CN114118142A - Method for identifying radar intra-pulse modulation type - Google Patents
Method for identifying radar intra-pulse modulation type Download PDFInfo
- Publication number
- CN114118142A CN114118142A CN202111310285.5A CN202111310285A CN114118142A CN 114118142 A CN114118142 A CN 114118142A CN 202111310285 A CN202111310285 A CN 202111310285A CN 114118142 A CN114118142 A CN 114118142A
- Authority
- CN
- China
- Prior art keywords
- layer
- training
- neural network
- signal
- radar
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a method for identifying a radar intra-pulse modulation type, which comprises the following steps of: generating a modulation signal data set by utilizing radar signals, wherein the modulation signal data set comprises 7 types of modulation signal samples, and each type of signal is at least provided with 10 signal-to-noise ratios; and (3) dividing a training set and a test set: at the signal-to-noise ratio, randomly selecting and forming a training set, a verification set and a test set according to a preset proportion; constructing a neural network model: the neural network model is sequentially provided with an input layer, a convolution layer, a pooling layer, a normalization layer, a channel attention module, a full-connection layer and an output layer from the input layer to the output layer; training and verifying a neural network model: inputting the training set into a neural network for training, and selecting a neural network model with the highest accuracy of the verification set as a final neural network model in the training result; identification of modulation type: and inputting a test set in the final neural network model, and outputting the recognition rate of the radar tuning type.
Description
Technical Field
The invention belongs to the technical field of signal processing, and particularly relates to a method for identifying a radar intra-pulse modulation type.
Background
Electronic countermeasure plays an important role in electronic information reconnaissance, electronic support and threat warning systems, and radar intra-pulse modulation type identification is an important link in electronic countermeasure. With the development and progress of science and technology, the system of radar is continuously updated, and the electronic environment is more intensive and complex, which increases the difficulty of extracting characteristic parameters from received radar signals. Moreover, the received radar intra-pulse modulation signals often have various noises, and the intra-pulse modulation signals have larger signal-to-noise ratio range due to larger difference of electromagnetic environments, so that the difficulty of intra-pulse modulation identification is greatly increased. However, in the current environment, the electromagnetic environment is gradually complicated, and new system radars are continuously available. The traditional radar intra-pulse modulation type identification method is low in identification rate and poor in effect. How to effectively identify the modulation type in the radar pulse becomes a key problem to be solved by radar reconnaissance signal processing. The high-precision identification of the radar intra-pulse modulation type is beneficial to obtaining the advantages on the electromagnetic battlefield of the own party.
In a published paper of Yewenqiang et al, "convolutional neural network-based radiation source signal identification algorithm" (computer simulation, 2019, 36(09):33-37.), a convolutional neural network-based radar intra-pulse modulation type classification and identification method is proposed. According to the method, time-frequency transformation is carried out on radar intra-pulse signals to obtain two-dimensional time-frequency images, a series of preprocessing is carried out on the time-frequency images, the processed images are input into a neural network model, a deep learning model is adjusted through pre-training, and finally extracted features are input into a classifier to complete an identification task. The method does not need to manually extract features, and the requirement of prior knowledge is low. The method has the disadvantages that a large amount of time is consumed for carrying out time-frequency transformation on the radar intra-pulse signals, the real-time performance is not strong, meanwhile, the parameters of the radar intra-pulse signals are fixed, and the identification effect is poor under the condition that the parameters have a large variation range.
In summary, under the current increasingly complex and variable electromagnetic environment, the existing radar intra-pulse modulation type identification method has poor identification effect and low identification rate, and is not beneficial to judgment of situations and adjustment of decisions.
Disclosure of Invention
The invention provides a method for identifying a radar intra-pulse modulation type, which combines a one-dimensional convolutional neural network with channel attention to realize the effect of high identification rate.
In order to achieve the technical effects, the invention is realized by the following technical scheme.
A method for identifying the type of radar intra-pulse modulation includes the following steps,
selection of modulation signal samples: generating a modulation signal data set in a radar pulse by using the acquired radar signals, wherein the modulation signal data set forms a modulation signal sample which at least comprises seven types of signals, and each type of signal is at least provided with 8 signal-to-noise ratio points;
and (3) dividing a training set and a test set: randomly selecting a training sample, a verification sample and a test sample at each signal-to-noise ratio point of each type of signal according to a preset proportion to respectively form a training set, a verification set and a test set;
constructing a neural network model: the neural network model is sequentially provided with an input layer, a convolution layer, a pooling layer, a normalization layer, a channel attention module, a full-connection layer and an output layer from the input layer to the output layer, wherein the channel attention module is provided with an addition layer, and the addition layer is provided with a Sigmoid activation function;
training and verifying a neural network model: setting training parameters and training rounds in a neural network model, inputting a training set into the neural network for training, and selecting the neural network model with the highest accuracy of a verification set as a final neural network model in a training result;
identification of modulation type: and inputting a test set in the final neural network model, and outputting the recognition rate of the radar tuning type.
In the technical scheme, the network for radar intra-pulse modulation identification is simple in structure, and due to the fact that a one-dimensional structure is adopted in the structure, the parameter scale is small, compared with a traditional two-dimensional convolution neural network which needs dimension transformation on radar signals, the data preprocessing of the method is simple, and the real-time performance of the network is excellent.
The invention has high identification accuracy: compared with a network without the channel attention module, the method has higher identification accuracy and better performance for 7 radar intra-pulse modulation types.
In the invention, through carrying out query investigation and classification summarization on related files, signals used by typical radar intra-pulse modulation types of 7 different modulation types are reasonably set, and parameters of the signals are reasonably set on the basis, so that the characteristic of complex and changeable electromagnetic environment at present is met.
As a further improvement of the present invention, in the selection of the modulation signal samples, the modulation signal data set of seven types of signals specifically includes: conventional pulse signals, chirp signals, non-chirp signals, bi-phase encoded signals, multi-phase encoded signals, two-frequency encoded signals, and four-frequency encoded signals.
In the invention, the 7 types of signals are common signals of the radar radiation source, so that the conventional network signals can be reflected, and the identification of the conventional number signals is realized.
As a further improvement of the invention, in the seven types of modulation signal data sets, the signal-to-noise ratio points of each type of signal are 11, and the number of samples of each signal at each signal-to-noise ratio point is 800-1200.
In the technical scheme, 8-10 signal-to-noise ratio points are selected, so that the condition of radar radiation source signals in various signal-to-noise ratio environments can be simulated as much as possible. And 10dB to-0 dB are selected and are separated by 1dB to 2dB, so that the operation can simulate the condition of the radar radiation source signal in the environment with the common signal-to-noise ratio as much as possible.
As a further improvement of the invention, the acquisition frequency of the radar signal is 1GHz, and the pulse width range of the radar signal is 2-10 us; the signal-to-noise ratio ranges from-14 dB to 0 dB.
In the technical scheme, the acquisition frequency of the radar signal is 1GHz, and the requirement of the Nyquist sampling rate is met; the pulse width range is 2-10us, and the radar radiation source signals with different pulse widths can be covered; the signal-to-noise ratio ranges from-14 dB to 0dB, and the condition of the radar radiation source signal in a low signal-to-noise ratio environment can be simulated as much as possible.
As a further improvement of the present invention, the predetermined ratio is training samples: and (3) verifying the sample: the test sample was 6:1: 3.
In the technical scheme, the proportion distribution is selected, so that the number of training samples can be ensured to be enough, the accuracy rate and the reliability during verification are high, and the actual test result can reflect the real situation.
As a further improvement of the present invention, the division of the training set and the test set further includes sample preprocessing, where the sample preprocessing: and performing sample pretreatment on the training sample, the verification sample and the test sample, wherein the sample pretreatment specifically comprises the steps of supplementing 0 at the tail to ensure that the lengths of all samples are the same, and performing discrete Fourier transform and normalization operation on the samples after 0 is supplemented at the tail in sequence to obtain a modulus result with the amplitude of 0-1.
In the technical scheme, the data are more standard through an operation method of 0 compensation at the end, discrete Fourier transform and the like, the influence of extreme values is removed, specifically, 0 compensation at the end is firstly carried out until the sample lengths are the same after all 0 compensation; performing discrete Fourier transform on the sample after 0 is supplemented, and taking a modulus value to obtain a modulus value result of the sample after the discrete Fourier transform; and (3) carrying out normalization operation on the modulus result of the sample after discrete Fourier transform, namely dividing each value in the modulus result of the sample after discrete Fourier transform by the maximum value in the modulus result of the sample after discrete Fourier transform, thereby obtaining the modulus normalization result of the sample after discrete Fourier transform, wherein the amplitude range is 0-1.
As a further improvement of the invention, in the building of the neural network model, the channel attention module comprises a global maximum pooling layer, a global average pooling layer, a perceptron, an addition layer, an activation layer and a multiplier layer.
In the technical scheme, the channel attention module can perform weight distribution on the features extracted by the convolutional layer, so that the weight of identifying important features is improved, and the weight of identifying unimportant features is reduced. Meanwhile, in the technical scheme, the one-dimensional convolutional layer is adopted, and compared with the traditional two-dimensional convolutional neural network, the convolutional neural network has the advantages of simple structure, small parameter scale and better real-time performance.
As a further improvement of the invention, the perceptron has weight sharing and hidden layers, the hidden layers comprise a first hidden layer close to the input layer and a second hidden layer close to the output layer, and the number of nodes of the first hidden layer is less than that of the second hidden layer.
In the technical scheme, the weight sharing aims to share the weight in the model after multiple times of training, so that later-stage comparison and selection are facilitated, the hidden layer aims to further extract feature information, and the number of nodes of the second hidden layer is the size matching of features.
As a further improvement of the invention, the mathematical model in the channel attention module is as follows:
wherein W represents the input FinC is the number of channels, FinPresentation input；
And
respectively represent FinThe results after the global maximum pooling layer and after the global average pooling layer,
and
to represent
And
processing by a multilayer perceptron with two layers sharing weights, sigma (·) representing a Sigmoid activation function, McTo represent
And
after the results of the multi-layer perceptron and the activation function sigma (-) of two layers of shared weight,
representing the multiplication of elements in the channel dimension, FoutIndicating the output result.
In the technical scheme, the mathematical model is selected, so that element multiplication is carried out after data passes through multiple steps of shared sensing and the like, and a result which is satisfactory and high in accuracy is output finally.
As a further improvement of the present invention, in the training and verification of the neural network model, training parameters including a learning rate, a loss function and a model optimization algorithm are set, and the learning rate is 0.001; setting a loss function as a cross entropy function; the model optimization algorithm selects the Adam function.
In the technical scheme, the learning rate is 0.001, and the learning rate can improve the convergence and reduce the divergence degree of the model in the training process; the loss function is set as a cross entropy function, so that the convergence speed of the model is ensured; the Adam algorithm is selected as the model optimization algorithm, and the convergence guarantee performance of the Adam algorithm on the model is stronger in the training process.
Drawings
Fig. 1 is a flowchart of a method for identifying a radar intra-pulse modulation type according to the present invention;
FIG. 2 is a flow chart in example 3 provided by the present invention;
fig. 3 is a structural diagram of a parallel prototype network provided by the present invention.
Detailed Description
The present invention is described in detail with reference to the embodiments shown in the drawings, but it should be understood that these embodiments are not intended to limit the present invention, and those skilled in the art should understand that functional, methodological, or structural equivalents or substitutions made by these embodiments are within the scope of the present invention.
In the description of the present embodiments, it is to be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of describing the present invention and simplifying the description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus are not to be construed as limiting the present invention
Furthermore, the terms "first," "second," "third," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicit to a number of indicated technical features. Thus, a feature defined as "first," "second," etc. may explicitly or implicitly include one or more of that feature. In the description of the invention, the meaning of "a plurality" is two or more unless otherwise specified. The terms "mounted," "connected," and "coupled" are to be construed broadly and may, for example, be fixedly coupled, detachably coupled, or integrally coupled; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meaning of the above terms in the creation of the present invention can be understood by those of ordinary skill in the art through specific situations.
Example 1
Referring to fig. 1, in the present embodiment, a method for identifying a radar intra-pulse modulation type includes the following steps,
selection of modulation signal samples: generating a modulation signal data set in a radar pulse by using the acquired radar signals, wherein the modulation signal data set forms a modulation signal sample which at least comprises seven types of signals, and each type of signal is at least provided with 8 signal-to-noise ratio points;
and (3) dividing a training set and a test set: randomly selecting a training sample, a verification sample and a test sample at each signal-to-noise ratio point of each type of signal according to a preset proportion to respectively form a training set, a verification set and a test set;
constructing a neural network model: the neural network model is sequentially provided with an input layer, a convolution layer, a pooling layer, a normalization layer, a channel attention module, a full-connection layer and an output layer from the input layer to the output layer, wherein the channel attention module is provided with an addition layer, and the addition layer is provided with a Sigmoid activation function;
training and verifying a neural network model: setting training parameters and training rounds in a neural network model, inputting a training set into the neural network for training, and selecting the neural network model with the highest accuracy of a verification set as a final neural network model in a training result;
identification of modulation type: and inputting a test set in the final neural network model, and outputting the recognition rate of the radar tuning type.
In the technical scheme, the network for radar intra-pulse modulation identification is simple in structure, and due to the fact that a one-dimensional structure is adopted in the structure, the parameter scale is small, compared with a traditional two-dimensional convolution neural network which needs dimension transformation on radar signals, the data preprocessing of the method is simple, and the real-time performance of the network is excellent.
The invention has high identification accuracy: compared with a network without the channel attention module, the method has higher identification accuracy and better performance for 7 radar intra-pulse modulation types.
In the invention, through carrying out query investigation and classification summarization on related files, signals used by typical radar intra-pulse modulation types of 7 different modulation types are reasonably set, and parameters of the signals are reasonably set on the basis, so that the characteristic of complex and changeable electromagnetic environment at present is met.
Example 2
In the present embodiment, the description is given in conjunction with the application.
Under the current electromagnetic environment, the system of the radar is continuously updated, the electronic environment is more intensive and complex, and the radar intra-pulse modulation signals are effectively identified with high precision, so that the radar intra-pulse modulation signals are a difficult problem in the current electronic warfare and an important subject. The passive radar system can be perfected, and the method has great research value for improving the performance of the active radar system. Many experts use two-dimensional convolutional neural networks to identify radar intra-pulse modulation signals. However, the structure of the conventional two-dimensional convolutional neural network is relatively complex, and dimension transformation is required to be performed on the acquired one-dimensional radar intra-pulse modulation signal. Meanwhile, the traditional two-dimensional convolutional neural network has poor identification effect on radar intra-pulse modulation and low identification accuracy. Aiming at the problems, the invention provides a radar intra-pulse modulation type identification method with high identification rate based on a one-dimensional convolutional neural network and channel attention.
The invention relates to a radar intra-pulse modulation type identification method based on a one-dimensional convolutional neural network and channel attention, which is shown in figure 1 and comprises the following steps:
1) collecting radar signals: collecting radar signals, and generating a radar intra-pulse modulation signal data set, wherein the data set signals comprise seven different modulation type signals, namely conventional pulse signals, linear frequency modulation signals, non-linear frequency modulation signals, two-phase coded signals, multi-phase coded signals, two-frequency coded signals and four-frequency coded signals, each signal is from-10 dB to-0 dB, the interval is 1dB, and the total number of signal-to-noise ratio points is 11. The number of samples of each signal at each signal-to-noise ratio point is 1000;
specifically, the acquisition frequency of the radar signal is 1GHz, and the requirement of the Nyquist sampling rate is met; the pulse width range is 2-10us, and the radar radiation source signals with different pulse widths can be covered; the signal-to-noise ratio ranges from-14 dB to 0dB, and the condition of the radar radiation source signal in a low signal-to-noise ratio environment can be simulated as much as possible.
2) Dividing a training set and a testing set: training samples are carried out on signals of a radar intra-pulse adjustment signal data set, and a verification sample and a test sample are divided, wherein the training samples and the test sample of each signal at each signal-to-noise ratio point are randomly selected, the proportion of the training samples to the verification sample is 6:1:3, a training set with the sample number of 33600, a verification set with the sample number of 5600, and a test set with the sample number of 16800 are obtained;
the proportion distribution of the verification samples and the test samples can ensure that the number of the training samples is enough, the accuracy rate during verification is high, and the actual test result can reflect the real situation.
3) Respectively preprocessing the training set, the verification set and the test set in the step 2): the data preprocessing comprises the steps of carrying out end 0 complementing on the samples of the training set, the verification set and the test set in the step 3) until the lengths of all the samples are identical after 0 complementing; performing discrete Fourier transform on the sample after 0 is supplemented, and taking a modulus value to obtain a modulus value result of the sample after the discrete Fourier transform; carrying out normalization operation on the modulus result of the sample after discrete Fourier transform, namely dividing each value in the modulus result of the sample after discrete Fourier transform by the maximum value in the modulus result of the sample after discrete Fourier transform, thereby obtaining the modulus normalization result of the sample after discrete Fourier transform, wherein the amplitude range is 0-1;
through an operation method of 0 compensation at the end, discrete Fourier transform and the like, data are enabled to be more standard, the influence of extreme values is removed, specifically, 0 compensation at the end is firstly carried out, and the sample length is the same after all 0 compensation are carried out; performing discrete Fourier transform on the sample after 0 is supplemented, and taking a modulus value to obtain a modulus value result of the sample after the discrete Fourier transform; and (3) carrying out normalization operation on the modulus result of the sample after discrete Fourier transform, namely dividing each value in the modulus result of the sample after discrete Fourier transform by the maximum value in the modulus result of the sample after discrete Fourier transform, thereby obtaining the modulus normalization result of the sample after discrete Fourier transform, wherein the amplitude range is 0-1.
4) Constructing a one-dimensional convolution neural network with a channel attention mechanism: the constructed one-dimensional convolutional neural network with the channel attention mechanism comprises four one-dimensional convolutional layers, four pooling layers, four normalization layers, four channel attention modules and a full-connection layer from an input layer to an output layer, wherein the one-dimensional convolutional layers, the pooling layers, the normalization layers and the channel attention modules are sequentially cascaded, and the full-connection layer is cascaded behind the last channel attention module; the activation function adopted by the output layer of the model is SoftMax, the activation function of the addition layer in the channel attention module is Sigmoid, and the activation functions adopted by the rest parts are ReLU;
5) training a one-dimensional convolutional neural network with a channel attention mechanism: setting a one-dimensional convolutional neural network training parameter with a channel attention mechanism, inputting the preprocessed data of a training set into the one-dimensional convolutional neural network with the channel attention mechanism for training to obtain a trained one-dimensional convolutional neural network with the channel attention mechanism;
5a) setting training parameters: the learning rate was set to 0.001; setting a loss function as a cross entropy function; selecting Adam by a model optimization algorithm;
5b) setting the number of training rounds; saving the model weight: after each round of training, the weight of the model is saved; after the set training round number is finished, selecting a model with the highest accuracy of the preprocessed verification set data from the stored model weights as a finally trained model, and loading the model into a network to obtain a trained one-dimensional convolution neural network with a channel attention mechanism;
6) and inputting the preprocessed data of the test set into a trained one-dimensional convolutional neural network with a channel attention mechanism, and outputting the integral identification rate of the radar intra-pulse modulation type.
The invention develops research aiming at the problems of complex network structure, poor recognition effect, low recognition rate and the like in the existing radar intra-pulse modulation type recognition technology. And the recognition rate of the network to the radar intra-pulse modulation type is improved by adding the attention of the channel.
The invention adopts the one-dimensional convolution neural network and the channel attention module, forms the whole technical scheme aiming at the identification of the radar intra-pulse modulation type and also obtains the technical effect of high identification rate aiming at the radar intra-pulse modulation type.
Example 2
In this embodiment, specific features in embodiment 1 are described.
Firstly, the introduction of the collected radar signals is specifically that the radar signals are of 7 different modulation types, and the corresponding parameters of each type of signals are set as follows:
the sampling frequency of the 7 radar signals with different modulation types is set to be 1GHz, and the pulse width range of the radar signals is 2-10 us;
the frequency range of the conventional pulse signal is 50-450 MHz;
the frequency range of the linear frequency modulation signal is 50-450MHz, wherein the bandwidth range is 10-400 MHz;
the nonlinear frequency modulation signal adopts cosine modulation, the frequency range is 50-450MHz, and the bandwidth range is 10-400 MHz;
the frequency range of the two-phase coded signal is 50-450MHz, and the coding mode adopts a 5, 7, 11, 13 bit Barker code;
the frequency range of the multiphase coding signals is 50-450MHz, and the coding mode adopts 36, 49 and 64 bit Frank codes;
the frequency ranges of the two-frequency coding signals are 50-450MHz and 50-450MHz respectively, wherein the interval between two frequency points is more than 50MHz, and the coding mode adopts 5, 7, 11 and 13-bit Barker codes;
the carrier frequency ranges of the four-frequency coding signals are respectively 50-450MHz, 50-450MHz and 50-450MHz, the interval of any two frequency points is more than 50MHz, and a 16-bit Frank code is adopted in the coding mode.
Specifically, through query investigation and classification summary of related files, signals used by typical radar intra-pulse modulation types of 7 different modulation types are reasonably set, and parameters of the signals are reasonably set on the basis, so that the characteristics of complexity and changeability of the current electromagnetic environment are met.
Secondly, the convolutional layer can extract the information of the data, and the pooling layer reduces the data size; the number of convolution kernels in the convolution layer increases with the number of layers, and data information can be extracted from multiple angles.
Referring to fig. 3, a specific network structure includes the following:
the first layer is an input layer, and the number of nodes is 10000;
the second layer is a one-dimensional convolution layer which contains 16 convolution kernels and the size of the convolution kernels is 9;
the third layer is a pooling layer with a pooling window of 6, a step length of 6 and a sampling maximum pooling mode;
the fourth layer is a batch normalization layer;
the fifth layer is a channel attention module which comprises a global maximum pooling layer, a global average pooling layer and a multilayer perceptron with shared weight and two hidden layers, wherein the number of nodes of the first hidden layer is 8, the number of nodes of the second hidden layer is 16, an addition layer, an activation layer and a multiplication layer;
the sixth layer is a one-dimensional convolution layer which contains 32 convolution kernels and the size of the convolution kernels is 9;
the seventh layer is a pooling layer with a pooling window of 6 and a step length of 6 and adopting a maximum pooling mode;
the eighth layer is a batch normalization layer;
the ninth layer is a channel attention module and comprises a global maximum pooling layer, a global average pooling layer and a multilayer perceptron with shared weight and two hidden layers, wherein the number of nodes of the first hidden layer is 16, the number of nodes of the second hidden layer is 32, the added layer, the activated layer and the multiplied layer are arranged in the ninth layer;
the tenth layer is a one-dimensional convolution layer which contains 64 convolution kernels and the size of the convolution kernels is 9;
the eleventh layer is a pooling layer with a pooling window of 6, a step length of 6 and a maximum pooling mode;
the twelfth layer is a batch normalization layer;
the tenth layer is a channel attention module and comprises a global maximum pooling layer, a global average pooling layer and a multilayer perceptron with shared weight and two hidden layers, wherein the number of nodes of the first hidden layer is 32, the number of nodes of the second hidden layer is 64, one addition layer, one activation layer and one multiplication layer;
the fourteenth layer is a one-dimensional convolutional layer which contains 128 convolutional kernels and has the convolutional kernel size of 9;
the fifteenth layer is a pooling layer with a pooling window of 6, a step length of 6 and a maximum pooling mode;
the sixteenth layer is a batch normalization layer;
the seventeenth layer is a channel attention module which comprises a global maximum pooling layer, a global average pooling layer and a multilayer perceptron with two hidden layers and shared weight, wherein the number of nodes of the first hidden layer is 64, the number of nodes of the second hidden layer is 128, the adding layer, the activating layer and the multiplying layer are arranged in sequence;
the eighteenth layer is a full connection layer, and the number of nodes is 128;
the nineteenth layer is an output layer, and the number of nodes is 7.
In the network, the second, sixth, tenth and fourteenth layers are all one-dimensional convolution layers; the third, seventh, eleventh and fifteenth layers are all pooling layers; the fourth, eighth, twelfth and sixteenth layers are normalization layers; the fifth layer, the ninth layer, the fourteenth layer and the seventeenth layer are all channel attention modules.
Compared with the traditional two-dimensional convolutional neural network, the network provided by the invention adopts a one-dimensional convolutional layer, and has the advantages of simple structure, small parameter scale and better real-time performance.
Thirdly, the algorithm of the channel attention module is described in detail, and the specific algorithm is expressed as follows:
wherein W represents the input FinC is the number of channels, FinRepresenting an input;
and
respectively represent FinThe results after the global maximum pooling layer and after the global average pooling layer,
and
to represent
And
processing by a multilayer perceptron with two layers sharing weights, sigma (·) representing a Sigmoid activation function, McTo represent
And
after the results of the multi-layer perceptron and the activation function sigma (-) of two layers of shared weight,
representing the multiplication of elements in the channel dimension, FoutIndicating the output result.
The channel attention module constructed by the method has the advantages of less parameters and low calculation required resource, and can improve the identification accuracy of the network for identifying the radar intra-pulse modulation type.
Example 3
In this embodiment, detailed steps are introduced in combination with specific practical cases.
Referring to fig. 1-3, the radar intra-pulse modulation type identification method based on the one-dimensional convolutional neural network and the channel attention of the invention comprises the following implementation steps:
step 1: a radar signal data set is generated.
In the example, MATLAB software is used for simulating and generating a radar intra-pulse modulation signal data set, wherein signals of the data set comprise seven different modulation type signals, namely a conventional pulse signal, a linear frequency modulation signal, a non-linear frequency modulation signal, a two-phase coding signal, a multi-phase coding signal, a two-frequency coding signal and a four-frequency coding signal, the signal-to-noise ratio of each signal ranges from-14 dB to 0dB, the interval is 2dB, and the signal-to-noise ratio is 8, so that the condition of various noise intensities can be met as far as possible. The number of samples per signal-to-noise ratio point is 1000 per signal.
The sampling frequency of the 7 radar signals with different modulation types is set to be 1GHz, and the pulse width range of the radar signals is 2-10 us;
the frequency range of the conventional pulse signal is 50-450 MHz;
the frequency range of the linear frequency modulation signal is 50-450MHz, wherein the bandwidth range is 10-400 MHz;
the nonlinear frequency modulation signal adopts cosine modulation, the frequency range is 50-450MHz, and the bandwidth range is 10-400 MHz;
the frequency range of the two-phase coded signal is 50-450MHz, and the coding mode adopts a 5, 7, 11, 13 bit Barker code;
the frequency range of the multiphase coding signals is 50-450MHz, and the coding mode adopts 36, 49 and 64 bit Frank codes;
the frequency ranges of the two-frequency coding signals are 50-450MHz and 50-450MHz respectively, wherein the interval between two frequency points is more than 50MHz, and the coding mode adopts 5, 7, 11 and 13-bit Barker codes;
the carrier frequency ranges of the four-frequency coding signals are respectively 50-450MHz, 50-450MHz and 50-450MHz, the interval of any two frequency points is more than 50MHz, and a 16-bit Frank code is adopted in a coding mode;
step 2: and (4) for the radar signal training sample, verifying the division of the sample and the test sample.
And (2) carrying out training sample division on the signals of the data set generated in the step (1), and dividing a verification sample and a test sample, wherein the training sample and the test sample of each signal at each signal-to-noise ratio point adopt a random selection mode, the training sample and the verification sample are in a ratio of 6:1:3, a training set with a sample number of 33600, a verification set with a sample number of 5600 and a test set with a sample number of 16800 are obtained.
And step 3: and (4) preprocessing data.
And (3) respectively preprocessing the training set, the verification set and the test set in the step (2): the data preprocessing comprises the steps of performing end 0 complementing on the samples of the training set, the verification set and the test set in the step 2 until the lengths of all the samples are the same after 0 complementing; performing discrete Fourier transform on the sample after 0 is supplemented, and taking a modulus value to obtain a modulus value result of the sample after the discrete Fourier transform; carrying out normalization operation on the modulus result of the sample after discrete Fourier transform, namely dividing each value in the modulus result of the sample after discrete Fourier transform by the maximum value in the modulus result of the sample after discrete Fourier transform, thereby obtaining the modulus normalization result of the sample after discrete Fourier transform, wherein the amplitude range is 0-1; the data after preprocessing has the extreme value removed, and is more beneficial to subsequent network identification.
And 4, step 4: and constructing a one-dimensional convolutional neural network with a channel attention mechanism.
Here, as in example 2, referring to fig. 3, the network includes an input layer, four one-dimensional convolutional layers, four pooling layers, four normalization layers, four channel attention modules, a full-link layer, and an output layer, and the specific structure thereof is as follows:
the first layer is an input layer, and the number of nodes is 10000;
the second layer is a one-dimensional convolution layer which contains 16 convolution kernels and the size of the convolution kernels is 9;
the third layer is a pooling layer with a pooling window of 6, a step length of 6 and a sampling maximum pooling mode;
the fourth layer is a batch normalization layer;
the fifth layer is a channel attention module which comprises a global maximum pooling layer, a global average pooling layer and a multilayer perceptron with shared weight and two hidden layers, wherein the number of nodes of the first hidden layer is 8, the number of nodes of the second hidden layer is 16, an addition layer, an activation layer and a multiplication layer;
the sixth layer is a one-dimensional convolution layer which contains 32 convolution kernels and the size of the convolution kernels is 9;
the seventh layer is a pooling layer with a pooling window of 6 and a step length of 6 and adopting a maximum pooling mode;
the eighth layer is a batch normalization layer;
the ninth layer is a channel attention module and comprises a global maximum pooling layer, a global average pooling layer and a multilayer perceptron with shared weight and two hidden layers, wherein the number of nodes of the first hidden layer is 16, the number of nodes of the second hidden layer is 32, the added layer, the activated layer and the multiplied layer are arranged in the ninth layer;
the tenth layer is a one-dimensional convolution layer which contains 64 convolution kernels and the size of the convolution kernels is 9;
the eleventh layer is a pooling layer with a pooling window of 6, a step length of 6 and a maximum pooling mode;
the twelfth layer is a batch normalization layer;
the tenth layer is a channel attention module and comprises a global maximum pooling layer, a global average pooling layer and a multilayer perceptron with shared weight and two hidden layers, wherein the number of nodes of the first hidden layer is 32, the number of nodes of the second hidden layer is 64, one addition layer, one activation layer and one multiplication layer;
the fourteenth layer is a one-dimensional convolutional layer which contains 128 convolutional kernels and has the convolutional kernel size of 9;
the fifteenth layer is a pooling layer with a pooling window of 6, a step length of 6 and a maximum pooling mode;
the sixteenth layer is a batch normalization layer;
the seventeenth layer is a channel attention module which comprises a global maximum pooling layer, a global average pooling layer and a multilayer perceptron with two hidden layers and shared weight, wherein the number of nodes of the first hidden layer is 64, the number of nodes of the second hidden layer is 128, the adding layer, the activating layer and the multiplying layer are arranged in sequence;
the eighteenth layer is a full connection layer, and the number of nodes is 128;
the sixteenth layer is an output layer, and the number of nodes is 7.
And 5: training a one-dimensional convolutional neural network with a channel attention mechanism, and inputting the data of a preprocessed training set into the one-dimensional convolutional neural network with the channel attention mechanism for training, wherein:
5a) the learning rate was set to 0.001; the loss function is set as a cross entropy function L (θ):
where θ represents the model weight, y is the one-hot encoding of the data,
representing the result of the input data computed via the network at the kth node of the output layer. The convergence speed of the model can be ensured by using the cross entropy function.
The model optimization algorithm selects Adam, and the specific algorithm is as follows:
m←β1m+(1-β1)g
v←β2v+(1-β2)g2
where θ represents a model weight of the network, L (θ) is a loss function, g is a gradient of L (θ),
representing a gradient operator, m being an estimate of the first moment of the gradient with an initial value of 0, v being an estimate of the second moment of the gradient with an initial value of 0, β1Is the exponential decay rate of the first moment estimation, and the value is 0.9, beta2Is the exponential decay rate of the second moment estimation, with the value of 0.999, alpha is the learning rate, and epsilon is the value of 10-8Is constant.
5b) Setting the number of training rounds to be 20, wherein the number of training rounds is generally 10-20, and the problem of insufficient model accuracy caused by insufficient number of training rounds can be avoided as much as possible by selecting 20 rounds; saving the model weight: after each round of training, the weight of the model is saved; after the set training round number is finished, selecting a model with the highest accuracy of the preprocessed verification set data from the stored model weights as a finally trained model, and loading the model into a network to obtain a trained one-dimensional convolution neural network with a channel attention mechanism;
step 6: and inputting the preprocessed data of the test set into a trained one-dimensional convolutional neural network with a channel attention mechanism, and outputting the integral identification rate of the radar intra-pulse modulation type.
The invention solves the problems that the dimension conversion processing is required to be carried out on the radar intra-pulse signals and the recognition rate is low in the prior art. The scheme is as follows: collecting radar intra-pulse signals and making a data set of the radar signals; dividing a data set of the radar signals into a training set, a verification set and a test set; carrying out data preprocessing on the training set, the verification set and the test set; constructing a one-dimensional convolution neural network with a channel attention mechanism; setting training parameters and training a network by using the preprocessed training set; selecting a model with the highest data accuracy of the preprocessed verification set as a finally trained model, and loading the model into a network to obtain a trained one-dimensional convolutional neural network with a channel attention mechanism; and inputting the preprocessed data of the test set into a trained one-dimensional convolutional neural network with a channel attention mechanism, and outputting the recognition rate of the whole test signal under different signal-to-noise ratios. The one-dimensional network structure used by the invention does not need dimension transformation on the radar intra-pulse modulation signal, has simple structure and less parameters, and simultaneously adopts the channel attention module to ensure that the network has higher recognition rate on the radar intra-pulse modulation type. Therefore, the method can be used for identifying the type of the radar intra-pulse modulation in the complex electromagnetic environment.
Comparative example 1
In the comparative examples, only the channel attention module was absent, and the other modules were identical to those of examples 1 to 3.
The predicted recognition rates of the examples and the comparative examples are compared, namely the predicted recognition rate test of the overall test signal of the one-dimensional convolutional neural network without the attention module under different signal-to-noise ratios is given at the same time, and the results are shown in table 1:
table 1: prediction recognition rate of integral test signal of one-dimensional convolutional neural network without channel attention module under different signal-to-noise ratios
Meanwhile, in the invention, the prediction recognition rate test of the whole test signal is also carried out under different signal-to-noise ratios, and the result is shown in Table 2,
table 2: the invention integrally tests the predictive recognition rate of the signal under different signal-to-noise ratios
As can be seen from tables 1-2, the average value of the test results without the channel attention module is obviously lower than that of the invention under different signal-to-noise ratios, that is, the recognition rate of the invention is higher; meanwhile, when the signal-to-noise ratio is more than or equal to-12 dB, the recognition rate of 7 radar intra-pulse modulation signals is more than 0.94; when the signal-to-noise ratio is-14 dB, the recognition rate of the invention to 7 radar intra-pulse modulation signals is still more than 0.85; the average recognition rate of the invention to 7 kinds of radar intra-pulse modulation signal can reach more than 0.97; compared with a method of a one-dimensional convolutional neural network without an attention module, the method has better recognition effect and higher recognition rate under the condition of low signal-to-noise ratio, particularly-14 dB and-12 dB.
In summary, the method for identifying the radar intra-pulse modulation type based on the one-dimensional convolutional neural network and the channel attention mainly solves the problems that dimension transformation processing needs to be carried out on radar signals and the identification rate is low in the prior art. The implementation scheme comprises the following steps: collecting radar intra-pulse signals and making a data set of the radar signals; dividing a data set of the radar signals into a training set, a verification set and a test set; carrying out data preprocessing on the training set, the verification set and the test set; constructing a one-dimensional convolution neural network with a channel attention mechanism; setting training parameters and training a network by using the preprocessed training set; selecting a model with the highest data accuracy of the preprocessed verification set as a finally trained model, and loading the model into a network to obtain a trained one-dimensional convolutional neural network with a channel attention mechanism; and inputting the preprocessed test set into a trained one-dimensional convolutional neural network with a channel attention mechanism, and outputting the recognition rate of the whole test signal under different signal-to-noise ratios. The method constructs the one-dimensional convolution neural network with the channel attention mechanism, has simple structure and less parameter quantity, saves the time for carrying out dimension transformation on the radar intra-pulse modulation signal, and has good real-time property. Meanwhile, the invention adopts the channel attention module, thereby improving the identification rate of the network to the radar intra-pulse modulation type. The method can be used for identifying the type of the radar intra-pulse modulation in the complex electromagnetic environment.
The above-listed detailed description is only a specific description of a possible embodiment of the present invention, and they are not intended to limit the scope of the present invention, and equivalent embodiments or modifications made without departing from the technical spirit of the present invention should be included in the scope of the present invention.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.
Claims (10)
1. A method for identifying the type of radar intra-pulse modulation comprises the following steps,
selection of modulation signal samples: generating a modulation signal data set in a radar pulse by using the acquired radar signals, wherein the modulation signal data set forms a modulation signal sample which at least comprises seven types of signals, and each type of signal is at least provided with 8 signal-to-noise ratio points;
and (3) dividing a training set and a test set: randomly selecting a training sample, a verification sample and a test sample at each signal-to-noise ratio point of each type of signal according to a preset proportion to respectively form a training set, a verification set and a test set;
constructing a neural network model: the neural network model is sequentially provided with an input layer, a convolution layer, a pooling layer, a normalization layer, a channel attention module, a full-connection layer and an output layer from the input layer to the output layer, wherein the channel attention module is provided with an addition layer, and the addition layer is provided with a Sigmoid activation function;
training and verifying a neural network model: setting training parameters and training rounds in a neural network model, inputting a training set into the neural network for training, and selecting the neural network model with the highest accuracy of a verification set as a final neural network model in a training result;
identification of modulation type: and inputting a test set in the final neural network model, and outputting the recognition rate of the radar tuning type.
2. The method according to claim 1, wherein in the selection of the modulation signal samples, the modulation signal data sets of seven types of signals are specifically: conventional pulse signals, chirp signals, non-chirp signals, bi-phase encoded signals, multi-phase encoded signals, two-frequency encoded signals, and four-frequency encoded signals.
3. The method as claimed in claim 2, wherein the number of the snr points of each of the seven types of the modulated signal data sets is 8-10, and the number of the samples of each signal at each snr point is 800-1200.
4. The method for identifying the type of radar intra-pulse modulation according to claim 1, wherein the radar signal is acquired at a frequency of 1GHz and has a pulse width in a range of 2-10 us; the signal-to-noise ratio ranges from-14 dB to 0 dB.
5. The method according to claim 1, wherein the predetermined ratio is training samples: and (3) verifying the sample: the test sample was 6:1: 3.
6. The method of claim 1, wherein the dividing of the training set and the test set further comprises sample preprocessing, wherein the sample preprocessing comprises: and performing sample pretreatment on the training sample, the verification sample and the test sample, wherein the sample pretreatment specifically comprises the steps of supplementing 0 at the tail to ensure that the lengths of all samples are the same, and performing discrete Fourier transform and normalization operation on the samples after 0 is supplemented at the tail in sequence to obtain a modulus result with the amplitude of 0-1.
7. The method according to claim 1, wherein in the building of the neural network model, the channel attention module comprises a global maximum pooling layer, a global average pooling layer, a perceptron, an addition layer, an activation layer and a multiplier layer.
8. The method of claim 7, wherein the perceptron has weight sharing and hidden layers, the hidden layers include a first hidden layer near the input layer and a second hidden layer near the output layer, and the number of nodes of the first hidden layer is less than the number of nodes of the second hidden layer.
9. The method of claim 8, wherein the mathematical model in the channel attention module is as follows:
wherein W represents the input FinC is the number of channels, FinRepresenting an input;
and
respectively represent FinThe results after the global maximum pooling layer and after the global average pooling layer,
and
to represent
And
processing by a multilayer perceptron with two layers sharing weights, sigma (-) representing Sigmoid activation function, McTo represent
And
after the results of the multi-layer perceptron and the activation function sigma (-) of two layers of shared weight,
representing the multiplication of elements in the channel dimension, FoutIndicating the output result.
10. The method for identifying the type of radar intra-pulse modulation according to any one of claims 1 to 9, wherein in the training and verification of the neural network model, training parameters are set, the training parameters include a learning rate, a loss function and a model optimization algorithm, the learning rate is 0.001, the loss function is set as a cross entropy function, and the model optimization algorithm selects an Adam algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111310285.5A CN114118142A (en) | 2021-11-05 | 2021-11-05 | Method for identifying radar intra-pulse modulation type |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111310285.5A CN114118142A (en) | 2021-11-05 | 2021-11-05 | Method for identifying radar intra-pulse modulation type |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114118142A true CN114118142A (en) | 2022-03-01 |
Family
ID=80381339
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111310285.5A Pending CN114118142A (en) | 2021-11-05 | 2021-11-05 | Method for identifying radar intra-pulse modulation type |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114118142A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114564982A (en) * | 2022-01-19 | 2022-05-31 | 中国电子科技集团公司第十研究所 | Automatic identification method for radar signal modulation type |
| CN115712867A (en) * | 2022-11-03 | 2023-02-24 | 哈尔滨工程大学 | Multi-component radar signal modulation identification method |
| CN117407785A (en) * | 2023-12-15 | 2024-01-16 | 西安晟昕科技股份有限公司 | Training method of radar signal recognition model, radar signal recognition method and device |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107220606A (en) * | 2017-05-22 | 2017-09-29 | 西安电子科技大学 | The recognition methods of radar emitter signal based on one-dimensional convolutional neural networks |
| CN112098957A (en) * | 2020-09-15 | 2020-12-18 | 西安电子科技大学 | Complex radar radiation source identification method based on one-dimensional self-walking convolution neural network |
| CN112577747A (en) * | 2020-12-07 | 2021-03-30 | 东南大学 | Rolling bearing fault diagnosis method based on space pooling network |
| US20210117737A1 (en) * | 2019-10-18 | 2021-04-22 | Korea University Research And Business Foundation | Earthquake event classification method using attention-based convolutional neural network, recording medium and device for performing the method |
| CN112731309A (en) * | 2021-01-06 | 2021-04-30 | 哈尔滨工程大学 | Active interference identification method based on bilinear efficient neural network |
| CN112881518A (en) * | 2021-01-08 | 2021-06-01 | 东冶及策河北能源技术有限公司 | Method for predicting residual life of dynamic filter compensator |
| CN113033452A (en) * | 2021-04-06 | 2021-06-25 | 合肥工业大学 | Lip language identification method fusing channel attention and selective feature fusion mechanism |
| CN113029327A (en) * | 2021-03-02 | 2021-06-25 | 招商局重庆公路工程检测中心有限公司 | Tunnel fan embedded foundation damage identification method based on metric attention convolutional neural network |
| CN113076878A (en) * | 2021-04-02 | 2021-07-06 | 郑州大学 | Physique identification method based on attention mechanism convolution network structure |
| CN113160246A (en) * | 2021-04-14 | 2021-07-23 | 中国科学院光电技术研究所 | Image semantic segmentation method based on depth supervision |
| CN113469196A (en) * | 2021-06-25 | 2021-10-01 | 南京航空航天大学 | Image classification method based on attention depth convolution neural classification network |
| CN113469198A (en) * | 2021-06-30 | 2021-10-01 | 南京航空航天大学 | Image classification method based on improved VGG convolutional neural network model |
| CN113591606A (en) * | 2021-07-08 | 2021-11-02 | 武汉理工大学 | Method and device for identifying hidden diseases of asphalt pavement, electronic equipment and storage medium |
-
2021
- 2021-11-05 CN CN202111310285.5A patent/CN114118142A/en active Pending
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107220606A (en) * | 2017-05-22 | 2017-09-29 | 西安电子科技大学 | The recognition methods of radar emitter signal based on one-dimensional convolutional neural networks |
| US20210117737A1 (en) * | 2019-10-18 | 2021-04-22 | Korea University Research And Business Foundation | Earthquake event classification method using attention-based convolutional neural network, recording medium and device for performing the method |
| CN112098957A (en) * | 2020-09-15 | 2020-12-18 | 西安电子科技大学 | Complex radar radiation source identification method based on one-dimensional self-walking convolution neural network |
| CN112577747A (en) * | 2020-12-07 | 2021-03-30 | 东南大学 | Rolling bearing fault diagnosis method based on space pooling network |
| CN112731309A (en) * | 2021-01-06 | 2021-04-30 | 哈尔滨工程大学 | Active interference identification method based on bilinear efficient neural network |
| CN112881518A (en) * | 2021-01-08 | 2021-06-01 | 东冶及策河北能源技术有限公司 | Method for predicting residual life of dynamic filter compensator |
| CN113029327A (en) * | 2021-03-02 | 2021-06-25 | 招商局重庆公路工程检测中心有限公司 | Tunnel fan embedded foundation damage identification method based on metric attention convolutional neural network |
| CN113076878A (en) * | 2021-04-02 | 2021-07-06 | 郑州大学 | Physique identification method based on attention mechanism convolution network structure |
| CN113033452A (en) * | 2021-04-06 | 2021-06-25 | 合肥工业大学 | Lip language identification method fusing channel attention and selective feature fusion mechanism |
| CN113160246A (en) * | 2021-04-14 | 2021-07-23 | 中国科学院光电技术研究所 | Image semantic segmentation method based on depth supervision |
| CN113469196A (en) * | 2021-06-25 | 2021-10-01 | 南京航空航天大学 | Image classification method based on attention depth convolution neural classification network |
| CN113469198A (en) * | 2021-06-30 | 2021-10-01 | 南京航空航天大学 | Image classification method based on improved VGG convolutional neural network model |
| CN113591606A (en) * | 2021-07-08 | 2021-11-02 | 武汉理工大学 | Method and device for identifying hidden diseases of asphalt pavement, electronic equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| SHIBO YUAN等: "Intra-Pulse Modulation Classification of Radar Emitter Signals Based on a 1-D Selective Kernel Convolutional Neural Network", 《REMOTE SENSING》 * |
| 皮骏等: "无人机目标分类的深度卷积网络设计与优化", 《计算机系统应用》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114564982A (en) * | 2022-01-19 | 2022-05-31 | 中国电子科技集团公司第十研究所 | Automatic identification method for radar signal modulation type |
| CN114564982B (en) * | 2022-01-19 | 2023-09-26 | 中国电子科技集团公司第十研究所 | Automatic identification method for radar signal modulation type |
| CN115712867A (en) * | 2022-11-03 | 2023-02-24 | 哈尔滨工程大学 | Multi-component radar signal modulation identification method |
| CN117407785A (en) * | 2023-12-15 | 2024-01-16 | 西安晟昕科技股份有限公司 | Training method of radar signal recognition model, radar signal recognition method and device |
| CN117407785B (en) * | 2023-12-15 | 2024-03-01 | 西安晟昕科技股份有限公司 | Training method of radar signal recognition model, radar signal recognition method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114118142A (en) | Method for identifying radar intra-pulse modulation type | |
| CN112098957A (en) | Complex radar radiation source identification method based on one-dimensional self-walking convolution neural network | |
| CN107220606B (en) | Radar radiation source signal identification method based on one-dimensional convolutional neural network | |
| CN110222748B (en) | OFDM radar signal identification method based on 1D-CNN multi-domain feature fusion | |
| CN110133599B (en) | Intelligent radar radiation source signal classification method based on long-time and short-time memory model | |
| CN109471074B (en) | Radar radiation source identification method based on singular value decomposition and one-dimensional CNN network | |
| CN111551992A (en) | Rock reservoir structure characterization method and device, computer-readable storage medium and electronic equipment | |
| DE2045508A1 (en) | Data processing method and device | |
| CN112115924A (en) | Radar radiation source identification method based on one-dimensional CNN and LSTM | |
| CN112882009B (en) | Radar micro Doppler target identification method based on amplitude and phase dual-channel network | |
| CN109116312A (en) | Radar signal classification method based on QMFB and convolutional neural networks | |
| CN111175718A (en) | Time-frequency domain combined ground radar automatic target identification method and system | |
| Ghadimi et al. | Deep learning-based approach for low probability of intercept radar signal detection and classification | |
| CN112149524B (en) | Radar signal sorting and identifying method, device, detector and storage medium | |
| CN112859014A (en) | Radar interference suppression method, device and medium based on radar signal sorting | |
| CN114895263A (en) | Radar active interference signal identification method based on deep migration learning | |
| CN111310700A (en) | Intermediate frequency sampling sequence processing method for radiation source fingerprint feature identification | |
| CN115565004A (en) | Raman spectrum analysis method based on two-dimensional Raman map combined with deep learning | |
| CN114021458B (en) | Small sample radar radiation source signal identification method based on parallel prototype network | |
| CN116797796A (en) | Signal identification method based on time-frequency analysis and deep learning under DRFM intermittent sampling | |
| CN116699526A (en) | Vehicle millimeter wave radar interference suppression method based on sparse and low-rank model | |
| CN112034434A (en) | Radar radiation source identification method based on sparse time-frequency detection convolutional neural network | |
| CN117671374A (en) | Method and device for identifying image target of inverse synthetic aperture radar | |
| CN117331031A (en) | LPI radar signal spectrogram fusion identification method | |
| Limin et al. | Low probability of intercept radar signal recognition based on the improved AlexNet model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220301 |
|
| RJ01 | Rejection of invention patent application after publication |