A cross-receiver radiation source identification method and system based on CNN-CBAM-BDA
By employing the CNN-CBAM-BDA method and utilizing convolutional neural networks and balanced distribution adaptive techniques, the problem of radiation source identification under the influence of receiver distortion was solved, achieving efficient and accurate radiation source identification across receivers.
Patent Information
- Application Number
- CN202211526466.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-30
- Publication Date
- 2025-11-28
- Estimated Expiration
- 2042-11-30
AI Technical Summary
In radiation source identification, existing technologies cannot effectively solve the problem of decreased recognition rate due to receiver distortion. Receiver distortion affects feature extraction, resulting in signal samples from different receivers not being compatible with each other in the same system.
A cross-receiver radiation source identification method based on CNN-CBAM-BDA is adopted. By combining convolutional neural network (CNN) with convolutional block attention mechanism (CBAM) and balanced distribution adaptation (BDA), the feature distribution of different receivers is matched and mapped to the Hilbert kernel space to reduce the influence of cross-receiver factors.
It improves the accuracy and speed of radiation source identification, overcomes the problems of low efficiency and low accuracy of traditional methods, and achieves improved identification performance across receivers.
Smart Images
Figure CN115712856B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of signal processing, and particularly relates to a cross-receiver emitter identification method and system based on CNN-CBAM-BDA. BACKGROUND
[0002] The emitter individual identification technology identifies a specific emitter device by extracting the subtle features of the emitter device, but the problem of emitter individual identification after replacing the receiver needs to be solved in practical application. Since the collected signal is actually affected by the distortion of the emitter and the distortion of the receiver, that is, when the receiving device is changed, the extracted features will deviate greatly, and the signal samples obtained by different receivers cannot be used in the same identification system, so the application of the traditional emitter identification method will be limited.
[0003] For specific emitter identification, previous studies are mostly based on expert experience feature extraction, such as using statistical feature quantities such as high-order spectrum and fractal features, and domain transformation such as discrete wavelet transform and Hilbert-Huang transform, using support vector machine, decision tree, and nearest neighbor classifier for classification and identification. In recent years, deep learning methods such as convolutional neural network and deep residual network have appeared for feature extraction and identification. However, both technical routes cannot avoid the influence of receiver distortion on features.
[0004] Therefore, it is of great significance to propose a cross-receiver emitter identification method and system based on CNN-CBAM-BDA. SUMMARY
[0005] In order to solve the problems of the prior art, the present application provides a cross-receiver emitter identification method and system based on CNN-CBAM-BDA to solve the technical defect problem that the recognition rate decreases due to the replacement of the receiver in the signal emitter identification of the prior art.
[0006] In a first aspect, the present application provides a cross-receiver emitter identification method based on CNN-CBAM-BDA, which comprises the following steps:
[0007] S1, collecting signals received by a plurality of different receivers to generate a cross-receiver emitter individual identification sample set;
[0008] S2, normalizing the sample set, dividing it into source domain samples and target domain samples, marking all samples of the source domain and part of the samples in the target domain, and taking the remaining samples in the target domain as a test sample set;
[0009] S3, further construct the CNN feature extraction module, and train the CNN network by using the source domain sample data, and introduce the CBAM attention mechanism to further extract deep features;
[0010] S4, input the source domain data and the target domain data into the CNN network trained in step S3, and extract feature vectors output by the last layer of the convolution layer respectively;
[0011] S5, construct the BDA module, and input the source domain features and the target domain features extracted in step S4 to obtain a training set and a test set;
[0012] S6, construct a KNN classifier to judge the target domain sample set.
[0013] Preferably, in S1, specifically comprising: using two different receivers to receive the radiation source signals sent by the same transmitter, and generating a cross-receiver radiation source individual identification sample set.
[0014] Preferably, in S2, specifically comprising:
[0015] S21, normalizing and preprocessing each signal sample in the sample set;
[0016] S22, taking the signal sample received by one of the receivers as the source domain sample, and taking the signal sample received by the other receiver as the target domain sample.
[0017] Preferably, in S3, further comprising:
[0018] The size of the convolution kernel in the first layer to the third convolution layer in the CNN network is set to 10, 5 and 3 respectively, the step is set to 1, and the number of convolution kernels is set to 32, 64 and 128; The first to third layer pooling layer adopts the maximum pooling method, and the pooling region size is 2, 2 and 2 respectively, and the step is set to 2;
[0019] Wherein, the activation function is ReLu function, and the Dropout value of the random inactivation layer is set to 0.5.
[0020] Further preferably, in S3, further comprising: the structure of the convolution block attention mechanism is set to: given mapping CBAM directly generates a one-dimensional channel attention mechanism And a two-dimensional spatial attention mechanism The calculation process of the whole CBAM can be summarized as:
[0021]
[0022]
[0023] Wherein, Default using pytorch tensor operation broadcast mechanism, F" is the output of CBAM.
[0024] Further preferably, S3 further comprises:
[0025] The feature block output by the last convolutional layer is denoted as F, followed by a global average pooling layer and a global maximum pooling layer, and the outputs are denoted as Favg and Fmax respectively. and A shared single-hidden-layer multi-layer perception (MLP) is then connected to further extract information, and the size of the hidden layer is The Sigmoid function is used as the activation function of the output layer, and the final output of the channel attention module is denoted as F. The expression is:
[0026]
[0027] Preferably, S5 specifically comprises:
[0028] S51, using z-score normalization algorithm, normalizing the source domain features and target domain features extracted in S4;
[0029] S52, constructing a BDA module, inputting the source domain features and target domain features extracted in step S4 into the BDA module to obtain new source domain and target domain sample features, which are used as the training set and test set of the KNN classifier respectively.
[0030] In a second aspect, the application further provides a cross-receiver emitter identification system based on CNN-CBAM-BDA, comprising:
[0031] A sample collection module is used to collect signals received by different receivers to generate a cross-receiver emitter individual identification sample set.
[0032] A sample processing module is used to normalize the sample set, divide it into source domain samples and target domain samples, label all samples in the source domain and part of the samples in the target domain, and use the remaining samples in the target domain as a test sample set.
[0033] A training module is used to further construct a CNN feature extraction module, train the CNN network using the source domain sample data, and introduce a CBAM attention mechanism to further extract deep features.
[0034] An extraction module is used to input the source domain data and target domain data into the CNN network trained by the training module, and extract the feature vectors output by the last layer of the convolutional layer.
[0035] The BDA construction module is used for constructing the BDA module, and inputting the source domain features and the target domain features extracted by the extraction module to obtain a training set and a test set.
[0036] The KNN classifier module is used for judging the target domain sample set.
[0037] In a third aspect, an electronic device is provided, including one or more processors; and a storage device storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner of the first aspect.
[0038] In a fourth aspect, a computer-readable storage medium is provided, and the computer-readable storage medium stores a computer program, when the computer program is executed by a processor, the computer program implements the method described in any implementation manner of the first aspect.
[0039] Compared with the prior art, the beneficial results of the present application are as follows:
[0040] (1) The radiation source individual identification model based on the convolutional neural network is used in the present application, and the convolution block attention mechanism is added, so that the problems of low efficiency, weak universality and low precision of the prior art are overcome, the deep features of the radiation source can be automatically extracted, and the identification speed and precision of the network are improved.
[0041] (2) The radiation source individual identification model based on the transfer learning is used in the present application, the feature distribution of different receivers is matched, the data of two domains is mapped to a Hilbert kernel space by using the balanced distribution adaptive method, the distance between the two domains is minimized, the internal properties of the two domains are preserved to the greatest extent, the influence of the cross-receiver factor on the identification performance is reduced, and the present application has certain engineering application value. BRIEF DESCRIPTION OF DRAWINGS
[0042] The accompanying drawings are included to provide a further understanding of 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 present application. Other embodiments and many of the intended advantages of the present application will be readily appreciated as the same becomes better understood by reference to the following detailed description. The elements of the drawings are not necessarily to scale relative to each other. Like reference numerals designate corresponding similar parts.
[0043] Figure 1 is an exemplary device architecture diagram in which an embodiment of the present application can be applied;
[0044] Figure 2 is a flowchart of a CNN-CBAM-BDA based cross-receiver radiation source identification method of an embodiment of the present application.
[0045] Figure 3 Network structure schematic diagram of CNN feature extraction module in the CNN-CBAM-BDA based cross-receiver radiation source identification method of the embodiment of the present application;
[0046] Figure 4-1 And 4-2 Identification test result graph of the simulation AIS signal in the CNN-CBAM-BDA based cross-receiver radiation source identification method of the embodiment of the present application;
[0047] Figure 5 Identification test result graph of the actual collected AIS signal in the CNN-CBAM-BDA based cross-receiver radiation source identification method of the embodiment of the present application;
[0048] Figure 6 Flow schematic diagram of the CNN-CBAM-BDA based cross-receiver radiation source identification system of the embodiment of the present application;
[0049] Figure 7 Structure schematic diagram of the computer device suitable for realizing the electronic device of the embodiment of the present application. DETAILED DESCRIPTION
[0050] In the following detailed description, reference is made to the accompanying drawings, which form a part hereof, and in which are shown by way of illustration specific embodiments in which the application can be practiced. In this regard, directional terminology, such as "top," "bottom," "left," "right," "above," "below," etc., is used with reference to the orientation of the Figure(s) being described. Because components of embodiments can be positioned in a number of different orientations, the directional terminology is used for purposes of illustration and is in no way limiting. It is to be understood that other embodiments can be utilized and structural or logical changes can be made without departing from the scope of the present application. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present application is defined by the appended claims.
[0051] It should be understood that Figure 1 The number of terminal devices, networks and servers in the system is only illustrative. According to the needs of implementation, there can be any number of terminal devices, networks and servers.
[0052] Figure 1 An exemplary system architecture 100 for a method of processing information or an apparatus for processing information to which embodiments of the present application can be applied is shown.
[0053] As Figure 1As shown, the system architecture 100 can include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is a medium for providing communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 can include various connection types, such as wired, wireless communication links or optical fiber cables, etc.
[0054] The users can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.
[0055] The terminal devices 101, 102, 103 can be various electronic devices with communication functions, including but not limited to smart phones, tablet computers, laptop computers and desktop computers, etc.
[0056] The server 105 can be a server providing various services, such as a background information processing server for processing the verification request information sent by the terminal devices 101, 102, 103. The background information processing server can analyze and process the received verification request information, and obtain a processing result (such as verification success information for representing that the verification request is a legal request).
[0057] It should be noted that the method for processing information provided by the embodiments of the present application is generally executed by the server 105, and correspondingly, the device for processing information is generally arranged in the server 105. In addition, the method for sending information provided by the embodiments of the present application is generally executed by the terminal devices 101, 102, 103, and correspondingly, the device for sending information is generally arranged in the terminal devices 101, 102, 103.
[0058] It should be noted that the server can be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software or software modules (for example, for providing distributed services), or as a single software or multiple software modules, which is not specifically limited here.
[0059] The technical idea of the present application is as follows: the individual identification of the radiation source mainly includes two aspects of feature extraction and classification identification. In the process of extracting features, the traditional feature extraction method needs to map the time domain signal to a certain transform domain for analysis through time-frequency or other transform, which greatly increases the calculation amount and complexity. In addition, when the existing technology adopts the one-dimensional convolution network to extract features, only the relationship between the sampling points in a short time of the signal is considered, and the local features are extracted.
[0060] The present application combines the convolution block attention mechanism in the feature extraction network, uses the correlation of the communication signal time sequence, and the added attention mechanism can improve the expression of the interested area and further extract important features. The domain adaptation in transfer learning is similar to the cross-receiver identification problem. Changing the receiver causes the feature to shift, and the domain adaptation wants to solve the problem that the feature distribution of the source domain and the target domain is different but the task is the same.
[0061] The present application first introduces the convolution block attention mechanism (CBMA) in the convolutional neural network (CNN) to extract deep features, and then uses the balanced distribution adaptive method to map the feature data of different distributions to a high-dimensional reproducing kernel Hilbert space for K-neighbor (KNN) classification.
[0062] Figure 2 The embodiment of the present application is shown, which discloses a cross-receiver radiation source identification method based on CNN-CBAM-BDA, as shown in Figure 2 The method comprises the following steps:
[0063] S1, collecting signals received by several different receivers to generate a cross-receiver radiation source individual identification sample set;
[0064] Preferably, in the present embodiment, the radiation source signals sent by the same transmitter are received by two different receivers to generate a cross-receiver radiation source individual identification sample set. The two receivers are respectively denoted as receiver A and receiver B. In other embodiments, the number of receivers can be modified as appropriate.
[0065] S2, normalizing the sample set, dividing it into source domain samples and target domain samples, marking all samples of the source domain and part of the samples in the target domain, and taking the remaining samples in the target domain as a test sample set;
[0066] Specifically, the specific implementation method in S2 includes:
[0067] S21, normalizing and preprocessing each signal sample in the sample set;
[0068] S22, the signal sample received by the receiver A is taken as the source domain sample, the signal sample received by the receiver B is taken as the target domain sample, all samples of the source domain and part of the samples in the target domain are labeled, and the target domain labeled samples are taken as the sample set to be tested.
[0069] S3, a CNN feature extraction module is further constructed, and the CNN network is trained by using the source domain sample data, and a CBAM attention mechanism is introduced to further extract deep features.
[0070] Specifically, referring to Figure 3 , the specific implementation method of S3 is:
[0071] S31, a CNN feature extraction module is constructed, which is different from the traditional CNN network structure, and a CBAM convolution block attention mechanism is added in the model, which can improve the expression of the region of interest by using the correlation of the communication signal time sequence, and further extract important features.
[0072] A CNN network with an accuracy of about 99% is trained by using the labeled source domain sample data. The network structure is: the first convolution layer→the batch normalization layer→the first pooling layer→the second convolution layer→the batch normalization layer→the second pooling layer→the third convolution layer→the batch normalization layer→the third pooling layer→the random inactivation layer→the convolution block attention mechanism→the binary adaptive mean pooling layer→the full connection layer.
[0073] S32, the convolution kernel size of the first layer to the third convolution layer is set to 10, 5 and 3 respectively, the step is 1, and the number of convolution kernels is set to 32, 64 and 128. The activation function is ReLu function. The first to third layer pooling layer adopts the maximum pooling mode, and the pooling region size is 2, 2 and 2, and the step is 2. The Dropout value of the random inactivation layer is set to 0.5.
[0074] S33, the structure of the convolution block attention mechanism is: given mapping CBAM directly generates a one-dimensional channel attention mechanism and a two-dimensional spatial attention mechanism
[0075] Further, the whole CBAM calculation process can be summarized as:
[0076]
[0077]
[0078] Among them, The pytorch tensor operation broadcast mechanism is used by default, and F" is the final output of the CBAM. The feature blocks output by the last convolutional layer are denoted as F, and a global uniform pooling layer and a global maximum pooling layer are connected thereto, and the outputs are denoted as A shared single-hidden-layer multi-layer perception (MLP) is connected thereafter to further extract information, and a τ is introduced herein to reduce the parameters of the hidden layer, and the size of the hidden layer is A sigmoid function is used as the activation function of the output layer, and the final output of the channel attention module is denoted as The expression is as follows:
[0079]
[0080] The spatial attention mechanism is connected to the average pooling layer and the maximum pooling layer in the channel dimension after the input feature block, and thus a two-dimensional feature block and The two 2D feature blocks are then concatenated in the channel dimension, and the concatenated feature block is mapped to HxW using a 7x7 convolutional layer, and the final output of the spatial attention module is denoted as The expression is as follows:
[0081]
[0082] S4, inputting the source domain data and the target domain data into the CNN network trained in step S3 to extract the feature vectors output by the last convolutional layer;
[0083] Specifically, the method of S4 is as follows: inputting the source domain data and the target domain data into the CNN network trained in step S3, changing the length of each signal sample output by the binary adaptive mean aggregation layer to 128, and extracting 128-dimensional feature vectors to construct the source domain features and the target domain features.
[0084] S5, constructing a BDA module and inputting the source domain features and the target domain features extracted in step S4 to obtain a training set and a test set;
[0085] Specifically, the features are first normalized, and then the BDA module is constructed, and the source domain features and the target domain features extracted in step S4 are inputted to obtain a training set and a test set. The method of S5 is as follows:
[0086] S51, using the z-score normalization algorithm to normalize the source domain features and the target domain features extracted in S4, and the specific steps are as follows:
[0087] First, calculate the mean value of each input sample according to the following formula:
[0088]
[0089] wherein, denotes the mean value of the jth input sample, N denotes the total number of sampling points of the jth input sample, denotes the amplitude value of the ith sampling point in the jth input sample;
[0090] Secondly, the standard deviation of each input sample is calculated according to the following formula:
[0091]
[0092] wherein, s j denotes the variance of the jth input sample;
[0093] Thirdly, the normalized value of each sampling point in each input sample is calculated according to the following formula:
[0094]
[0095] wherein, denotes the normalized value of the ith sampling point in the jth input sample.
[0096] S52, a BDA module is constructed, and the source domain feature set extracted in step S4 is denoted as X S , the domain is the target domain feature set is X T , the domain is D T , and a mapping Φ is found such that the training sample and the test sample distribution The minimum distance between the source domain and the target domain in this mapping space is obtained, and the calculation expression is as follows:
[0097]
[0098] wherein, is a Hilbert kernel space, c={1, 2, …, C} is a label of each class, and denote the source domain sample and the target domain sample belonging to the cth class, n c and m c denote the number of source domain samples and the number of target domain samples belonging to the cth class, respectively.
[0099] Further, the optimization target obtained through the kernel matrix is:
[0100]
[0101] wherein, μ is a balance factor, λ is a regularization parameter, A is a transformation matrix, X is an input matrix composed of X S and X T , I is an identity matrix, is an identity matrix, is the center matrix, M0and M c are MMD matrices, whose expressions are as follows, respectively:
[0102]
[0103]
[0104] The transformation matrix A is obtained, and the new source domain and target domain sample features after mapping can be obtained by multiplying the original sample features with the matrix A, wherein A is the smallest d eigenvalues of the following formula:
[0105]
[0106] S53, input the source domain features and target domain features extracted in step S4 into the BDA module to obtain new source domain and target domain sample features, which are respectively taken as the training set and test set of the KNN classifier.
[0107] S6, constructing a KNN classifier to judge the target domain sample set to be measured.
[0108] Specifically, the implementation method of S6 is: constructing a KNN classifier to judge the target domain sample set to be measured. The expression of the KNN algorithm is as follows:
[0109] S61, according to the feature vector of each test sample extracted in step S53, the similarity between the vector and the feature vector in the training set is calculated, and the calculation formula is:
[0110]
[0111] Wherein, d i is the feature vector of the test sample, d j is the center vector in the jth class, M is the dimension of the feature vector, and W k is the kth dimension of the vector. Generally, the value of k is determined by an initial value, and then the value of k is changed according to the test result. The initial value is generally several hundred to several thousand.
[0112] S62, according to the similarity, the k samples most similar to the test sample vector are selected in the training set.
[0113] S63, in the k neighbors of the test sample vector, the weight of each class is calculated in turn, and the calculation formula is:
[0114]
[0115] Wherein, x is the feature vector of the test sample, Sim(x, d i ) is the similarity, b is the threshold value, and y(d i , C j ) is 0 or 1, if di ∈C j If the function value is 1, otherwise 0.
[0116] S64, compare the weights of the classes, and divide the signal samples into the class with a larger weight, to obtain the recognition result of each sample.
[0117] The technical effects of the present application will be described in detail below with examples, referring to Figure 4-1 , Figure 4-2 and Figure 5 .
[0118] Specifically, according to the unintentional modulation parameters in Table 1, the radiation source individuals 1 to 8 are simulated, the carrier frequency of the signal in the experiment is 100 MHz, the pulse width is 5 μs, the sampling frequency is 1 GHz, and the signal bandwidth is 20 MHz. All the radiation sources emit LFM signals at a signal-to-noise ratio of 30 dB. Each radiation source individual randomly generates 100 groups of signals. According to the ratio of 5:5, the training sample set and the test set are constructed, and the feature extraction network is trained. The receiver is essentially the inverse process of the transmitter, and its analog hardware contains similar distortion to the radiation source distortion, so the receiver B adds a set of phase noise to the original distortion, so that the feature distribution of the receiver A and the receiver B is different. According to the unintentional modulation simulated by the phase noise, the data of the eight radiation source individuals collected by two different receivers are simulated, 50 samples of each class of signals in the source domain and the target domain are taken for transfer learning, and the recognition accuracy finally reaches 100%. The BDA parameter settings are shown in Table 2, and the cross-receiver recognition rate is shown in Figure 5 .
[0119] Table 1
[0120]
[0121] Table 2
[0122]
[0123] According to the power amplifier nonlinear distortion parameter simulation of the source individual 1 to the source individual 8 in table 3, the carrier frequency of the signal in the experiment is 10MHz, the pulse width is 3us, the sampling frequency is 500MHz, and the signal bandwidth is 20MHz. All the sources emit LFM signals at a signal-to-noise ratio of 30dB. The Taylor series model is a commonly used power amplifier nonlinear model, which can be used to simulate different radar source individuals. 100 groups of signals are randomly generated for each source individual. According to the ratio of 5:5, the training sample set and the test set are constructed. Similarly, when simulating the nonlinear distortion of the power amplifier, the receiver also manually considers the influence of the distortion. Therefore, according to the approximation of the cascade of two power amplifiers, the receiver B superimposes a set of power amplifier distortions on the original power amplifier distortion, so that the feature distribution of receiver A and receiver B is different. According to the power amplifier nonlinear distortion simulation of the 8 source individuals collected by two different receivers, 50 samples of each type of signal in the source domain and the target domain are taken for transfer learning, and the recognition accuracy finally reaches 96.75%. The BDA parameter settings are shown in table 4, and the cross-receiver recognition rate is shown in Figure 5
[0124] Table 3
[0125]
[0126]
[0127] Table 4
[0128]
[0129] The existing 8 types of AIS signals received by two different receivers are used for experiments. The AIS signal adopts Gaussian minimum frequency shift keying modulation with a rate of 9600kBd. The sampling frequency is 48KHz and the signal length is 1280 during the actual measurement. In order to prove the effectiveness of CNN-CBAM-BDA, five methods of Cepstrum-TCA, Cepstrum-JDA, Cepstrum-BDA, CNN-BDA and CNN-CBAM-BDA are compared in the experiment, in which Cepstrum is used to extract the nonlinear features of the source by using the cyclic spectrum.
[0130] The receiver A (source domain) collects 800 samples in total, and the receiver B (target domain) collects 800 samples in total, wherein 100 samples of each type of signal are included. The source domain samples are used to construct a training sample set and a test set in a ratio of 8:2, a feature extraction network is trained, the accuracy rate is 99%, and the accuracy rate is reduced to 77% after the receiver is replaced (that is, the data collected by the receiver B is tested). The 640 labeled training samples of the source domain and the 640 unlabeled test samples of the target domain are input into the network to extract feature vectors, and then are sent to the BDA module for transfer learning, and finally, the recognition rate can be increased to 97.50% by using the method of the application. The BDA parameter settings are shown in Table 5, and the recognition rates of different transfer learning methods across receivers are shown in Table 6. Figure 5
[0131] Table 5
[0132]
[0133] The application has the following beneficial effects:
[0134] Since the application uses a radiation source individual recognition model based on a convolutional neural network, a convolutional block attention mechanism is added, and the problems of low efficiency, weak universality and low precision of the prior art are overcome, so that the application can automatically extract the deep features of the radiation source, and the recognition speed and precision of the network are improved.
[0135] The application uses a radiation source individual recognition model based on transfer learning, matches the feature distributions of different receivers, maps the data of the two domains to a Hilbert kernel space by using a balanced distribution adaptive method, minimizes the distance between the two domains, and at the same time, the internal properties of the two domains are preserved to the greatest extent, the influence of the cross-receiver factor on the recognition performance is reduced, and the application has certain engineering application value.
[0136] In a second aspect, the application further provides a cross-receiver radiation source recognition system based on CNN-CBAM-BDA, as shown in Figure 6
[0137] The sample collection module 61 is used to collect signals received by a plurality of different receivers to generate a cross-receiver radiation source individual recognition sample set.
[0138] The sample processing module 62 is used to normalize the sample set, divide the sample set into source domain samples and target domain samples, label all samples of the source domain and part of the samples in the target domain, and take the remaining samples of the target domain as a test sample set.
[0139] The training module 63 is configured to further construct the CNN feature extraction module, train the CNN network by using the source domain sample data, and introduce the CBAM attention mechanism to further extract deep features.
[0140] The extraction module 64 is configured to input the source domain data and the target domain data into the CNN network trained by the training module, and extract feature vectors output by the last layer of the convolutional layer.
[0141] The BDA construction module 65 is configured to construct the BDA module, and input the source domain features and the target domain features extracted by the extraction module to obtain a training set and a test set.
[0142] The KNN classifier module 66 is configured to make a decision on the target domain sample set to be measured.
[0143] Reference will now be made to the following description Figure 7 , which shows a structural schematic diagram of a computer device 600 of an electronic device (for example Figure 1 , a server or a terminal device) suitable for implementing embodiments of the present application. Figure 7 The electronic device shown is merely an example, and should not impose any limitation on the functions and use range of the embodiments of the present application.
[0144] As shown in Figure 7 , the computer device 600 includes a central processing unit (CPU) 601 and a graphics processing unit (GPU) 602, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 603 or programs loaded into a random access memory (RAM) 604 from a storage portion 609. In the RAM 604, various programs and data required for the operation of the device 600 are also stored. The CPU 601, the GPU 602, the ROM 603, and the RAM 604 are connected to each other through a bus 605. An input / output (I / O) interface 606 is also connected to the bus 605.
[0145] The following components are connected to the I / O interface 606: an input portion 607 including a keyboard, a mouse, and the like; an output portion 608 including a display such as a cathode ray tube (CRT) display, a liquid crystal display (LCD), and the like, and a speaker, and the like; a storage portion 609 including a hard disk, and the like; and a communication portion 610 including a network interface card such as a LAN card, a modem, and the like. The communication portion 610 performs communication processing via a network such as the Internet. A drive 611 can also be connected to the I / O interface 606 as needed. A removable medium 612 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like is mounted on the drive 611 as needed, so that a computer program read therefrom is installed in the storage portion 609 as needed.
[0146] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program in accordance with embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for executing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 610, and / or installed from the removable medium 612. When the computer program is executed by the central processing unit (CPU) 601 and the graphics processor (GPU) 602, the above-described functions defined in the methods of the present invention are executed.
[0147] It should be noted that the computer readable medium described in the present invention can be a computer readable signal medium or a computer readable storage medium or any combination of the two. The computer readable medium may, for example, be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor device, device or means, or any combination of the above. More specific examples of the computer readable medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution device, device or means. In the present invention, the computer readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer readable program code. Such a propagated data signal can take many forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer readable signal medium can also be any computer readable medium other than the computer readable medium that can send, propagate or transmit the program for use by or in conjunction with an instruction execution device, device or means. The program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.
[0148] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).
[0149] The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functionalities, and operations of possible implementations of apparatuses, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flow diagrams or block diagrams can represent a module, a procedure, or a part of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or in the reverse order, depending on the functionality involved. It is also noted that each block of the block diagrams and / or flow diagrams and combinations of blocks in the block diagrams and / or flow diagrams can be implemented by dedicated hardware-based apparatuses or by a combination of dedicated hardware-based apparatuses and computer instructions.
[0150] The modules described in the embodiments of the present application can be implemented by software, or by hardware. The described modules can also be arranged in a processor.
[0151] As another aspect, the present application also provides a computer readable medium, which can be included in the electronic device described in the above embodiments, or can exist separately without being assembled into the electronic device. The above computer readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method steps described in the first aspect of the present application.
[0152] The above description is only the preferred embodiment of the present application and the explanation of the applied technical principles. It should be understood by those skilled in the art that the inventive scope of the present application is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by the combinations of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by the mutual replacement of the above features and the technical features with similar functions disclosed in the present application (but not limited to) without departing from the above inventive concept.
Claims
1. A CNN-CBAM-BDA-based cross-receiver radiation source identification method, characterized in that, The method comprises the following steps: S1, collecting signals received by several different receivers to generate a cross-receiver radiation source individual identification sample set; S2, normalizing the sample set, dividing it into source domain samples and target domain samples, marking all samples in the source domain and part of the samples in the target domain, and taking the remaining samples in the target domain as a sample set to be tested; S3, constructing a CNN feature extraction module, training the CNN network using source domain sample data, and introducing a CBAM attention mechanism to further extract deep features; S4, inputting the source domain data and the target domain data into the CNN network trained in step S3 to extract the feature vectors output by the last layer of the convolution layer; S5, constructing a BDA module, inputting the source domain features and the target domain features extracted in step S4 to obtain a training set and a test set; S6, constructing a KNN classifier to judge the target domain sample set to be tested; In S3, the structure of the convolution block attention mechanism is set as follows: given a mapping CBAM directly generates a one-dimensional channel attention mechanism and a two-dimensional spatial attention mechanism The entire CBAM calculation process can be summarized as follows: ， wherein, By default, the pytorch tensor operation broadcast mechanism is used, is the last output of the CBAM; The feature blocks outputted by the last convolutional layer are denoted as , followed by a global average pooling layer and a global max pooling layer, and the outputs are denoted as , followed by a shared single-hidden-layer multi-layer perceptron (MLP) to further extract information, and the size of the hidden layer is , and the Sigmoid function is used as the activation function of the output layer, and the final output of the channel attention module is denoted as , and the expression is: ； The network structure levels of the CNN network are in turn: a first convolution layer, a batch normalization layer, a first pooling layer, a second convolution layer, a batch normalization layer, a second pooling layer, a third convolution layer, a batch normalization layer, a third pooling layer, a random deactivation layer, a convolution block attention mechanism, a binary adaptive mean pooling layer, and a full connection layer.
2. The CNN-CBAM-BDA-based cross-receiver radiation source identification method according to claim 1, characterized in that, In S1, specifically comprising: using two different receivers to receive the radiation source signals sent by the same transmitter to generate a cross-receiver radiation source individual identification sample set.
3. The CNN-CBAM-BDA based cross receiver radiation source identification method according to claim 1, wherein, In S2, specifically comprising: S21, normalizing and preprocessing each signal sample in the sample set; S22, taking the signal samples received by one of the receivers as the source domain samples and the signal samples received by the other receiver as the target domain samples.
4. The CNN-CBAM-BDA-based cross-receiver radiation source identification method according to claim 1, characterized in that, In S3, further comprising: The convolution kernel sizes in the first to third convolution layers in the CNN network are set to 10, 5, and 3 respectively, the step size is set to 1, and the number of convolution kernels is set to 32, 64, and 128; the first to third pooling layers all use the maximum pooling method, and the pooling region sizes are 2, 2, and 2 respectively, and the step size is set to 2; The activation function is a ReLu function, and the Dropout value of the random deactivation layer is set to 0.
5.
5. The CNN-CBAM-BDA based cross receiver radiation source identification method according to claim 1, wherein, In S5, specifically comprising: S51, using the z-score normalization algorithm to normalize the source domain features and the target domain features extracted in S4; S52, constructing a BDA module, inputting the source domain features and the target domain features extracted in step S4 into the BDA module to obtain new source domain and target domain sample features, which are respectively used as the training set and the test set of the KNN classifier.
6. A CNN-CBAM-BDA-based cross-receiver radiation source identification system, characterized in that, The method comprises the following steps: A sample collection module is used to collect signals received by several different receivers to generate a cross-receiver radiation source individual identification sample set; A sample processing module is used to normalize the sample set, divide it into source domain samples and target domain samples, mark all samples in the source domain and part of the samples in the target domain, and take the remaining samples in the target domain as a sample set to be tested; The training module is configured to further construct the CNN feature extraction module, train the CNN network by using the source domain sample data, and introduce a CBAM attention mechanism to further extract deep features. The extraction module is configured to input the source domain data and the target domain data into the CNN network trained by the training module, and extract feature vectors output by a last layer of a convolutional layer. The BDA construction module is configured to construct a BDA module, and input the source domain features and the target domain features extracted by the extraction module to obtain a training set and a test set. The KNN classifier module is configured to make a decision on the target domain sample set. 7.An electronic device, comprising: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1-5.
8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1-5.