CN113902974B - Air combat threat target identification method based on convolutional neural network - Google Patents

Air combat threat target identification method based on convolutional neural network Download PDF

Info

Publication number
CN113902974B
CN113902974B CN202111134920.9A CN202111134920A CN113902974B CN 113902974 B CN113902974 B CN 113902974B CN 202111134920 A CN202111134920 A CN 202111134920A CN 113902974 B CN113902974 B CN 113902974B
Authority
CN
China
Prior art keywords
air combat
threat
threat target
layer
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.)
Active
Application number
CN202111134920.9A
Other languages
Chinese (zh)
Other versions
CN113902974A (en
Inventor
张�成
李帆
许悦雷
周清
张兆祥
回天
胡璐娟
崔祺
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN202111134920.9A priority Critical patent/CN113902974B/en
Publication of CN113902974A publication Critical patent/CN113902974A/en
Application granted granted Critical
Publication of CN113902974B publication Critical patent/CN113902974B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/048Activation functions
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Aiming, Guidance, Guns With A Light Source, Armor, Camouflage, And Targets (AREA)

Abstract

The invention discloses an air combat threat target identification method based on a convolutional neural network, which comprises the steps of firstly adding a attention mechanism into a backbone network of a ResNet model, enhancing the characteristic expression capability of a region of interest, reducing the influence of useless information on an identification result, and improving the region focusing capability of a detection model; and finally, changing the structure of the single full-connection layer into the structure of the double full-connection layer at the end of the ResNet model, and respectively carrying out feature mapping on the threat type and the guidance type of the air combat threat target so as to improve the recognition capability of the model and accelerate the convergence of the model. Through the two methods, an improved ResNet model is obtained, and the improved ResNet model is used for identifying the type of the air combat threat target, so that the task of identifying the air combat threat target is realized. The invention fully utilizes the channel attention and the space attention mechanism to guide the transmission of important characteristic information, and effectively improves the perception capability of the recognition model on global information and the recognition precision of the type of the air combat threat target.

Description

Air combat threat target identification method based on convolutional neural network
Technical Field
The invention belongs to the technical field of pattern recognition, and particularly relates to an air combat threat target recognition method.
Background
The fighter plane and the fighter plane in the modern air combat are increasingly hard against the fighter and fighter plane threat targets, and the problem of how to effectively improve the fighter plane terminal fighter efficacy and the fighter plane terminal defending viability against the threat targets (such as missiles) with continuously enhanced capability is a difficult problem to be solved urgently. The air combat threat target type identification is an important component of airborne terminal threat perception, fully utilizes various attribute characteristic information of the air combat threat target acquired by a plurality of information sources, and performs combined reasoning according to related criteria so as to acquire accurate and reliable type estimation. The type of accurately identifying the threat target of the air combat is the basis of the fighter to carry out the defense decision, and the speed and the accuracy of the identification have important influence on the generation of the defense strategy and the effect of the defense measures. It is difficult to find a complete and effective identification method or identification system in foreign publications. The related technology of the type identification of the threat targets of the air combat in China starts later, the type identification of the threat targets of the air combat in China is often regarded as a dynamic uncertainty reasoning process, the credibility based on entropy gain description evidence is built and integrated into a Bayesian network reasoning process, and a certain result is obtained. However, under the complex condition of external interference and multiple targets, the Bayesian network reasoning model is low in recognition precision of the type of the air combat threat target and difficult to meet the requirements due to poor learning ability and sensitivity to the form of input data facing the requirement of high confidence in the complex environment. Under complex conditions, the technology for identifying the type of the threat target of the air combat in a long distance and high precision is urgent to break through.
In recent years, deep learning research is continuously focused by students at home and abroad, along with continuous upgrading and updating of air combat threat targets (such as air-air/surface-air missiles), the intelligent and anti-interference capabilities of the air combat threat targets are continuously enhanced, so that terminal countermeasure environments are increasingly complex, and a great challenge is brought to terminal countermeasure viability of fighters. The problems of incomplete sensor observation information, low recognition confidence and the like in the airborne terminal countermeasure process are solved by utilizing an advanced intelligent technology, an intelligent fighter terminal defense perception decision-making integrated technology is developed, the terminal countermeasure defense efficiency is improved, and therefore, a target recognition method for multi-source sensor information needs to be developed by combining a neural network algorithm.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an air combat threat target identification method based on a convolutional neural network, which comprises the steps of firstly adding a attention mechanism into a backbone network of a ResNet model, enhancing the characteristic expressive power of a region of interest, reducing the influence of useless information on an identification result, and improving the region focusing power of a detection model; and finally, changing the structure of the single full-connection layer into the structure of the double full-connection layer at the end of the ResNet model, and respectively carrying out feature mapping on the threat type and the guidance type of the air combat threat target so as to improve the recognition capability of the model and accelerate the convergence of the model. Through the two methods, an improved ResNet model is obtained, and the improved ResNet model is used for identifying the type of the air combat threat target, so that the task of identifying the air combat threat target is realized. The invention fully utilizes the channel attention and the space attention mechanism to guide the transmission of important characteristic information, and effectively improves the perception capability of the recognition model on global information and the recognition precision of the type of the air combat threat target.
The technical scheme adopted by the invention for solving the technical problems comprises the following steps:
step 1: acquiring air combat threat target data to be identified, and constructing an air combat threat target data set;
Step 2: two labels are assigned to each air combat threat target data in the air combat threat target data set, threat type labels and guidance type labels of the air combat threat targets are respectively given, and different types of air combat threat targets are distinguished by adopting the labels;
Step 3: carrying out data link on the data of N different moments of the same air combat threat target in the air combat threat target data set to obtain an enhanced air combat threat target data set;
step 4: constructing an improved ResNet model;
step 4-1: attention modules are added after the 7 th layer, the 15 th layer, the 27 th layer and the 33 th layer of the ResNet model, so that the characteristic expression capacity of the ResNet model on the region of interest is enhanced;
Step 4-2: changing the last single full-connection layer of ResNet model into two parallel full-connection layers, wherein the two parallel full-connection layers respectively correspond to threat type identification and guidance type identification of the air combat threat target;
Taking an output characteristic diagram of the ResNet model as input of the two parallel full-connection layers, and respectively obtaining the identification output of the threat type and the guidance type of the final air combat threat target after mapping of the two full-connection layers;
Step 5: taking the enhanced air combat threat target data set obtained in the step 3 as input of an improved ResNet model, taking the label obtained in the step 2 as an air combat threat target data label, and training the improved ResNet model to obtain an improved ResNet model after training;
Step 6: and inputting the air combat threat target data into the improved ResNet model with the completed training to obtain an air combat threat target type identification result.
Further, the attention module comprises a channel attention module and a space attention module which are two parallel modules;
in the channel attention module, the feature map passes through the maximum pooling layer, passes through the hidden layer with two full-connection layers respectively, passes through the hidden layer with two full-connection layers again after being activated by Relu functions, and finally passes through the Sigmoid function to obtain a one-dimensional channel attention feature map;
in the spatial attention module, feature images respectively pass through two channels of a maximum pooling layer and an average pooling layer, then pass through a full-connection layer, vectors of the two channels are subjected to dimension combination, after a high-dimensional vector image is obtained, a connection convolution layer further extracts features, and finally a final spatial attention feature image is obtained through a Sigmoid function.
Further, the channel attention module and the spatial attention module are expressed as:
Mc(F)=σ(MLP(MaxPool(F))) (3)
Where F is a feature map, F 'is a channel attention mechanism, F' is a spatial attention mechanism, σ is a Sigmoid function, For dot product, M c (F) is a one-dimensional channel attention map, M s (F) is a two-dimensional spatial attention map, MLP is a hidden layer structured as a fully connected layer, F 7*7 is the size of the convolution layer and convolution kernel, avgPool is the average pooling layer, and MaxPool is the maximum pooling layer.
Further, the n=8.
The beneficial effects of the invention are as follows:
1. The feature extraction mechanism based on the attention module designed by the method fully utilizes the channel attention and the space attention mechanism to guide the transmission of important feature information, enhances the extraction of high-weight features in feature vectors, effectively improves the perception capability of an identification model on global information, and effectively improves the identification precision of the type of the air combat threat target.
2. The feature mapping mechanism based on the parallel full-connection layer designed by the method can accelerate the convergence of the model, and compared with the traditional serial full-connection structure which performs feature mapping from a single angle, the parallel full-connection layer structure can perform different feature mapping from multiple angles, so that more feature information is reserved, and the classification precision of the air combat threat target type is improved.
3. The improved ResNet neural network recognition model provided by the invention has good generalization and universality, and can be widely applied to real-time automatic recognition of the type of the threat target of the air combat and other actual scenes.
Drawings
FIG. 1 is a schematic diagram of an air combat threat target type identification model based on ResNet neural networks.
Fig. 2 is a schematic diagram of an identification model module of the present invention, fig. 2 (a) is a schematic diagram of a residual module structure, and fig. 2 (b) is a schematic diagram of a convolution module.
Fig. 3 is a schematic view of an attention module according to the present invention, fig. 3 (a) is a schematic view of a channel attention module, and fig. 3 (b) is a schematic view of a spatial attention module.
Detailed Description
The invention will be further described with reference to the drawings and examples.
Along with the continuous improvement of the autonomy level of the air combat threat targets, the autonomy capability of the discovery targets, the identification targets and the attack targets is improved, the dependence on information guarantee is greatly reduced, the intelligence and anti-interference capability is greatly enhanced, and the advanced air combat threat targets (such as air-to-air/surface-to-air missiles) become the greatest threat of an airborne platform. As the main basis for selecting the bait type by the defense decision algorithm, the accuracy of identifying the target type of the air combat threat directly influences the efficacy of the defense decision. Aiming at the specific requirements of fighter terminal defense decisions on the recognition of the air combat threat target types, the method is used for researching and establishing an air combat threat target type and a characteristic model and analyzing the comprehensive recognition flow of the air combat threat target types. By combining the characteristics of the neural network, an air combat threat target type recognition algorithm based on the neural network is researched, and under the condition of complex situation, compared with the traditional air combat threat target type recognition algorithm, a recognition result with higher accuracy and higher confidence is obtained.
In summary, in order to improve the accuracy of identifying the type of the threat object of the air combat, the problem to be solved is mainly: how to effectively improve the accuracy of target type identification. The measures taken are as follows: (1) The ResNet network is selected as a main convolution neural network, and a attention mechanism is added into a main network of the Resnet model, so that the characteristic expression capability of the region of interest is enhanced, the influence of useless information such as noise on a recognition result is reduced, the region focusing capability of the detection model is improved, and the recognition accuracy of the overall target is effectively improved; (2) And 2 full-connection layers are arranged at the tail part of the ResNet model, and belong to parallel relations, and the two full-connection layers respectively point to threat types and guidance types of the air combat threat targets, namely, a unified backbone network is adopted for feature extraction, and different full-connection layers are arranged according to different attributes of the target types when the features are mapped into the target types, so that the identification accuracy of the targets is improved.
As shown in fig. 1, the air combat threat target identification method based on the convolutional neural network comprises the following steps:
step 1: acquiring air combat threat target data to be identified, and constructing an air combat threat target data set;
Step 2: two labels are assigned to each air combat threat target data in the air combat threat target data set, threat type labels and guidance type labels of the air combat threat targets are respectively given, and different types of air combat threat targets are distinguished by adopting the labels;
Step 3: carrying out data link on 8 data of the same air combat threat target in the air combat threat target data set at different moments to obtain an enhanced air combat threat target data set;
step 4: constructing an improved ResNet model;
step 4-1: attention modules are added after the 7 th layer, the 15 th layer, the 27 th layer and the 33 th layer of the ResNet model, so that the characteristic expression capacity of the ResNet model on the region of interest is enhanced;
Step 4-2: changing the last single full-connection layer of ResNet model into two parallel full-connection layers, wherein the two parallel full-connection layers respectively correspond to threat type identification and guidance type identification of the air combat threat target;
Taking an output characteristic diagram of the ResNet model as input of the two parallel full-connection layers, and respectively obtaining the identification output of the threat type and the guidance type of the final air combat threat target after mapping of the two full-connection layers;
Step 5: taking the enhanced air combat threat target data set obtained in the step 3 as input of an improved ResNet model, taking the label obtained in the step 2 as an air combat threat target data label, and training the improved ResNet model to obtain an improved ResNet model after training;
Step 6: and inputting the air combat threat target data into the improved ResNet model with the completed training to obtain an air combat threat target type (comprising threat type and guidance type) identification result.
In fig. 3, a convolution-based attention module is designed. Typically, the attention mechanism in deep learning is used to remove redundant information, selecting information that is more important to the current goal. The attention mechanism can effectively learn the weight distribution of different parts on the input data or the feature map, reduce the influence caused by noise or useless information, and improve the recognition capability and the robustness of the model. As in fig. 2, the residual attention network constructs the network using residual mechanisms, guaranteeing the depth of the network while introducing the attention mechanism. The convolution attention module simultaneously utilizes the channel information and the space information of the feature map to design an attention module, so that the model can focus on more useful information, and the classification capability of the model on data is further enhanced. The invention designs an attention module by referring to a double-channel attention mechanism of a convolution attention module, wherein the attention module comprises a channel attention module, a feature map passes through a maximum pooling layer, then passes through a hidden layer with two full-connection layers, the hidden layer shares features, and finally obtains a one-dimensional channel attention map after Sigmiod functions; for the spatial attention module, the feature map passes through two paths of a maximum pooling layer and an average pooling layer respectively, and then passes through a full connection layer sharing features, and unlike the channel attention model, vectors of the two paths are combined in dimensions to obtain vectors with higher dimensions, and then a convolution layer is connected to perform further feature extraction, so that a final spatial attention map is obtained through a Sigmoid function.
The channel attention module and the spatial attention module can be expressed as:
Mc(F)=σ(MLP(MaxPool(F))) (3)
Where F is a feature map, F 'is a channel attention mechanism, F' is a spatial attention mechanism, σ is a Sigmoid function, For dot product, M c (F) is a one-dimensional channel attention map, M s (F) is a two-dimensional spatial attention map, MLP is a hidden layer structured as a fully connected layer, F 7*7 is the size of the convolution layer and convolution kernel, avgPool is the average pooling layer, and MaxPool is the maximum pooling layer.
The last parallel full-connection layer structure of the model can reduce the dimension of the information space of the air combat threat target recognition task, is beneficial to the model to better recognize different label types independently, and is beneficial to the model to accelerate the convergence rate. Specifically, compared with the conventional method of identifying the threat object types of the air combat, the method has the advantages that the threat object types of the air combat are classified into label types such as "air-air radar guidance", "air-air infrared guidance", "ground air/ship air radar guidance", and the like, although the label types are different, the label types are more similar to each other, for example: the "air-air radar guidance" and the "air-air infrared guidance" are of the air-air type in terms of threat type, while the "air-air radar guidance" and the "ground air/ship air radar guidance" are of the radar guidance type in terms of brake type. In view of this, there is a great similarity between the two sets of data, which is unfavorable for the neural network model to accurately identify it. The multi-parallel full-connection layer structure adopted by the invention can respectively and individually identify the threat type and the guidance type of the air combat threat target in a targeted manner, reduces the redundancy phenomenon of data in an information space caused by the arrangement of the data labels, reduces the dimension of the information space of the data, reduces the similarity between the data of different labels in the data set, and can accelerate the convergence speed of the model and effectively improve the identification precision of the neural network model.
Fig. 1 is a general structural diagram of the type identification of threat targets for air combat. Firstly, acquiring data of an air combat threat target to be identified and splicing different sensor data of the same air combat threat target at a plurality of moments; and then taking the obtained air combat threat target data as the input of a pre-trained air combat threat target type recognition neural network model based on ResNet, extracting the characteristics of the air combat threat target data based on the ResNet air combat threat target type recognition neural network model, respectively inputting the extracted characteristics into two full-connection layers at the tail, and outputting the guidance type and threat type recognition result of the air combat threat target after the mapping of the full-connection layers. The training sample which is pre-trained and based on ResNet air combat threat target type recognition neural network model is marked with two classification labels respectively, wherein one of the training sample is a guidance type label which comprises radar guidance and infrared guidance; and the other is threat type labels, which comprise three types of empty, ground empty/ship empty and non-elastic. The model for identifying the air combat threat target type based on ResNet comprises a convolution layer module, an attention mechanism module, a parallel full-connection layer module and a loss function calculation module, wherein the convolution layer module performs feature extraction on input data of the neural network, the attention mechanism module adaptively improves the proportion of forward calculation of a certain amount of feature values in feature vectors, the feature values belong to feature values which greatly contribute to the air combat threat target type classification result, the loss function calculation module determines parameters of the model for identifying the neural network based on ResNet air combat threat target type based on labels of training samples and a plurality of type identification results, so that the model for identifying the neural network based on ResNet air combat threat target type can detect data aiming at the air combat threat target to be identified, and outputs a corresponding air combat threat target type identification result. The attention mechanism feature extraction module is utilized to fully combine the channel attention and the space attention to guide the transmission of feature information, enhance the extraction of high-weight features in feature vectors, improve the expression capability of the feature information, effectively improve the grabbing capability of a detection model on key information, and effectively improve the identification accuracy of the type of the air combat threat target; the method for mapping and identifying the output feature map of the backbone neural network by utilizing the multi-parallel full-connection layer structure can reduce the dimension of the information space of data, can accurately identify the threat type and the guidance type of the air combat threat target in a targeted manner respectively, and compared with the traditional single full-connection layer structure, the multi-parallel full-connection layer structure accelerates the convergence speed of the model and effectively improves the identification precision of the type of the air combat threat target.
Specific examples:
In order to verify the effectiveness of the algorithm, 5000 pieces of air combat threat target data at different moments are generated through the simulation system, and the algorithm is verified.
In order to verify the effectiveness of the algorithm before and after improvement, the original ResNet neural network model and the improved ResNe neural network model in the invention are respectively used for verification on the air combat threat target data set, and the experimental results are shown in table 1. As can be seen from the table, under the same dataset, the improved ResNet neural network model has a certain improvement on the accuracy compared with the original ResNet neural network model, because the improved ResNet neural network model has better feature extraction capability, the recognition accuracy of the deep learning detection network on the air combat threat target is effectively improved. Therefore, the improved ResNet neural network model provided by the invention effectively improves the recognition rate of the type of the threat target of the air combat, and the invention has great effectiveness.
Table 1ResNet and comparative experiments on air combat threat target datasets with ResNet after modification

Claims (4)

1. The air combat threat target identification method based on the convolutional neural network is characterized by comprising the following steps of:
step 1: acquiring air combat threat target data to be identified, and constructing an air combat threat target data set;
Step 2: two labels are assigned to each air combat threat target data in the air combat threat target data set, threat type labels and guidance type labels of the air combat threat targets are respectively given, and different types of air combat threat targets are distinguished by adopting the labels;
Step 3: carrying out data link on the data of N different moments of the same air combat threat target in the air combat threat target data set to obtain an enhanced air combat threat target data set;
step 4: constructing an improved ResNet model;
step 4-1: attention modules are added after the 7 th layer, the 15 th layer, the 27 th layer and the 33 th layer of the ResNet model, so that the characteristic expression capacity of the ResNet model on the region of interest is enhanced;
Step 4-2: changing the last single full-connection layer of ResNet model into two parallel full-connection layers, wherein the two parallel full-connection layers respectively correspond to threat type identification and guidance type identification of the air combat threat target;
Taking an output characteristic diagram of the ResNet model as input of the two parallel full-connection layers, and respectively obtaining the identification output of the threat type and the guidance type of the final air combat threat target after mapping of the two full-connection layers;
Step 5: taking the enhanced air combat threat target data set obtained in the step 3 as input of an improved ResNet model, taking the label obtained in the step 2 as an air combat threat target data label, and training the improved ResNet model to obtain an improved ResNet model after training;
Step 6: and inputting the air combat threat target data into the improved ResNet model with the completed training to obtain an air combat threat target type identification result.
2. The method for identifying the air combat threat target based on the convolutional neural network according to claim 1, wherein the attention module comprises two parallel modules, namely a channel attention module and a space attention module;
in the channel attention module, the feature map passes through the maximum pooling layer, passes through the hidden layer with two full-connection layers respectively, passes through the hidden layer with two full-connection layers again after being activated by Relu functions, and finally passes through the Sigmoid function to obtain a one-dimensional channel attention feature map;
in the spatial attention module, feature images respectively pass through two channels of a maximum pooling layer and an average pooling layer, then pass through a full-connection layer, vectors of the two channels are subjected to dimension combination, after a high-dimensional vector image is obtained, a connection convolution layer further extracts features, and finally a final spatial attention feature image is obtained through a Sigmoid function.
3. An air combat threat target identification method based on a convolutional neural network in accordance with claim 2, wherein the channel attention module and the spatial attention module are represented as:
Mc(F)=σ(MLP(MaxPool(F))) (3)
Where F is a feature map, F 'is a channel attention mechanism, F' is a spatial attention mechanism, σ is a Sigmoid function, For dot product, M c (F) is a one-dimensional channel attention map, M s (F) is a two-dimensional spatial attention map, MLP is a hidden layer structured as a fully connected layer, F 7*7 is the size of the convolution layer and convolution kernel, avgPool is the average pooling layer, and MaxPool is the maximum pooling layer.
4. An air combat threat target identification method based on a convolutional neural network according to claim 1, wherein N = 8.
CN202111134920.9A 2021-09-27 2021-09-27 Air combat threat target identification method based on convolutional neural network Active CN113902974B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111134920.9A CN113902974B (en) 2021-09-27 2021-09-27 Air combat threat target identification method based on convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111134920.9A CN113902974B (en) 2021-09-27 2021-09-27 Air combat threat target identification method based on convolutional neural network

Publications (2)

Publication Number Publication Date
CN113902974A CN113902974A (en) 2022-01-07
CN113902974B true CN113902974B (en) 2024-05-17

Family

ID=79029633

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111134920.9A Active CN113902974B (en) 2021-09-27 2021-09-27 Air combat threat target identification method based on convolutional neural network

Country Status (1)

Country Link
CN (1) CN113902974B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117073848A (en) * 2023-10-13 2023-11-17 中国移动紫金(江苏)创新研究院有限公司 Temperature measurement method, device, equipment and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110472296A (en) * 2019-07-22 2019-11-19 中北大学 A kind of air objective intimidation estimating method based on the full connection residual error network of standardization
WO2021073528A1 (en) * 2019-10-18 2021-04-22 华中光电技术研究所(中国船舶重工集团有限公司第七一七研究所) Intelligent decision-making method and system for unmanned surface vehicle

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7904396B2 (en) * 2006-03-28 2011-03-08 The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration Systems, methods and apparatus for quiesence of autonomic safety devices with self action
US20210273953A1 (en) * 2018-02-20 2021-09-02 Darktrace Holdings Limited ENDPOINT AGENT CLIENT SENSORS (cSENSORS) AND ASSOCIATED INFRASTRUCTURES FOR EXTENDING NETWORK VISIBILITY IN AN ARTIFICIAL INTELLIGENCE (AI) THREAT DEFENSE ENVIRONMENT

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110472296A (en) * 2019-07-22 2019-11-19 中北大学 A kind of air objective intimidation estimating method based on the full connection residual error network of standardization
WO2021073528A1 (en) * 2019-10-18 2021-04-22 华中光电技术研究所(中国船舶重工集团有限公司第七一七研究所) Intelligent decision-making method and system for unmanned surface vehicle

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
刘峰 ; 沈同圣 ; 马新星 ; .特征融合的卷积神经网络多波段舰船目标识别.光学学报.(10),全文. *
基于Elman神经网络的空战威胁排序研究;俞卫树;姜长生;;电光与控制;20080815(08);全文 *
密集挤压激励网络的多标签胸部X光片疾病分类;张智睿;李锵;关欣;;中国图象图形学报;20201016(10);全文 *
标准化全连接残差网络空战目标威胁评估;翟翔宇;杨风暴;吉琳娜;李书强;吕红亮;;火力与指挥控制;20200615(06);全文 *

Also Published As

Publication number Publication date
CN113902974A (en) 2022-01-07

Similar Documents

Publication Publication Date Title
US20220197281A1 (en) Intelligent decision-making method and system for unmanned surface vehicle
Jung et al. ResNet-based vehicle classification and localization in traffic surveillance systems
CN110941794B (en) Challenge attack defense method based on general inverse disturbance defense matrix
CN111783020B (en) Battlefield entity target grouping method and system with multidimensional features
CN113806746B (en) Malicious code detection method based on improved CNN (CNN) network
CN109655815A (en) Sonar target detection method based on SSD
Peng et al. Battlefield image situational awareness application based on deep learning
CN114255403A (en) Optical remote sensing image data processing method and system based on deep learning
Wang et al. Recognition in label and discrimination in feature: A hierarchically designed lightweight method for limited data in sar atr
CN113902974B (en) Air combat threat target identification method based on convolutional neural network
CN113361565A (en) Countermeasure sample generation method and system for laser radar
Dai et al. GCD-YOLOv5: An armored target recognition algorithm in complex environments based on array lidar
CN107273914B (en) Efficient fusion identification method based on information source self-adaptive dynamic selection
CN113988164A (en) Representative point self-attention mechanism-oriented lightweight point cloud target detection method
CN112668662A (en) Outdoor mountain forest environment target detection method based on improved YOLOv3 network
CN111950635A (en) Robust feature learning method based on hierarchical feature alignment
CN109597044B (en) Broadband polarization radar seeker target identification method based on hierarchical decision tree
Zhao et al. Deep learning-based laser and infrared composite imaging for armor target identification and segmentation in complex battlefield environments
CN115049842B (en) Method for detecting damage of aircraft skin image and positioning 2D-3D
Xiao et al. A Shadow Capture Deep Neural Network for Underwater Forward‐Looking Sonar Image Detection
Zhang et al. Detection of bird’s nest on transmission lines from aerial images based on deep learning model
Zhang et al. Research on intelligent target recognition integrated with knowledge
Sun et al. A Safety-Helmet Detection Algorithm Based on Attention Mechanism
CN114494819A (en) Anti-interference infrared target identification method based on dynamic Bayesian network
CN114417713A (en) Grouping method for multiple targets

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