CN110472296B - Air combat target threat assessment method based on standardized full-connection residual error network - Google Patents
Air combat target threat assessment method based on standardized full-connection residual error network Download PDFInfo
- Publication number
- CN110472296B CN110472296B CN201910662593.0A CN201910662593A CN110472296B CN 110472296 B CN110472296 B CN 110472296B CN 201910662593 A CN201910662593 A CN 201910662593A CN 110472296 B CN110472296 B CN 110472296B
- Authority
- CN
- China
- Prior art keywords
- data
- network
- residual error
- layer
- error network
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an air combat target threat assessment method based on a standardized full-connection residual error network, and belongs to the field of battlefield situation assessment. Firstly, marking data by a simulation experiment, constructing a training set and a testing set, storing the training set and the testing set in a CSV file, secondly, constructing a standardized full-connection residual error network under a TensorFlow database, including constructing a graph for reading CSV file data, a residual error network layer and a standardized full-connection residual error network graph, and finally, establishing a TensorFlow session, training and testing a network model, analyzing network performance and verifying the model. The method solves the problem that the evaluation result is inaccurate due to the lack of self-learning reasoning capability on large sample data in other air combat target threat evaluation methods, can self-learn the distribution of input data, excavate the rules hidden in the data, and can accurately evaluate the air combat target threat by a trained model. The invention is mainly used for (but not limited to) battlefield situation assessment.
Description
Technical Field
The invention belongs to the field of air battlefield situation analysis, and particularly relates to an air combat target threat assessment method based on a standardized fully-connected residual error network.
Background
The air combat target threat assessment is used as an important air combat auxiliary means, is an important basis for pilots to realize leading air combat situation and achieve rapid defeat in air combat, and mainly utilizes enemy situation information obtained by the local and combines expert experience and mathematical theory to assess the killing capacity of enemy and the threat degree of enemy to our aircraft. The accurate assessment of the threat degree of the air combat target can provide reliable decision basis for pilots, realize the rapid attack on the large target of the threat degree of the self, and improve the combat efficiency and the survival probability of the aircraft.
At present, fully mining the situation information and the rule of the air combat target is a necessary premise for carrying out air combat target threat assessment. Common threat assessment methods are: the method for combining the Bayesian network with the Gaussian cloud model comprises the steps of describing the ambiguity of the attributes of the enemy plane by using the Gaussian cloud model, and repeatedly reasoning by using the Bayesian network, so that the situation threat assessment probability of the enemy plane can be accurately given; the method for evaluating the air combat target threat based on the interval support vector regression analyzes the characteristics of air combat situation data, and simultaneously utilizes expert experience to improve the accuracy of an evaluation result.
The method has certain effect in the evaluation of the air combat target threat, but the following problems still exist: the method only analyzes the characteristics of the surface of the small sample data of the air combat target situation, is suitable for simple air combat situation scenes, but cannot face complex and changeable air combat situation environments, and the fundamental reason is that the self-learning reasoning capability of the large sample data is lacked, the recessive law of the air combat target threat assessment accumulated in the large sample data cannot be found, and the assessment result is inaccurate. Based on the problems, the invention provides a standardized full-connection residual error network air combat target threat assessment method based on deep learning, which has strong self-learning capability, can deeply dig out rules hidden in data, improves the accuracy of air combat target threat assessment, and lays a foundation for realizing leading battlefield situation and achieving rapid defeating for our party.
Disclosure of Invention
The invention provides an air battle field target threat assessment method based on a deep learning standardized fully-connected residual error network, aiming at the problem that the assessment result is inaccurate due to the fact that the air battle field target threat assessment method lacks self-learning reasoning capability on large sample data. The method adopts a fully-connected network to map input data to a high-dimensional space, finally maps the input data to a sample mark space, realizes data classification, utilizes a batch normalization algorithm to make the network focus on learning nonlinearity, accelerates the network convergence speed, and adds a residual error network to continuously improve the training precision and the testing precision of the network when the original network is close to saturation.
The invention is realized by adopting the following technical scheme: an air combat target threat assessment method based on a standardized full-connection residual error network comprises the following steps:
s1: marking data by a simulation experiment, evaluating the attributes of a target including missile attack distance, course angle, distance, speed, altitude, type and interference capability, wherein the missile attack distance, the course angle and the distance all use real data, quantizing the speed, the altitude, the type and the interference capability to obtain quantized data of the speed, the altitude, the type and the interference capability, arranging and combining the 7 attributes of the data to form a plurality of groups of simulation data, outputting the simulation data as threat level of the simulation data as sample marks of the group of simulation data by simulation, and adding corresponding sample marks to each group of simulation data to form sample data,
s2: dividing the sample data in the S1 into a training set and a test set, and storing the training set and the test set in a CSV file;
s3: constructing a standardized full-connection residual error network on the basis of a TensorFlow database, wherein the drawing of CSV file data, the drawing of a residual error network layer and the structure drawing of the standardized full-connection residual error network are read;
s4: training the standardized fully-connected residual error network built in the S3, constructing all graphs in the session execution S3, calculating and constructing arrays to store the accuracy and loss of each batch of data, exiting from the loop and drawing a change graph when the training number reaches the self-defined training data in the program, analyzing the network training effect, and judging whether to modify the hyper-parameter repeated training or finely adjust the network model;
s5: and testing the trained network model in the S4, analyzing and verifying the network model, if the accuracy of the standardized full-connection residual error network meets the use requirement, carrying out threat assessment on the air combat target by using the network, and if the accuracy of the standardized full-connection residual error network does not meet the use requirement, continuing training the network until the use requirement is met.
According to the air combat target threat assessment method based on the standardized full-connection residual error network, after the simulation of S1 is finished, expert experience can be considered, and threat levels of partial simulation data can be modified.
The air combat target threat assessment method based on the standardized fully-connected residual error network quantifies the speed according to the characteristics of rapidness, generality, slowness and slowness, the height attribute concerns the height difference between a target and an airplane of the local party, the height difference is quantified according to the characteristics of superelevation, height, middle height, lower height, low height and ultralow height, the type of the target is quantified according to reconnaissance aircraft, unmanned aerial vehicles, small targets and large targets, and the interference capability of the target is quantified according to the characteristics of no interference, weak interference, middle interference and strong interference.
The method for evaluating the air combat target threat based on the standardized fully-connected residual error network specifically comprises the following steps of: constructing a file queue to read and store sample data in a CSV file in S2, calling a text reader to separate values after dequeuing the file queue, decoding according to a given corresponding data format and writing into a corresponding sample queue, combining 7 attribute values after decoding into a two-dimensional tensor, converting a sample mark into a one _ hot mark, finally batching the read sample data, wherein every n groups of data are one batch, and storing the 7 attribute values and the sample mark of the same batch of data in two different variables respectively, thereby facilitating the next calculation in a TensorFlow session.
The air combat target threat assessment method based on the standardized fully-connected residual error network comprises the following steps of: self-defining a residual error network layer function, wherein the function comprises three layers, the two adjacent layers are connected end to end, the first two layers of networks are all full-connected layers, all the first two layers of networks comprise 1024 nodes and are subjected to batch standardization, the input of the function is added with the output of the second layer in the third layer, the result is used as the input of the excitation function of the third layer, the excitation functions of the three layers of networks in the function all use ReLu functions, the output of the third layer is the output of the residual error network layer, and the last output value is used as the return value of the excitation function of the residual error network layer.
The method for evaluating the air combat target threat based on the standardized full-connection residual error network comprises the following steps of: the network comprises 11 layers, including 1 input layer, 7 full connection layers, 2 residual error network layers and 1 output layer, the output of the previous layer is used as the input of the next layer, the input layer contains 7 nodes, only the input data is not calculated, the network structures from the 2 nd layer to the 7 th layer are all full connection layers, the number of the nodes is 32, 64, 128, 256, 512 and 1024 respectively, the 8 th layer and the 9 th layer are residual error network layers, the residual error network layer function in the graph of the residual error network layer needs to be called, the 10 th layer network structure is a full connection layer, the number of the nodes is 3, the 11 th layer is an output layer of the network, the output result of the previous layer is normalized by using a Softmax function, the output of the network is the normalized result, the 7 full connection layers in the network are standardized in batch, all excitation functions use ReLu functions, the network output is the probability corresponding to the 7 corresponding air target attributes belonging to the input, the probability that a certain group of 7 attribute data output by the network belongs to each corresponding threat level and the cross entropy of the attribute data sample mark are used as training loss functions, the output probability and the sample mark are compared, the change of the learning rate is controlled by using an optimization function, and finally the loss is reduced by definition.
Compared with the prior art, the invention has the following advantages:
1) the invention provides a standardized full-connection residual error network air combat target threat assessment method based on deep learning, which solves the problem that the assessment result is inaccurate due to the lack of self-learning reasoning capability on large sample data in the traditional air combat target threat assessment method.
2) The invention realizes the organic combination of the batch normalization algorithm and the self-defined residual error network, accelerates the network convergence speed, simplifies the parameter adjusting process, leads the network to be trained by using larger learning rate and improves the network performance.
Drawings
FIG. 1 is a flow chart of the model algorithm of the present invention.
FIG. 2 shows the change of accuracy when a model is trained on 20 ten thousand data.
FIG. 3 shows the loss variation when the model is trained with 20 ten thousand data.
FIG. 4 is a graph of the change in accuracy of a model after training with 20 million data when tested on a test set containing 3 million data.
FIG. 5 shows the change in loss for a model trained with 20 million data when tested on a test set containing 3 million data.
FIG. 6 shows the variation of accuracy when training a model with 3 ten thousand samples.
Fig. 7 shows the change of loss when the model is trained with 3 ten thousand samples.
FIG. 8 shows the variation of accuracy when training a model with 5 ten thousand samples.
Fig. 9 shows the change in loss when 5 ten thousand samples were used to train the model.
Fig. 10 is a comparison graph of the test accuracy change of the trained model on the test set when the training set samples are 3 ten thousand data, 5 ten thousand data and 20 ten thousand data.
Fig. 11 is a graph showing a comparison of loss and change of the trained model when the training set samples are 3 ten thousand data, 5 ten thousand data, and 20 ten thousand data.
Detailed Description
Referring to the flow chart of fig. 1, experiments were conducted with raw data samples and network structure and optimization as study subjects.
S1: simulation experiment marking data: the air battlefield data were simulated using MATLAB R2014b, and the raw data samples focused on the following 7 factors of the air battle target from the perspective of the fighter pilot: missile attack distance, course angle, distance, speed, altitude, type, and interference capability. The missile attack distance, the course angle and the distance are all real data, and the speed, the altitude, the type and the interference capacity are quantized. The speeds are quantified at once as 9, 8, 7, 6, 5, 4, 3, very fast, normal, slow, very slow. The height index focuses on the height difference between the target and the airplane at our place, and the height difference is 8, 7, 6, 5, 4, 3 and 2 in the sequence of super-high, medium, low and super-low. The target types are sequentially quantized into 2, 4, 6 and 8 according to a reconnaissance plane, an unmanned plane, a small target and a large target. The interference capability of the target is quantized into 2, 4, 6 and 8 according to the sequence of no interference capability, weak interference capability, medium interference capability and strong interference capability. The data of 7 attributes are arranged and combined to form a plurality of groups of simulation data, the threat level of the simulation data is output as the simulation output and is used as the sample mark of the group of simulation data, and the corresponding sample mark is added to each group of simulation data to form sample data. After the simulation is finished, expert experience can be considered, and the threat level of a part of sample labels can be modified.
S2: constructing a training set and a testing set: and storing the sample data generated by the S1 in CSV files, disordering and randomly dividing the sample data into a training set and a testing set, wherein the training set sample comprises 20 thousands of data, and the testing set sample comprises 3 thousands of data, which are respectively stored in different CSV files.
S3: constructing a batch standardized full-connection residual error network: namely, constructing a TensorFlow static calculation diagram of the network, wherein the TensorFlow static calculation diagram comprises a diagram for reading CSV file data, a diagram of a residual error network layer and a structure diagram of a standardized full-connection residual error network:
s31: constructing a graph for reading CSV file data: the method comprises the steps of constructing a file queue file _ queue by using a TensorFlow library function string _ input _ producer (), reading and storing data in a CSV file in S2, calling a text reader TextLineReader () function to distinguish values of a text queue after dequeuing, defining a format of the text data after decoding, wherein 7 attribute values are 32-bit floating point types, samples are marked as 32-bit integer types, calling a function decode _ CSV () function to decode and write the function decode _ CSV () function into a corresponding sample queue, using a stack () function to combine 7 decoded attribute values into a two-dimensional tensor, and converting a sample mark into a one _ hot mark. Sample data is batched by using a slice _ input _ producer () function and a shuffle _ batch () function, the size of each batch of data is 1000 groups, and finally 7 attribute values and sample marks of the same batch of data are respectively stored in two variables, namely an example and a layer, so that the calculation in a Tensflow session in the next step is facilitated.
S32: constructing a graph of residual network layers: a residual error network layer function is defined by user, an input parameter is x, the function comprises 3 layers, the two adjacent layers are connected end to end, the 3-layer network excitation functions in the function all use ReLu functions, the first two layers of networks are all full-connection layers, all the networks comprise 1024 nodes and are standardized in batches. The layer 2 output is defined as bn2, the layer 3 input x of the function is added to the layer 2 output bn2, the result is taken as the layer 3 excitation function input, the excitation function output is taken as the layer 3 output and also as the residual network layer output, and the return function is used to return the value of the last output of the residual network layer function.
S33: constructing a structure diagram of a standardized full-connection residual error network: the definition input is a placeholder for 1000 rows and 7 columns, and the sample is labeled as a placeholder for 1000 rows and 3 columns. The network comprises 11 layers including 1 input layer, 7 fully connected layers, 2 residual network layers and 1 output layer, and the output of the previous layer is used as the input of the next layer. The input layer comprises 7 nodes, and only data is input without calculation. The network structures from the layer 2 to the layer 7 are all full connection layers, and the number of nodes is 32, 64, 128, 256, 512 and 1024 respectively. Both layers 8 and 9 are residual network layers, and the residual network layer function in S32 needs to be called. The layer 10 network structure is a full connection layer, and the number of nodes is 3. And the 11 th layer is an output layer of the network, the output result of the last layer is normalized by using a Softmax function, and the output of the network is the normalized result. And 7 full connection layers in the network are subjected to batch standardization, all excitation functions use ReLu functions, and the network output is the probability that 7 air target attributes which are correspondingly input belong to each corresponding threat level. In the process, each node weight parameter w of the full connection layer is initialized randomly, each node bias b is initialized to be zero, and each full connection layer is subjected to batch standardization by using a Tensorflow function. The probability that a certain group of 7 attribute data output by the network belongs to each corresponding threat level and the cross entropy of the attribute data sample mark are used as training loss functions, the output probability and the sample mark are compared, the change of the learning rate is controlled by using an optimization function, and finally, the operation of reducing the loss by using a minimize () function is defined as train _ op.
TABLE 1 standardized full-connection residual error network model structure
S4: training a network model: creating a TensorFlow session, performing operations in the network, initializing global and local variables in the session, and starting a queue of incoming data. The graph of read data is executed in a loop, X and Y store the results of the data batches in the data read graph as variables, run the operation train _ op, while preserving the values of the losses and the calculation results of the accuracy. And executing the operation of the network structure chart once every batch of data is read, outputting the accuracy and loss of each batch of data, constructing data storage, drawing a change chart when training is finished, exiting from the cycle finishing training when the training quantity reaches the self-defined training data in the program, storing the network model, outputting the position of the network model, and judging whether to modify the hyper-parameter repeated training or finely adjust the network model. The network hyper-parameters can be continuously adjusted in the training process, and the network can be optimized and finely adjusted according to the convergence condition or the convergence speed of the network, and the final hyper-parameters are set as follows: the learning rate is an exponential decay learning rate, the initial value is 0.1, the decay rate of each round of learning is 0.1, the number of steps of each round of learning is 5000, the size of each batch of data is 1000 groups, and the optimization function used by the network is an Adam optimization algorithm.
S5: test S4 saved network model: creating a python file, reserving a graph for reading CSV file data and a network structure graph, creating a session, loading the model saved in S4, starting a queue for inputting data, reading the data of the CSV file in the test set in S2, operating train _ op, simultaneously constructing the calculation results of the values of array storage loss and accuracy, and drawing a change graph after training. And observing the test accuracy and loss change, analyzing the network performance, and judging whether the network is over-fitted or not, and whether the hyper-parameter repeated training needs to be modified or the network model needs to be finely adjusted.
Claims (6)
1. An air combat target threat assessment method based on a standardized full-connection residual error network is characterized by comprising the following steps:
s1: marking data by a simulation experiment, evaluating the attributes of a target including missile attack distance, course angle, distance, speed, altitude, type and interference capability, wherein the missile attack distance, the course angle and the distance all use real data, quantizing the speed, the altitude, the type and the interference capability to obtain quantized data of the speed, the altitude, the type and the interference capability, arranging and combining the 7 attributes of the data to form a plurality of groups of simulation data, outputting the simulation data as threat level of the simulation data as sample marks of the group of simulation data by simulation, and adding corresponding sample marks to each group of simulation data to form sample data,
s2: dividing the sample data in the S1 into a training set and a test set, and storing the training set and the test set in a CSV file;
s3: constructing a standardized full-connection residual error network on the basis of a TensorFlow database, wherein the drawing of CSV file data, the drawing of a residual error network layer and the structure drawing of the standardized full-connection residual error network are read;
s4: training the standardized fully-connected residual error network built in the S3, constructing all graphs in the session execution S3, calculating and constructing arrays to store the accuracy and loss of each batch of data, exiting from the loop and drawing a change graph when the training number reaches the self-defined training data in the program, analyzing the network training effect, and judging whether to modify the hyper-parameter repeated training or finely adjust the network model;
s5: and testing the trained network model in the S4, analyzing and verifying the network model, if the accuracy of the standardized full-connection residual error network meets the use requirement, carrying out threat assessment on the air combat target by using the network, and if the accuracy of the standardized full-connection residual error network does not meet the use requirement, continuing training the network until the use requirement is met.
2. The method for assessing the threat of the air combat target based on the standardized full-connection residual error network as claimed in claim 1, wherein after the simulation of S1 is finished, expert experience can be considered, and the threat level of partial simulation data can be modified.
3. The method for assessing threat to air combat targets based on standardized fully-connected residual error networks as claimed in claim 1 or 2, wherein the speed is quantified in terms of very fast, general, slow and very slow, the attribute of altitude concerns the altitude difference between the target and the airplane of our party, and is quantified in terms of superelevation, altitude, higher, medium, lower, low and ultralow, the type of target is quantified in terms of reconnaissance plane, unmanned plane, small target and large target, and the interference capability of the target is quantified in terms of none, weak, medium and strong.
4. The air combat target threat assessment method based on the standardized fully-connected residual error network according to claim 2, wherein the step of constructing a graph for reading CSV file data specifically comprises the following steps: constructing a file queue to read and store sample data in a CSV file in S2, calling a text reader to separate values after dequeuing the file queue, decoding according to a given corresponding data format and writing into a corresponding sample queue, combining 7 attribute values after decoding into a two-dimensional tensor, converting a sample mark into a one _ hot mark, finally batching the read sample data, wherein every n groups of data are one batch, and storing the 7 attribute values and the sample mark of the same batch of data in two different variables respectively, thereby facilitating the next calculation in a TensorFlow session.
5. The air combat target threat assessment method based on the standardized fully-connected residual error network as claimed in claim 4, characterized in that the process of constructing the graph of the residual error network layer is performed according to the following steps: self-defining a residual error network layer function, wherein the function comprises three layers, the two adjacent layers are connected end to end, the first two layers of networks are all full-connected layers, all the first two layers of networks comprise 1024 nodes and are subjected to batch standardization, the input of the function is added with the output of the second layer in the third layer, the result is used as the input of the excitation function of the third layer, the excitation functions of the three layers of networks in the function all use ReLu functions, the output of the third layer is the output of the residual error network layer, and the last output value is used as the return value of the excitation function of the residual error network layer.
6. The method for assessing the threat of the air combat target based on the standardized full-connection residual error network as claimed in claim 5, wherein the step of constructing a structure diagram of the standardized full-connection residual error network comprises the following steps: the network comprises 11 layers, including 1 input layer, 7 full connection layers, 2 residual error network layers and 1 output layer, the output of the previous layer is used as the input of the next layer, the input layer contains 7 nodes, only the input data is not calculated, the network structures from the 2 nd layer to the 7 th layer are all full connection layers, the number of the nodes is 32, 64, 128, 256, 512 and 1024 respectively, the 8 th layer and the 9 th layer are residual error network layers, the residual error network layer function in the graph of the residual error network layer needs to be called, the 10 th layer network structure is a full connection layer, the number of the nodes is 3, the 11 th layer is an output layer of the network, the output result of the previous layer is normalized by using a Softmax function, the output of the network is the normalized result, the 7 full connection layers in the network are standardized in batch, all excitation functions use ReLu functions, the network output is the probability corresponding to the 7 corresponding air target attributes belonging to the input, the probability that a certain group of 7 attribute data output by the network belongs to each corresponding threat level and the cross entropy of the attribute data sample mark are used as training loss functions, the output probability and the sample mark are compared, the change of the learning rate is controlled by using an optimization function, and finally the loss is reduced by definition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910662593.0A CN110472296B (en) | 2019-07-22 | 2019-07-22 | Air combat target threat assessment method based on standardized full-connection residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910662593.0A CN110472296B (en) | 2019-07-22 | 2019-07-22 | Air combat target threat assessment method based on standardized full-connection residual error network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110472296A CN110472296A (en) | 2019-11-19 |
| CN110472296B true CN110472296B (en) | 2022-04-15 |
Family
ID=68508169
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910662593.0A Active CN110472296B (en) | 2019-07-22 | 2019-07-22 | Air combat target threat assessment method based on standardized full-connection residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110472296B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046497B (en) * | 2019-12-24 | 2023-04-07 | 中国航空工业集团公司沈阳飞机设计研究所 | Rapid assessment device for high-altitude high-speed airplane penetration viability |
| CN112149715B (en) * | 2020-08-31 | 2024-04-02 | 南京航空航天大学 | Unmanned aerial vehicle air combat threat assessment method based on deep learning |
| CN112598046B (en) * | 2020-12-17 | 2023-09-26 | 沈阳航空航天大学 | Target tactical intent recognition method in multi-machine cooperative air combat |
| CN112801403A (en) * | 2021-02-10 | 2021-05-14 | 武汉科技大学 | Method and system for predicting potential threat degree of aerial target based on SSA-BP |
| CN113902974B (en) * | 2021-09-27 | 2024-05-17 | 西北工业大学 | Air combat threat target identification method based on convolutional neural network |
| CN115222199B (en) * | 2022-05-31 | 2023-11-14 | 中国电子科技集团公司第五十二研究所 | Air combat key node determining method and device based on win-plant rate prediction |
| CN115952909A (en) * | 2022-12-31 | 2023-04-11 | 中国电子科技集团公司信息科学研究院 | Target threat estimation method and device based on combined empowerment and LSTM |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0892356A2 (en) * | 1997-07-18 | 1999-01-20 | Informix Software, Inc. | Data processing apparatus and method |
| CN107563446A (en) * | 2017-09-05 | 2018-01-09 | 华中科技大学 | A kind of micro OS object detection method |
-
2019
- 2019-07-22 CN CN201910662593.0A patent/CN110472296B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0892356A2 (en) * | 1997-07-18 | 1999-01-20 | Informix Software, Inc. | Data processing apparatus and method |
| CN107563446A (en) * | 2017-09-05 | 2018-01-09 | 华中科技大学 | A kind of micro OS object detection method |
Non-Patent Citations (2)
| Title |
|---|
| 基于RBF神经网络的空战威胁评估;范隽等;《微计算机信息》;20080725(第21期);第209-210页 * |
| 基于动态贝叶斯网络的空战目标威胁等级评估;晏师励等;《计算机与数字工程》;20151220(第12期);第2150-2154页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110472296A (en) | 2019-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110472296B (en) | Air combat target threat assessment method based on standardized full-connection residual error network | |
| Wang et al. | Combination of hyperband and Bayesian optimization for hyperparameter optimization in deep learning | |
| Gao et al. | Low-shot learning via covariance-preserving adversarial augmentation networks | |
| Lundberg et al. | An unexpected unity among methods for interpreting model predictions | |
| CN111429340A (en) | Cyclic image translation method based on self-attention mechanism | |
| Erlan | C4. 5 Algorithm Application for Prediction of Self Candidate New Students in Higher Education | |
| CN111340076B (en) | Zero sample identification method for unknown mode of radar target of new system | |
| Jing et al. | Community detection in sparse networks using the symmetrized laplacian inverse matrix (slim) | |
| CN111694974A (en) | Depth hash vehicle image retrieval method integrating attention mechanism | |
| Du et al. | Improving the performance of feature selection and data clustering with novel global search and elite-guided artificial bee colony algorithm | |
| CN113344174A (en) | Efficient neural network structure searching method based on probability distribution | |
| Li et al. | Enhancing the competitive swarm optimizer with covariance matrix adaptation for large scale optimization | |
| CN114912741A (en) | Effectiveness evaluation method and device for combat system structure and storage medium | |
| Morawski et al. | Detecting dense-matter phase transition signatures in neutron star mass-radius measurements as data anomalies using normalizing flows | |
| Laube et al. | What to expect of hardware metric predictors in NAS | |
| CN114219017A (en) | Implicit semantic data enhancement method | |
| CN110807291B (en) | On-site situation future guiding technology based on mimicry countermeasure learning mechanism | |
| CN112749516A (en) | System combination model reliability intelligent evaluation method suitable for multi-type data characteristics | |
| Zuin et al. | Deep learning techniques for explainable resource scales in collectible card games | |
| Ashlock et al. | Techniques for analysis of evolved prisoner's dilemma strategies with fingerprints | |
| CN110930054A (en) | Data-driven battle system key parameter rapid optimization method | |
| Vejdemo-Johansson et al. | Multiple testing with persistent homology | |
| Michalenko et al. | Multimodal Data Fusion via Entropy Minimization | |
| Li et al. | Defensive Few-shot Learning | |
| CN117312810B (en) | Incomplete information attack and defense game opponent identification method based on game history tree |
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 |