CN111352086B - Unknown target identification method based on deep convolutional neural network - Google Patents

Unknown target identification method based on deep convolutional neural network Download PDF

Info

Publication number
CN111352086B
CN111352086B CN202010152743.6A CN202010152743A CN111352086B CN 111352086 B CN111352086 B CN 111352086B CN 202010152743 A CN202010152743 A CN 202010152743A CN 111352086 B CN111352086 B CN 111352086B
Authority
CN
China
Prior art keywords
target
neural network
dimensional range
range profile
difference
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
CN202010152743.6A
Other languages
Chinese (zh)
Other versions
CN111352086A (en
Inventor
周代英
张同梦雪
李粮余
胡晓龙
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.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
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 University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202010152743.6A priority Critical patent/CN111352086B/en
Publication of CN111352086A publication Critical patent/CN111352086A/en
Application granted granted Critical
Publication of CN111352086B publication Critical patent/CN111352086B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/411—Identification of targets based on measurements of radar reflectivity
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks

Abstract

The invention belongs to the technical field of unknown target identification, and particularly relates to an unknown target identification method based on a deep convolutional neural network. The method comprises the steps of preprocessing one-dimensional range profile data (HRRP) obtained by a broadband radar, and reducing the amplitude sensitivity of the one-dimensional range profile; secondly, extracting features by utilizing a deep convolution neural network; and finally, processing the identification probability of the known target data by a difference probability method to obtain a discrimination threshold, and discriminating the output vector of the neural network so as to identify the unknown target. The method introduces the discrimination threshold obtained by adopting the difference probability method, effectively describes the statistical distribution area boundary of the data set of the known target and the unknown target, and solves the problem that the conventional convolutional neural network cannot identify the unknown target.

Description

Unknown target identification method based on deep convolutional neural network
Technical Field
The invention belongs to the technical field of unknown target identification, and particularly relates to an unknown target identification method based on a deep convolutional neural network.
Background
Since the middle of the last century, radar target identification technology has gradually developed and matured, and the target to be identified by the radar is judged mainly according to the radar target cross-sectional area (RCS) or the one-dimensional range profile (HRRP). The high-resolution one-dimensional range image is the vector sum of target scattering center echoes acquired by the broadband radar, and not only provides the geometric shape and structural characteristics of a target, but also contains more relevant information required by target identification.
In recent years, the deep learning theory is gradually mature, the convolutional neural network is widely applied to the field of radar target identification, and the convolutional neural network has the characteristics of translation insensitivity, nonlinearity and self-learning, so that a good identification effect is obtained. However, the conventional convolutional neural network needs to train a large amount of data of known targets in advance, which means that the conventional convolutional neural network can only recognize targets of the known targets (i.e. targets that have participated in training), however, in practical applications, it is impossible to acquire one-dimensional range profile data of all targets in advance, construct a complete convolutional neural network for recognition, and when the network inputs data of unknown targets (i.e. targets that have not participated in training), the data will be forcibly recognized as classes of the known targets, resulting in erroneous recognition.
Disclosure of Invention
The invention provides a radar unknown target identification method based on a deep convolutional neural network, aiming at the problems. The method is based on a conventional deep convolution neural network, the training one-dimensional range profile data of the known target is used for obtaining the identification threshold, and the statistical distribution area boundary of the known target and the unknown target data set is effectively described, so that the unknown target is identified, and the problem that the conventional neural network cannot identify the unknown target is solved.
The technical scheme of the invention is as follows: an unknown target identification method based on a deep convolutional neural network comprises the following steps:
s1, setting a single target one-dimensional range profile sample obtained by the broadband radar as x ═ x based on the target scattering center model 1 ,x 2 ,...,x i ,...,x N ]Where N is the number of distance units, x i Representing the amplitude of the ith distance unit, highlighting the contrast effect of the strong scattering point and other scattering points in order to reduce the influence of the amplitude sensitivity of the one-dimensional distance image on the identification performance, and carrying out beta-mean value standardization treatment on the one-dimensional distance image:
Figure GDA0003673933450000021
wherein
Figure GDA0003673933450000022
Represents the normalized amplitude of the ith distance element, beta is a constant, E x The mean value of the single range profile is shown, and the single one-dimensional range profile after beta-mean value standardization processing is
Figure GDA0003673933450000023
S2, constructing a deep convolutional neural network model, and learning under strong supervisionNext, a Deep Convolutional Neural Network (DCNN) improved based on AlexNet is used for extracting high-dimensional features of the radar one-dimensional range profile. The deep convolution neural network is formed by stacking a plurality of convolution modules, and a Dropout layer is added in the middle of the deep convolution neural network to enable part of neurons to be inactivated randomly so as to reduce training parameters and reduce the risk of model overfitting. Based on a neural network principle, a Back Propagation (BP) algorithm and a random gradient descent (SGD) algorithm are adopted for model convergence and training, as shown in FIG. 1, the deep convolution neural network has 13 layers in total, and a convolution module 1, a Dropout layer 1, a convolution module 2, a convolution module 3, a Dropout layer 2, a convolution module 4, a convolution module 5, a full connection layer 1, a batch normalization layer 1, a Dropout layer 3, a full connection layer 2, a batch normalization layer 2 and a classifier are sequentially arranged; the input of the deep convolutional neural network is
Figure GDA0003673933450000024
Outputting identification label given for classifier
Figure GDA0003673933450000025
Each convolution module is composed of a convolution layer, an activation function, a batch normalization layer and a pooling layer, the radar minimum resolution unit and one-dimensional range profile data characteristics are considered, the size of a convolution kernel is 1 x 3, each convolution layer is provided with 64 convolution kernels, the pooling kernel is 1 x 2, and because an Exponential Linear Unit (ELU) has better differential characteristics, the conventional activation function RELU is replaced by the Exponential Linear Unit (ELU), the model fitting capability is improved, and the robustness to input changes is improved. The BN layer is added, so that the training speed is increased, the model is converged quickly, and the activation function is as follows:
Figure GDA0003673933450000026
s3, determining an identification threshold: the conventional convolutional neural network target identification needs to acquire characteristic parameters of all classes in training learning, and an unknown class target is forced to be identified as a certain known class in classification. In order to identify unknown types, a difference probability method is introduced to obtain an identification threshold;
in deep convolutional neural networksIn the learning stage, the probability output p of the ith one-dimensional distance image obtained from the classifier belonging to the jth known class ij The probability vector corresponding to each one-dimensional range profile is p i ＝[p i1 ,p i2 ,...,p iM ]Where M is the number of known classes in the probability vector p i To choose the maximum value p m And the sub-maximum p sm Obtaining the difference probability v d ＝p m -p sm ；
Inputting single one-dimensional range profiles of different types of targets, wherein each type of target can obtain a difference vector:
Figure GDA0003673933450000027
wherein d is m Is the difference vector for the mth class of objects,
Figure GDA0003673933450000031
the difference probability of the ith one-dimensional range profile of the mth type target is shown, and the superscript T represents a transpose symbol;
the difference vector d of all known targets m Calculating a histogram of the difference probabilities in the image, and selecting one difference probability from the histogram of the difference probabilities as an identification threshold tau according to the predetermined correct discrimination rate of the known target;
s4, unknown target identification:
inputting the obtained single one-dimensional range profile of the unknown target into a trained deep convolutional neural network model to obtain a corresponding difference vector
Figure GDA0003673933450000032
Wherein
Figure GDA0003673933450000033
The difference probability corresponding to the c test one-dimensional range profile data is obtained, and P represents the number of unknown target data;
will d t Comparing the difference probability of the inner one-dimensional range profile with the identification threshold tau, and if the difference probability is more than or equal to the threshold
Figure GDA0003673933450000034
Identifying the c-th one-dimensional range profile data as a known target; if the difference probability is less than the threshold
Figure GDA0003673933450000035
Identifying the c-th one-dimensional range profile data as an unknown target, namely the identification rule is as follows:
Figure GDA0003673933450000036
wherein
Figure GDA0003673933450000037
Indicating that the ith unknown target one-dimensional range image data belongs to a known target,
Figure GDA0003673933450000038
and the ith unknown target one-dimensional range profile data belongs to the unknown target P.
The method has the advantages that the discrimination threshold obtained by adopting a difference probability method is introduced, so that the statistical distribution area boundary of the known target and the unknown target data set is effectively described, and the problem that the conventional convolutional neural network cannot identify the unknown target is solved.
Drawings
FIG. 1 is a schematic diagram of a deep convolutional neural network model structure.
Detailed Description
The effectiveness of the invention is demonstrated below in connection with the simulation example.
Experiments are carried out by utilizing simulated one-dimensional distance images of five different types of military aircrafts including AH64, AN26, F15, B1B and B52 obtained by a special electromagnetic simulation characteristic scene. The experimental simulation radar parameters comprise: the radar carrier frequency is 6GHz, and the radar bandwidth is 400 MHz. In the simulation scene, a simulation target collects a one-dimensional range profile at an interval of 0.1 degrees in the range of 0-180 degrees of azimuth angle at an elevation angle of 3 degrees, each type of airplane collects 1801 one-dimensional range profiles, each one-dimensional range profile contains 320 range units, namely, input data of each type of airplane is a 1801 × 320 one-dimensional range profile matrix.
In the process of training the update parameter W, the random initialization weight W ═ W 1 ,w 2 ,w 3 ]And bias B ═ B 1 ,b 2 ,...,b N ]And selecting a cross entropy loss function as a loss function and optimization parameters of an Adam optimizer with an adaptive learning rate, wherein the learning rate is initialized to 0.0001.
The identification results of the above 5 types of simulated radar target data using the conventional convolutional neural network method and the method herein are shown in table 1:
TABLE 1 recognition results of two methods on unknown targets
Figure GDA0003673933450000041
From experimental results, under the condition that three types of airplanes are randomly extracted as known targets and the other two types of airplanes are as unknown targets, the unknown targets cannot be identified by using a conventional CNN network, but the unknown targets can be well identified by the method due to the introduction of the unknown target distinguishing threshold, and the average correct identification rate of the unknown targets is over 80 percent, so that the method is verified to be effective.

Claims (1)

1. An unknown target identification method based on a deep convolutional neural network is characterized by comprising the following steps:
s1, setting a single target one-dimensional range profile sample obtained by the broadband radar as x ═ x based on the target scattering center model 1 ,x 2 ,...,x i ,...,x N ]Where N is the number of distance units, x i Representing the amplitude of the ith range bin, and carrying out beta-mean normalization processing on the one-dimensional range profile:
Figure FDA0003673933440000011
wherein
Figure FDA0003673933440000012
Represents the normalized amplitude of the ith distance element, beta is a constant, E x The mean value of the single range profile is shown, and the single one-dimensional range profile after beta-mean value standardization processing is
Figure FDA0003673933440000013
S2, constructing a deep convolution neural network model, wherein the deep convolution neural network has 13 layers in total, namely a convolution module 1, a Dropout layer 1, a convolution module 2, a convolution module 3, a Dropout layer 2, a convolution module 4, a convolution module 5, a full connection layer 1, a batch normalization layer 1, a Dropout layer 3, a full connection layer 2, a batch normalization layer 2 and a classifier; the input of the deep convolutional neural network is
Figure FDA0003673933440000014
Outputting identification label given for classifier
Figure FDA0003673933440000015
Each convolution module is composed of convolution layers, an activation function, a batch normalization layer and a pooling layer, wherein the size of a convolution kernel is 1 x 3, each convolution layer is provided with 64 convolution kernels, the pooling kernel is 1 x 2, and the activation function is as follows:
Figure FDA0003673933440000016
s3, determining an identification threshold: in the learning phase of the deep convolutional neural network, the probability output p of the ith one-dimensional distance image obtained from the classifier belonging to the jth known class ij The probability vector corresponding to each one-dimensional range profile is p i ＝[p i1 ,p i2 ,...,p iM ]Where M is the number of known classes in the probability vector p i To choose the maximum value p m And the sub-maximum p sm To obtain a difference value thereofProbability v d ＝p m -p sm ；
Inputting single one-dimensional range profiles of different types of targets, wherein each type of target can obtain a difference vector:
Figure FDA0003673933440000017
wherein d is m Is the difference vector for the mth class of objects,
Figure FDA0003673933440000018
the difference probability of the ith one-dimensional range profile of the mth type target is shown, and the superscript T represents a transpose symbol;
the difference vector d of all known targets m Calculating a histogram of the difference probabilities in the image, and selecting one difference probability from the histogram of the difference probabilities as an identification threshold tau according to the predetermined correct discrimination rate of the known target;
s4, unknown target identification:
inputting the obtained single one-dimensional range profile of the unknown target into a trained deep convolutional neural network model to obtain a corresponding difference vector
Figure FDA0003673933440000021
Wherein
Figure FDA0003673933440000022
The difference probability corresponding to the c test one-dimensional range profile data is obtained, and P represents the number of unknown target data;
will d t Comparing the difference probability of the inner one-dimensional range profile with the identification threshold tau, and if the difference probability is more than or equal to the threshold
Figure FDA0003673933440000023
Identifying the c-th one-dimensional range profile data as a known target; if the difference probability is less than the threshold
Figure FDA0003673933440000024
Identifying the c-th one-dimensional range profile data as an unknown target, namely the identification rule is as follows:
Figure FDA0003673933440000025
wherein
Figure FDA0003673933440000026
Indicating that the ith unknown target one-dimensional range image data belongs to a known target,
Figure FDA0003673933440000027
and indicating that the ith unknown target one-dimensional range profile data belongs to the unknown target.
CN202010152743.6A 2020-03-06 2020-03-06 Unknown target identification method based on deep convolutional neural network Active CN111352086B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010152743.6A CN111352086B (en) 2020-03-06 2020-03-06 Unknown target identification method based on deep convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010152743.6A CN111352086B (en) 2020-03-06 2020-03-06 Unknown target identification method based on deep convolutional neural network

Publications (2)

Publication Number Publication Date
CN111352086A CN111352086A (en) 2020-06-30
CN111352086B true CN111352086B (en) 2022-08-02

Family

ID=71194330

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010152743.6A Active CN111352086B (en) 2020-03-06 2020-03-06 Unknown target identification method based on deep convolutional neural network

Country Status (1)

Country Link
CN (1) CN111352086B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112014821B (en) * 2020-08-27 2022-05-17 电子科技大学 Unknown vehicle target identification method based on radar broadband characteristics
CN112163510B (en) * 2020-09-25 2022-04-22 电子科技大学 Human body action classification recognition method based on multi-observation variable HMM model
CN112163636B (en) * 2020-10-15 2023-09-26 电子科技大学 Unknown mode identification method of electromagnetic signal radiation source based on twin neural network
CN113156416B (en) * 2021-05-17 2022-05-17 电子科技大学 Unknown target discrimination method based on multi-kernel dictionary learning
CN113640764B (en) * 2021-08-09 2023-08-11 中国人民解放军海军航空大学航空作战勤务学院 Radar one-dimensional range profile identification method and device based on multi-dimension one-dimensional convolution
CN114821335B (en) * 2022-05-20 2023-04-25 电子科技大学 Unknown target discrimination method based on fusion of depth features and linear discrimination features

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107341447A (en) * 2017-06-13 2017-11-10 华南理工大学 A kind of face verification mechanism based on depth convolutional neural networks and evidence k nearest neighbor
CN107729986A (en) * 2017-09-19 2018-02-23 平安科技（深圳）有限公司 Driving model training method, driver's recognition methods, device, equipment and medium
CN107766893A (en) * 2017-11-03 2018-03-06 电子科技大学 Target identification method based on label multilevel coding neutral net
CN109086700A (en) * 2018-07-20 2018-12-25 杭州电子科技大学 Radar range profile's target identification method based on depth convolutional neural networks
WO2019155214A1 (en) * 2018-02-07 2019-08-15 Green Running Limited Method and apparatus for power signal disaggragation using a convolutional neural network
CN114333774A (en) * 2021-12-15 2022-04-12 腾讯科技（深圳）有限公司 Speech recognition method, speech recognition device, computer equipment and storage medium

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107341447A (en) * 2017-06-13 2017-11-10 华南理工大学 A kind of face verification mechanism based on depth convolutional neural networks and evidence k nearest neighbor
CN107729986A (en) * 2017-09-19 2018-02-23 平安科技（深圳）有限公司 Driving model training method, driver's recognition methods, device, equipment and medium
CN107766893A (en) * 2017-11-03 2018-03-06 电子科技大学 Target identification method based on label multilevel coding neutral net
WO2019155214A1 (en) * 2018-02-07 2019-08-15 Green Running Limited Method and apparatus for power signal disaggragation using a convolutional neural network
CN109086700A (en) * 2018-07-20 2018-12-25 杭州电子科技大学 Radar range profile's target identification method based on depth convolutional neural networks
CN114333774A (en) * 2021-12-15 2022-04-12 腾讯科技（深圳）有限公司 Speech recognition method, speech recognition device, computer equipment and storage medium

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ImageNet Classification with Deep Convolutional Neural Networks;Krizhevsky. A, et al;《Communications of the ACM》;20171231;84-91 *
基于卷积神经网络提取冬小麦空间分布信息的方法研究;宋德娟;《中国优秀硕士学位论文全文数据库 工程科技II辑》;20200115;C028-162 *
基于深度卷积神经网络的极化雷达目标识别;盖晴晴;《电波科学学报》;20181031;第22卷(第5期);575-582 *

Also Published As

Publication number Publication date
CN111352086A (en) 2020-06-30

Similar Documents

Publication Publication Date Title
CN111352086B (en) Unknown target identification method based on deep convolutional neural network
CN108154118B (en) A kind of target detection system and method based on adaptive combined filter and multistage detection
CN106599797B (en) A kind of infrared face recognition method based on local parallel neural network
CN110334741B (en) Radar one-dimensional range profile identification method based on cyclic neural network
CN108256436B (en) Radar HRRP target identification method based on joint classification
CN111368885B (en) Gas circuit fault diagnosis method for aircraft engine
CN111273288B (en) Radar unknown target identification method based on long-term and short-term memory network
CN104408469A (en) Firework identification method and firework identification system based on deep learning of image
CN112802054B (en) Mixed Gaussian model foreground detection method based on fusion image segmentation
Ahmadi et al. Iris tissue recognition based on GLDM feature extraction and hybrid MLPNN-ICA classifier
CN103886334A (en) Multi-index fused hyperspectral remote sensing image dimensionality reduction method
Xiao et al. Specific emitter identification of radar based on one dimensional convolution neural network
CN110033043B (en) Radar one-dimensional range profile rejection method based on condition generation type countermeasure network
CN110412548B (en) Radar multi-target recognition method based on high-resolution one-dimensional range profile
Guo et al. Two-level K-nearest neighbors approach for invasive plants detection and classification
ul Haq et al. Neural network based adaboosting approach for hyperspectral data classification
CN111368653B (en) Low-altitude small target detection method based on R-D graph and deep neural network
CN114821335B (en) Unknown target discrimination method based on fusion of depth features and linear discrimination features
CN111310838A (en) Drug effect image classification and identification method based on depth Gabor network
Senthilnath et al. A novel harmony search-based approach for clustering problems
Eleyan Breast cancer classification using moments
CN114137518A (en) Radar high-resolution range profile open set identification method and device
CN114624715A (en) Radar echo extrapolation method based on self-attention space-time neural network model
CN109946667B (en) Convolutional neural network-based method for identifying high-resolution range image of morphing aircraft
Sahay et al. Uncertainty quantification-based unmanned aircraft system detection using deep ensembles

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