Radiation source identification method and system based on multi-task learning

Abstract

The embodiment of the invention relates to a radiation source identification method and system based on multi-task learning. The method comprises the following steps: receiving signals from a plurality of communication radiation sources, and preprocessing the signals to generate a baseband signal sequence; inputting the baseband signal sequence into a pre-established multi-task classification model, wherein the multi-task classification model comprises a shared feature extraction network and a dual-task classifier; extracting shared features from the baseband signal sequence through the shared feature extraction network; classifying the shared features through the dual-task classifier, and respectively outputting a modulation mode identification result and a radiation source identification result; wherein the dual-task classifier comprises a modulation mode classifier and a radiation source classifier, and the output of the radiation source classifier depends on the output of the modulation mode classifier.

Description

Technical Field

[0001] Multiple embodiments of this specification relate to, specifically to a method and system for radiation source identification based on multi-task learning. Background Art

[0002] Radiation source identification generally refers to Specific Emitter Identification (SEI), which is of great significance in the field of wireless communication. Specifically, it generally involves the identification of the fingerprints of radiation sources. The fingerprint of a radiation source refers to the unique and distinctive characteristics of each radiation source when emitting electromagnetic signals. These characteristics can be used to identify and distinguish different radiation sources. Even if two devices use the same model and technical parameters, there will be subtle differences in the signals they emit, and these differences constitute the so-called "fingerprints". The core of SEI technology lies in extracting and identifying the unique characteristics in the signals emitted by radiation sources, which are usually referred to as "fingerprints". Since the fingerprint of each radiation source is unique and cannot be replicated or forged. Using radiation source fingerprint identification technology can effectively prevent spoofing attacks because this "fingerprint" is more difficult to imitate than traditional key authentication. SEI technology generally includes the following steps: signal acquisition, where the electromagnetic signals received from radiation sources are collected; feature extraction, where the signals are processed to extract unique features that can be used for identification, such as non-linear characteristics, transient signals, etc.; classification and identification, where a classifier is used to compare the extracted features with a known database to determine the signal source; database management, where a database containing the characteristics of known radiation sources is established and maintained, and continuously updated. However, the existing observation methods are still in the experimental stage, and it is difficult to ensure that all radiation source fingerprints can be accurately captured. And when facing different working conditions, there may be problems of unstable feature classification performance and insufficient adaptability. Summary of the Invention

[0003] Multiple embodiments of this specification describe a method and system for radiation source identification based on multi-task learning.

[0004] In a first aspect, an embodiment of this specification provides a method for radiation source identification based on multi-task learning, including the following steps:

[0005] Receive signals from multiple communication radiation sources, and preprocess the signals to generate a baseband signal sequence;

[0006] Input the baseband signal sequence into a pre-established multi-task classification model, where the multi-task classification model includes a shared feature extraction network and a dual-task classifier;

[0007] Extract shared features from the baseband signal sequence through the shared feature extraction network;

[0008] Classify the shared features through the dual-task classifier, and respectively output the modulation mode recognition result and the radiation source recognition result;

[0009] Among them, the dual-task classifier includes a modulation mode classifier and a radiation source classifier, and the output of the radiation source classifier depends on the output of the modulation mode classifier.

[0010] In a second aspect, an embodiment of this specification provides a radiation source recognition system based on multi-task learning, including:

[0011] A signal receiving module, which receives signals from multiple communication radiation sources, preprocesses the signals, and generates a baseband signal sequence;

[0012] A multi-task classification module, including a shared feature extraction network and a dual-task classifier, is used to extract shared features from the baseband signal sequence, and respectively output a modulation mode recognition result and a radiation source recognition result;

[0013] Among them, the dual-task classifier includes a modulation mode classifier and a radiation source classifier, and the output of the radiation source classifier depends on the output of the modulation mode classifier.

[0014] In a third aspect, an embodiment of this specification provides an electronic device, including a processor and a memory;

[0015] The processor is connected to the memory;

[0016] The memory is used to store executable program code;

[0017] The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the method described in any of the above aspects.

[0018] In a fourth aspect, an embodiment of this specification provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described in any of the above aspects is implemented.

[0019] In a fifth aspect, an embodiment of this specification provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method described in any of the above aspects is implemented.

[0020] The beneficial effects brought by the technical solutions provided in some embodiments of this specification at least include:

[0021] In multiple embodiments of this specification, the provided radiation source identification method and system based on multi-task learning utilize the characteristic that the AMC task and the SEI task are two tasks with related features. Establishing a multi-task learning model can make full use of this correlation to improve the accuracy of the radiation source classifier. The dual-task classifier provided in this specification can not only give full play to the advantage of multi-task learning in improving performance, but also has a certain advantage in computational efficiency, achieving improvements in detection accuracy and computational efficiency. The structure of the shared feature extraction network is the same under the AMC task and the SEI task, which can simplify the training and deployment of the monitoring system under multiple tasks.

[0022] Other features and advantages of multiple embodiments of this specification will be further revealed in the following specific implementation manners and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] To more clearly illustrate the technical solutions in the embodiments of this specification, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of this specification. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0024] Figure 1 It is a schematic diagram of the radiation source identification scenario provided for the embodiments of this specification.

[0025] Figure 2 It is a schematic diagram of the radiation source identification process provided for the embodiments of this specification.

[0026] Figure 3 It is a schematic diagram of the flow of the radiation source identification method provided for the embodiments of this specification.

[0027] Figure 4 It is a schematic diagram of the Transformer encoder provided for the embodiments of this specification.

[0028] Figure 5 It is a schematic diagram of the mask matrix provided for the embodiments of this specification.

[0029] Figure 6 It is a schematic diagram of calculating the multi-task cross-entropy loss function provided for the embodiments of this specification.

[0030] Figure 7 It is a schematic diagram of the radiation source identification system provided for the embodiments of this specification.

[0031] Figure 8 It is a schematic diagram of the electronic device provided for the embodiments of this specification. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] The technical solutions of the embodiments of this specification will be explained and described below with reference to the accompanying drawings of the embodiments of this specification. However, the following embodiments are only the preferred embodiments of this specification, not all of them. Based on the embodiments in the implementation manners, other embodiments obtained by those skilled in the art without creative efforts all fall within the protection scope of this specification.

[0033] The terms "first", "second", "third", etc. in the specification, claims and the above-mentioned drawings of this specification are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices.

[0034] In the following description, the appearance of terms such as "inner", "outer", "upper", "lower", "left", "right", etc. indicating directions or positional relationships is only for the convenience of describing the embodiments and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation of this specification.

[0035] The data involved in this application are all information and data authorized by users or fully authorized by all parties, and the collection of relevant data complies with the relevant laws, regulations and standards of relevant countries and regions.

[0036] Before introducing the technical solutions described in this specification, the application scenarios of the technical solutions and related technologies will be introduced.

[0037] Please refer to the attached Figure 1 , in an environment with multiple electromagnetic radiation sources 60, a receiving antenna 20 is set, and the received radio signals are finally input into the server 30 after being processed. The radiation source 60 identification method provided in this embodiment is run on the server 30 to identify multiple electromagnetic radiation sources 60 in the electromagnetic radiation environment 10. The identification and marking of the electromagnetic radiation sources 60 are realized, providing conditions for subsequent applications. Exemplarily, subsequent applications can be electromagnetic environment supervision, electromagnetic environment monitoring or electromagnetic environment perception. The server 30 involved in this specification can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0038] The identification of electromagnetic radiation source 60 is called the SEI (Specific Emitter Identification) task. In specific implementations, the impact of the AMC (Adaptive Modulation and Coding) task also needs to be considered. AMC is a key technology used to dynamically adjust transmission parameters according to channel conditions to optimize the performance of communication systems. In existing signal feature recognition research, the AMC task and the SEI task are often implemented as two independent tasks. That is, each task corresponds to a dedicated model, which not only wastes the computing resources of the monitoring end but also does not fully utilize the association between the two tasks to improve the recognition performance. As wireless communication systems become more complex, the signals received by the SEI task are no longer generated by a single communication modulation method but change over time. At this time, the monitoring system needs to consider the impact of the change in modulation method on the SEI task during design.

[0039] In the field of deep learning, end-to-end model design has become a mainstream method. The advantage of this method is that the model can directly learn the mapping relationship from the original input data to the target output, avoiding the need for staged processing and manual feature engineering. Multi-Task Learning (MTL) is a learning algorithm that incorporates multiple tasks into the same end-to-end model. It allows the model to share information between different tasks, thereby improving the performance of each task. For the AMC and SEI tasks in the field of signal monitoring, whether from the perspective of model structure design or data input form, the two tasks are interoperable and are therefore very suitable for sharing the same feature extraction network in an end-to-end model. From existing research, it can be seen that both the AMC task and the SEI task follow the same model design criteria, and both tasks can use constellation diagrams, IQ data, or Amplitude-Phase (AP) data as inputs. Therefore, the two tasks have a common model basis. In addition, from the perspective of the signal feature generation mechanism, the AMC task and the SEI task also have similarities. According to the analysis results of the signal model, the AMC task is a deliberate modulation process, while the SEI task is an unintentional modulation process, and both change the waveform of the signal to different degrees. The recognition of waveform features can be subdivided into the recognition of signal modulation features and the recognition of signal radio frequency fingerprint features. Therefore, the two tasks have a common feature basis.

[0040] To improve the accuracy of the SEI task by leveraging the aforementioned characteristics, this embodiment specifically provides an improved backbone feature extraction model to increase the capacity of the backbone feature extraction model, so that it has sufficient capacity to extract the morphological features of the waveform.

[0041] On the other hand, this embodiment also specifically designs a dual-task classifier 52, and combines a masking technique in the dual-task classifier 52. By combining the masking technique, the classification accuracy is improved. Please refer to the attached Figure 2 , which is a schematic diagram of the classifier connection used in the radiation source 60 recognition method based on multi-task learning provided in this embodiment.

[0042] Specifically, please refer to the attached Figure 3 , a radiation source recognition method based on multi-task learning provided in this embodiment includes the following steps:

[0043] Step S101) Receive signals from multiple communication radiation sources 60, and preprocess the signals to generate a baseband signal sequence 40.

[0044] In the radiation source 60 recognition scenario of this embodiment, there are multiple similar communication devices at the transmitting end. Only one transmitting end device is transmitting signals during a certain period of time, and the single-carrier modulation method adopted by the transmitting end device is not fixed but will be continuously changed. The signal monitoring end first captures the signals in the air, and then identifies the radiation source information contained in the signals. Among them, the modulation method information is used for subsequent signal content parsing, and the radio frequency fingerprint information is used to identify whether the information transmitted by the signal is secure and reliable. In the figure, the N-point complex baseband signal sequence 40 obtained by the receiver sampling is represented by x(n). The real part of the baseband signal sequence 40 represents the in-phase component of the baseband signal, and the imaginary part of the sequence represents the quadrature component of the baseband signal.

[0045] Assume that the baseband signal output by the upconverter is s(t), let PA[] represent the equivalent baseband model of a non-linear power amplifier, and fe be the center frequency of the carrier wave. Then the radio frequency signal RF(t) output by the transmitting end antenna can be expressed as:

[0046] RF(t) = PA[s(t)]e2πfet;

[0047] At the receiving end, the output y(t) of the baseband signal passing through an ideal downconverter with a center frequency of fo is:

[0048] y(t) = [h(t)*RF(t)]e-j2πfot + 80 + n(t) = h(t)*PA[s(t)]e-j2π(fo - fe)t + 60 + n(t). Where h(t) represents the impulse response of the channel, and n(t) represents additive white Gaussian noise.

[0049] Step S102) Input the baseband signal sequence 40 into a pre-established multi-task classification model 50. The multi-task classification model 50 includes a shared feature extraction network 51 and a dual-task classifier 52.

[0050] The shared feature extraction network 51 includes a convolutional neural network and a self-attention encoder. The convolutional neural network outputs preset local features of the baseband signal sequence 40, and the self-attention encoder outputs preset global features of the baseband signal sequence 40.

[0051] The convolutional neural network sequentially includes a DenseCNN backbone network, a dimension transformation network, a Transformer encoder 511, and a dual-task classifier 52. The DenseCNN backbone network includes 15 sequentially connected convolutional modules. The input of the DenseCNN backbone network is a vector with a shape of [N×2×1×128], where each dimension represents the data Batch Size N, the number of feature channels (2 in this embodiment), and the height and width of the feature map of each channel (1×128 in this embodiment).

[0052] The dimension of the input features of each convolutional module is represented as [N×Cin×1×128]. Inside the convolutional module, the input features first go through a one-dimensional batch normalization, which is used to make the distribution of each channel of the input features more stable and accelerate the convergence of the model. Then it goes through the ReLU function activation to provide the convolutional module with the ability of non-linear representation. Finally, it is a one-dimensional convolutional layer, which has a large convolutional kernel of 1×15, a sliding step of 1, and the convolutional layer ensures that the length of the input features and the output features are the same through zero-padding operations at both ends. The number of output channels of all convolutional layers in the backbone network is fixed and is set to 16 in this embodiment, which represents the growth rate of the number of channels of the feature map every time it goes through a convolutional module. The dimension of the output feature map of the basic convolutional unit is [N×16×1×128]. It can be seen that there is only a difference in the number of channels between the input features and the output features, and the length and width of the features do not change, which can realize cross-layer connection in the subsequent channel splicing module. The dimension of the feature map after splicing of the last convolutional module is [N×(Cin + 16)×1×128]. In the DenseCNN backbone network, no pooling layer is used to shorten the width of the feature map to ensure that the finally output feature map of the network contains the features output by each convolutional module. Please refer to Table 1, which shows the changes in the feature dimensions output by each convolutional module. During the signal waveform feature extraction process, the features of the shallow convolution represent some basic features of the waveform, and its receptive field is small, while the deep features represent some semantic features, and its receptive field is larger. Through the form of cross-layer connection, the finally output feature map contains receptive field features of multiple scales, so the feature representation ability of the network is more abundant.

[0053] Table 1 Feature Dimensions Output by Convolutional Modules

[0054] Layer Name Dimension Input Layer N×2×1×128 BaseConvBlock1 Nx18×1×128 BaseConvBlock2 N×34×1x128 BaseConvBlock15 N×242×1×128

[0055] In some cases, it is necessary to change the number of channels of the output features of the DenseCNN backbone network to meet the requirements of the downstream Transformer network. For example, it is desired that the number of channels can be divisible by the number of heads to ensure that the feature dimensions of each head are the same, or it is necessary to increase or decrease the number of input feature channels to change the computational load of the downstream network. In the previous introduction of the backbone network, the number of channels of the final output can also be controlled by changing the growthrate parameter. However, it is found in experiments that there are many deficiencies in this control method. The number of channels of the final output of the backbone network is mainly determined by multiplying the growthrate and the number of basic convolutional units. Among them, the number of convolutional units corresponds to the model depth. It is hoped that the model can have sufficient depth to ensure the ability of non-linear mapping. The growthrate is the level at which each convolutional unit fuses the information of each channel to obtain a new expression channel. If this value is set too small, it will obviously inhibit the feature extraction ability of the model. If it is set too large, the number of channels of the final output will increase exponentially. Therefore, in this embodiment, the growthrate is fixed to a typical value, such as 16 for example. In this embodiment, the number of channels of the final output of the DenseCNN backbone network is flexibly changed by adding a Transition layer, and the Transition layer constitutes a dimensionality transformation network. The Transition layer is a 1×1 convolutional layer and a dimensionality transformation module. The convolutional layer can achieve arbitrary changes in the number of output channels, and at the same time can maximize the performance of retaining the local features extracted by the DenseCNN backbone network. The dimensionality transformation module transforms the multi-channel features of the DenseCNN backbone network into a time series feature.

[0056] Please refer to the attached Figure 4 , in this embodiment, three Transformer encoders 511 are stacked to form a time series feature extraction network, as the Transformer encoder 511. The schematic diagram of the network structure of the Transformer encoder 511 refers to Figure 3 . The main parameters set for the Transformer encoder 511 are the number of heads, the output dimension of each head, the Dropout dropout rate, and the parameters of the feed-forward layer. In Table 2, the change process of the feature dimensions and the parameter settings of the first Transformer encoder are given. The change processes of the feature dimensions and the parameter settings of the other two stacked Transformer encoders 511 are the same as those of the first Transformer encoder. In the feature dimensions, N represents the BatchSize size, and M represents the number of channels output in the Transition layer. Exemplarily, the value of M is 128.

[0057] Table 2 Feature Dimension Change Process and Parameter Settings of Transformer Encoder

[0058] Layer Name Parameter Description Output Feature Dimension CNN Feature Input Nx128×M Add Classification Label Nx129×M Add Position Embedding Nx129×M LayerNorml Nx129×M Multi-Head Dimension Split 2 heads, input feature dimension halved (N / 2)x129×(M / 2) Multi-Head Self-Attention Calculation Output dimension remains unchanged (N / 2)x129×(M / 2) Multi-Head Result Merging Nx129×M Node Suppression Suppression Rate 0.1 Nx129×M Residual Connection Nx129×M LayerNorm2 Nx129×M FF Linear Layer 1 + ReLU Dimension transformed to 1024 Nx129×1024 FFDropout Layer Dropout Probability 0.5 Nx129×1024 FF Linear Layer 2 Output dimension restored Nx129×M Residual Connection Nx129×M

[0059] The dual - task classifier 52 includes three parts, namely the DenseNet part, the Transformer part in the shared backbone network, and a multi - classifier head MDHC. The shared backbone network is based on the CTDNN model. The CTDNN model has high - performance feature extraction capabilities and flexible model scaling capabilities. In the design of the downstream multi - classifier head MDHC, in order to improve the performance of the two tasks, classifier heads are designed for the AMC task and the SEI task respectively. The two classifier heads map the shared features provided upstream into class probability vectors corresponding to the respective tasks. The output of the classification task is the one with the highest probability in the vector, representing the serial numbers in the pre - established set of transmitter modulation methods and the set of transmitter individuals.

[0060] Step S103) Extract shared features from the baseband signal sequence 40 through the shared feature extraction network 51. Input the baseband signal sequence 40 into the shared feature extraction network 51, and the output of the shared feature extraction network 51 is the shared feature. After being trained through the previous steps, the shared feature extraction network 51 has the ability to extract features.

[0061] Step S104) Classify the shared features through the dual - task classifier 52, and output the modulation mode recognition result and the radiation source recognition result respectively. The modulation mode recognition result is used as the result of the AMC task, and the radiation source recognition result is used as the result of the SEI task.

[0062] Among them, the dual - task classifier 52 includes a modulation mode classifier 521 and a radiation source classifier 522, and the output of the radiation source classifier 522 depends on the output of the modulation mode classifier 521.

[0063] In this embodiment, the features and radiation source markers of the known radiation source 60 are pre - recorded in the system where this method runs. After extracting the shared features of the baseband signal sequence 40, input the shared features into the modulation mode classifier 521 to obtain the modulation mode recognition result. Through the modulation mode recognition result, the signal of the radiation source 60 can be demodulated to obtain the information carried by the signal of the radiation source 60.

[0064] Suppose the transmitter has a total of U modulation methods and V different radiation sources 60. What the radiation source classifier 522 gives is the probability that the signal belongs to a certain radiation source 60 under the assumption of each modulation method. Therefore, the radiation source classifier 522 will output a total of U probabilities. The final output of the radiation source classifier 522 is selected according to the result of the modulation method classifier 521, that is, from the U probabilities, the probability corresponding to the modulation method recognition result of the modulation method classifier 521 is selected as the final output. That is, the probability that the signal belongs to the corresponding radiation source is the probability output by the radiation source classifier 522. When the probability output by the radiation source classifier 522 is not lower than the preset threshold, it indicates that the radiation source corresponding to the signal is a known radiation source and the information carried by the signal is credible.

[0065] When the probability output by the radiation source classifier 522 is lower than the preset threshold, it indicates that the radiation source corresponding to the signal is an unknown radiation source and the information carried by the signal is not credible. It may be an interference signal or an attack signal.

[0066] On the other hand, in another embodiment, the radiation source classifier 522 in the dual-task classifier 52 adopts a masking mechanism, and the masking mechanism screens the output of the radiation source classifier 522 according to the output of the modulation method classifier 521.

[0067] The masking mechanism is implemented through the following steps:

[0068] Generate a masking matrix according to the output of the modulation method classifier 521;

[0069] Perform a dot product of the masking matrix and the output matrix of the radiation source classifier 522 to obtain the screened radiation source recognition result.

[0070] The dimension of the masking matrix is U×V. Each row of the masking matrix corresponds to an element position of the output of the modulation method classifier 521. If the element at this position in the modulation method classifier 521 obtains the maximum value, the corresponding row of the masking matrix takes all 1s, otherwise it is all 0s. Please refer to the appendix Figure 5 , taking 3 as an example for both the values of U and V. After the vector output by the radiation source classifier 522 is rearranged, it also forms a U×V matrix. After performing a dot product with the masking matrix generated in the previous step and adding by column, the matrix will become a vector of length V, which is the final output result of the radiation source classifier 522.

[0071] On the other hand, in another embodiment, the multi-task classification model 50 is trained by a multi-task cross-entropy loss function, and the multi-task cross-entropy loss function includes a modulation method classification loss and a radiation source classification loss.

[0072] Among them, please refer to the appendixFigure 6 , the method for calculating the multi-task cross-entropy loss function includes:

[0073] Step S201) Calculate the modulation mode classification loss and the radiation source classification loss respectively;

[0074] Step S202) Add the modulation mode classification loss and the radiation source classification loss to obtain the total loss;

[0075] Step S203) Minimize the total loss through an optimization algorithm and update the parameters of the multi-task classification model 50.

[0076] Exemplarily, the loss function uses the cross-entropy loss. The optimization algorithm can adopt the gradient descent method to minimize the total loss function, thereby adjusting the model parameters to achieve better performance. In this embodiment, learning multiple related tasks simultaneously in one model is more efficient than training each task separately and can improve the generalization ability of the model to a certain extent.

[0077] On the other hand, this specification provides a radiation source identification system based on multi-task learning. Please refer to the attached Figure 7 , including:

[0078] A signal receiving module 100 that receives signals from multiple communication radiation sources, preprocesses the signals, and generates a baseband signal sequence 40;

[0079] A multi-task classification module 200, including a shared feature extraction network 51 and a dual-task classifier 52, which is used to extract shared features from the baseband signal sequence 40 and respectively output a modulation mode recognition result and a radiation source recognition result;

[0080] Among them, the dual-task classifier 52 includes a modulation mode classifier 521 and a radiation source classifier 522, and the output of the radiation source classifier 522 depends on the output of the modulation mode classifier 521.

[0081] Please refer to Figure 8 The structural schematic diagram of an electronic device provided by the embodiment of this specification shown.

[0082] As Figure 8As shown in the figure, the electronic device 1100 may include: at least one processor 1101, at least one network interface 1104, a user interface 1103, a memory 1105, and at least one communication bus 1102. Among them, the communication bus 1102 can be used to realize the connection and communication of the above-mentioned components. Among them, the user interface 1103 may include buttons, and the optional user interface may further include a standard wired interface and a wireless interface. Among them, the network interface 1104 may but is not limited to include a Bluetooth module, an NFC module, a Wi-Fi module, etc. Among them, the processor 1101 may include one or more processing cores. The processor 1101 connects various parts within the entire electronic device 1100 through various interfaces and lines, and executes various functions of the routing device 1100 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 1105, and calling the data stored in the memory 1105. Optionally, the processor 1101 may be implemented in at least one of the hardware forms of DSP, FPGA, and PLA. The processor 1101 may integrate one or a combination of several of CPU, GPU, and modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communication.

[0083] It can be understood that the above-mentioned modem may not be integrated into the processor 1101 and may be implemented separately through a single chip.

[0084] Among them, the memory 1105 may include RAM and may also include ROM. Optionally, the memory 1105 includes a non-transitory computer-readable medium. The memory 1105 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 1105 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned method embodiments, etc.; the data storage area may store the data involved in the above-mentioned method embodiments. Optionally, the memory 1105 may also be at least one storage device located far from the aforementioned processor 1101. The memory 1105, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and application programs. The processor 1101 may be used to call the application programs stored in the memory 1105 and execute the methods in the above-mentioned multiple embodiments.

[0085] The embodiments of this specification also provide a computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer or a processor, the computer or the processor is caused to execute multiple steps in the above embodiments. If each component module of the above electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in the computer-readable storage medium.

[0086] The embodiments of this specification also provide a computer program product, including a computer program. When the computer program is executed by a processor, multiple steps in the above embodiments are implemented.

[0087] Without conflict, the technical features in this embodiment and the implementation solutions can be combined arbitrarily.

[0088] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes multiple computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this specification are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server 30, or data center to another website, computer, server 30, or data center in a wired manner (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or a wireless manner (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server 30, a data center, etc. that integrates multiple available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a Digital Versatile Disc (DVD)), or a semiconductor medium (for example, a Solid State Disk (SSD)), etc.

[0089] When implemented by hardware or firmware, the foregoing method flow is programmed into a hardware circuit to obtain a corresponding hardware circuit structure and implement the corresponding functions. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is an integrated circuit whose logical function is determined by a user's programming of the device. A designer can program on their own to "integrate" a digital system onto a PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a hardware description language (HDL), and there are not only one but many kinds of HDLs. Those skilled in the art should also be clear that as long as the method flow is slightly logically programmed using the above-mentioned several hardware description languages and programmed into an integrated circuit, it is easy to obtain a hardware circuit that implements the logical method flow.

[0090] The embodiments described above are merely described in the preferred embodiment mode of this specification and do not limit the scope of this specification. Without departing from the design spirit of this specification, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of this specification shall fall within the protection scope determined by the claims of this specification.

Claims

1. A radiation source identification method based on multi-task learning, characterized in that, It includes the following steps: Receiving signals from multiple communication radiation sources, and preprocessing the signals to generate a baseband signal sequence; Inputting the baseband signal sequence into a pre-established multi-task classification model, where the multi-task classification model includes a shared feature extraction network and a dual-task classifier; Extracting shared features from the baseband signal sequence through the shared feature extraction network; Classifying the shared features through the dual-task classifier to respectively output a modulation mode recognition result and a radiation source recognition result; Wherein, the dual-task classifier includes a modulation mode classifier and a radiation source classifier, and the output of the radiation source classifier depends on the output of the modulation mode classifier.

2. The radiation source recognition method based on multi-task learning according to claim 1, wherein The shared feature extraction network includes a convolutional neural network and a self-attention encoder. The convolutional neural network outputs preset local features of the baseband signal sequence, and the self-attention encoder outputs preset global features of the baseband signal sequence.

3. The radiation source recognition method based on multi-task learning according to claim 1 or 2, wherein The radiation source classifier in the dual-task classifier adopts a masking mechanism, and the masking mechanism screens the output of the radiation source classifier according to the output of the modulation mode classifier.

4. The radiation source recognition method based on multi-task learning according to claim 3, wherein The masking mechanism is implemented through the following steps: Generating a masking matrix according to the output of the modulation mode classifier; Performing a dot product of the masking matrix and the output matrix of the radiation source classifier to obtain a screened radiation source recognition result.

5. The radiation source recognition method based on multi-task learning according to claim 1 or 2, wherein The multi-task classification model is trained through a multi-task cross-entropy loss function, and the multi-task cross-entropy loss function includes a modulation mode classification loss and a radiation source classification loss.

6. The radiation source recognition method based on multi-task learning according to claim 5, wherein The method for calculating the multi-task cross-entropy loss function includes: Respectively calculating the modulation mode classification loss and the radiation source classification loss; Adding the modulation mode classification loss and the radiation source classification loss to obtain a total loss; Minimizing the total loss through an optimization algorithm to update the parameters of the multi-task classification model.

7. A radiation source recognition system based on multi-task learning, characterized in that, It includes: A signal receiving module that receives signals from multiple communication radiation sources and preprocesses the signals to generate a baseband signal sequence; A multi-task classification module that includes a shared feature extraction network and a dual-task classifier, and is used to extract shared features from the baseband signal sequence and respectively output a modulation mode recognition result and a radiation source recognition result; Wherein, the dual-task classifier includes a modulation mode classifier and a radiation source classifier, and the output of the radiation source classifier depends on the output of the modulation mode classifier.

8. An electronic device, characterized in that, It includes a processor and a memory; The processor is connected to the memory; The memory is used to store executable program codes; The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, the method according to any one of claims 1-6 is implemented.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, the method according to any one of claims 1-6 is implemented.

Images