CN114078214A

CN114078214A - Radar target RCS identification method and device based on complex neural network

Info

Publication number: CN114078214A
Application number: CN202111393850.9A
Authority: CN
Inventors: 冯雪健; 霍超颖; 毛冠乔; 邓浩川; 韦笑; 殷红成
Original assignee: Beijing Institute of Environmental Features
Current assignee: Beijing Institute of Environmental Features
Priority date: 2021-11-23
Filing date: 2021-11-23
Publication date: 2022-02-22
Anticipated expiration: 2041-11-23
Also published as: CN114078214B

Abstract

The invention relates to a radar target RCS identification method and device based on a plurality of neural networks. Identifying and classifying the Complex RCS data of the target by adopting a Complex neural network, wherein the Complex convolution layer performs Complex convolution on the Complex RCS data, then performs activation on the Complex RCS data by using a Complex Relu activation function, and then performs maximum pooling operation; the first real convolution layer performs real convolution on the complex features, then performs activation by a Relu activation function, and then performs maximum pooling operation; and the second real convolution layer performs real convolution on the maximum real characteristic, then performs average pooling operation after being activated by a Relu activation function, and finally outputs a classification characteristic vector. The invention realizes the deep learning identification of the plural RCS data with richer target characteristic information and can effectively improve the identification accuracy of the target.

Description

Radar target RCS identification method and device based on complex neural network

Technical Field

The invention relates to the technical field of radar target identification, in particular to a radar target RCS identification method and device based on a plurality of neural networks.

Background

Since the military aerial target detection of the two-war radar, the military radar with all-weather and long-distance detection function is a research hotspot of researchers in various countries for decades as a battlefield with 'thousand eyes'.

Target RCS (radar cross section) is a main feature of radar detection, and is often used for detection, tracking and identification for military purposes. The original RCS data are complex data, compared with single real part amplitude RCS information, the complex RCS has richer representation capability, the real part amplitude of the complex RCS can express a great deal of target size information, the imaginary part phase of the complex RCS also contains rich target distance information, and the target identification accuracy rate can be effectively improved if the complex RCS can be completely used for target identification. The existing radar target RCS identification method based on the neural network only utilizes real part amplitude data of the target RCS, and how to further improve the utilization rate of original RCS information and further improve the identification accuracy of the target has great significance for identification application of military targets.

Disclosure of Invention

The invention aims to solve the technical problem that a neural network cannot be used for identifying and applying a plurality of RCS data in the prior art, and provides a radar target RCS identification method and device based on the plurality of neural networks aiming at the defects in the prior art.

In order to solve the technical problem, the invention provides a radar target RCS identification method based on a plurality of neural networks, which comprises the following steps:

performing Complex convolution on Complex RCS data of a target by a Complex convolution layer of a Complex neural network, outputting Complex eigenvectors, activating the Complex eigenvectors by using a Complex Relu activation function to obtain Complex activated eigenvectors, and performing maximum pooling operation on the Complex activated eigenvectors to obtain Complex characteristics; wherein, the filter of the plural convolution layer is W ═ A + iB, the plural RCS data of the target is h ═ x + iy, the plural feature vector is W ═ h ═ (A × -B × + i (B × x + A ×) and the Complex Relu activation function is W ═ x + A ═ x, the Complex Relu activation function is

The complex activation feature vector is

A. B, x and y are real matrices;

the method comprises the steps that a first real convolution layer of a complex neural network conducts real convolution on complex features, a first real feature vector is output, a Relu activation function is used for activating the first real feature vector to obtain a first real activation feature vector, and maximum pooling operation is conducted on the real activation feature vector to obtain maximum real features;

the second real convolution layer of the complex neural network performs real convolution on the maximum real feature, outputs a second real feature vector, activates the second real feature vector by using a Relu activation function to obtain a second real activation feature vector, and performs average pooling operation on the second real activation feature vector to obtain an average real feature;

the full-connection layer of the complex neural network processes the average real number characteristic to obtain a full-connection characteristic vector;

a softmax classifier of the complex neural network classifies the fully connected feature vectors and outputs classified feature vectors; the classification feature vector is used to determine the type of the target.

Optionally, the training of the complex neural network comprises the following steps:

constructing a complex vector data set based on complex RCS data of different types of targets; wherein the complex vector data set comprises a training data set and a test data set;

carrying out multiple batch normalization processing on the training data set and the test data set to obtain multiple batches of training data sets and multiple batches of test data sets;

constructing a plurality of neural networks based on a plurality of training data sets; the Complex neural network comprises a Complex convolutional layer, a first real convolutional layer, a second real convolutional layer, a full connection layer and a softmax classifier, wherein the Complex convolutional layer adopts a Complex Relu activation function and maximum pooling, the first real convolutional layer adopts a Relu activation function and maximum pooling, and the second real convolutional layer adopts a Relu activation function and average pooling;

carrying out learning training on the plurality of neural networks through a plurality of training data sets, and stopping training after the maximum iteration times is reached;

and inputting the multiple test data sets into the trained multiple neural networks for identification and verification, finishing the training if the identification accuracy is greater than or equal to a preset threshold, and retraining if the identification accuracy is less than the preset threshold.

Optionally, the complex RCS data of different states of the target are divided into a training data set and a training test data set based on the following data set division formula:

where rand is the random number generated for each complex RCS data, the random number is for (0,1), r1 is the data set construction parameter, r1 is for (0, 1).

Optionally, the training data set and the test data set are subjected to a batch normalization process based on the following formula:

wherein,

is the scaling data of the complex RCS data,

x is complex RCS data, E [ X ]]Is the mean of X;

v is a covariance matrix which is a function of the covariance,

gamma is a scaling parameter that is a function of,

β is the translation factor.

In order to solve the above technical problem, the present invention further provides a radar target RCS identification apparatus based on a plurality of neural networks, including:

the Complex convolution module is used for carrying out Complex convolution on Complex RCS data of a target by using a Complex convolution layer of a Complex neural network, outputting a Complex feature vector, activating the Complex feature vector by using a Complex Relu activation function to obtain a Complex activation feature vector, and carrying out maximum pooling operation on the Complex activation feature vector to obtain Complex features; wherein, the filter of the plural convolution layer is W ═ A + iB, the plural RCS data of the target is h ═ x + iy, the plural feature vector is W ═ h ═ (A × -B × + i (B × x + A ×) and the Complex Relu activation function is W ═ x + A ═ x, the Complex Relu activation function is

The complex activation feature vector is

A. B, x and y are real matrices;

the first real convolution module is used for carrying out real convolution on complex characteristics by utilizing a first real convolution layer of a complex neural network, outputting a first real characteristic vector, activating the first real characteristic vector by utilizing a Relu activation function to obtain a first real activation characteristic vector, and carrying out maximum pooling operation on the real activation characteristic vector to obtain maximum real characteristics;

the second real convolution module is used for carrying out real convolution on the maximum real feature by utilizing a second real convolution layer of the complex neural network, outputting a second real feature vector, activating the second real feature vector by utilizing a Relu activation function to obtain a second real activation feature vector, and carrying out average pooling operation on the second real activation feature vector to obtain an average real feature;

the full-connection module is used for processing the average real number characteristic by using a full-connection layer of the complex neural network to obtain a full-connection characteristic vector;

the classification module is used for classifying the fully-connected feature vectors by using a softmax classifier of the complex neural network and outputting the classified feature vectors; the classification feature vector is used to determine the type of the target.

wherein,

is the scaling data of the complex RCS data,

x is complex RCS data, E [ X ]]Is the mean of X;

v is a covariance matrix which is a function of the covariance,

gamma is a scaling parameter that is a function of,

β is the translation factor.

In order to solve the technical problem, the invention further provides a radar target RCS identification terminal based on the plurality of neural networks.

The invention relates to a radar target RCS identification terminal based on a plurality of neural networks, which comprises: one or more processors; a storage device, configured to store one or more programs, which when executed by the one or more processors, cause the one or more processors to implement a complex neural network-based radar target RCS identification method of the present invention.

In order to solve the above technical problem, the present invention further provides a computer-readable storage medium.

A computer-readable storage medium of the present invention has stored thereon a computer program which, when executed by a processor, implements a radar target RCS recognition method of the present invention based on a complex neural network.

The radar target RCS identification method and device based on the plurality of neural networks have the following beneficial effects: the deep learning identification of the complex RCS data with richer target characteristic information is realized based on the complex neural network, the method can be applied to the field of radar target identification, the identification accuracy of the target can be effectively improved, and the method has important practical application value.

Drawings

Fig. 1 is a schematic diagram of a radar target RCS identification method based on a complex neural network according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a complex neural network training system provided by an embodiment of the present invention;

FIG. 3 is a schematic diagram of a plurality of neural networks provided by an embodiment of the present invention;

FIG. 4 is a schematic diagram of the recognition accuracy of a complex neural network according to the variation of iteration times provided by the embodiment of the present invention;

FIG. 5 is a diagram one of the output feature vectors of the target;

FIG. 6 is a second schematic of the output feature vector of the target;

FIG. 7 is a third schematic of the output feature vector of the target;

FIG. 8 is a diagram four of the output feature vector of the target;

FIG. 9 is a schematic diagram five of the output feature vector of the target;

FIG. 10 is a schematic diagram of the main modules of a complex neural network based radar target RCS identification apparatus according to an embodiment of the present invention;

FIG. 11 is an exemplary system architecture diagram in which embodiments of the present invention may be employed;

fig. 12 is a schematic structural diagram of a computer system suitable for implementing a terminal device or a server according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.

Because the complex RCS data has richer representation capability, the complex RCS data not only can express very much target size information on the real part amplitude, but also has abundant target distance information on the imaginary part phase, and if the complex RCS data can be completely used for target identification, the target identification accuracy rate can be effectively improved, but the prior art cannot use a neural network for identification application of the complex RCS data, and therefore a new target identification method capable of utilizing the complex RCS data is urgently needed.

According to the radar target RCS identification method and device based on the plurality of neural networks, the deep learning identification of the plurality of RCS data with richer target characteristic information is realized based on the plurality of neural networks, and the identification accuracy of the target can be effectively improved.

As shown in fig. 1, a radar target RCS identification method based on a plurality of neural networks provided by the embodiment of the present invention mainly includes the following steps:

s101, Complex convolution layers of a Complex neural network perform Complex convolution on Complex RCS data of a target, Complex feature vectors are output, Complex Relu activation functions are used for activating the Complex feature vectors to obtain Complex activation feature vectors, and maximum pooling operation is performed on the Complex activation feature vectors to obtain Complex features.

The complex RCS data is original RCS data of the object, the complex RCS data not only can express very much object size information in real part amplitude, but also contains abundant object distance information in imaginary part phase, the complex RCS data can be expressed as h ═ x + iy, x and y are real matrixes, and x and y respectively express the real part and the imaginary part of the complex RCS data.

In the embodiment of the present invention, the complex neural network is a 3-layer neural network obtained based on an improvement on a common Convolutional Neural Network (CNN), and includes one complex convolutional layer and two real convolutional layers (a first real convolutional layer and a second real convolutional layer), the complex convolutional layers are added on the basis of the convolutional neural network, a filter of the complex convolutional layers is W ═ a + iB, a and B are real matrices, a and B correspond to a real part and an imaginary part, respectively, the complex convolutional layers perform complex convolution on complex RCS data and output complex eigenvectors, which can be represented as W ═ h ═ a x-B × + i (B x + a ×), and if the real part and the imaginary part of the convolutional operation are represented by a matrix representation, there are real parts and imaginary parts of the convolutional operation

The ComplexRelu activation function is a function that applies a single ReLU to both the real and imaginary parts, and the ComplexRelu activation function used by the Complex convolution layer can be expressed as

After activation by the Complex Relu activation function, the Complex activation feature vector can be expressed as

Set of points removed after activation

In addition, the Cauchy-Riemann equation is satisfied everywhere, and z is W h.

S102, a first real convolution layer of the complex neural network performs real convolution on complex features, outputs a first real feature vector, activates the first real feature vector by using a Relu activation function to obtain a first real activation feature vector, and performs maximum pooling operation on the real activation feature vector to obtain maximum real features.

Here, the real convolution is a convolution operation on a complex feature. It should be noted that the real convolution, Relu activation function, and max pooling adopted by the first real convolution layer are the same as those of the conventional convolutional neural network, and are not described herein again.

S103, carrying out real number convolution on the maximum real number feature by a second real number convolution layer of the complex neural network, outputting a second real number feature vector, activating the second real number feature vector by using a Relu activation function to obtain a second real number activation feature vector, and carrying out average pooling operation on the second real number activation feature vector to obtain an average real number feature.

Here, the real convolution is a convolution operation on the maximum real feature. It should be noted that the real convolution, Relu activation function, and average pooling adopted by the second real convolution layer are the same as those of the conventional convolutional neural network, and are not described herein again.

And S104, processing the average real number characteristic by the full-connection layer of the complex neural network to obtain a full-connection characteristic vector.

The fully-connected layer and softmax classifier of the complex neural network are consistent with the method of the common convolutional neural network, and are not described herein in detail.

And S105, classifying the fully connected feature vectors by a softmax classifier of the complex neural network, and outputting the classified feature vectors. The classification feature vector is used for determining the type of the target, so that a classification result of the target is obtained.

As shown in fig. 2, the training of the complex neural network may include the following steps:

s201, constructing a complex vector data set based on complex RCS data of different types of targets; the complex vector data set comprises a training data set and a testing data set, wherein the training data set is used for training the complex neural network, and the testing data set is used for checking whether the trained complex neural network can be used for subsequent target identification;

s202, carrying out multiple batch normalization processing on the training data set and the test data set; obtaining a plurality of training data sets and a plurality of testing data sets after a plurality of normalization processes; before the constructed complex vector data set is used, complex batch normalization processing needs to be carried out on the complex vector data set in order to improve the accuracy of final classification and identification and facilitate network parameter setting;

s203, constructing a plurality of neural networks based on a plurality of training data sets; the Complex neural network mainly comprises a Complex convolutional layer, a first real convolutional layer, a second real convolutional layer, a full connection layer and a softmax classifier, wherein the Complex convolutional layer adopts a Complex Relu activation function and maximum pooling, the first real convolutional layer adopts a Relu activation function and maximum pooling, and the second real convolutional layer adopts a Relu activation function and average pooling;

s204, learning and training the plurality of neural networks through a plurality of training data sets; the maximum iteration times can be preset during training, one training is completed when the maximum iteration times are reached, and the training is stopped; the maximum iteration number can be set according to actual conditions, and as a preferred embodiment, the maximum iteration number can be set to be 50;

s205, inputting multiple batches of test data sets into the trained multiple neural networks for identification and verification; finishing training if the recognition accuracy is greater than or equal to a preset threshold, and retraining if the recognition accuracy is less than the preset threshold; the identification accuracy rate can be adopted for the training evaluation of the plurality of neural networks, when the identification accuracy rate is greater than or equal to a preset threshold value, the trained plurality of neural networks can be used for target identification after meeting the use requirements, and the preset threshold value can be set according to actual conditions.

In the embodiment of the present invention, the complex RCS data of different states of the target may be divided into a training data set and a training test data set based on the following data set division formula:

Assuming that there are N types of target complex RCS data, each type of target complex RCS data having M states, and the length of each state complex RCS data is P, the total complex RCS data of the N types of targets can be represented by a matrix as [ N × M × P ], and correspondingly, the complex RCS data of a single target can be represented by a matrix as [ M × P ]. When the training data set and the training test data set are divided, the complex RCS data of each state in the matrix [ N M P ] are divided according to a data set division formula, namely the complex RCS data of each state of all targets are divided into the training data set or the test data set by the data set division formula.

The complex Batch Normalization, is a Batch Normalization that selects a whitening two-dimensional vector, i.e., scales the data along each of the two principal components by the square root of its variance. In the embodiment of the invention, the training data set and the test data set are subjected to multiple batch normalization processing based on the following formula:

wherein

Is the scaling data of the complex RCS data,

x is complex RCS data, E [ X ]]Is the mean value of X, and it should be noted that the Complex RCS data is activated by the Complex Relu activation function;

v is a covariance matrix which is a function of the covariance,

gamma is a scaling parameter that is a function of,

β is the translation factor.

In order to further illustrate the technical idea of the present invention, the technical solution of the present invention will now be described with reference to specific application scenarios.

The first step is as follows: constructing complex vector data sets

Assuming that there are N types of target complex RCS data, each type of target complex RCS data having M states, and the length of the complex RCS data for each state is P, the total complex RCS data for the N types of targets may be represented by a matrix as [ N × M × P ], and the complex RCS data for the corresponding individual target may be represented by a matrix as [ M × P ]. When constructing the training data set and the test data set, it is determined whether the different state data is part of the training data set or part of the test data set by the criteria in the following disclosure.

Wherein rand is a random number generated for each complex RCS data, and takes values between (0,1), and r 1E (0,1) is a data set construction parameter set manually.

The second step is that: construction of complex neural networks

A 3-layer complex neural network is constructed by complex convolution and real convolution, as shown in fig. 3.

The method comprises the steps that Complex RCS data of various targets are input into a constructed Complex neural network, a first layer is a Complex convolution layer, the Complex RCS data are activated by a Complex Relu activation function after being subjected to Complex convolution, then the Complex RCS data enter a second layer (namely a first real convolution layer) and a third layer (namely a second real convolution layer) after being subjected to maximum pooling operation, the second layer and the third layer are activated by the Relu activation function after the real convolution, the second layer and the third layer are subjected to maximum pooling after the activation of the second layer activation function and then enter the third layer, average pooling is performed after the activation of the third layer activation function, the second layer and the third layer are connected with a full-connection layer and are classified by a softmax classifier, and a final classification result is output.

The Relu activation function, the maximum pooling, the real convolution, the average pooling, the full connection layer and the softmax classifier in the constructed 3-layer Complex neural network are all consistent with the modules in the general convolutional neural network, and particularly two modules, namely the Complex convolution and the Complex Relu activation function, need to be noticed.

With respect to complex convolution

To perform two-dimensional convolution in the complex domain equivalent to the conventional real-valued convolution, assume that the convolved complex filter matrix is W-a + iB and the vector of the input complex RCS data is denoted as h-x + iy, where a and B, x and y are real matrices. Since the convolution operator is of the partition type, the vector h is convolved with the filter W to obtain:

W*h＝(A*x-B*y)+i(B*x+A*y)

if the real and imaginary parts of the convolution operation are represented in a matrix representation, then there are:

for complete Relu activation function

The Complex Relu activation function, also called Complex activation or CRelu, is the simultaneous application of individual relus to the real and imaginary parts of a neuron (i.e., a Complex eigenvector), i.e.:

it should be noted that, the reactivation division point set based on the ReLU

In addition, the Cauchy-Riemann equation is satisfied anywhere.

In addition, before the constructed complex vector data set is input into the constructed complex neural network, in order to improve the accuracy of final classification and identification and facilitate the setting of network parameters, the complex vector data set needs to be subjected to multiple batch normalization processing to obtain multiple batches of training data sets and multiple batches of test data sets.

Normalization on multiple batches

A whitening two-dimensional vector is selected, scaling the data by the square root of its variance along each of the two principal components. This can be achieved by multiplying the zero-centered data (X-E [ X ]) by the inverse square root of the 2X2 covariance matrix V:

is scaling data of complex RCS data, X is complex RCS data, E [ X [ ]]Is the mean value of X, and it should be noted that the Complex RCS data is activated by the Complex Relu activation function;

wherein the covariance matrix V is:

the scaling parameters are:

multiple batch normalization is defined as:

β is the translation factor.

The third step: and (5) training a plurality of neural networks.

And inputting the constructed multiple batches of training data sets into the established multiple neural networks, jumping out of the networks after the preset maximum network iteration times are reached, and finishing the training of the multiple neural networks.

The fourth step: and (5) identifying and verifying.

And inputting the constructed multiple batches of test data sets into the trained multiple neural networks, and outputting classification recognition results.

For example, assuming that the multiple training data sets and the multiple testing data sets have multiple RCS data of five types of targets, namely pentahedral data (bihedral), cubic data (cubic), cylindrical data (cylinder), sphere-cone data (sphere-cone), and trigonal data (trihedral), the multiple RCS data are classified and identified by using a common 3-layer neural network and the multiple neural network according to the embodiment of the present invention, the maximum iteration number is set to be 50 times, and the obtained identification result is shown in table 1, for example, it can be seen that the complex nerves constructed herein can effectively improve the identification accuracy of the targets.

TABLE 1

Method	Rate of accuracy
		Traditional 3-layer neural network	0.932
Plural neural networks	0.979

Continuing with the above example, the identification accuracy of the complex neural network constructed in the embodiment of the present invention along with the change of the number of iterations is shown in fig. 4.

Continuing with the above example, the examples of the input (i.e. Complex RCS data) of the 5-class target, the feature vector (i.e. Complex feature) finally output by the layer 1 (via Complex convolution, Complex Relu activation function activation and maximum pooling operation), the feature vector (i.e. maximum real feature) finally output by the layer 2 (via real convolution, Relu activation function activation and maximum pooling operation), the feature vector (i.e. average real feature) finally output by the layer 3 (via real convolution, Relu activation function activation and average pooling operation), the feature vector (i.e. full-connected feature vector) finally output by the layer 4 (i.e. full-connected layer), and the feature vector (i.e. classification feature vector) finally output (i.e. softmax classifier) are respectively shown in fig. 5-9.

Fig. 10 is a schematic diagram of main blocks of a radar target RCS recognition apparatus based on a complex neural network according to an embodiment of the present invention.

As shown in fig. 10, a radar target RCS recognition apparatus 1000 based on a complex neural network according to an embodiment of the present invention includes: a complex convolution module 1001, a first real convolution module 1002, a second real convolution module 1003, a fully-connected module 1004, and a classification module 1005.

Wherein,

the Complex convolution module 1001 is configured to perform Complex convolution on Complex RCS data of a target by using a Complex convolution layer of a Complex neural network, output a Complex feature vector, activate the Complex feature vector by using a Complex Relu activation function to obtain a Complex activation feature vector, and perform maximum pooling operation on the Complex activation feature vector to obtain a Complex feature; wherein, the filter of the plural convolution layer is W ═ A + iB, the plural RCS data of the target is h ═ x + iy, the plural feature vector is W ═ h ═ (A × -B × + i (B × x + A ×) and the Complex Relu activation function is W ═ x + A ═ x, the Complex Relu activation function is

The complex activation feature vector is

A. B, x and y are real matrices;

the first real convolution module 1002 is configured to perform real convolution on a complex feature by using a first real convolution layer of a complex neural network, output a first real feature vector, activate the first real feature vector by using a Relu activation function to obtain a first real activation feature vector, and perform maximum pooling operation on the real activation feature vector to obtain a maximum real feature;

a second real convolution module 1003, configured to perform real convolution on the maximum real feature by using a second real convolution layer of the complex neural network, output a second real feature vector, activate the second real feature vector by using a Relu activation function to obtain a second real activation feature vector, and perform average pooling operation on the second real activation feature vector to obtain an average real feature;

a full-connection module 1004, configured to process the average real number feature by using a full-connection layer of the complex neural network to obtain a full-connection feature vector;

a classification module 1005, configured to classify the fully-connected feature vectors by using a softmax classifier of the complex neural network, and output the classified feature vectors; the classification feature vector is used to determine the type of the target.

In the embodiment of the present invention, the training of the complex neural network includes the following steps:

In the embodiment of the present invention, the training data set and the test data set may be subjected to multiple batch normalization processing based on the following formulas:

wherein,

is the scaling data of the complex RCS data,

x is complex RCS data, E [ X ]]Is the mean of X;

v is a covariance matrix which is a function of the covariance,

gamma is a scaling parameter that is a function of,

β is the translation factor.

Fig. 11 illustrates an exemplary system architecture 1100 to which the complex neural network-based radar target RCS identification method or the complex neural network-based radar target RCS identification apparatus of the embodiments of the present invention may be applied. The statements made in the text preceding the figures relate to the introduction of the present solution when implemented by a computer, are a general expression and do not suggest any major modifications.

As shown in fig. 11, the system architecture 1100 may include

terminal devices

1101, 1102, 1103, a network 1104, and a server 1105. The network 1104 is a medium to provide communication links between the

terminal devices

1101, 1102, 1103 and the server 1105. Network 1104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, to name a few.

A user may use

terminal devices

1101, 1102, 1103 to interact with a server 1105 over a network 1104 to receive or send messages or the like. Various communication client applications may be installed on the

terminal devices

1101, 1102, 1103.

The

terminal devices

1101, 1102, 1103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like.

The server 1105 may be a server that provides various services, such as a background management server that supports shopping websites browsed by users using the

terminal apparatuses

1101, 1102, 1103. The background management server can analyze and process the received data such as the product information inquiry request and feed back the processing result to the terminal equipment.

It should be noted that the radar target RCS identification method based on the plurality of neural networks provided by the embodiment of the present invention is generally executed by the server 1105, and accordingly, the radar target RCS identification apparatus based on the plurality of neural networks is generally disposed in the server 1105.

It should be understood that the number of terminal devices, networks, and servers in fig. 11 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

Referring now to FIG. 12, shown is a block diagram of a computer system 1200 suitable for use with a terminal device implementing an embodiment of the present invention. The terminal device shown in fig. 12 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiment of the present invention.

As shown in fig. 12, the computer system 1200 includes a Central Processing Unit (CPU)1201, which can perform various appropriate actions and processes in accordance with a program stored in a Read Only Memory (ROM)1202 or a program loaded from a storage section 1208 into a Random Access Memory (RAM) 1203. In the RAM 1203, various programs and data necessary for the operation of the system 1200 are also stored. The CPU 1201, ROM 1202, and RAM 1203 are connected to each other by a bus 1204. An input/output (I/O) interface 1205 is also connected to bus 1204.

The following components are connected to the I/O interface 1205: an input section 1206 including a keyboard, a mouse, and the like; an output portion 1207 including a display device such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage section 1208 including a hard disk and the like; and a communication section 1209 including a network interface card such as a LAN card, a modem, or the like. The communication section 1209 performs communication processing via a network such as the internet. A driver 1210 is also connected to the I/O interface 1205 as needed. A removable medium 1211, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like, is mounted on the drive 1210 as necessary, so that a computer program read out therefrom is mounted into the storage section 1208 as necessary.

In particular, according to the embodiments of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication section 1209, and/or installed from the removable medium 1211. The computer program performs the above-described functions defined in the system of the present invention when executed by the Central Processing Unit (CPU) 1201.

It should be noted that the computer readable medium shown in the present invention can be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present invention, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present invention, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The modules described in the embodiments of the present invention may be implemented by software or hardware. The described modules may also be provided in a processor, which may be described as: a processor includes a complex convolution module, a first real convolution module, a second real convolution module, a full-connect module, and a classification module. The names of these modules do not limit the module itself in some cases, for example, the sending module may also be described as a "module that processes the average real number feature using the fully-connected layer of the complex neural network to obtain a fully-connected feature vector".

As another aspect, the present invention also provides a computer-readable medium that may be contained in the apparatus described in the above embodiments; or may be separate and not incorporated into the device. The computer readable medium carries one or more programs which, when executed by a device, cause the device to comprise: s101, Complex convolution layers of a Complex neural network perform Complex convolution on Complex RCS data of a target, Complex feature vectors are output, Complex Relu activation functions are used for activating the Complex feature vectors to obtain Complex activation feature vectors, and maximum pooling operation is performed on the Complex activation feature vectors to obtain Complex features; s102, performing real number convolution on complex number features by a first real number convolution layer of a complex number neural network, outputting a first real number feature vector, activating the first real number feature vector by using a Relu activation function to obtain a first real number activation feature vector, and performing maximum pooling operation on the real number activation feature vector to obtain maximum real number features; s103, carrying out real number convolution on the maximum real number feature by a second real number convolution layer of the complex neural network, outputting a second real number feature vector, activating the second real number feature vector by using a Relu activation function to obtain a second real number activation feature vector, and carrying out average pooling operation on the second real number activation feature vector to obtain an average real number feature; s104, processing the average real number characteristic by a full-connection layer of the complex neural network to obtain a full-connection characteristic vector; and S105, classifying the fully connected feature vectors by a softmax classifier of the complex neural network, and outputting the classified feature vectors.

In summary, the radar target RCS identification method and device based on the complex neural network provided by the invention can be used for deep learning and identification of complex RCS data with richer target feature information, can be applied to the field of radar target identification, can effectively improve the identification accuracy of targets, and have important practical application value.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. A radar target RCS identification method based on a plurality of neural networks is characterized by comprising the following steps:

The complex activation feature vector is

A. B, x and y are real matrices;

2. The method of claim 1, wherein the training of the complex neural network comprises the steps of:

3. The method of claim 2, wherein:

the complex RCS data of different states of the target are divided into a training data set and a training test data set based on the following data set division formula:

4. The method of claim 2, wherein the training dataset and the test dataset are subjected to a multiple batch normalization process based on the following formula:

wherein,

is the scaling data of the complex RCS data,

x is complex RCS data, E [ X ]]Is the mean of X;

v is a covariance matrix which is a function of the covariance,

gamma is a scaling parameter that is a function of,

β is the translation factor.

5. A radar target RCS recognition device based on a plurality of neural networks is characterized by comprising the following components:

The complex activation feature vector is

A. B, x and y are real matrices;

6. The apparatus of claim 5, wherein the training of the complex neural network comprises the steps of:

7. The apparatus of claim 6, wherein:

8. The apparatus of claim 6, wherein the training dataset and the test dataset are subjected to a multiple batch normalization process based on the following equation:

wherein,

is the scaling data of the complex RCS data,

x is complex RCS data, E [ X ]]Is the mean of X;

v is a covariance matrix which is a function of the covariance,

gamma is a scaling parameter that is a function of,

β is the translation factor.

9. A radar target RCS identification terminal based on a plurality of neural networks is characterized by comprising:

one or more processors;

a storage device for storing one or more programs,

when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-4.

10. A computer-readable medium, on which a computer program is stored, which, when being executed by a processor, carries out the method according to any one of claims 1-4.