention;
FIG. 3 is a schematic diagram of a decision tree according to another embodiment of the present invention;
FIG. 4 is a flow chart of a decision tree generation in accordance with another embodiment of the present invention;
FIG. 5 is a schematic diagram illustrating the height division of an early warning device according to another embodiment of the present invention;
fig. 6 is a flowchart of a bayesian network structure learning algorithm according to another embodiment of the present invention.
Detailed Description
The present invention will be described in further detail below with reference to the drawings and detailed description for the purpose of better understanding of the technical solution of the present invention to those skilled in the art.
The invention mainly provides an aerial target intention recognition model based on decision tree and static Bayesian network based on decision tree construction and static Bayesian network construction. Training a decision tree by using a training set composed of data of all types of aircrafts, inputting a decision tree model by using a testing set composed of data of all types of aircrafts in a testing process, completing type division of the types of aircrafts, and selecting a proper reasoning model according to a division result.
A bayesian network is a graph network based on probabilistic reasoning. The network is supported by variable nodes and builds a hierarchical model of the problem based on causal relationships between events. The process of solving the problem according to this model is called bayesian reasoning. During the reasoning process, a set of variables with known probabilities provides evidence for solving the variables. In fact, the process of solving the variables is the process of deriving a known prior probability. The learning of the bayesian network includes structure learning and parameter learning. Structure learning is to learn the dependency relationship among variables and display the dependency relationship in a graphical mode; parameter learning is to learn the probability dependence of a variable with respect to its parent node set and then obtain a local conditional probability distribution function. Compared with structure learning, the parameter learning of the Bayesian network is more standard and simpler, and the accuracy of the Bayesian network structure directly influences the accuracy of the parameter learning and the reasoning result. Therefore, the learning direction of the current bayesian network mainly focuses on learning of the bayesian network structure. At present, structure learning is still mainly data driven, but bayesian network structures through data learning have proven to be NP-hard, and many problems still need to be solved.
In a bayesian network, directed Acyclic Graphs (DAGs) are formed between nodes. How to build an efficient and optimal DAG becomes the basis and core for offline reasoning using bayesian networks. Generally, there are two methods of establishing a DAG. One based on expert knowledge and the other learning from data. Although experts have rich knowledge and high precision in projects, as the number of nodes in a network increases as the network is built using expert knowledge, the potential number of structures G of the corresponding DAG graph will increase exponentially. Thus, expert knowledge is included in the data during scene simulation and the network structure is learned using methods learned from the data.
The structure learning of a static bayesian network can be converted into the learning of a priori networks and transfer networks under the conditions of stationarity and markov properties. Under the complete dataset, the scoring function of the bayesian network can be broken down into independent factors for the variables and all their parents. If the parent-child structure of one node changes, the scoring results of other nodes will not be affected. Thus, in calculating the scoring function, only the statistical factors of the local structure associated with each node need be calculated. Based on this idea, genetic algorithm can be introduced for structure learning aiming at the characteristics of the non-time sequence edge of the transfer network, and fig. 1 shows the situation of the non-time sequence edge.
As can be seen from the left half of fig. 1, the edges are broken down into temporal edges and non-temporal edges as a whole. The static Bayes only need to search edges in the time sequence, so that the time complexity of search work and algorithm operation is greatly reduced. On the basis, an optimal network structure model is learned by adopting an adaptive genetic algorithm and a BIC scoring function.
The information gain represents the change of information before and after dividing the data set, and the characteristic of high information gain is better choice; the information gain ratio is represented by the information gain ratio of the feature to the training data set, and is the ratio of the information gain to the experience entropy of the training data set; gini exponents represent the uncertainty of a collection, features are typically selected and partitioned using Gini exponent minimization criteria, and Gini exponents are calculated without logarithmic operations, all more efficient, and favoring the attribute of continuity. For node t, the Gini index calculation formula is as follows:
Gini(t)=1-∑ k [p(c k |t)] 2
wherein c k For a class, the greater the Gini index, the greater the degree of uncertainty of the sample set, and the nature of the class learning is a reduction in the degree of uncertainty of the sample, so the smallest Gini index is selected for feature splitting. The sample set corresponding to the father node is D, the selected feature A is split into two child nodes, and the corresponding set is D L And D R The Gini index after splitting is defined as follows:
the decision tree is typically generated using the maximum information gain, the maximum information gain ratio, and the minimum base index as criteria for feature selection. Starting from the root node, a decision tree is recursively generated, which amounts to continuously selecting locally optimal features, or dividing the training set into subsets that can be substantially correctly classified.
An air target intention recognition model construction method according to an embodiment of the present invention will be specifically described below.
As shown in fig. 2, the method includes:
1. and building an aircraft type identification model based on the feature selection and the decision tree.
In the training process, data of all airplane types are mixed to be used as a training set of a decision tree, data of various different airplanes under different graphs are used as the training set to carry out parameter learning, and in the testing process, data of all airplane types are mixed to be used as a testing set. For all detected data, firstly, an aircraft type recognition engine based on a decision tree is used for recognizing the aircraft type, a corresponding aircraft type label is added, reasonable feature selection is carried out for Bayesian networks of different types, and finally, a proper reasoning engine is selected for input.
A decision tree is a tree structure classification method, also called decision tree or classification tree, consisting of root nodes, internal nodes, leaf nodes and directed edges. The root node represents a first characteristic attribute, each internal node represents the judgment of the attribute, each branch represents the result of the judgment of the attribute, each leaf node represents the classification result, and the decision tree schematic diagram is shown in fig. 3.
The decision tree is usually built in three steps: feature selection, decision tree generation, decision tree pruning:
step 1: and (5) selecting characteristics. The aim is to select features that can classify the training set. The key to feature selection is that the criteria are information gain, information gain ratio, gini index.
Step 2: and (6) generating a decision tree. The generation of the decision tree for aircraft type recognition requires firstly constructing a root node, placing the original situation training data of all aerial targets in the root node, selecting the optimal characteristics thereof, and dividing the training data set into subsets according to the characteristics, so that each subset has the optimal classification under the current condition. If the subsets can be correctly classified, leaf nodes are constructed and the subsets are partitioned into corresponding leaf nodes. If there are subsets that cannot be classified correctly, new best features are selected for the subsets, they continue to be partitioned, and corresponding nodes are constructed. If recursion is performed until all subsets of training data are substantially correctly classified, or there are no suitable features, each subset is partitioned into leaf nodes, i.e. there are explicit classes, and thus decision trees are generated, the process of which is shown in fig. 4.
Step 3: pruning of decision trees. The pruning of the decision tree is to prevent the overfitting of the tree and enhance the generalization capability of the tree.
2. Data processing
Because the early warning machine, the fighter plane, the bomber plane and the electronic fighter plane execute tasks with different rules, the specific rules are as follows: the early warning machine can execute patrol and early warning detection/command tasks, the fighter can execute patrol and attack tasks, the bomber can execute patrol and attack tasks, and the electronic fighter can execute patrol, electronic reconnaissance and electronic interference tasks. In order to ensure high accuracy of task intention recognition, training sets are respectively divided by airplane types, and four Bayesian networks are trained to respectively conduct intention reasoning. The original data generated according to the intention regulation scene is divided into four airplane type data, the four sets of data are discretized and preprocessed, and then the four sets of data are used as input of each Bayesian network model and output as possible intention at each moment.
Step 1: data discretization. In order to enable data to meet the requirement of model input, discretization processing is required for the data, and specifically, discretized attributes are shown in table 1.
After discretizing the data, the data attributes are 9-dimensional (meaning) and are respectively altitude, speed, radar type, radar state, communication state, distance, maneuvering mode, acceleration and intention. Data is classified into 4 main categories by aircraft type: early warning machine, electronic fighter, bomber.
Step 2: and (5) data noise adding. In the discretized data of the early warning machine, the electronic fighter, the fighter and the bomber, the values of the height, the speed and the distance are uncertain because the discretized data value fluctuates when the airplane is at the critical point of high altitude/hollow, hollow/low altitude, high speed/medium speed, medium speed/low speed, far/medium speed and medium/near; the radar status data is uncertain because noise is added to the radar status before discretization (the radar type data has 20% probability of identifying errors as other values), for example, the discretization value of the radar status of an electronic fighter when patrol is 0,1,2,3, wherein 0,1,2 is the error value generated by adding noise.
Step 3: and (5) preprocessing data. The electronic fighter, fighter and bomber are high-altitude when executing patrol task, and the pre-warning machine is hollow when executing patrol task, because the pre-warning machine patrol and pre-warning detection/command are found to be executed at high altitude when re-carding rules, the high altitude is divided into three ranges again for subdividing the difference of the altitude, namely, high altitude in high altitude, hollow in high altitude and low altitude in high altitude. The early warning machine performs patrol task in the air, and performs early warning detection task in the air, as shown in fig. 5.
Step 4: and (5) selecting characteristics. In the process of parameter learning and reasoning, the Bayesian network can cause data redundancy for part of dimensional data, so that the recognition algorithm cannot accurately learn the data relationship between key data and intention. Thus, data dimensions less relevant to the intended node information are removed by using mutual information and maximum information coefficients, chi-square test, analysis of variance (F-value).
3. Construction of a static Bayesian network model
The construction of the static Bayesian network model mainly comprises three steps:
step 1: and the structure learning is responsible for extracting corresponding rules from the data and learning the structure of the network from the data.
Firstly, constructing an initial transfer network by using time mutual information, then, carrying out BIC scoring, coding the transfer network into a chromosome capable of being used for a genetic algorithm, carrying out optimal searching by using a self-adaptive genetic algorithm, defining a crossover operator and a mutation operator, and obtaining the optimal network by using the algorithm flow in FIG. 6.
Step 2: and (3) parameter learning, namely learning the probability of each node value by using maximum likelihood estimation.
Step 3: and (3) reasoning, namely using the learned network with parameters and a junction tree algorithm to perform reasoning, and finally outputting the intention by comparing the posterior probability of each intention value output.
Step 4: combining the data after the aircraft type identification based on the decision tree according to the steps, and generating 4 different Bayesian networks according to all tactical rules contained in different types of aircraft, wherein the structure learning and the parameter learning are contained. In the parallel reasoning process, the obtained group of data are respectively put into four different networks for learning.
Step 5: knowing the labels of each bayesian network, the maximum posterior probability that each network infers a tactical intent is calculated.
Step 6: and reasoning threat levels of tactical intentions according to each network to carry out comprehensive evaluation. Such as: only fighter can make attack intention, bomber can make attack intention, so that Bayesian network can infer that the attack intention is more than 30% of the actual intention of the fighter. Under the real situation, four airplanes can make patrol intention, so that all networks deduce the patrol intention and consider the real intention as patrol intention. Since the early warning machine can make intention for patrol and early warning detection/command, the intention is considered to be early warning detection at this time as long as the network containing the early warning machine judges that the intention is not patrol. The intent of electronic reconnaissance and electronic interference will only occur in the rule network of the electronic warplane, so the rule network containing the electronic warplane can be used to determine whether it is electronic reconnaissance or electronic interference.
The method uses a plurality of static Bayesian networks to build the intention recognition model, uses the decision tree model as an auxiliary decision to use which static Bayesian network to carry out reasoning, and can accurately predict different intentions of different types of airplanes.
It is to be understood that the above embodiments are merely illustrative of the application of the principles of the present invention, but not in limitation thereof. Various modifications and improvements may be made by those skilled in the art without departing from the spirit and substance of the invention, and are also considered to be within the scope of the invention.
Claims (7)
1. An aerial target intention recognition model construction method, which is characterized by comprising the following steps:
acquiring type data of all different types of aircrafts in the air and intention data of all different types of aircrafts under different intentions, and taking the type data and the intention data as a training set;
establishing an aircraft type recognition model based on feature selection and a decision tree, training the decision tree by using the type data, and performing parameter learning by using the intention data to obtain a trained aircraft type recognition model;
processing different types of data predicted and output by the aircraft category recognition model;
establishing a plurality of static Bayesian network models; each static Bayesian network model corresponds to one type of data, the processed data of different types are input into the corresponding static Bayesian network model to be trained, and finally a trained air target intention recognition model based on a decision tree and the static Bayesian network model is obtained.
2. The method according to claim 1, wherein the establishment of the decision tree comprises the steps of:
and (3) feature selection: selecting features capable of classifying training sets, wherein key criteria for feature selection comprise information gain, information gain ratio and Gini index;
generating a decision tree: firstly, constructing a root node, placing original situation training data of all aerial target airplanes in the root node, selecting optimal characteristics in the original situation training data, and dividing a training data set into subsets according to the characteristics so that each subset has optimal classification under the current condition; if the subsets can be correctly classified, constructing leaf nodes and dividing the subsets into corresponding leaf nodes; if subsets which cannot be classified correctly exist, selecting new optimal features for the subsets, continuing partitioning the subsets, and constructing corresponding nodes; if recursion is performed until all subsets of training data are substantially correctly classified, or there are no suitable features, each subset is partitioned into leaf nodes, i.e., there are explicit classes, generating a decision tree;
pruning a decision tree: the pruning of the decision tree is to prevent the overfitting of the tree and enhance the generalization capability of the tree.
3. The method of claim 1, wherein the processing the data comprises: data discretization, data noise adding, data preprocessing and feature selection.
4. A method according to any one of claims 1 to 3, wherein said building a plurality of static bayesian network models comprises:
and (3) structure learning: is responsible for extracting corresponding rules from the data and learning the structure of the network from the data;
parameter learning: learning the probability of each node value using maximum likelihood estimation;
reasoning: and (3) reasoning by using the learned network with parameters and the junction tree algorithm, and finally outputting the intention by comparing the posterior probability of each intention value.
5. The method according to claim 4, wherein the step of structure learning specifically comprises:
constructing an initial transfer network by using time mutual information, then performing BIC scoring, encoding the transfer network into a chromosome capable of being used for a genetic algorithm, performing optimal searching by using a self-adaptive genetic algorithm, defining a crossover operator and a mutation operator, and obtaining the optimal network by adopting a preset learning algorithm.
6. The method of claim 5, wherein the pre-configuring the maximum evolution times, population sizes, and variation probability parameters, the obtaining the optimal network by using a pre-set learning algorithm comprises:
s1, initializing a population, and coding;
s2, calculating fitness, namely BIC scores, of all individuals in the population, arranging the individuals in descending order according to the fitness, and reserving optimal individuals based on elite strategies;
s3, judging whether the difference value between the optimal fitness obtained in the iteration and the optimal fitness obtained in the previous iteration is smaller than a set threshold value; if yes, ending; otherwise, turning to step S4;
s4, selecting parent individuals from the better individuals, generating new individuals through cross mutation, and placing the new individuals into the population to replace the worse individuals to obtain a new population;
s5, judging whether the iteration times reach the maximum evolution times or not; if yes, ending; otherwise go to step S2.
7. The method of claim 6, wherein the method further comprises:
knowing the label of each Bayesian network, calculating the maximum posterior probability of each network for reasoning out a tactical intention;
and reasoning threat levels of tactical intentions according to each network to carry out comprehensive evaluation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211737280.5A CN116029379B (en) | 2022-12-31 | 2022-12-31 | Method for constructing air target intention recognition model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211737280.5A CN116029379B (en) | 2022-12-31 | 2022-12-31 | Method for constructing air target intention recognition model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116029379A true CN116029379A (en) | 2023-04-28 |
CN116029379B CN116029379B (en) | 2024-01-02 |
Family
ID=86070447
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211737280.5A Active CN116029379B (en) | 2022-12-31 | 2022-12-31 | Method for constructing air target intention recognition model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116029379B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117347961A (en) * | 2023-12-04 | 2024-01-05 | 中国电子科技集团公司第二十九研究所 | Radar function attribute identification method based on Bayesian learning |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050246307A1 (en) * | 2004-03-26 | 2005-11-03 | Datamat Systems Research, Inc. | Computerized modeling method and a computer program product employing a hybrid Bayesian decision tree for classification |
US20130346346A1 (en) * | 2012-06-21 | 2013-12-26 | Microsoft Corporation | Semi-supervised random decision forests for machine learning |
CN109784387A (en) * | 2018-12-29 | 2019-05-21 | 天津南大通用数据技术股份有限公司 | Multi-level progressive classification method and system based on neural network and Bayesian model |
CN111694939A (en) * | 2020-04-28 | 2020-09-22 | 平安科技(深圳)有限公司 | Method, device and equipment for intelligently calling robot and storage medium |
CN112232396A (en) * | 2020-10-08 | 2021-01-15 | 西北工业大学 | Fusion identification method for ship formation intention based on LSTM and D-S evidence theory |
AU2020103407A4 (en) * | 2020-11-12 | 2021-01-28 | Army Academy of Armored Forces | Intention recognition method based on normal cloud generator-bayesian network |
CN112598046A (en) * | 2020-12-17 | 2021-04-02 | 沈阳航空航天大学 | Target tactical intention identification method in multi-machine collaborative air combat |
CN114266355A (en) * | 2021-03-29 | 2022-04-01 | 中国人民解放军空军工程大学 | Tactical intention identification method based on BilSTM-Attention |
CN114330716A (en) * | 2021-12-24 | 2022-04-12 | 徐州工程学院 | University student employment prediction method based on CART decision tree |
US20220114343A1 (en) * | 2020-12-25 | 2022-04-14 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method of training model, device, and storage medium |
CN114818853A (en) * | 2022-03-10 | 2022-07-29 | 中国人民解放军空军工程大学 | Intention identification method based on bidirectional gating cycle unit and conditional random field |
CN114997306A (en) * | 2022-05-31 | 2022-09-02 | 电子科技大学长三角研究院(湖州) | Target intention identification method based on dynamic Bayesian network |
-
2022
- 2022-12-31 CN CN202211737280.5A patent/CN116029379B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050246307A1 (en) * | 2004-03-26 | 2005-11-03 | Datamat Systems Research, Inc. | Computerized modeling method and a computer program product employing a hybrid Bayesian decision tree for classification |
US20130346346A1 (en) * | 2012-06-21 | 2013-12-26 | Microsoft Corporation | Semi-supervised random decision forests for machine learning |
CN109784387A (en) * | 2018-12-29 | 2019-05-21 | 天津南大通用数据技术股份有限公司 | Multi-level progressive classification method and system based on neural network and Bayesian model |
CN111694939A (en) * | 2020-04-28 | 2020-09-22 | 平安科技(深圳)有限公司 | Method, device and equipment for intelligently calling robot and storage medium |
CN112232396A (en) * | 2020-10-08 | 2021-01-15 | 西北工业大学 | Fusion identification method for ship formation intention based on LSTM and D-S evidence theory |
AU2020103407A4 (en) * | 2020-11-12 | 2021-01-28 | Army Academy of Armored Forces | Intention recognition method based on normal cloud generator-bayesian network |
CN112598046A (en) * | 2020-12-17 | 2021-04-02 | 沈阳航空航天大学 | Target tactical intention identification method in multi-machine collaborative air combat |
US20220114343A1 (en) * | 2020-12-25 | 2022-04-14 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method of training model, device, and storage medium |
CN114266355A (en) * | 2021-03-29 | 2022-04-01 | 中国人民解放军空军工程大学 | Tactical intention identification method based on BilSTM-Attention |
CN114330716A (en) * | 2021-12-24 | 2022-04-12 | 徐州工程学院 | University student employment prediction method based on CART decision tree |
CN114818853A (en) * | 2022-03-10 | 2022-07-29 | 中国人民解放军空军工程大学 | Intention identification method based on bidirectional gating cycle unit and conditional random field |
CN114997306A (en) * | 2022-05-31 | 2022-09-02 | 电子科技大学长三角研究院(湖州) | Target intention identification method based on dynamic Bayesian network |
Non-Patent Citations (5)
Title |
---|
YONGSHENG GUAN; YINGPING WANG: "《Joint Detection and Tracking Scheme for Target Tracking in Moving Platform》", 《2020 IEEE RADAR CONFERENCE (RADARCONF20)》, pages 1 - 4 * |
ZHENG ZHANG, YANG QU, HONGKUN, LIU: "《Air Target Intention Recognition Based on Further Clustering and Sample Expansion》", 《 2018 37TH CHINESE CONTROL CONFERENCE (CCC)》, pages 3565 - 3569 * |
关永胜,左群声: "《基于特征谱的空间锥体目标识别方法》", 《中国电子科学研究院学报》, vol. 9, no. 02, pages 161 - 168 * |
李宁安;张剑;周倜;: "一种TensorFlow平台上目标意图识别模型设计与实现", 舰船电子工程, no. 05, pages 31 - 34 * |
胡智勇,刘华丽,龚淑君,彭超: "《基于随机森林的目标意图识别》", 《现代电子技术》, vol. 45, no. 19, pages 1 - 8 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117347961A (en) * | 2023-12-04 | 2024-01-05 | 中国电子科技集团公司第二十九研究所 | Radar function attribute identification method based on Bayesian learning |
CN117347961B (en) * | 2023-12-04 | 2024-02-13 | 中国电子科技集团公司第二十九研究所 | Radar function attribute identification method based on Bayesian learning |
Also Published As
Publication number | Publication date |
---|---|
CN116029379B (en) | 2024-01-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Choudhary et al. | Comprehensive review on supervised machine learning algorithms | |
US11627048B2 (en) | Systems and methods for network pattern matching | |
CN109657945B (en) | Industrial production process fault diagnosis method based on data driving | |
US6397200B1 (en) | Data reduction system for improving classifier performance | |
US20210374279A1 (en) | Differentially private dataset generation and modeling for knowledge graphs | |
CN114816909A (en) | Real-time log detection early warning method and system based on machine learning | |
CN112327903B (en) | Aircraft trajectory generation method based on deep mixed density network | |
CN116029379B (en) | Method for constructing air target intention recognition model | |
CN113157800A (en) | Identification method for discovering dynamic target in air in real time | |
CN113341919A (en) | Computing system fault prediction method based on time sequence data length optimization | |
CN114116829A (en) | Abnormal data analysis method, abnormal data analysis system, and storage medium | |
CN115659244A (en) | Fault prediction method, device and storage medium | |
US11914672B2 (en) | Method of neural architecture search using continuous action reinforcement learning | |
KR100895481B1 (en) | Method for Region Based on Image Retrieval Using Multi-Class Support Vector Machine | |
CN114492741A (en) | Deep Bayesian network modeling training method | |
CN112115996B (en) | Image data processing method, device, equipment and storage medium | |
Han et al. | Classifying uncertain and evolving data streams with distributed extreme learning machine | |
KR20230119464A (en) | Method, server and system for deep metric learning per hierarchical steps of multi-labels and few-shot inference using thereof | |
Sadhukhan | Can reverse nearest neighbors perceive unknowns? | |
Wendl et al. | Novelty detection in very high resolution urban scenes with density forests | |
Chakrabarti et al. | Modeling and Characterizing Aircraft Trajectories Near Airports Using Extracted Control Actions | |
Okafor | Entropy based techniques with applications in data mining | |
Brabec | Decision forests in the task of semi-supervised learning | |
CN113516189B (en) | Website malicious user prediction method based on two-stage random forest algorithm | |
US20220138632A1 (en) | Rule-based calibration of an artificial intelligence model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |