CN114048774A - Se-block-based resnet communication radiation source identification method and system - Google Patents
Se-block-based resnet communication radiation source identification method and system Download PDFInfo
- Publication number
- CN114048774A CN114048774A CN202111327199.5A CN202111327199A CN114048774A CN 114048774 A CN114048774 A CN 114048774A CN 202111327199 A CN202111327199 A CN 202111327199A CN 114048774 A CN114048774 A CN 114048774A
- Authority
- CN
- China
- Prior art keywords
- signal
- radiation source
- steady
- block
- communication radiation
- 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
- 230000005855 radiation Effects 0.000 title claims abstract description 78
- 238000004891 communication Methods 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000001052 transient effect Effects 0.000 claims abstract description 38
- 230000004913 activation Effects 0.000 claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims abstract description 19
- 238000000605 extraction Methods 0.000 claims abstract description 18
- 238000007781 pre-processing Methods 0.000 claims abstract description 15
- 238000005070 sampling Methods 0.000 claims description 20
- 238000007906 compression Methods 0.000 claims description 15
- 230000006835 compression Effects 0.000 claims description 15
- 230000005284 excitation Effects 0.000 claims description 14
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 12
- 230000002159 abnormal effect Effects 0.000 claims description 10
- 239000000284 extract Substances 0.000 claims description 10
- 230000007246 mechanism Effects 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 5
- 238000001514 detection method Methods 0.000 claims description 4
- 238000013144 data compression Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 abstract description 11
- 230000000694 effects Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 14
- 238000013528 artificial neural network Methods 0.000 description 9
- 230000008901 benefit Effects 0.000 description 4
- 238000011176 pooling Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
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
- G06F2218/12—Classification; Matching
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/045—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- 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
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Signal Processing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention provides a respet communication radiation source identification method and system based on se-block, which comprises the steps of preprocessing a training sample of a communication radiation source signal, modulating a lead code, then performing cross correlation with the communication radiation source, finding the position of the lead code, intercepting a transient state signal and a steady state signal, splicing important features of the two radiation sources, putting the spliced important features into the se-block, obtaining a weight matrix and an input product of a channel, putting the product into a residual error network, continuously extracting the features, and finally classifying the communication radiation sources through an activation function and a full connection layer. The method is convenient for extracting effective radiation source characteristics, realizes characteristic extraction by combining se-block with strong classification and a residual module, and obtains good classification effect.
Description
Technical Field
The invention relates to the technical field of communication radiation source individual identification, in particular to a respet communication radiation source identification method and system based on se-block.
Background
With the rapid development of wireless communication, the types of radiation sources are various and complex, signals present diversity, and the traditional radiation source identification method may not meet the social requirements. In order to obtain information about a communication entity, researchers have proposed different identification methods, which have different effects. The most critical task for communication radiation source identification is to extract valid features. If the characteristics are extracted manually, the problems of time consumption, high difficulty, high complexity and the like can be encountered in the extraction process due to the large signal data volume. If the signal essential characteristics with strong discriminative power can be extracted, the method has important significance for the design and the identification performance improvement of a subsequent classifier.
In the field of individual identification of communication radiation sources, the traditional processing mode has deviated from the era, and the traditional high-order spectrum, which typically represents bispectral transformation, and the classical short-time fourier transformation, yellow hilbert transformation, wavelet transformation, and further waveform entropy, bispectral entropy are processed by traditional classifiers such as: and (4) carrying out recognition classification by a Support Vector Machine (SVM). The calculated amount far exceeds that of the signal, the accuracy is low, and the universality is not realized.
Disclosure of Invention
The invention provides a respet communication radiation source identification method and system based on se-block, and aims to overcome the defects in the prior art.
In one aspect, the invention provides a respet communication radiation source identification method based on se-block, which comprises the following steps:
s1: knowing that a preamble of a communication radiation source signal is a certain fixed sequence, finding out a position of a sequence which is most similar to the fixed sequence in the communication radiation source signal as a preamble position, extracting a signal with a length of L1 before the preamble position as a transient signal, and extracting a signal with a length of L2 after the preamble position as a steady-state signal;
s2: sequentially extracting the real part and the imaginary part of the transient signal and normalizing the real part and the imaginary part so as to eliminate amplitude difference caused by noise and errors of individual abnormal points and finally obtain a preprocessed transient signal;
s3: down-sampling the steady-state signal so as to keep the characteristics of the steady-state signal, sequentially extracting the real part and the imaginary part of the steady-state signal after down-sampling and normalizing the real part and the imaginary part, thereby eliminating the amplitude difference caused by the noise and the error of individual abnormal points and finally obtaining the steady-state signal after preprocessing;
s4: splicing the preprocessed transient signal and the preprocessed steady-state signal end to end, thereby combining the signals into a signal and inputting the signal serving as an effective characteristic into a se-block;
s5: and updating and iterating the channel weight of the effective characteristic based on the se-block and a channel attention mechanism according to the importance degree of the channel to obtain a channel weight matrix, weighting the channel weight matrix to the effective characteristic through multiplication, inputting the weighted result into a residual error network, and finally classifying the communication radiation source signal by sequentially passing the output result of the residual error network through an activation layer and a full connection layer.
The method adopts the advantage of se-block, a sub-network of the se-block comprises a module of the se-block, effective characteristics are better extracted by a channel attention mechanism according to corresponding parameters of the number of channels allocated according to priority, and finally, residual error networks are added, each network comprises two residual error networks, functions are activated through pooling, and finally, full connection classification is realized. The invention creates a novel method for combining a se-block residual block with a resnet network, and effectively classifies communication radiation sources.
In a specific embodiment, the finding a position of a sequence most similar to the fixed sequence in the communication radiation source signal as a preamble position specifically includes:
modulating the fixed sequence to be consistent with the form of the communication radiation source signal;
then, performing a cross-correlation algorithm on the modulated fixed sequence and the communication radiation source signal;
and finding out the position of the peak value highest point in the result obtained by the cross-correlation algorithm as the starting position of the lead code in the communication radiation source signal, thereby determining the position of the lead code.
In a specific embodiment, L1 is equal to or greater than 100.
In a specific embodiment, L2 is equal to or greater than 4000.
In a specific embodiment, the down-sampling the steady-state signal specifically includes: and performing down-sampling on the steady-state signal at intervals of a plurality of points, wherein the selection of the plurality of points comprises ensuring that the length of a signal generated after the steady-state signal passes the down-sampling is consistent with the length of the transient signal.
In a specific embodiment, the se-block structure comprises a compression stage and an excitation stage:
the compression stage comprises:
wherein, XcThe value distribution of C characteristic graphs, namely global information, H and W are the length and width of the input, ucDenotes the C-th feature in u, C denotes channel, FsqDenoted as compression operation;
and performing data compression and information summarization on the Embedding vector of each feature in the compression stage as follows:
where k is the dimension of the input feature, viExpressed as input features, ziExpressed as the input features after compression, i is expressed as the serial number of the features;
the excitation phase comprises:
s＝Fex(z,W)＝σ(g(z,W))＝σ)W2δ(W1z))
where δ represents the activation function (Relu), σ represents the activation function (Sigmoid), W1And W2Expressed as fully connected layers, z is the compressed feature of the input, FexDenoted as actuation operation;
in the excitation stage, important features in the effective features are dynamically strengthened and unimportant features in the effective features are weakened by constructing an operation mode, wherein the unimportant features comprise noise features, and the operation mode is shown as the following formula:
S＝Fex(z,W)＝δ(W2δ(W1z))
where δ represents the activation function (Relu), FexDenoted as excitation operation, W1And W2Denoted as fully connected layers, and z represents the compressed features of the input.
In a specific embodiment, the updating and iteration based on the channel attention mechanism is performed on the channel weights of the effective features based on the se-block and according to the importance degree of the channel to obtain a channel weight matrix, and the specific steps include:
adding the dimension of the effective feature to two dimensions to obtain a two-dimensional effective feature;
compressing and information summarizing the Embedding vector of each feature in the two-dimensional effective features, and exciting through two full-connection layers;
and scaling the result obtained by the excitation so as to reduce the number of channels and obtain the channel weight matrix. A sub-network (se-block) is introduced by using a se-block module, channel weights are updated and iterated through a channel attention mechanism, after a group of threshold values are obtained, after cross-layer identical paths are carried out on data behind a first layer, residual paths are shrunk, and the process is a good process for deleting redundant information.
In a specific embodiment, the residual network comprises 6 residual blocks with identity mapping. And sub-modules of the residual error network with the identity mapping are superposed to deepen the network and avoid gradient disappearance.
According to a second aspect of the present invention, a computer-readable storage medium is proposed, on which a computer program is stored, which computer program, when being executed by a computer processor, carries out the above-mentioned method.
According to a third aspect of the present invention, a respet communication radiation source identification system based on se-block is provided, the system comprising:
the signal detection and extraction module: configuring a preamble of a known communication radiation source signal as a certain fixed sequence, finding out a position of a sequence most similar to the fixed sequence in the communication radiation source signal as a preamble position, extracting a signal with a length of L1 before the preamble position as a transient signal, and extracting a signal with a length of L2 after the preamble position as a steady-state signal;
the transient signal preprocessing module: the method is configured to sequentially extract a real part and an imaginary part of the transient signal and normalize the real part and the imaginary part, so as to eliminate amplitude difference caused by noise and errors of individual abnormal points and finally obtain a preprocessed transient signal;
the steady-state signal preprocessing module: the configuration is used for carrying out down-sampling on the steady-state signal so as to keep the characteristics of the steady-state signal, then extracting the real part and the imaginary part of the steady-state signal after down-sampling in sequence and normalizing the real part and the imaginary part, thereby eliminating the amplitude difference caused by the noise and the error of individual abnormal points and finally obtaining the preprocessed steady-state signal;
a feature extraction module: the system is configured to splice the preprocessed transient signal and the preprocessed steady-state signal end to end, so as to merge the signals into a signal and input the signal into se-block as an effective characteristic;
the communication radiation source identification and classification module comprises: and the channel weighting unit is configured and used for updating and iterating the channel weights of the effective characteristics based on the se-block and according to the importance degree of the channel to obtain a channel weight matrix, weighting the channel weight matrix to the effective characteristics through multiplication, inputting the weighted result into a residual error network, and finally classifying the communication radiation source signals through an activation layer and a full connection layer in sequence according to the output result of the residual error network.
The invention adopts the advantages of se-block modules, better extracts effective characteristics by allocating corresponding parameters to the number of channels according to priority through a channel attention mechanism, and finally adds residual error networks, wherein each network comprises 2 residual error modules, and full connection classification is finally realized through pooling and activating functions. The invention creatively adopts a se-block module and then combines with a residual block, and effectively classifies the communication radiation source signals.
Drawings
The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments and together with the description serve to explain the principles of the invention. Other embodiments and many of the intended advantages of embodiments will be readily appreciated as they become better understood by reference to the following detailed description. Other features, objects and advantages of the present application will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, made with reference to the accompanying drawings in which:
FIG. 1 is a flow chart of a se-block based resnet communication radiation source identification method according to an embodiment of the present invention;
fig. 2 is a diagram of peaks of a data-lookup preamble for a specific embodiment of the present invention;
FIG. 3 is a waveform diagram of a transient signal intercept of a specific embodiment of the present invention;
FIG. 4 is a steady state signal truncated waveform diagram of a specific embodiment of the present invention;
FIG. 5 is a diagram of a communication radiation source information frame format according to a specific embodiment of the present invention;
FIG. 6 is a flow diagram of a se-block of a specific embodiment of the present invention;
FIG. 7 is a block diagram of a feature extraction neural network according to an embodiment of the present invention;
FIG. 8 is a graph of the accuracy of the inventive scheme, resnet and lstm _ fcn and acceptance without se-block;
FIG. 9 is a block diagram of a communication radiation source identification system based on se-block resnet according to an embodiment of the present invention.
Detailed Description
The present application will be described in further detail with reference to the following drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the relevant invention and not restrictive of the invention. It should be noted that, for convenience of description, only the portions related to the related invention are shown in the drawings.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
Fig. 1 shows a flowchart of a resnet communication radiation source identification method based on se-block according to an embodiment of the present invention. As shown in fig. 1, the method comprises the steps of:
s1: knowing that a preamble of a communication radiation source signal is a certain fixed sequence, finding out a position of a sequence which is most similar to the fixed sequence in the communication radiation source signal as a preamble position, extracting a signal with a length of L1 before the preamble position as a transient signal, and extracting a signal with a length of L2 after the preamble position as a steady-state signal;
s2: sequentially extracting the real part and the imaginary part of the transient signal and normalizing the real part and the imaginary part so as to eliminate amplitude difference caused by noise and errors of individual abnormal points and finally obtain a preprocessed transient signal;
s3: down-sampling the steady-state signal so as to keep the characteristics of the steady-state signal, sequentially extracting the real part and the imaginary part of the steady-state signal after down-sampling and normalizing the real part and the imaginary part, thereby eliminating the amplitude difference caused by the noise and the error of individual abnormal points and finally obtaining the steady-state signal after preprocessing;
s4: splicing the preprocessed transient signal and the preprocessed steady-state signal end to end, thereby combining the signals into a signal and inputting the signal serving as an effective characteristic into a se-block;
s5: and updating and iterating the channel weight of the effective characteristic based on the se-block and a channel attention mechanism according to the importance degree of the channel to obtain a channel weight matrix, weighting the channel weight matrix to the effective characteristic through multiplication, inputting the weighted result into a residual error network, and finally classifying the communication radiation source signal by sequentially passing the output result of the residual error network through an activation layer and a full connection layer.
The detailed flow of the scheme shown in FIG. 1 is illustrated below by the specific example:
step 1: and after preprocessing the communication radiation source with the modulation rate of 120KBaud/s, the sampling rate of 2.4Mps and the modulation mode of BPSK, intercepting the training sample and splicing the transient signal and the steady signal to be used as the effective characteristic of the characteristic network to be input.
In a particular embodiment, the communication radiation source is to meet the IEEE802.11 standard. IEEE802.11 is a standard common to wireless local area networks today, and is a standard for wireless network communications defined by the Institute of Electrical and Electronics Engineers (IEEE).
Fig. 2 shows a process of finding a preamble, and a point with the highest similarity to the preamble is found by using a cross-correlation algorithm, which is an initial point of the preamble. In this embodiment, the radiation source signal is first read in, and then several radiation source signals of 12000 length are cross-correlated with the preamble as a fixed value code.
Fig. 3 shows a waveform diagram of a transient signal truncation according to an embodiment of the present invention, fig. 4 shows a waveform diagram of a steady-state signal truncation before sampling according to an embodiment of the present invention, and fig. 5 shows a frame format diagram of a communication radiation source information frame according to an embodiment of the present invention, wherein the communication signal comprises 200 bits of a preamble, and is stored in an interleaving manner in the form of int16, and a baseband IQ signal.
At the position where the preamble 0101 … … 0011 is found in the sequence of 12000, 200 points are sequentially intercepted forwards as an initial oscillation signal, and 4000 points are intercepted backwards as a steady-state signal.
Step two: updating the weight of the channel of the effective characteristics by using se-block, inputting the result passing through the channel into a next-layer resnet network to realize further characteristic extraction, further extracting the characteristics by using a1 × 1 convolution layer in a residual block, and classifying communication radiation source signals through an activation function and a full connection layer after passing through 6 residual blocks;
the flow chart of the se-block module proposed by the present invention and the structure of the neural network used will be explained below by using fig. 6 and fig. 7, respectively.
FIG. 6 shows a flow diagram of se-block, which includes a compression phase and an excitation phase:
the compression stage comprises:
wherein Z iscThe value distribution of C characteristic graphs, namely global information, H and W are the length and width of the input, ucDenotes the C-th feature in u, C denotes channel, FsqDenoted as compression operation;
and performing data compression and information summarization on the Embedding vector of each feature in the compression stage as follows:
where k is the dimension of the input feature, viExpressed as input features, ziExpressed as the input features after compression, i is expressed as the serial number of the features;
the excitation phase comprises:
s＝Fex(z,W)＝σ(g(z,W))＝σ(W2δ(W1z))
where δ represents the activation function (Relu), σ represents the activation function (Sigmoid), W1And W2Expressed as fully connected layers, z is the compressed feature of the input, FexDenoted as actuation operation;
in the excitation stage, important features in the effective features are dynamically strengthened and unimportant features in the effective features are weakened by constructing an operation mode, wherein the unimportant features comprise noise features, and the operation mode is shown as the following formula:
S＝Fex(z,W)＝δ(W2δ(W1z))
where δ represents the activation function (Relu), FexDenoted as excitation operation, W1And W2Denoted as fully connected layers, and z represents the compressed features of the input.
Channel weight a obtained by channel attention mechanismiAnd then multiplying the characteristic back to the Embedding corresponding to the characteristic, thereby realizing the weighting operation of the characteristic. Wherein the channel weight aiRepresenting the importance degree of the ith channel obtained by judging the effective characteristics through a se-block module.
As shown in fig. 6, the deep neural network established by weighting each feature channel according to the importance degree specifically includes the following operations performed on the input signal X in sequence: extracting features by using the residual block; compressing the data to 1 × 1 × C by using global average pooling, wherein C is the number of channels, changing the data to 1 × 1 × C × SERadio through the full link layer FC, wherein the SERadio is a contraction parameter, reducing the data to 1 × 1 × C through another full link layer FC after the data passes through the activation function, outputting 1 × 1 × C through a sigmod activation function, and finally completing the fusion among the channels as shown by a Scale block in the figure.
In a specific embodiment, the residual block includes three convolution kernels with a length of 3, a step of 1, and a fill of 1.
In a specific embodiment, the residual block includes parallel 1 × 1 convolutional layers, which facilitates better extraction of the original features of the signal.
Fig. 7 shows an architecture diagram of a feature extraction network according to an embodiment of the present invention, which includes transient signal and steady-state signal extraction, and is implemented by passing through a se-block, which includes a channel attention mechanism, and then passing through three residual networks, each of which includes two residual blocks, and then passing through global pooling, an activation function, and finally full-link classification.
In a specific embodiment, each of the 6 residual blocks adopts two convolution kernels, the length of the convolution kernel is 3, the step is 1, the padding is 1, the number, the length and the width of input channels obtained through the convolution are still 1 × 1 × 400, and the input channels are ensured to be consistent with the input length.
Here, one more 1 × 1 convolutional layer is added, followed by two batch normalization layers, to solve the challenge of deep model training. During model training, the intermediate output of the neural network is continuously adjusted by batch normalization by using the mean value and the standard deviation on a small batch, so that the output of the previous layer is normalized, and the numerical value of the intermediate output of the whole neural network on each layer is more stable. And finally accumulating the result output by the residual block and the result directly subjected to the 1 multiplied by 1 convolution layer, so that each local feature of the radiation source feature can be well extracted.
In a specific embodiment, the specificity of a transient signal and the fingerprint specificity of hardware are grasped, the steady-state characteristic is added, the radiation source characteristics are gathered together, the deep characteristics are extracted through a neural network with extremely strong classification capability, and finally classification is carried out, so that the accuracy is greatly improved compared with that of the traditional method.
In this example, the data sets are from a blind signal processing focus laboratory. The training environment and the parameter settings in the training process of the feature extraction network are as follows:
1. training environment
A programming environment: python
A rear end: pythrch 1.8.0 cuda11.1 and TensorFlow 1.18.0
GPU model: NVIDIA GeForce RTX 2060 Ti
2. Parameter setting
Batch size: 32
Loss function: categorical _ cross _ sense
An optimizer: adam optimizer
The embodiment compares the performance of the results through different neural networks by using 300 training samples in each class and 15 training samples in each class.
Fig. 8 is a diagram illustrating a comparison of accuracy of identification of a communication radiation source according to a specific embodiment of the present invention, which uses three methods, lstm _ fcn and inclusion, to compare with a respet residual error network based on se-block disclosed in the present invention. It can be known that the recognition accuracy of all methods based on the neural network is higher than that of the traditional recognition methods such as SVM and MLP, and the neural network can extract more remarkable communication radiation source characteristics. The method described herein has the highest accuracy, and still has a high accuracy in the case of 15 sample classes, which can reach more than 95%.
Fig. 9 shows a block diagram of a resnet communication radiation source identification system based on se-block according to an embodiment of the present invention. The system comprises a signal detection and extraction module 901, a transient signal preprocessing module 902, a steady-state signal preprocessing module 903, a feature extraction module 904 and a communication radiation source identification and classification module 905.
In a specific embodiment, the signal detection and extraction module 901 is configured to know that a preamble of a communication radiation source signal is a certain fixed sequence, find a position of a sequence in the communication radiation source signal that is most similar to the fixed sequence as a preamble position, extract a signal with a length of L1 before the preamble position as a transient signal, and extract a signal with a length of L2 after the preamble position as a steady-state signal;
the transient signal preprocessing module 902 is configured to sequentially extract and normalize the real part and the imaginary part of the transient signal, so as to eliminate amplitude differences caused by noise and errors of individual outliers, and finally obtain a preprocessed transient signal;
the steady-state signal preprocessing module 903 is configured to perform down-sampling on the steady-state signal so as to retain the characteristics of the steady-state signal, sequentially extract the real part and the imaginary part of the steady-state signal after down-sampling, and normalize the real part and the imaginary part, so as to eliminate amplitude differences caused by errors of noise and individual outliers, and finally obtain a preprocessed steady-state signal;
the feature extraction module 904 is configured to splice the preprocessed transient signal and the preprocessed steady-state signal end-to-end, thereby merging into a signal and inputting the signal into a se-block as a valid feature;
the communication radiation source identification and classification module 905 is configured to update and iterate channel weights of the effective features based on the se-block and according to channel importance degrees to obtain a channel weight matrix, weight the channel weight matrix to the effective features through multiplication, input weighted results into a residual error network, and finally classify the communication radiation source signals through an activation layer and a full connection layer in sequence according to output results of the residual error network.
The system provides a respet communication radiation source identification method and system of se-block, and the method comprises the steps of preprocessing a training sample of a communication radiation source signal, carrying out BPSK modulation on a lead code, carrying out cross correlation on the lead code and the communication radiation source, finding out the position of the lead code, intercepting a transient state signal and a steady state signal, splicing important characteristics of the two radiation sources, putting the spliced important characteristics into the se-block, obtaining a weight matrix and an input product of a channel, putting the product into a residual error network to continuously extract the characteristics, and finally classifying the communication radiation source through an activation function and a full connection layer. The method is convenient for extracting effective radiation source characteristics, realizes characteristic extraction by combining se-block with strong classification and a residual module, and obtains good classification effect.
Embodiments of the present invention also relate to a computer-readable storage medium having stored thereon a computer program which, when executed by a computer processor, implements the method above. The computer program comprises program code for performing the method illustrated in the flow chart. It should be noted that the computer readable medium of the present application can be a computer readable signal medium or a computer readable medium or any combination of the two.
The invention provides a respet communication radiation source identification method and system of se-block, which comprises the steps of preprocessing a training sample of a communication radiation source signal, carrying out BPSK modulation on a lead code, carrying out cross correlation on the lead code and the communication radiation source, finding out the position of the lead code, intercepting a transient state signal and a steady state signal, splicing important characteristics of the two radiation sources, putting the spliced important characteristics into the se-block, obtaining a weight matrix and an input product of a channel, putting the product into a residual error network to continuously extract the characteristics, and finally classifying the communication radiation source through an activation function and a full connection layer. The method is convenient for extracting effective radiation source characteristics, realizes characteristic extraction by combining se-block with strong classification and a residual module, and obtains good classification effect.
The above description is only a preferred embodiment of the application and is illustrative of the principles of the technology employed. It will be appreciated by those skilled in the art that the scope of the invention herein disclosed is not limited to the particular combination of features described above, but also encompasses other arrangements formed by any combination of the above features or their equivalents without departing from the spirit of the invention. For example, the above features may be replaced with (but not limited to) features having similar functions disclosed in the present application.
Claims (10)
1. A respet communication radiation source identification method based on se-block is characterized by comprising the following steps:
s1: knowing that a preamble of a communication radiation source signal is a certain fixed sequence, finding out a position of a sequence which is most similar to the fixed sequence in the communication radiation source signal as a preamble position, extracting a signal with a length of L1 before the preamble position as a transient signal, and extracting a signal with a length of L2 after the preamble position as a steady-state signal;
s2: sequentially extracting the real part and the imaginary part of the transient signal and normalizing the real part and the imaginary part so as to eliminate amplitude difference caused by noise and errors of individual abnormal points and finally obtain a preprocessed transient signal;
s3: down-sampling the steady-state signal so as to keep the characteristics of the steady-state signal, sequentially extracting the real part and the imaginary part of the steady-state signal after down-sampling and normalizing the real part and the imaginary part, thereby eliminating the amplitude difference caused by the noise and the error of individual abnormal points and finally obtaining the steady-state signal after preprocessing;
s4: splicing the preprocessed transient signal and the preprocessed steady-state signal end to end, thereby combining the signals into a signal and inputting the signal serving as an effective characteristic into a se-block;
s5: and updating and iterating the channel weight of the effective characteristic based on the se-block and a channel attention mechanism according to the importance degree of the channel to obtain a channel weight matrix, weighting the channel weight matrix to the effective characteristic through multiplication, inputting the weighted result into a residual error network, and finally classifying the communication radiation source signal by sequentially passing the output result of the residual error network through an activation layer and a full connection layer.
2. The method according to claim 1, wherein the step of finding the position of the sequence most similar to the fixed sequence in the communication radiation source signal as a preamble position comprises:
modulating the fixed sequence to be consistent with the form of the communication radiation source signal;
then, performing a cross-correlation algorithm on the modulated fixed sequence and the communication radiation source signal;
and finding out the position of the peak value highest point in the result obtained by the cross-correlation algorithm as the starting position of the lead code in the communication radiation source signal, thereby determining the position of the lead code.
3. The method of claim 1, wherein L1 is equal to or greater than 100.
4. The method of claim 1, wherein L2 is equal to or greater than 4000.
5. The method according to claim 1, wherein the down-sampling the steady-state signal specifically comprises: and performing down-sampling on the steady-state signal at intervals of a plurality of points, wherein the selection of the plurality of points comprises ensuring that the length of a signal generated after the steady-state signal passes the down-sampling is consistent with the length of the transient signal.
6. The method of claim 1, wherein the se-block structure comprises a compression stage and an excitation stage:
the compression stage comprises:
wherein Z iscThe value distribution of C characteristic graphs, namely global information, H and W are the length and width of the input, ucDenotes the C-th feature in u, C denotes channel, FsqDenoted as compression operation;
and performing data compression and information summarization on the Embedding vector of each feature in the compression stage as follows:
where k is the dimension of the input feature, viExpressed as input features, ziExpressed as the input features after compression, i is expressed as the serial number of the features;
the excitation phase comprises:
s＝Fex(z，W)＝σ(g(z，W))＝σ(W2δ(W1z))
where δ represents the activation function (Relu), σ represents the activation function (Sigmoid), W1And W2Expressed as fully connected layers, z is the compressed feature of the input, FexDenoted as actuation operation;
in the excitation stage, important features in the effective features are dynamically strengthened and unimportant features in the effective features are weakened by constructing an operation mode, wherein the unimportant features comprise noise features, and the operation mode is shown as the following formula:
S＝Fex(z，W)＝δ(W2δ(W1z))
where δ represents the activation function (Relu), FexDenoted as excitation operation, W1And W2Denoted as fully connected layers, and z represents the compressed features of the input.
7. The method according to claim 1, wherein the channel weights of the valid features are updated and iterated based on a channel attention mechanism based on the se-block and according to the importance degree of the channel to obtain a channel weight matrix, and the specific steps include:
adding the dimension of the effective feature to two dimensions to obtain a two-dimensional effective feature;
compressing and information summarizing the Embedding vector of each feature in the two-dimensional effective features, and exciting through two full-connection layers;
and scaling the result obtained by the excitation so as to reduce the number of channels and obtain the channel weight matrix.
8. The method of claim 1, wherein the residual network comprises 6 residual blocks with identity mapping.
9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a computer processor, carries out the method of any one of claims 1 to 8.
10. A se-block based resnet communication radiation source identification system, comprising:
the signal detection and extraction module: configuring a preamble of a known communication radiation source signal as a certain fixed sequence, finding out a position of a sequence most similar to the fixed sequence in the communication radiation source signal as a preamble position, extracting a signal with a length of L1 before the preamble position as a transient signal, and extracting a signal with a length of L2 after the preamble position as a steady-state signal;
the transient signal preprocessing module: the method is configured to sequentially extract a real part and an imaginary part of the transient signal and normalize the real part and the imaginary part, so as to eliminate amplitude difference caused by noise and errors of individual abnormal points and finally obtain a preprocessed transient signal;
the steady-state signal preprocessing module: the configuration is used for carrying out down-sampling on the steady-state signal so as to keep the characteristics of the steady-state signal, then extracting the real part and the imaginary part of the steady-state signal after down-sampling in sequence and normalizing the real part and the imaginary part, thereby eliminating the amplitude difference caused by the noise and the error of individual abnormal points and finally obtaining the preprocessed steady-state signal;
a feature extraction module: the system is configured to splice the preprocessed transient signal and the preprocessed steady-state signal end to end, so as to merge the signals into a signal and input the signal into se-block as an effective characteristic;
the communication radiation source identification and classification module comprises: and the channel weighting unit is configured and used for updating and iterating the channel weights of the effective characteristics based on the se-block and according to the importance degree of the channel to obtain a channel weight matrix, weighting the channel weight matrix to the effective characteristics through multiplication, inputting the weighted result into a residual error network, and finally classifying the communication radiation source signals through an activation layer and a full connection layer in sequence according to the output result of the residual error network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111327199.5A CN114048774A (en) | 2021-11-10 | 2021-11-10 | Se-block-based resnet communication radiation source identification method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111327199.5A CN114048774A (en) | 2021-11-10 | 2021-11-10 | Se-block-based resnet communication radiation source identification method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114048774A true CN114048774A (en) | 2022-02-15 |
Family
ID=80208346
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111327199.5A Pending CN114048774A (en) | 2021-11-10 | 2021-11-10 | Se-block-based resnet communication radiation source identification method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114048774A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115618271A (en) * | 2022-05-05 | 2023-01-17 | 腾讯科技（深圳）有限公司 | Object type identification method, device, equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019170093A1 (en) * | 2018-03-07 | 2019-09-12 | 西安电子科技大学 | Method for generating frequency spectrum state, device and computer storage medium |
| CN112183300A (en) * | 2020-09-23 | 2021-01-05 | 厦门大学 | AIS radiation source identification method and system based on multi-level sparse representation |
| CN113156376A (en) * | 2021-04-15 | 2021-07-23 | 中国人民解放军空军航空大学 | Radar radiation source signal identification method based on SACNN |
-
2021
- 2021-11-10 CN CN202111327199.5A patent/CN114048774A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019170093A1 (en) * | 2018-03-07 | 2019-09-12 | 西安电子科技大学 | Method for generating frequency spectrum state, device and computer storage medium |
| CN112183300A (en) * | 2020-09-23 | 2021-01-05 | 厦门大学 | AIS radiation source identification method and system based on multi-level sparse representation |
| CN113156376A (en) * | 2021-04-15 | 2021-07-23 | 中国人民解放军空军航空大学 | Radar radiation source signal identification method based on SACNN |
Non-Patent Citations (1)
| Title |
|---|
| 高欣宇;张文博;姬红兵;欧阳成;: "新型雷达辐射源识别", 中国图象图形学报, no. 06, 16 June 2020 (2020-06-16), pages 107 - 115 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115618271A (en) * | 2022-05-05 | 2023-01-17 | 腾讯科技（深圳）有限公司 | Object type identification method, device, equipment and storage medium |
| CN115618271B (en) * | 2022-05-05 | 2023-11-17 | 腾讯科技（深圳）有限公司 | Object category identification method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107316013B (en) | Hyperspectral image classification method based on NSCT (non-subsampled Contourlet transform) and DCNN (data-to-neural network) | |
| CN111695467B (en) | Spatial spectrum full convolution hyperspectral image classification method based on super-pixel sample expansion | |
| CN112702294B (en) | Modulation recognition method for multi-level feature extraction based on deep learning | |
| CN107506822B (en) | Deep neural network method based on space fusion pooling | |
| Burges et al. | Extracting noise-robust features from audio data | |
| US20060217968A1 (en) | Noise-robust feature extraction using multi-layer principal component analysis | |
| CN111507047B (en) | Inverse scattering imaging method based on SP-CUnet | |
| CN101763514A (en) | Image segmentation method based on characteristic importance sorting spectral clustering | |
| CN111901069B (en) | Multi-user detection method based on neural network and approximate message transfer algorithm | |
| Jdid et al. | Robust automatic modulation recognition through joint contribution of hand-crafted and contextual features | |
| CN111723874A (en) | Sound scene classification method based on width and depth neural network | |
| CN113642445B (en) | Hyperspectral image classification method based on full convolution neural network | |
| CN115982613A (en) | Signal modulation identification system and method based on improved convolutional neural network | |
| CN114998958A (en) | Face recognition method based on lightweight convolutional neural network | |
| CN114048774A (en) | Se-block-based resnet communication radiation source identification method and system | |
| CN114218984A (en) | Radio frequency fingerprint identification method based on sample multi-view learning | |
| CN116843975A (en) | Hyperspectral image classification method combined with spatial pyramid attention mechanism | |
| Baris et al. | Modulation classification of MFSK modulated signals using spectral centroid | |
| US20230214845A1 (en) | Deep learning based method and system for detecting abnormal cryptocurrency transaction between computing devices in a blockchain network | |
| CN106646347A (en) | Ecological niche differential evolution-based multi-signal classification spectral peak search method | |
| US20140343945A1 (en) | Method of visual voice recognition by following-up the local deformations of a set of points of interest of the speaker's mouth | |
| CN115859131A (en) | Group partner discovery method and device | |
| CN114818823A (en) | Electroencephalogram channel selection method based on squeezing and activation graph convolution neural network | |
| CN104166837A (en) | Method of visual voice recognition with selection of groups of most relevant points of interest | |
| CN113869238A (en) | Cognitive Internet of vehicles intelligent frequency spectrum sensing method and system |
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 |