Radiation source blind identification method and device based on domain self-adaption, equipment and medium

Abstract

The invention provides a radiation source blind identification method, device and equipment based on domain self-adaption and a medium, and the method comprises the following steps: generating a sharp space frequency spectrum related to a radiation source through an A3C network; a self-adaptive sub-domain network layer is introduced, and signal data of a target domain are distinguished from the mixed signals; estimating the mixed signal by using an optimization function according to the spatial frequency spectrum peak information of the radiation source; based on a KNN network, according to the weight of the loss function and the estimation matrix, separating a radiation source signal from the signal data of the target domain; and comparing the observation value with the radiation source signal to calculate the accuracy rate of radiation source identification, and taking the accuracy rate as a final output result. According to the invention, based on the adaptive sub-domain network layer and the A3C algorithm, the radiation source signal can be identified more intelligently, meanwhile, based on the adaptive sub-domain network algorithm, the influence of noise on the original signal is reduced, and the radiation source identification precision and stability are significantly improved under the complex signal condition.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication signal processing, and in particular to a method, device, equipment and medium for blind identification of radiation sources based on domain adaptation. Background Art

[0002] With the rapid development of wireless communication technology, the number of radiation sources in the electromagnetic space has increased dramatically, the signal density has increased significantly, and the working mode and signal structure of the radiation sources have become increasingly diversified, resulting in extensive overlap between signals and even signal loss. The highly complex electromagnetic environment poses a severe test to the radiation source identification technology.

[0003] The radiation source identification methods known to the inventors usually rely on accurate measurement of signal characteristic parameters (such as carrier frequency, pulse width, amplitude, arrival time and azimuth, etc.) for matching and identification. However, in the current environment, the measurement errors of the multi-source intelligence acquisition channels and equipment of the signal monitoring system itself lead to increased uncertainty in the radiation source characteristic parameters stored in the signal database, which seriously affects the accuracy and efficiency of identification. In addition, in the case of rapid changes in noise ratio, the rapid changes in signal characteristics in complex environments and noise pollution make it difficult for methods such as energy detection based on fixed thresholds to adapt, and are easily disturbed by fluctuations in noise levels, resulting in a significant attenuation of identification performance. Summary of the invention

[0004] The present invention provides a domain-adaptive based radiation source blind identification method, device, equipment and medium, which solves the problem of insufficient identification accuracy of related radiation source identification technologies in highly complex electromagnetic environments.

[0005] In order to achieve the above purpose, this application adopts the following technical solutions:

[0006] In a first aspect, a method for blindly identifying a radiation source based on domain adaptation is provided, comprising the following steps:

[0007] Generate a sharp spatial spectrum about the radiation source through an A3C network; wherein the input of the A3C network is a symmetric covariance matrix, and the output is a probability matrix of mixed signals at different arrival angles;

[0008] An adaptive subdomain network layer is introduced, the mixed signal probability matrix is ​​input, each convolution layer in the adaptive subdomain network layer shares parameters, and the signal data of the target domain is distinguished from the mixed signal according to the subdomain adaptive function;

[0009] According to the spatial spectrum peak information of the radiation source output by the adaptive sub-domain network layer, the mixed signal is estimated by using an optimization function;

[0010] Based on the KNN network, the radiation source signal is separated from the signal data of the target domain according to the loss function and the weight of the estimation matrix;

[0011] By comparing the observed value with the radiation source signal, the accuracy of radiation source identification is calculated as the final output result.

[0012] In a first possible implementation manner of the first aspect, the symmetric covariance matrix is ​​r 1×J =[r 11 ,...,r 1M ,r 22 ,...r 2M ,...,r MM ], J = M (M + 1) / 2 is the dimension of the input matrix;

[0013] The features of the input matrix are extracted through different convolution kernels. Assume that the convolution kernel is Then the convolution operation of the i-th layer can be expressed as:

[0014] When the input matrix and the convolution kernel dimensions are different, automatic zero padding is performed; if the zero padding dimension is N p And the step length is N s , then i in z(i) satisfies

[0015] Among them, N h is the convolution kernel vector dimension.

[0016] Based on the first possible implementation manner of the first aspect, in a second possible implementation manner of the first aspect, the A3C network is configured to newly define the state space, action space, and reward function to achieve estimation and separation of mixed radiation sources:

[0017] State space: The input of the A3C network is r 1×J , there are two possible states of the unprocessed matrix, either the target signal takes the value of 0 or the target signal takes the value of 1. The state space is defined as S:

[0018] S={s=(s 1 ,s 2 ,…,s J )|s i ∈{0,1},i∈{1,2,…,J}};

[0019] Action space: The action is used to select the probability of the signal in a certain direction. Dividing 360° into K angles, the action space is defined as:

[0020] A={a=(a 1 ,a 2 ,…,aK )|a i ∈[0,1],i∈{1,2,…,K}};

[0021] Reward function: According to the recognition accuracy η t Set the expected recognition accuracy Reward value r t Set to:

[0022] When η t =1, r t The values ​​are as follows:

[0023]

[0024] Among them, R is the adjustable reward base, which can be adjusted according to the actual training situation; η t-1 Indicates the recognition accuracy at the last moment.

[0025] In a third possible implementation manner of the first aspect, in the adaptive subdomain network layer, an A distance index is used to measure the similarity between the mixed radiation source and the target radiation source, and is defined as: d A =2(1-2ε c ), where ε c Represented as the c-class subset of the training set;

[0026] Define the target domain X t and radiation source domain X s The maximum average difference is:

[0027]

[0028] According to the maximum average difference, the subdomain adaptation function is defined as:

[0029]

[0030] The objective function is:

[0031]

[0032] In a fourth possible implementation manner of the first aspect, separating the radiation source signal from the signal data of the target domain based on the KNN network according to the loss function and the weight of the estimation matrix includes:

[0033] By comparing the loss function to derive the network weight W, we can identify the target radiation source d j and the radiation source d output by the adaptive subdomain network layer i whether they belong to the same category;

[0034] The comparison loss function is defined as:

[0035]

[0036] Derivative of this formula yields:

[0037]

[0038] in,

[0039] In a fifth possible implementation manner of the first aspect, the accuracy of the radiation source identification is calculated by i,j and the observed value O t get:

[0040]

[0041] Among them, ⊙ is the exclusive OR operator.

[0042] In a second aspect, a radiation source blind identification device based on domain adaptation is provided, comprising:

[0043] A deep learning network architecture module, used to generate a sharp spatial spectrum about a radiation source through an A3C network; wherein the input of the A3C network is a symmetric covariance matrix, and the output is a probability matrix of mixed signals at different arrival angles;

[0044] An adaptive subdomain network layer module introduces an adaptive subdomain network layer for inputting the mixed signal probability matrix, each convolution layer in the adaptive subdomain network layer shares parameters, and distinguishes the signal data of the target domain from the mixed signal according to the subdomain adaptive function;

[0045] A mixed signal estimation module, used to estimate the mixed signal using an optimization function according to the spatial spectrum peak information of the radiation source output by the adaptive sub-domain network layer;

[0046] A signal separation module, used for separating the radiation source signal from the signal data of the target domain based on a KNN network and according to a loss function and a weight of an estimation matrix;

[0047] The judgment output module is used to calculate the accuracy of radiation source identification by comparing the observed value and the radiation source signal as the final output result.

[0048] In a first possible implementation manner of the second aspect, the deep learning network architecture module is specifically used to:

[0049] The features of the input matrix are extracted through different convolution kernels. Assume that the convolution kernel is Then the convolution operation of the i-th layer can be expressed as:

[0050] When the input matrix and the convolution kernel dimensions are different, automatic zero padding is performed; if the zero padding dimension is N p And the step length is N s , then i in z(i) satisfies

[0051] Among them, the symmetric covariance matrix is ​​r 1×J =[r 11 ,...,r 1M ,r 22 ,...r 2M ,...,r MM ], J = M (M + 1) / 2 is the dimension of the input matrix; N h is the convolution kernel vector dimension.

[0052] According to a third aspect, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for blind identification of radiation sources based on domain adaptation as described in the first aspect.

[0053] In a fourth aspect, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the blind radiation source identification method based on domain adaptation as described in the first aspect are implemented.

[0054] This application adopts blind identification of radiation sources based on adaptive networks, and based on adaptive subdomain network layers and A3C algorithms, improves the accuracy and robustness of radiation source identification; designs a specific A3C network algorithm based on the signal separation matrix, redefines the state space and action space in the A3C network, and can more intelligently identify radiation source signals. At the same time, based on the adaptive subdomain network algorithm, further reduces the impact of noise on the original signal and improves the performance of radiation source identification. In response to low signal-to-noise ratio conditions, even if the number of radiation sources is unknown, the radiation source can be well identified, further improving the accuracy of radiation source identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic flowchart of a method for blindly identifying a radiation source based on domain adaptation provided in an embodiment of the present application;

[0056] Figure 2 A schematic flowchart of another method for blindly identifying a radiation source based on domain adaptation provided in an embodiment of the present application;

[0057] Figure 3 A schematic diagram of the architecture of A3C provided in an embodiment of the present application;

[0058] Figure 4A schematic diagram of an adaptive network architecture provided in an embodiment of the present application;

[0059] Figure 5 A schematic structural block diagram of a radiation source blind identification device based on domain adaptation provided in an embodiment of the present application;

[0060] Figure 6 A schematic structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0061] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose, the technical solutions in the embodiments of the present application are clearly described. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present application.

[0062] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0063] The description of the method flow in the specification of the present application and the steps of the flowchart in the drawings of the present specification do not have to be strictly executed according to the step numbers, and the method steps can be executed in a different order. Moreover, some steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps.

[0064] In conjunction with the accompanying drawings and preferred embodiments, the domain-adaptive blind radiation source identification method, device, equipment and medium provided in the embodiments of the present application are described in detail below.

[0065] First, the application scenarios of the domain-adaptive blind radiation source identification method, device, equipment and medium of the embodiments of the present application are described in detail.

[0066] Among the related technologies known to the inventors, the most commonly used radiation source identification method is the characteristic parameter matching identification method, which mainly associates and matches the channel information obtained by the detection equipment, such as carrier frequency, pulse width, pulse amplitude, arrival time, azimuth, etc., with the characteristics in the target signal to realize the identification of the radiation source signal and then obtain the attribute information of the radiation source. The advantages of this method are simple algorithm, fast calculation speed and easy implementation.

[0067] However, with the increasing complexity of the electromagnetic environment, the radiation source identification method based on the five characteristic parameters no longer meets the needs of the signal monitoring system for fast and accurate identification. The multi-source intelligence acquisition channels of the signal monitoring system and the measurement errors of the equipment itself also bring challenges to the uncertainty of the radiation source characteristic parameters in the signal database, restricting the development of radiation source signal identification technology.

[0068] In order to solve the technical problems of optimizing the measurement of radiation source signal parameters, extracting new features, and comprehensively evaluating features in complex electromagnetic environments, the introduction of methods such as deep learning and reinforcement learning into radiation source identification provides new solutions to the above problems. For example, Zheng Boyuan et al. proposed a small sample radar radiation source identification method based on self-supervision and dual-stream fusion, which is significantly improved compared with the traditional one-dimensional feature method and the method based on long short-term memory (LSTM). Chen Shikai et al. proposed a self-expanding identification method for non-cooperative space radiation sources, which improved the classification accuracy of radiation source equipment. However, even with the support of deep learning technology, how to achieve efficient and accurate identification of radiation sources in situations where the noise level changes rapidly and the signal quality is unstable is still a key problem that needs to be solved.

[0069] Therefore, this application constructs a deep learning network architecture for blind identification of radiation sources, designs a complex A3C network algorithm based on the signal separation matrix, can more intelligently identify radiation source signals, and based on the adaptive subdomain network algorithm, further reduce the impact of noise on the original signal, and improve the performance of radiation source identification. In the case of low signal-to-noise ratio, even if the number of radiation sources is unknown, the radiation source can be well identified, further improving the accuracy of radiation source identification.

[0070] See also Figure 1-2 , the embodiment of the present application provides a method for blindly identifying radiation sources based on domain adaptation, such as Figure 1-2 As shown, the method for automatically generating preference data in the embodiment of the present application includes the following steps:

[0071] Step 110, generating a sharp spatial spectrum about the radiation source through an A3C network; wherein the input of the A3C network is a symmetric covariance matrix, and the output is a probability matrix of mixed signals at different arrival angles.

[0072] Firstly, a deep learning network architecture for blind identification of radiation sources is constructed. Through the A3C network, a sharp spatial spectrum of the radiation source is generated. The input is a symmetric covariance matrix, and the output is the mixed signal probability of each arrival angle.

[0073] Recover the radiation source signal from the mixed signal, aiming to find a signal separation matrix B P×M , according to the formula Separation estimation signal can be obtained

[0074] Considering the unknown number of radiation sources, the A3C network is used to estimate the mixed radiation source signal. The input of the network is the symmetric covariance matrix r 1×J =[r 11 ,...,r 1M ,r 22 ,...r 2M ,...,r MM ], J = M (M + 1) / 2 is the dimension of the input matrix. The symmetric covariance matrix is ​​a mathematical expression that describes the statistical characteristics of a signal.

[0075] See also Figure 3 The A3C (Asynchronous Advantage Actor-Critic) network extracts signal features and estimates the probability distribution of the radiation source signal through a multi-layer convolutional structure. The A3C network contains the following key components:

[0076] Global Network: This is the initial model shared by all parallel threads, including the strategy π(s) and the value function V(s), which is responsible for decision-making and value evaluation. The parameters of the global network are updated regularly to provide the latest model basis for each thread.

[0077] Threads: The figure shows threads 1 to n, each thread represents an independent exploration agent, each of which interacts with the environment. Each thread maintains a local copy of the policy π and value function V(s) network, and based on these copies interacts with its own environment to collect experience.

[0078] Environments: Environments 1 to n represent different simulated or real-world situations. Each thread interacts with an environment, collects states, performs actions, receives rewards and observations, and then feeds this data back to the local network for updating.

[0079] In this network, different convolution kernels are used to extract the features of the input matrix. Assume that the convolution kernel is Then the convolution operation of the i-th layer can be expressed as: When the input matrix and convolution kernel dimensions are different, automatic zero padding is performed. If the zero padding dimension is N p And the step length is N s , then i in z(i) satisfies Among them, N h is the convolution kernel vector dimension.

[0080] The mixed radiation sources are estimated and separated through the A3C network, and new definitions are given for the state space, action space, and reward function, thus achieving the estimation and separation of the mixed radiation sources. Specifically:

[0081] State space: Assume that the input of the A3C network is r 1×J , there are two possible states of the unprocessed matrix, the values ​​are 0 (not the target signal) and 1 (the target signal), and the state space is defined as S:

[0082] S={s=(s 1 ,s 2 ,…,s J )|s i ∈{0,1},i∈{1,2,…,J}}.

[0083] Action space: The action is used to select the probability of the signal in a certain direction. Dividing 360° into K angles, the action space is defined as:

[0084] A={a=(a 1 ,a 2 ,…,a K )|a i ∈[0,1],i∈{1,2,…,K}}.

[0085] Reward function: The reward function is not only related to the observation value O t , State S t 、Action a t It is also related to the recognition accuracy η at the previous moment. t-1 According to the recognition accuracy η t Set the expected recognition accuracy Reward value r t Set to:

[0086] When η t =1, r t The values ​​are as follows:

[0087]

[0088] Among them, R is an adjustable reward base, which can be adjusted according to the actual training situation.

[0089] Step 120, introducing an adaptive sub-domain network layer, inputting the mixed signal probability matrix, each convolution layer in the adaptive sub-domain network layer shares parameters, and distinguishing the signal data of the target domain from the mixed signal according to the sub-domain adaptive function.

[0090] This step introduces an adaptive subdomain network layer after the A3C network: the mixed signal probability matrix is ​​input, each convolution layer in the adaptive subdomain network layer shares parameters, and the signal data of the target domain is distinguished from the mixed signal according to the subdomain adaptive function.

[0091] The adaptive subdomain network layer is applied to separate the target signal data from the mixed signal through the convolution operation with shared parameters according to the adaptive function. The domain characteristics of the signal are used to specifically reduce the interference of non-target signals.

[0092] See also Figure 4 , the adaptive subdomain can better separate the target radiation source under different signal-to-noise ratios. The adaptive network framework shown in the figure shows how the input signal is processed by the convolutional layer and the fully connected layer to finally generate the loss function. In this process, there is parameter sharing between the convolutional layer and the fully connected layer. They can use the same parameters together to better capture the patterns and features in the input signal. Finally, by calculating the loss function, we can evaluate the performance of the network and make corresponding optimizations.

[0093] In this network, the A distance index is used to measure the similarity between the mixed radiation source and the target radiation source, which can be defined as: d A =2(1-2ε c ), where ε c Denote the c-class subset of the training set. Define the target domain X t (target domain, where pure signal data is located) and the radiation source domain X s The maximum average difference is:

[0094]

[0095] According to the maximum average difference, the subdomain adaptation function is defined:

[0096]

[0097] The objective function is:

[0098]

[0099] The role of the adaptive subdomain network layer is to further refine and purify the signal estimate obtained from the A3C network, and to significantly reduce the impact of noise and improve the quality of signal separation by dynamically adjusting to different signal-to-noise ratio conditions. Using the subdomain adaptive function as a method strategy, the signal data of the target domain is distinguished from the mixed signal to achieve the purpose of reducing noise.

[0100] Step 130: Estimate the mixed signal using an optimization function according to the spatial spectrum peak information of the radiation source output by the adaptive sub-domain network layer.

[0101] The purpose of this step is to estimate the mixed signal based on the optimization function: according to the peak value and number of the spatial spectrum of the radiation source output by the adaptive network layer, the optimization function is used to estimate the mixed signal.

[0102] According to the output of the adaptive subdomain network, the precise location and number of the radiation source in the mixed signal are estimated. By using optimization functions, such as minimizing the loss function, mathematical optimization methods are used to make the signal estimation closer to the true value, thereby enhancing the accuracy and stability of recognition.

[0103] Step 140 , based on the KNN network, the radiation source signal is separated from the signal data in the target domain according to the loss function and the weight of the estimation matrix.

[0104] This step is based on the KNN network to separate the signal: by comparing the loss function to derive the network weight W, the target radiation source d is identified. j and the radiation source d output by the adaptive subdomain network layer i Are they of the same category?

[0105] The comparison loss function is defined as:

[0106]

[0107] Derivative of this formula yields:

[0108]

[0109] in,

[0110] This step is the application of the k-nearest neighbors algorithm to analyze the optimized target signal data. KNN calculates the distance between the observed samples and the signal samples of known categories to find the nearest neighbors for classification, identification and separation of different signal sources.

[0111] Step 150, by comparing the observed value and the radiation source signal, the accuracy of radiation source identification is calculated as the final output result.

[0112] This step is the process of determining the output: that is, by calculating the observed values ​​and the optimized recognition results, the accuracy of radiation source recognition is obtained as the final output result.

[0113] The final radiation source identification accuracy η is calculated by i,j and O t get:

[0114]

[0115] Among them, ⊙ is the exclusive OR operator.

[0116] Compared with the related art known to the inventor, the present application adopts blind identification of radiation sources based on adaptive network, and based on adaptive subdomain network layer and A3C algorithm, improves the accuracy and robustness of radiation source identification. A specific A3C network algorithm is designed based on the signal separation matrix, and the state space and action space are redefined in the A3C network, which can more intelligently identify the radiation source signal. At the same time, based on the adaptive subdomain network algorithm, the influence of noise on the original signal is further reduced, and the performance of radiation source identification is improved. In response to low signal-to-noise ratio conditions, even if the number of radiation sources is unknown, the radiation source can be well identified, further improving the accuracy of radiation source identification.

[0117] See also Figure 5 , corresponding to the above-mentioned embodiment of the method for blind identification of radiation sources based on domain adaptation, the embodiment of the present application provides a device for blind identification of radiation sources based on domain adaptation, and the device for automatically generating preference data includes:

[0118] A deep learning network architecture module 1001 is used to generate a sharp spatial spectrum about a radiation source through an A3C network; wherein the input of the A3C network is a symmetric covariance matrix, and the output is a probability matrix of mixed signals at different arrival angles;

[0119] An adaptive subdomain network layer module 1002 is introduced to introduce an adaptive subdomain network layer, which is used to input the mixed signal probability matrix, each convolution layer in the adaptive subdomain network layer shares parameters, and distinguishes the signal data of the target domain from the mixed signal according to the subdomain adaptive function;

[0120] A mixed signal estimation module 1003, configured to estimate the mixed signal using an optimization function according to the spatial spectrum peak information of the radiation source output by the adaptive sub-domain network layer;

[0121] A signal separation module 1004 is used to separate the radiation source signal from the signal data of the target domain based on the KNN network according to the loss function and the weight of the estimation matrix;

[0122] The decision output module 1005 is used to calculate the accuracy of radiation source identification by comparing the observed value and the radiation source signal as the final output result.

[0123] The deep learning network architecture module 1001 is specifically used for:

[0124] The features of the input matrix are extracted through different convolution kernels. Assume that the convolution kernel is Then the convolution operation of the i-th layer can be expressed as:

[0125] When the input matrix and the convolution kernel dimensions are different, automatic zero padding is performed; if the zero padding dimension is N p And the step length is N s , then i in z(i) satisfies

[0126] Among them, the symmetric covariance matrix is ​​r 1×J =[r 11 ,...,r 1M ,r 22 ,...r 2M ,...,r MM ], J = M (M + 1) / 2 is the dimension of the input matrix; N h is the convolution kernel vector dimension.

[0127] The above-mentioned domain-adaptive-based blind identification device for radiation sources implements the steps and various processes of the above-mentioned domain-adaptive-based blind identification method embodiment for radiation sources, and can achieve the same technical effect. To avoid repetition, they will not be described here.

[0128] See also Figure 6 Corresponding to the above-mentioned embodiment of the blind identification method of radiation sources based on domain adaptation, an embodiment of the present application provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the above-mentioned embodiment of the blind identification method of radiation sources based on domain adaptation and the various processes of the embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0129] The memory 1009 can be used to store software programs and various data. The memory 1009 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory 1009 may include a volatile memory or a non-volatile memory, or the memory 1009 may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM). The memory 1009 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

[0130] The processor 1010 may include one or more processing units; optionally, the processor 1010 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 1010.

[0131] Corresponding to the above-mentioned embodiment of the blind identification method of radiation sources based on domain adaptation, the embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by the processor, the steps of the above-mentioned embodiment of the blind identification method of radiation sources based on domain adaptation and the various processes of the embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0132] The processor is the processor in the electronic device described in the above embodiment of the present application. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0133] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0134] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0135] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A blind radiation source identification method based on domain adaptation, characterized in that: The steps include: Generate a sharp spatial spectrum about the radiation source through an A3C network; wherein the input of the A3C network is a symmetric covariance matrix, and the output is a probability matrix of mixed signals at different arrival angles; An adaptive subdomain network layer is introduced, the mixed signal probability matrix is ​​input, each convolution layer in the adaptive subdomain network layer shares parameters, and the signal data of the target domain is distinguished from the mixed signal according to the subdomain adaptive function; According to the spatial spectrum peak information of the radiation source output by the adaptive sub-domain network layer, the mixed signal is estimated by using an optimization function; Based on the KNN network, the radiation source signal is separated from the signal data of the target domain according to the loss function and the weight of the estimation matrix; By comparing the observed value with the radiation source signal, the accuracy of radiation source identification is calculated as the final output result.

2. The method for blind identification of radiation sources based on domain adaptation according to claim 1, characterized in that: The symmetric covariance matrix is ​​r 1×J =[r 11 ,...,r 1M ,r 22 ,...r 2M ,...,r MM ], J = M (M + 1) / 2 is the dimension of the input matrix; The features of the input matrix are extracted through different convolution kernels. Assume that the convolution kernel is Then the convolution operation of the i-th layer can be expressed as: When the input matrix and convolution kernel dimensions are different, automatic zero padding is performed; if the zero padding dimension is N p And the step length is N s , then i in z(i) satisfies Among them, N h is the convolution kernel vector dimension.

3. The method for blind identification of radiation sources based on domain adaptation according to claim 2, characterized in that: Configure the A3C network to define the state space, action space, and reward function to estimate and separate mixed radiation sources: State space: The input of the A3C network is r 1×J , there are two possible states of the unprocessed matrix, either the target signal takes the value of 0 or the target signal takes the value of 1. The state space is defined as S: S={s=(s1,s2,…,s J )|s i ∈{0,1},i∈{1,2,…,J}}; Action space: The action is used to select the probability of the signal in a certain direction. Dividing 360° into K angles, the action space is defined as: A={a=(a1,a2,…,a K )|a i ∈[0,1],i∈{1,2,…,K}}; Reward function: According to the recognition accuracy η t Set the expected recognition accuracy Reward value r t Set to: When η t =1, r t The values ​​are as follows: Among them, R is the adjustable reward base, which can be adjusted according to the actual training situation; η t-1 Indicates the recognition accuracy at the last moment.

4. The method for blind identification of radiation sources based on domain adaptation according to claim 1, characterized in that: In the adaptive subdomain network layer, the A distance index is used to measure the similarity between the mixed radiation source and the target radiation source, and is defined as: d A =2(1-2ε c ), where ε c Represented as the c-class subset of the training set; Define the target domain X t and radiation source domain X s The maximum average difference is: According to the maximum average difference, the subdomain adaptation function is defined as: The objective function is:

5. The method for blind identification of radiation sources based on domain adaptation according to claim 1, characterized in that: The method of separating the radiation source signal from the signal data of the target domain based on the KNN network according to the loss function and the weight of the estimation matrix includes: By comparing the loss function to derive the network weight W, we can identify the target radiation source d j and the radiation source d output by the adaptive subdomain network layer i whether they belong to the same category; The comparison loss function is defined as: Derivative of this formula yields: in, 6. The method for blind identification of radiation sources based on domain adaptation according to claim 1, characterized in that: The accuracy of the radiation source identification is calculated by i,j and the observed value O t get: Among them, ⊙ is the exclusive OR operator.

7. A radiation source blind identification device based on domain adaptation, characterized in that: include: A deep learning network architecture module, used to generate a sharp spatial spectrum about a radiation source through an A3C network; wherein the input of the A3C network is a symmetric covariance matrix, and the output is a probability matrix of mixed signals at different arrival angles; An adaptive subdomain network layer module introduces an adaptive subdomain network layer, which is used to input the mixed signal probability matrix, each convolution layer in the adaptive subdomain network layer shares parameters, and distinguishes the signal data of the target domain from the mixed signal according to the subdomain adaptive function; A mixed signal estimation module, used to estimate the mixed signal using an optimization function according to the spatial spectrum peak information of the radiation source output by the adaptive sub-domain network layer; A signal separation module, used for separating the radiation source signal from the signal data of the target domain based on a KNN network and according to a loss function and a weight of an estimation matrix; The judgment output module is used to calculate the accuracy of radiation source identification by comparing the observed value and the radiation source signal as the final output result.

8. The radiation source blind identification device based on domain adaptation according to claim 7, characterized in that: The deep learning network architecture module is specifically used for: The features of the input matrix are extracted through different convolution kernels. Assume that the convolution kernel is Then the convolution operation of the i-th layer can be expressed as: When the input matrix and convolution kernel dimensions are different, automatic zero padding is performed; if the zero padding dimension is N p And the step length is N s , then i in z(i) satisfies Among them, the symmetric covariance matrix is ​​r 1×J =[r 11 ,...,r 1M ,r 22 ,...r 2M ,...,r MM ], J = M (M + 1) / 2 is the dimension of the input matrix; N h is the convolution kernel vector dimension.

9. An electronic device, characterized in that: The electronic device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for blind identification of radiation sources based on domain adaptation as described in any one of claims 1 to 6.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the method for blind identification of radiation sources based on domain adaptation as claimed in any one of claims 1 to 6 are implemented.

Images