Communication radio station individual identification method, device and equipment based on automatic deep learning

Abstract

The invention relates to a communication radio station individual identification method, device and equipment based on automatic deep learning, develops a set of method for solving a radio station individual identification problem by utilizing an automatic deep learning framework AutoKeras, remarkably reduces the manual participation cost through an automatic implementation mode, simplifies the working process of learning and identification, and improves the working efficiency. And the individual identification performance of the communication radio station is effectively and greatly improved. Through experimental verification, compared with a traditional neural network, a neural architecture more suitable for individual identification of the communication radio station can be obtained, so that more representative radio frequency features can be effectively extracted, and finally higher individual identification accuracy of the communication radio station is realized; and greater operation convenience and efficiency advantages can be brought to communication radio station identification in practical application.

Description

Technical Field

[0001] The present invention belongs to the technical field of radiation source target recognition, and relates to a method, device and equipment for individual recognition of communication stations based on automatic deep learning. Background Art

[0002] Radio station individual identification has important research significance in the field of communication. By studying the subtle characteristics of radio station signals, it can obtain important information such as information characteristics, signal format and target individual characteristics. This information is crucial for data fusion, decryption and decoding, pattern and target recognition and analysis, etc. For a long time, the recognition of communication signal fingerprint features has been of great significance. However, due to the extremely small individual differences of radio stations and the difficulty of feature extraction, the research on individual characteristics of communication radio stations is still in the preliminary research stage, and there is a large gap between the recognition effect and practical application.

[0003] At present, radio station individual identification already has mature identification methods based on statistics and neural networks. For example, by taking the characteristic parameters of the measured signal as a pattern vector and matching it with the established radar model database, the radar emitter can be identified. Another statistical method is the expert system method, which is to construct the inference rules of radar signal identification based on the professional knowledge provided by experts, and to infer and identify the radar emitter data. For example, the time domain characteristics of the radar emitter signal, such as instantaneous amplitude, instantaneous frequency and instantaneous phase, are extracted. It is found that these characteristics are rich in individual difference information and can be used as unintentional features for identification. For example, the individual characteristics of the communication emitter are extracted by wavelet transform. This method can capture the local changes and nonlinear characteristics of the signal. In recent years, deep learning methods have been widely used in the identification of individual emitters. These methods realize identification through signal preprocessing, feature extraction and classification, and use models such as convolutional neural network (CNN) and deep belief network (DBN). Most of the above-mentioned identification methods based on traditional neural network architecture are based on image structure, have poor adaptability and are relatively complex in model parameter optimization. There is a technical problem of insufficient accuracy of individual radio station identification. Summary of the invention

[0004] In response to the problems existing in the above-mentioned traditional methods, the present invention proposes a communication station individual identification method based on automatic deep learning, a communication station individual identification device based on automatic deep learning, and a computer device, which can greatly improve the accuracy of individual station identification.

[0005] In order to achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: On the one hand, a method for identifying individual communication stations based on automatic deep learning is provided, comprising the steps of: Get radio station signal data; Perform data preprocessing on the acquired radio signal data to generate a data set; the data set includes a training set and a test set; The open source AutoKeras architecture is used for network construction and processing to obtain the constructed automatic deep learning neural network model; the network construction process includes defining supervised learning channels, neural architecture search, neural network verification and feature extraction; Train the automatic deep learning neural network model on the training set; Automated deep learning neural network model hyperparameter optimization using Bayesian optimization; After model evaluation of the automatic deep learning neural network model on the test set, the individual communication stations corresponding to the station signal data are detected and output.

[0006] On the other hand, a communication station individual identification device based on automatic deep learning is also provided, comprising: A data acquisition module, used to acquire radio station signal data; The preprocessing module is used to perform data preprocessing on the acquired radio signal data to generate a data set; the data set includes a training set and a test set; The network building module is used to use the open source AutoKeras architecture to build the network and obtain the constructed automatic deep learning neural network model; the network building process includes defining the supervised learning channel, neural architecture search, neural network verification and feature extraction; The network training module is used to train the automatic deep learning neural network model on the training set; Parameter Optimization module, which is used to optimize the hyperparameters of automatic deep learning neural network models using Bayesian optimization; The evaluation output module is used to detect and output individual radio stations corresponding to the radio station signal data after the automatic deep learning neural network model is evaluated on the test set.

[0007] On the other hand, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned method for individual identification of communication stations based on automatic deep learning are implemented.

[0008] One of the above technical solutions has the following advantages and beneficial effects: The above-mentioned communication radio station individual identification method, device and equipment based on automatic deep learning, by developing a method using the automatic deep learning framework AutoKeras to solve the problem of radio station individual identification, significantly reduces the cost of manual participation through its automated implementation, simplifies the learning and identification workflow, and effectively improves the performance of communication radio station individual identification. After experimental verification, compared with traditional neural networks, a neural architecture that is more suitable for communication radio station individual identification can be obtained, so that more representative radio frequency features can be effectively extracted, and ultimately a higher communication radio station individual identification accuracy is achieved, and it can also bring greater operational convenience and efficiency advantages to communication radio station identification in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0010] Figure 1 It is a flowchart of a method for identifying individual communication stations based on automatic deep learning in one embodiment; Figure 2 A schematic diagram of a core process framework of an automatic deep learning neural network model in one embodiment; Figure 3 is a demodulated time domain image in one embodiment; Figure 4 A schematic diagram of the network structure of an automatic deep learning neural network model in one embodiment; Figure 5 is a schematic diagram of a confusion matrix in one embodiment; Figure 6 A schematic diagram showing a comparison of the accuracy of three models after training in an embodiment; Figure 7 A schematic diagram of a visualization result after the network automatically extracts features and reduces their dimensionality in one embodiment; Figure 8 A schematic diagram of a visualization result after manually extracting features and reducing their dimensionality in one embodiment; Fig. 9 Schematic diagram of the module structure of a communication station individual identification device based on automatic deep learning in one embodiment. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and Examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0012] It should be noted that the reference to "embodiment" in this article means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present invention. The presentation of this phrase at various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It will be appreciated by those skilled in the art that the embodiments described herein may be combined with other embodiments. The term "and / or" used in the specification of the present invention and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0013] The following will describe the implementation of the present invention in detail with reference to the accompanying drawings in the embodiment diagram of the present invention.

[0014] Automated Machine Learning (AutoML) is a concept proposed in recent years. It can automatically implement data preprocessing, feature construction, feature selection, algorithm selection, and hyperparameter tuning. Compared with traditional neural network architectures, automatic deep learning models have great advantages.

[0015] Specifically, the automatic deep learning model can automatically clean data, remove abnormal data and perform preprocessing; it can select and build appropriate features, and automatically explore how to automatically extract, select and transform data features to enhance the performance of the model; it automatically explores and optimizes the neural network structure through algorithms, searches and evaluates a large number of existing network structures, and automatically discovers efficient neural network architectures in a similar way to stacking blocks; in the process of automatic hyperparameter optimization, Bayesian optimization, grid search, and random search can be used to automatically adjust hyperparameters to improve the performance and generalization ability of the model. For certain specific scenarios, automatic deep learning can almost fully automatically implement processes such as data cleaning, feature selection, modeling, hyperparameter optimization, and model evaluation.

[0016] In one embodiment, Figure 1 As shown, a method for identifying individual communication stations based on automatic deep learning is provided, which may include the following processing steps S10 to S20: S10, obtaining radio station signal data; S12, performing data preprocessing on the acquired radio station signal data to generate a data set; the data set includes a training set and a test set; S14, using the open source AutoKeras architecture to build the network and obtain the constructed automatic deep learning neural network model; the network building process includes defining the supervised learning channel, neural architecture search, neural network verification and feature extraction; S16, training the automatic deep learning neural network model on the training set; S18, Hyperparameter optimization of automatic deep learning neural network models using Bayesian optimization; S20, after evaluating the automatic deep learning neural network model on the test set, detect and output the communication radio station individual corresponding to the radio station signal data.

[0017] It can be understood that the open source AutoKeras architecture is used in this embodiment to build and train the neural network, and its core process framework can be as follows: Figure 2 shown.

[0018] First, when using the open source AutoKeras architecture for network construction, first define the supervised learning pipeline: The supervised learning pipeline creation in this embodiment refers to finding a pipeline structure, as well as the algorithms that constitute the pipeline and the hyperparameter combination of the algorithms to minimize the loss. The formula of the loss function is as follows:

[0019] in, Indicates that in the training set The pipeline created on In the test set The loss value for validation on represents the weight parameter of the model, represents the model hyperparameters, represents the regularization parameter, k Indicates the number of training sets or test sets.

[0020] For a machine learning pipeline , represents the feature vector, represents the corresponding target value, . Given a dataset And the loss metric function L , we can get Performance:

[0021] in, Represents a pipeline For a given sample The predicted output.

[0022] After defining the machine learning pipeline, perform a neural architecture search (NAS) on the network to be built. The first step is to define the search space and determine the scope of the neural architecture search. You can set the search space according to the actual search requirements according to the guidance of the AutoKeras architecture. The second step is to select a suitable search strategy and use the Efficient Neural Network Architecture Search (ENAS) strategy through the AutoKeras architecture to automatically design the neural network model. ENAS is a technology that improves the efficiency of neural architecture search by sharing parameters. The formula for training shared parameters is as follows:

[0023] in, m is from (One parameter The model sampled from the probability distribution of the control) is shown on the right. The unbiased formula for the gradient is: represents shared parameters, Represents the architecture parameters, M represents the number of models sampled, Indicates the parameters w The gradient of Indicates that the model has parameters The loss function under . Update the weights first w , and then fix the weight w , so that expectations The largest, among which, For the model in parameters The reward function under the test set. Repeating the above two steps can make the model The accuracy on the training set is the best, and the optimal neural architecture search is realized.

[0024] After the neural architecture search is completed, that is, after the neural network model is built, the neural network model needs to be verified. Robustness is usually used as an evaluation indicator of the anti-interference performance of the neural network model. Robustness refers to the ability of the neural network model to maintain good performance when facing disturbances, noise or outliers in the input data. The robustness boundary is usually used to quantify the robustness of the model, and the corresponding formula is as follows:

[0025] in, represents the sample space, represents the selected sample, Representative The model with parameters is Represents the loss function. The smaller this value is, the less sensitive the neural network model is to input disturbances, that is, the higher the robustness is, and the better the network performance is. After the neural network model is verified to meet the set performance requirements, it can be used as the constructed automatic deep learning neural network model, and the training set data can be input into the automatic deep learning neural network model to train the model.

[0026] There are usually two types of parameters in machine learning models. One type is the parameters that can be learned and estimated from the data; the other type of parameters is difficult to estimate from the data and needs to be designed and specified based on expert experience. The latter are called hyperparameters, such as learning rate, number of neural network layers, number of neurons, regularization parameters, choice of activation function, choice of optimization algorithm, and Dropout ratio (refers to randomly deleting a certain proportion of neurons in the neural network to prevent overfitting and improve the generalization ability of the model).

[0027] What algorithm can be used in the search space to quickly and accurately find the optimal hyperparameter configuration is what the search strategy does, which includes the classic exploration-exploitation trade-off. On the one hand, it is necessary to quickly find a hyperparameter configuration with good performance, and on the other hand, it should avoid converging to the suboptimal area too early.

[0028] There are two main methods for AutoML to implement hyperparameter search optimization: grid search and Bayesian optimization. Grid search automatically defines a complete parameter combination grid within the network based on the classification task. For each parameter combination in the grid, cross-validation is used to evaluate the performance of the model. For each hyperparameter combination, performance indicators (such as accuracy) are calculated to evaluate the stability and effectiveness of the model. Bayesian optimization uses statistical models (usually Gaussian processes) to approximate the objective function, and automatically trains the network based on the existing hyperparameter configuration and the corresponding performance evaluation results through a proxy model, using historical information to guide the search, thereby more intelligently selecting new hyperparameter combinations.

[0029] In comparison, in automatic deep learning, Bayesian optimization automatic search is more commonly used within the network. Its core idea is to regard the objective function as a random process and model it using a Gaussian process. In the optimization process, the Gaussian process must first be defined:

[0030] in, is the prior mean of the function, As the kernel function, you can choose the square exponential kernel and the radial basis function kernel. The calculation formula of the kernel function is as follows:

[0031] in, l is the length scale parameter.

[0032] By minimizing the negative improvement of the posterior mean NI Find the parameters that minimize the objective function x , the corresponding formula is as follows:

[0033] in, Indicates at point x The posterior mean at , which is an estimate of the expected value of the objective function at that point, Indicates at point x The posterior standard deviation at , which measures the uncertainty of the objective function at that point, It is a hyperparameter used to balance exploration and utilization. The optimization of each hyperparameter can be completed through the above-mentioned Bayesian optimization, so as to obtain a more complete automatic deep learning neural network model, which can effectively enhance the effect of individual recognition of communication stations.

[0034] Finally, a model evaluation was performed: The confusion matrix is ​​a table used to evaluate the performance of a classification model in supervised learning. It is suitable for classification tasks. It compares the model's prediction results with the actual category labels to show the model's classification of each category. The rows of the confusion matrix represent the true labels and the columns represent the predicted labels.

[0035] In this embodiment, the confusion matrix output by the automatic deep learning neural network model on the test set and the known category label distribution can be used to directly calculate the recall rate of each category label predicted by the model and obtain the overall accuracy of the model. The recall rate refers to the proportion of samples correctly identified by the system among all relevant samples, reflecting the probability that each category of the model is correctly identified; while the accuracy rate refers to the proportion of correct samples to total samples, measuring the overall recognition performance of the model for the radio station. Furthermore, the calculation formulas for the recall rate and accuracy rate of each communication radio station are as follows:

[0036]

[0037] in, TP Representation sample i The number of correct predictions, TN Indicates that other samples are predicted as samples i The number of FN Indicates that the sample i The number of other samples predicted, FP Indicates the number of correct predictions for other samples.

[0038] The above-mentioned communication radio station individual identification method based on automatic deep learning has developed a method to solve the problem of individual radio station identification using the automatic deep learning framework AutoKeras. Its automated implementation significantly reduces the cost of manual participation, simplifies the learning and identification workflow, and effectively improves the performance of individual communication radio station identification. Experimental verification shows that compared with traditional neural networks, a neural architecture that is more suitable for individual communication radio station identification can be obtained, so that more representative RF features can be effectively extracted, and ultimately a higher accuracy rate for individual communication radio station identification is achieved. It can also bring greater operational convenience and efficiency advantages to communication radio station identification in practical applications.

[0039] In one embodiment, in order to more intuitively demonstrate the implementation process and effect of the above-mentioned communication station individual identification method based on automatic deep learning, one of the experimental examples is given. Those skilled in the art can understand that the experimental example is only a schematic auxiliary explanation and is not the only limitation on the implementation method of the communication station individual identification method based on automatic deep learning.

[0040] Experimental setup: The experimental environment was built by deploying four communication radios of the same model to complete the collection of experimental data sets. The radio data details are shown in Table 1: Table 1

[0041] After 8PSK demodulation, we get Figure 3 The time domain image shown, the horizontal axis X is time, the unit is seconds; the vertical axis Y is amplitude, the unit is volts or dimensionless, through 8PSK demodulation technology, the received radio signal is divided into I-way (co-directional) and Q-way (orthogonal), and the modulus values ​​of these two channels are extracted. The IQ two-way is spliced ​​together to form a signal matrix similar to the picture, so as to facilitate subsequent network processing. In the training stage, this embodiment divides 2000 training samples and 400 test samples, and uses the AutoKeras architecture to automatically build and optimize the deep learning network.

[0042] While using the AutoKeras architecture to build an automatic deep learning neural network model, we used the traditional ResNet (residual) network and CNN network as experimental groups, input the training set into the built neural network, updated the network parameters, and used the test set to verify the model. The output accuracy of the test set was compared with that of the automatic deep learning neural network model. The network structure of the automatic deep learning neural network model can be as follows: Figure 4As shown in the figure, from the network structure diagram, we can see that the automatic deep learning neural network model first upgrades the dimension of the input image and standardizes it, converts the data into a form that the network can process, and ensures that the input data is in an appropriate numerical range; the RandomFlip layer randomly flips the input data with a certain probability during the training process to enhance data diversity and reduce overfitting; the Reisizing layer adjusts the size of the input data to make it meet the input requirements of the neural network model; the Concatenate layer combines the features of the previous layer output into a tensor for further processing by subsequent layers; then it passes through the residual layer (ResNet-50 network) and the pooling layer, and through the fully connected Dense layer and the appropriate activation function (such as the relu function) to generate the output tensor, increase the network's expressive power, and enable the neural network to learn and capture more complex patterns and relationships.

[0043] In addition, in order to save training time and resources, the AutoKeras architecture implements an early stopping strategy, that is, terminating certain training tasks early when it is observed that the model performance is no longer improving. This strategy can be dynamically adjusted according to the model's training curve to improve efficiency and performance. After model training, the confusion matrix is ​​as follows Figure 5 As shown, the rows represent the true labels of the radio stations and the columns represent the radio station labels predicted by the model.

[0044] The training set is input into the automatic deep learning neural network model, and after several rounds of training, the trained updated weights are used to verify the test set. The construction process of the residual network is similar to that of the convolutional network. Figure 6 This is a comparison of the test accuracy of the three models after training. It can be found from the figure that the CNN network has the lowest test accuracy, which means that ordinary convolutional networks are not suitable for the task of radio individual identification; the residual network Resnet has a higher test accuracy, but it is difficult to find hyperparameters suitable for the task through manual parameter adjustment; the automatic deep learning neural network model based on the AutoKeras architecture automatically returns the most suitable hyperparameters for the task through continuous parameter adjustment, so it has the highest accuracy.

[0045] After calculation, the recall rate of each sample in the original data is shown in Table 2: Table 2

[0046] From Table 2, we can see that the recall rate of Radio 0 and Radio 1 is 1.00, which means that the model recognizes all samples of Radio 0 and Radio 1 well.

[0047] The basic principle of communication station fingerprint recognition is to extract the equipment differences of communication stations from the communication signals to complete the identification of the transmitter of the communication radiation source or the target individual associated with it. This embodiment selects to extract the output of the penultimate layer of the trained automatic deep learning neural network model as the fingerprint feature, and uses the dimensionality reduction algorithm t-SNE to reduce the dimension and display it. t-SNE is a nonlinear technology for data dimensionality reduction and visualization. It realizes dimensionality reduction and visualization of high-dimensional data by calculating the similarity between each pair of data points. The visualization result of the radio station feature extracted by the neural network after dimensionality reduction is shown in the figure below. Figure 7 As shown in the figure, another set of RF features based on wavelet matrix is ​​visualized after t-SNE dimension reduction. Figure 8 shown.

[0048] Figure 7 and Figure 8 Shown are two commonly used feature extraction methods. Figure 7 The fingerprint corresponding to the output features of the penultimate layer of the trained neural network is extracted. Figure 8 Corresponding to manual extraction of RF features. The theory of extracting the output features of the penultimate layer of the neural network as fingerprints is based on the hierarchical structure of feature representation and the principle of information encoding in the deep learning model. According to the graphical effect, it can be concluded that the features automatically extracted by the network can well identify the slight differences between each radio station and serve as the RF fingerprint of each radio station; while the manually extracted features represented by wavelet transform lose a lot of original data information. Figure 8 The scatter points are rather scattered, and it is not possible to distinguish the radio stations well.

[0049] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of the steps, and the steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0050] In one embodiment, Fig. 9As shown, a communication station individual identification device 100 based on automatic deep learning is also provided, including a data acquisition module 11, a preprocessing module 13, a network construction module 15, a network training module 17, a parameter optimization module 19 and an evaluation output module 21. Among them, the data acquisition module 11 is used to obtain station signal data. The preprocessing module 13 is used to perform data preprocessing on the acquired station signal data to generate a data set; the data set includes a training set and a test set. The network construction module 15 is used to use the open source AutoKeras architecture to perform network construction processing to obtain the constructed automatic deep learning neural network model; the network construction processing includes defining supervised learning channels, neural architecture search, neural network verification and feature extraction. The network training module 17 is used to train the automatic deep learning neural network model on the training set. The parameter optimization module 19 is used to optimize the hyperparameters of the automatic deep learning neural network model using Bayesian optimization. The evaluation output module 21 is used to detect and output the radio station individual corresponding to the radio signal data after the automatic deep learning neural network model is evaluated on the test set.

[0051] The above-mentioned communication station individual identification device 100 based on automatic deep learning has developed a method for solving the problem of individual station identification using the automatic deep learning framework AutoKeras. Its automated implementation method significantly reduces the cost of manual participation, simplifies the learning and identification workflow, and effectively improves the performance of individual communication station identification. Experimental verification shows that compared with traditional neural networks, a neural architecture that is more suitable for individual communication station identification can be obtained, so that more representative radio frequency features can be effectively extracted, and ultimately a higher accuracy rate for individual communication station identification is achieved. It can also bring greater operational convenience and efficiency advantages to communication station identification in practical applications.

[0052] In one embodiment, the recall rate of each communication station is:

[0053] The accuracy of each communication station is:

[0054] in, TP Representation sample i The number of correct predictions, TN Indicates that other samples are predicted as samples i The number of FN Indicates that the sample i The number of other samples predicted, FP Indicates the number of correct predictions for other samples.

[0055] For the specific definition of the communication station individual identification device 100 based on automatic deep learning, please refer to the definition of the communication station individual identification method based on automatic deep learning in the above text, which will not be repeated here. Each module in the above-mentioned communication station individual identification device 100 based on automatic deep learning can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0056] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the following processing steps when executing the computer program: obtaining radio station signal data; performing data preprocessing on the obtained radio station signal data to generate a data set; the data set includes a training set and a test set; using the open source AutoKeras architecture to perform network construction processing to obtain a constructed automatic deep learning neural network model; the network construction processing includes defining supervised learning channels, neural architecture search, neural network verification and feature extraction; training the automatic deep learning neural network model on the training set; optimizing the hyperparameters of the automatic deep learning neural network model using Bayesian optimization; after performing model evaluation on the automatic deep learning neural network model on the test set, detecting and outputting the communication radio station individuals corresponding to the radio signal data.

[0057] It can be understood that in addition to the memory and processor mentioned above, the above-mentioned computer device may also include other software and hardware components not listed in this specification. The specific details can be determined according to the model of the specific communication identification device in different application scenarios, and this specification will not list them one by one in detail.

[0058] In one embodiment, when the processor executes the computer program, it can also implement the steps or sub-steps added in each embodiment of the above-mentioned communication station individual identification method based on automatic deep learning.

[0059] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following processing steps are implemented: obtaining radio station signal data; performing data preprocessing on the obtained radio station signal data to generate a data set; the data set includes a training set and a test set; using the open source AutoKeras architecture to perform network construction processing to obtain a constructed automatic deep learning neural network model; the network construction processing includes defining supervised learning channels, neural architecture search, neural network verification and feature extraction; training the automatic deep learning neural network model on the training set; using Bayesian optimization to optimize the hyperparameters of the automatic deep learning neural network model; after performing model evaluation on the automatic deep learning neural network model on the test set, detecting and outputting the communication radio station individuals corresponding to the radio signal data.

[0060] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus dynamic random access memory (RambusDRAM, referred to as RDRAM) and interface dynamic random access memory (DRDRAM).

[0061] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0062] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of protection of the invention. It should be pointed out that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the attached claims.

Claims

1. A method for identifying individual communication stations based on automatic deep learning, characterized in that: Includes steps: Get radio station signal data; Performing data preprocessing on the acquired radio signal data to generate a data set; the data set includes a training set and a test set; The open source AutoKeras architecture is used to perform network construction processing to obtain the constructed automatic deep learning neural network model; the network construction processing includes defining supervised learning channels, neural architecture search, neural network verification and feature extraction; Training the automatic deep learning neural network model on the training set; Optimizing hyperparameters of the automatic deep learning neural network model using Bayesian optimization; After performing model evaluation on the automatic deep learning neural network model on the test set, the communication radio station individual corresponding to the radio station signal data is detected and output.

2. The method for identifying individual communication stations based on automatic deep learning according to claim 1, characterized in that: The recall rates of each communication station are: The accuracy of each communication station is: in, TP Representation sample i The number of correct predictions, TN Indicates that other samples are predicted as samples i The number of FN Indicates that the sample i The number of other samples predicted, FP Indicates the number of correct predictions for other samples.

3. A communication station individual identification device based on automatic deep learning, characterized in that: include: A data acquisition module, used to acquire radio station signal data; A preprocessing module is used to perform data preprocessing on the acquired radio signal data to generate a data set; The data set includes a training set and a test set; A network building module, which is used to use the open source AutoKeras architecture to perform network building processing and obtain the constructed automatic deep learning neural network model; the network building processing includes defining supervised learning channels, neural architecture search, neural network verification and feature extraction; A network training module, used for training the automatic deep learning neural network model on the training set; A parameter optimization module, used for optimizing hyperparameters of the automatic deep learning neural network model using Bayesian optimization; An evaluation output module is used to detect and output individual radio stations corresponding to the radio station signal data after performing model evaluation on the automatic deep learning neural network model on the test set.

4. The communication station individual identification device based on automatic deep learning according to claim 3 is characterized in that: The recall rates of each communication station are: The accuracy of each communication station is: in, TP Representation sample i The number of correct predictions, TN Indicates that other samples are predicted as samples i The number of FN Indicates that the sample i The number of other samples predicted, FP Indicates the number of correct predictions for other samples.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the communication station individual identification method based on automatic deep learning described in claim 1 or 2 are implemented.

Images