Model training method and device, equipment, storage medium and program product

Abstract

The invention discloses a model training method and device, equipment, a storage medium and a program product, and relates to the technical field of computers, and the model training method comprises the steps that a Wi-Fi signal data set of communication equipment is acquired, the Wi-Fi signal data set comprises an auxiliary data set, and the auxiliary data set is used for processing the Wi-Fi signal data set; signal samples in the auxiliary data set are obtained by performing signal phase rotation on Wi-Fi signals, and each signal sample has a rotation label value; inputting the auxiliary data set into a preset neural network model for signal rotation prediction to obtain signal features and model parameters; loss values of the signal features and the model parameters are determined, the neural network model is updated according to the loss values, an equipment identification model is obtained, and the equipment identification model is used for identifying the type of the communication equipment. According to the method and the device, the dependence on the marked data is reduced, and the detection efficiency of the Wi-Fi signal and the identification and authentication precision of the equipment are improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a model training method, device, equipment, storage medium and program product. Background Art

[0002] With the development of the sixth generation of mobile communication technology, network security issues have become increasingly prominent. Specific Emission Identification (SEI) technology can identify and authenticate communication devices based on the RF fingerprint of the extracted signal. Existing specific emission identification methods are mainly based on deep learning and rely on a large amount of labeled data. In scenarios where the labeled data is limited, the accuracy of communication device identification and authentication is low. Summary of the invention

[0003] The main purpose of this application is to provide a model training method, device, equipment, storage medium and program product, aiming to solve the technical problem that the existing specific radiation source identification method is mainly based on deep learning and relies on a large amount of labeled data. In scenarios where the labeled data is limited, the identification and authentication accuracy of communication equipment is low.

[0004] To achieve the above objectives, the present application proposes a model training method, which includes:

[0005] Acquire a Wi-Fi signal dataset of a communication device, where the Wi-Fi signal dataset includes an auxiliary dataset, where signal samples in the auxiliary dataset are obtained by performing a signal phase rotation on the Wi-Fi signal, and each of the signal samples has a rotation label value;

[0006] Inputting the auxiliary data set into a preset neural network model to perform signal rotation prediction to obtain signal features and model parameters;

[0007] Determine the loss values ​​of the signal characteristics and the model parameters, and update the neural network model according to the loss values ​​to obtain a device identification model, wherein the device identification model is used to identify the category of the communication device.

[0008] In one embodiment, the step of obtaining a Wi-Fi signal data set of a communication device includes:

[0009] Encode the original samples in sequence and determine the instance number label value of each original sample;

[0010] Performing a preset number of signal phase rotations on each original sample according to the instance number label value, and generating a time-frequency diagram based on the samples after the signal phase rotation;

[0011] The rotation label value corresponding to the time-frequency diagram is obtained, and the auxiliary data set is generated according to the time-frequency diagram and the rotation label value.

[0012] In one embodiment, the Wi-Fi signal data set includes a target data set, each signal sample in the target data set has a device category label value, and after the step of determining the signal feature and the loss value of the model parameter, and updating the neural network model according to the loss value to obtain the device recognition model, the method further includes:

[0013] Inputting the target data set into the device identification model with a new classifier to identify the device category corresponding to the signal sample;

[0014] The device identification model is updated according to the identified device category and the device category tag value, and the updated device identification model is saved.

[0015] In one embodiment, the signal features include model encoding features, high-dimensional mean features, and low-dimensional mean features, the model parameters include model prediction values ​​and model weights, and the step of inputting the auxiliary data set into a preset neural network model for signal rotation prediction to obtain signal features and model parameters also includes:

[0016] The auxiliary data set is input into a neural network model, a feature encoder of the neural network model extracts image coding features of a preset dimension, and divides the image coding features of the preset dimension into image coding features of at least two sub-dimensions;

[0017] Determine a high-dimensional mean feature according to an average value of the image encoding features of at least two sub-dimensions;

[0018] Mapping the high-dimensional mean feature to a low-dimensional space to obtain a low-dimensional mean feature;

[0019] Obtain the image coding features of the neural network model based on preset dimensions, the predicted model prediction value and the associated model weight value.

[0020] In one embodiment, the signal features include model encoding features, high-dimensional mean features, and low-dimensional mean features, the model parameters include model prediction values ​​and model weights, and the step of determining the loss values ​​of the signal features and the model parameters includes:

[0021] Obtaining a classification loss value according to a cross entropy loss function, the model prediction value, and the rotation label value;

[0022] Obtaining a mean square error loss value according to the mean square error function, the high-dimensional mean feature and the model encoding feature of the sub-dimension;

[0023] A noise contrast estimation loss value is obtained according to the noise contrast estimation loss function, the low-dimensional mean feature and the instance number label value.

[0024] In one embodiment, the step of updating the neural network model according to the loss value to obtain the device identification model includes:

[0025] Obtaining an adaptive loss weight, weighting and accumulating the cross entropy loss value, the mean square error loss value, and the noise contrast estimation loss value respectively, to obtain a joint loss value;

[0026] Perform back propagation based on the joint loss value to obtain the gradient;

[0027] The parameters of the neural network model are updated according to the gradient until a device recognition model is obtained.

[0028] In addition, to achieve the above objectives, the present application also proposes a model training device, which includes:

[0029] an acquisition module, configured to acquire a Wi-Fi signal dataset of a communication device, wherein the Wi-Fi signal dataset includes an auxiliary dataset, wherein signal samples in the auxiliary dataset are obtained by performing a signal phase rotation on the Wi-Fi signal, and each of the signal samples has a rotation label value;

[0030] A prediction module, used for inputting the auxiliary data set into a preset neural network model to perform signal rotation prediction to obtain signal features and model parameters;

[0031] An updating module is used to determine the loss value of the signal characteristics and the model parameters, and update the neural network model according to the loss value to obtain a device identification model, wherein the device identification model is used to identify the category of the communication device.

[0032] In addition, to achieve the above-mentioned purpose, the present application also proposes a model training device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the model training method described above.

[0033] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the model training method described above are implemented.

[0034] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the model training method described above.

[0035] One or more technical solutions proposed in this application have at least the following technical effects:

[0036] Obtain the Wi-Fi signal data set of the communication device; input the auxiliary data set into the preset neural network model for signal rotation prediction to obtain signal features and model parameters; determine the loss value of the signal features and model parameters, and update the neural network model according to the loss value to obtain the device identification model, which is used to identify the category of the communication device. The neural network model learns the decoupled representation of the signal phase rotation correlation and rotation independence from the auxiliary data set, thereby extracting the signal features of the RF fingerprint that are more robust and more beneficial to the identification of specific radiation sources, and then completes the identification and authentication of the Wi-Fi device based on the extracted signal features, reducing the dependence on the labeled data, improving the detection efficiency of the Wi-Fi signal and the accuracy of device identification and authentication in sample-restricted scenarios, and improving the versatility in different environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0039] Figure 1 A flowchart of the first embodiment of the model training method of the present application is provided;

[0040] Figure 2 A schematic diagram of the model training process of the first embodiment of the model training method of the present application;

[0041] Figure 3 A flowchart of the second embodiment of the model training method of the present application is provided;

[0042] Figure 4 A flowchart of the third embodiment of the model training method of the present application is provided;

[0043] Figure 5 A flowchart of the fourth embodiment of the model training method of the present application is provided;

[0044] Figure 6 A flowchart of the fifth embodiment of the model training method of the present application is provided;

[0045] Figure 7A flowchart of the sixth embodiment of the model training method of the present application is provided;

[0046] Figure 8 This is a schematic diagram of the device structure of the hardware operating environment involved in the model training method in the embodiment of the present application.

[0047] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0048] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0049] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0050] The main solution of the embodiment of the present application is: obtain the Wi-Fi signal data set of the communication device; input the auxiliary data set into the preset neural network model to predict the signal rotation to obtain the signal characteristics and model parameters; determine the loss value of the signal characteristics and model parameters, and update the neural network model according to the loss value to obtain the device recognition model, which is used to identify the category of the communication device.

[0051] In this embodiment, for ease of description, the following description is made with the model training device as the execution entity.

[0052] Since existing specific radiation source identification methods are mainly based on deep learning and rely on a large amount of labeled data, the accuracy of identification and authentication of communication equipment is low in scenarios where labeled data is limited.

[0053] The present application provides a solution, which uses a neural network model to learn the decoupled representation of signal phase rotation correlation and rotation independence from an auxiliary data set, thereby extracting signal features of radio frequency fingerprints that are more robust and more beneficial to the identification of specific radiation sources, and then completing the identification and authentication of Wi-Fi devices based on the extracted signal features, thereby reducing the dependence on labeled data, improving the detection efficiency of Wi-Fi signals and the accuracy of device identification and authentication in sample-restricted scenarios, and improving the versatility in different environments.

[0054] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a model training device, etc. The following takes the model training device as an example to illustrate this embodiment and the following embodiments.

[0055] Based on this, the present application embodiment provides a model training method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the model training method of the present application.

[0056] In this embodiment, the model training method includes steps S10 to S30:

[0057] Step S10: acquiring a Wi-Fi signal data set of a communication device, wherein the Wi-Fi signal data set includes an auxiliary data set, wherein signal samples in the auxiliary data set are obtained by performing a signal phase rotation on the Wi-Fi signal, and each of the signal samples has a rotation label value.

[0058] It should be noted that the Wi-Fi signal of the communication device is obtained and preprocessed into a Wi-Fi signal dataset. The Wi-Fi signal dataset includes a target dataset and an auxiliary dataset, such as Figure 2 Optionally, the number of samples in the target dataset is smaller than the number of samples in the auxiliary dataset.

[0059] The signal samples in the target data set have category label values, which can be obtained by manual classification, and each signal sample is a time-frequency diagram obtained by short-time Fourier transform of the Wi-Fi signal. Optionally, the target data set is divided into a training data set and a test data set according to a preset ratio, and the preset ratio is 8:2 for example.

[0060] The samples in the auxiliary data set are obtained by performing signal phase rotation on the Wi-Fi signal, and each sample has a rotation label value and an instance number label value.

[0061] Step S20, inputting the auxiliary data set into a preset neural network model to perform signal rotation prediction to obtain signal features and model parameters.

[0062] It should be noted that the neural network model may include deep learning models such as convolutional neural network models and recurrent neural network models.

[0063] Use convolutional network and PyTorch code to build a convolutional neural network model. The built model is the basic model. The basic model does not involve increments. The loss function used in this model is the cross entropy loss function, and its formula is: No freezing is performed.

[0064] Set the number of categories and training parameters required for training the basic model, input the training set data of the selected number of categories into the model for training, and obtain the optimal basic model; the training parameters are preset as shown in the following table:

[0065] parameter value Training equipment GeForce GTX 2080Ti Training Environment PyTorch Optimizer SGD Batch size 32 Learning Rate The initial value is 0.001 and it is decreased by 90% every 50 iterations. Number of training sessions 250

[0066] Optionally, the signal feature includes at least one of a model encoding feature, a high-dimensional mean feature, a low-dimensional mean feature, etc.

[0067] It should be noted that model encoding features refer to high-level features extracted by a series of convolutional layers and pooling layers when a deep learning model processes input data. The above features capture important information of the input data, such as shape, texture, etc., and provide a basis for subsequent classification or other tasks. High-dimensional mean features are the average processing of multiple feature vectors, i.e., encoding features, of the same category or sample instance to obtain representative features of the category or instance. Mean features help reduce the dimension of features while retaining important category information. Low-dimensional mean features are features obtained by further reducing the dimensionality of high-dimensional mean features. For example, through dimensionality reduction techniques such as principal component analysis or autoencoders, features are mapped to a lower dimensional space while retaining important information of the original features as much as possible.

[0068] Optionally, the model features include at least one of a model prediction value, a model weight, etc.

[0069] It should be noted that the model prediction value is the output obtained by the model through forward propagation calculation based on the parameters learned by the model, such as weights and biases, after given input data. For example, in classification tasks, it refers to the category probability output by the model, and in regression tasks, it refers to the predicted continuous value. In deep learning models, weights refer to the parameters connecting each neuron in the network. They are learned and updated through the back-propagation algorithm during training. The weights determine how the model maps input features to output results.

[0070] Step S30, determining the loss values ​​of the signal characteristics and the model parameters, and updating the neural network model according to the loss values ​​to obtain a device identification model, wherein the device identification model is used to identify the category of the communication device.

[0071] It should be noted that the loss values ​​of signal features and model parameters are determined according to a preset loss function. Optionally, the loss values ​​of signal features and model parameters include classification loss values, mean square error loss values, noise contrast estimation loss values, etc.

[0072] Optionally, the model encoding features, high-dimensional mean features, low-dimensional mean features, model prediction values, sample rotation label values ​​and instance number label values ​​are respectively substituted into the cross entropy loss function, mean square error function, and noise contrast estimation loss function to obtain classification loss values, mean square error loss values, and noise contrast estimation loss values. The parameters of the neural network model are updated according to the classification loss value, mean square error loss value, and noise contrast estimation loss value until the optimal neural network model is obtained.

[0073] In one embodiment, if Figure 2As shown, an auxiliary dataset without device category labels is used to predict signal rotation, obtain signal features and model parameters, and update the neural network model.

[0074] In the technical solution of this embodiment, a Wi-Fi signal data set of a communication device is obtained; an auxiliary data set is input into a preset neural network model for signal rotation prediction to obtain signal features and model parameters; the loss values ​​of the signal features and model parameters are determined, and the neural network model is updated according to the loss values ​​to obtain a device identification model, which is used to identify the category of the communication device. The neural network model is used to learn the decoupled representation of the signal phase rotation correlation and the rotation independence from the auxiliary data set, thereby extracting the signal features of the RF fingerprint that are more robust and more beneficial to the identification of specific radiation sources, and then completing the identification and authentication of the Wi-Fi device based on the extracted signal features, reducing the dependence on the tag data, improving the detection efficiency of the Wi-Fi signal and the accuracy of device identification and authentication in sample-restricted scenarios, and improving the versatility in different environments.

[0075] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above embodiment can refer to the above introduction, and will not be repeated later. Figure 3 , step S10 comprises:

[0076] Step S11, encoding the original samples in sequence, and determining the instance number label value of each original sample;

[0077] Step S12, performing a preset number of signal phase rotations on each original sample according to the instance number label value, and generating a time-frequency diagram based on the samples after the signal phase rotation;

[0078] Step S13, obtaining the rotation label value corresponding to the time-frequency diagram, and generating the auxiliary data set according to the time-frequency diagram and the rotation label value.

[0079] It should be noted that a first proportion of original samples are selected from the original signal data set, the instance number label value of each original sample is determined, and a preset number of signal phase rotations are performed on each original sample, for example, the preset number of times is 4 times. Optionally, the sample transformation after the signal phase rotation is processed into a time-frequency diagram using short-time Fourier. The rotation label value and the instance number label value of each time-frequency diagram are manually annotated to obtain an auxiliary data set, wherein the instance number label value is consistent with the instance number label value of the original sample.

[0080] In the technical solution of this embodiment, in various scenarios of complex real Internet data environments, by processing auxiliary data sets and updating the neural network model, the problem of low detection efficiency and low accuracy for small sample Wi-Fi signal data in sample-restricted scenarios is solved.

[0081] Based on the first or second embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the above embodiment can be referred to the above introduction, and will not be repeated in the following. Figure 4 , after step S30, further comprising:

[0082] Step S40, inputting the target data set into the device identification model with a new classifier to identify the device category corresponding to the signal sample;

[0083] Step S50: updating the device identification model according to the identified device category and the device category tag value, and saving the updated device identification model.

[0084] It should be noted that the Wi-Fi signal data set includes a target data set, and the signal samples in the target data set have device category label values. Optionally, the category label values ​​of the Wi-Fi signal samples of a second ratio are obtained; the Wi-Fi signal samples are processed into a time-frequency graph to prepare a target data set, wherein the target data set can also be divided into a training data set and a test data set according to a preset ratio. Optionally, the second ratio is smaller than the first ratio.

[0085] Freeze the module for predicting rotation labels in the device identification model and add a new category classifier, such as Figure 2 As shown, the ability to detect device categories is obtained; according to the feature encoder and the new category classifier in the device recognition model, the model parameters are fine-tuned on the target data set to obtain the final device recognition model.

[0086] In the technical solution of this embodiment, unsupervised pre-training is performed on Wi-Fi data without category annotation information to obtain an initialized neural network model, and in the later training process, the model is fine-tuned on the target Wi-Fi data to obtain the fine-tuned neural network model, thereby further improving the accuracy of model training.

[0087] Based on any one of the first to third embodiments of the present application, in the fourth embodiment of the present application, the same or similar contents as those in the above embodiments can refer to the above introduction, and will not be repeated later. Figure 5 , step S20 further includes:

[0088] Step S21, inputting the auxiliary data set into a neural network model, wherein a feature encoder of the neural network model extracts image coding features of a preset dimension, and divides the image coding features of the preset dimension into image coding features of at least two sub-dimensions;

[0089] Step S22, determining a high-dimensional mean feature according to the average values ​​of the image coding features of at least two sub-dimensions;

[0090] Step S23, mapping the high-dimensional mean feature to a low-dimensional space to obtain a low-dimensional mean feature;

[0091] Step S24, obtaining the image coding features of the neural network model based on preset dimensions, the predicted model prediction value and the associated model weight value.

[0092] Optionally, the image data of the time-frequency diagram in the auxiliary data set is input into a convolutional neural network model with a large channel and a large convolution kernel, wherein the 4096-dimensional image coding features extracted by the feature encoder are divided into two to obtain 2048-dimensional coding features of the front and back halves; each image is classified according to the original sample instance number label, and the average value of the four feature vectors of the latter 2048 dimensions of the same instance number is calculated to obtain a high-dimensional mean feature vector; the high-dimensional mean feature vector is mapped to a low-dimensional space of 128 dimensions, and the mean vector is regularized using a two-norm to obtain a low-dimensional mean feature vector; the classifier in the convolutional neural network model predicts a signal rotation label based on the feature vector of the first 2048 dimensions, and obtains a model weight value associated with the model prediction value.

[0093] In the technical solution of this embodiment, in various scenarios of complex real Internet data environments, by processing auxiliary data sets, extracting signal features and model features, and updating the neural network model, the problem of low detection efficiency and low accuracy for small sample Wi-Fi signal data in sample-limited scenarios is solved.

[0094] Based on any one of the first to fourth embodiments of the present application, in the fifth embodiment of the present application, the same or similar contents as those in the above embodiments can be referred to the above introduction, and will not be repeated in the following. Figure 6 , step S30 further includes:

[0095] Step S31, obtaining a classification loss value according to a cross entropy loss function, the model prediction value and the rotation label value;

[0096] Step S32, obtaining a mean square error loss value according to the mean square error function, the high-dimensional mean feature and the model encoding feature of the sub-dimension;

[0097] Step S33, obtaining a noise contrast estimation loss value according to the noise contrast estimation loss function, the low-dimensional mean feature and the instance number label value.

[0098] In a feasible embodiment, the cross entropy loss value is shown in the following formula:

[0099]

[0100] Among them, H CE(y i ,p i ) is the cross entropy loss function, L CE is the cross entropy loss value, N is the number of input samples, R is the number of rotations of the sample, y i,j is the rotation label value of the i-th rotation copy of the i-th sample, p i,j The model's predicted value for the jth rotation of the ith sample.

[0101] In a feasible embodiment, the mean square error loss value is shown in the following formula:

[0102]

[0103] Among them, H MSE is the mean square error function, L MSE is the mean square error loss value, N is the number of input samples, R is the number of rotations of the sample, is the encoding feature f of the model's i-th rotation copy of the i-th sample i,j The second half of is the mean vector of the encoded features of the R rotated copies of the i-th sample.

[0104] In a feasible embodiment, the noise contrast estimation loss value is shown in the following formula:

[0105]

[0106] In the formula, H NCE is the noise contrast estimation function, L NCE is the noise contrast estimation loss value, N is the number of input samples, is the mean vector In the low-dimensional mapping of 128-dimensional modulo-2 regularization, The values ​​at the current and previous moments, The sum of the squares of the differences, is the prior probability In m times Gaussian noise P n The posterior probability under (i).

[0107] According to the multi-task learning strategy, an adaptive loss weight is obtained, and the cross entropy loss value, the mean square error loss value, and the noise contrast estimation loss value are weighted and accumulated respectively to obtain a joint loss value; back propagation is performed according to the joint loss value to obtain a gradient; and the parameters of the neural network model are updated according to the gradient until a device recognition model is obtained.

[0108] In the technical solution of this embodiment, by introducing rotational decoupling representation, the feature extractor learns to extract more robust RF fingerprint information that is more beneficial to the identification of specific radiation sources, reduces dependence on label data, improves the detection efficiency of Wi-Fi signals and the accuracy of device identification and authentication in sample-restricted scenarios, and improves versatility in different environments.

[0109] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the model training method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0110] This application also provides a model training device, please refer to Figure 7 , the model training device comprises:

[0111] An acquisition module 10 is configured to acquire a Wi-Fi signal dataset of a communication device, wherein the Wi-Fi signal dataset includes an auxiliary dataset, wherein signal samples in the auxiliary dataset are obtained by performing a signal phase rotation on the Wi-Fi signal, and each of the signal samples has a rotation label value;

[0112] A prediction module 20, used for inputting the auxiliary data set into a preset neural network model to perform signal rotation prediction to obtain signal features and model parameters;

[0113] The updating module 30 is used to determine the loss value of the signal characteristics and the model parameters, and update the neural network model according to the loss value to obtain a device identification model, wherein the device identification model is used to identify the category of the communication device.

[0114] Optionally, the step of acquiring a Wi-Fi signal data set of a communication device includes:

[0115] Encode the original samples in sequence and determine the instance number label value of each original sample;

[0116] Performing a preset number of signal phase rotations on each original sample according to the instance number label value, and generating a time-frequency diagram based on the samples after the signal phase rotation;

[0117] The rotation label value corresponding to the time-frequency diagram is obtained, and the auxiliary data set is generated according to the time-frequency diagram and the rotation label value.

[0118] Optionally, the Wi-Fi signal data set includes a target data set, each signal sample in the target data set has a device category label value, and after the step of determining the signal feature and the loss value of the model parameter, and updating the neural network model according to the loss value to obtain the device recognition model, the step further includes:

[0119] Inputting the target data set into the device identification model with a new classifier to identify the device category corresponding to the signal sample;

[0120] The device identification model is updated according to the identified device category and the device category tag value, and the updated device identification model is saved.

[0121] Optionally, the signal features include model encoding features, high-dimensional mean features, and low-dimensional mean features, the model parameters include model prediction values ​​and model weights, and the step of inputting the auxiliary data set into a preset neural network model for signal rotation prediction to obtain signal features and model parameters further includes:

[0122] The auxiliary data set is input into a neural network model, a feature encoder of the neural network model extracts image coding features of a preset dimension, and divides the image coding features of the preset dimension into image coding features of at least two sub-dimensions;

[0123] Determine a high-dimensional mean feature according to an average value of the image encoding features of at least two sub-dimensions;

[0124] Mapping the high-dimensional mean feature to a low-dimensional space to obtain a low-dimensional mean feature;

[0125] Obtain the image coding features of the neural network model based on preset dimensions, the predicted model prediction value and the associated model weight value.

[0126] Optionally, the signal feature includes a model encoding feature, a high-dimensional mean feature, and a low-dimensional mean feature, the model parameter includes a model prediction value and a model weight, and the step of determining the loss value of the signal feature and the model parameter includes:

[0127] Obtaining a classification loss value according to a cross entropy loss function, the model prediction value, and the rotation label value;

[0128] Obtaining a mean square error loss value according to the mean square error function, the high-dimensional mean feature and the model encoding feature of the sub-dimension;

[0129] A noise contrast estimation loss value is obtained according to the noise contrast estimation loss function, the low-dimensional mean feature and the instance number label value.

[0130] Optionally, the step of updating the neural network model according to the loss value to obtain a device identification model includes:

[0131] Obtaining an adaptive loss weight, weighting and accumulating the cross entropy loss value, the mean square error loss value, and the noise contrast estimation loss value respectively, to obtain a joint loss value;

[0132] Perform back propagation based on the joint loss value to obtain the gradient;

[0133] The parameters of the neural network model are updated according to the gradient until a device recognition model is obtained.

[0134] The model training device provided by the present application adopts the model training method in the above embodiment, which can solve the technical problem that the existing specific radiation source identification method is mainly based on deep learning and relies on a large amount of labeled data. In the scenario where the labeled data is limited, the recognition and authentication accuracy of the communication device is low. Compared with the prior art, the beneficial effects of the model training device provided by the present application are the same as the beneficial effects of the model training method provided by the above embodiment, and the other technical features in the model training device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0135] The present application provides a model training device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the model training method in the above-mentioned embodiment one.

[0136] Reference below Figure 8 , which shows a schematic diagram of the structure of a model training device suitable for implementing an embodiment of the present application. The model training device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), etc., and fixed terminals such as computers, etc. Figure 8 The model training device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0137] like Figure 8As shown, the model training device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the model training device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the model training device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a model training device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or provided instead.

[0138] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0139] The model training device provided by the present application adopts the model training method in the above embodiment, which can solve the technical problem that the existing specific radiation source identification method is mainly based on deep learning and relies on a large amount of labeled data. In the scenario where the labeled data is limited, the recognition and authentication accuracy of the communication device is low. Compared with the prior art, the beneficial effects of the model training device provided by the present application are the same as the beneficial effects of the model training method provided by the above embodiment, and the other technical features in the model training device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0140] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0141] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0142] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the model training method in the above-mentioned embodiment.

[0143] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0144] The above-mentioned computer-readable storage medium may be included in the model training device; or it may exist independently without being assembled into the model training device.

[0145] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the model training device, the model training device learns the decoupled representation of signal phase rotation correlation and rotation independence from the auxiliary data set through a neural network model, thereby extracting signal features of radio frequency fingerprints that are more robust and more beneficial to the identification of specific radiation sources, and then completing the identification and authentication of Wi-Fi devices based on the extracted signal features, thereby reducing the dependence on labeled data, improving the detection efficiency of Wi-Fi signals and the accuracy of device identification and authentication in sample-restricted scenarios, and improving the versatility in different environments.

[0146] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0147] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0148] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0149] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned model training method, and can solve the technical problem that the existing specific radiation source identification method is mainly based on deep learning and relies on a large amount of labeled data. In the scenario where the labeled data is limited, the recognition and authentication accuracy of the communication device is low. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the model training method provided by the above-mentioned embodiment, which will not be repeated here.

[0150] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned model training method when executed by a processor.

[0151] The computer program product provided by the present application can solve the technical problem that the existing specific radiation source identification method is mainly based on deep learning and relies on a large amount of labeled data. In the scenario where the labeled data is limited, the recognition and authentication accuracy of the communication device is low. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the model training method provided by the above embodiment, which will not be repeated here.

[0152] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A model training method, characterized in that: The method comprises: Acquire a Wi-Fi signal dataset of a communication device, where the Wi-Fi signal dataset includes an auxiliary dataset, where signal samples in the auxiliary dataset are obtained by performing a signal phase rotation on the Wi-Fi signal, and each of the signal samples has a rotation label value; Inputting the auxiliary data set into a preset neural network model to perform signal rotation prediction to obtain signal features and model parameters; Determine the loss values ​​of the signal characteristics and the model parameters, and update the neural network model according to the loss values ​​to obtain a device identification model, wherein the device identification model is used to identify the category of the communication device.

2. The method according to claim 1, characterized in that The step of obtaining a Wi-Fi signal data set of a communication device includes: Encode the original samples in sequence and determine the instance number label value of each original sample; Performing a preset number of signal phase rotations on each original sample according to the instance number label value, and generating a time-frequency diagram based on the samples after the signal phase rotation; The rotation label value corresponding to the time-frequency diagram is obtained, and the auxiliary data set is generated according to the time-frequency diagram and the rotation label value.

3. The method according to claim 1, characterized in that The Wi-Fi signal data set includes a target data set, each signal sample in the target data set has a device category label value, and after the step of determining the signal feature and the loss value of the model parameter, and updating the neural network model according to the loss value to obtain the device recognition model, the step further includes: Inputting the target data set into the device identification model with a new classifier to identify the device category corresponding to the signal sample; The device identification model is updated according to the identified device category and the device category tag value, and the updated device identification model is saved.

4. The method according to claim 1, characterized in that The signal features include model coding features, high-dimensional mean features, and low-dimensional mean features, and the model parameters include model prediction values ​​and model weights. The step of inputting the auxiliary data set into a preset neural network model for signal rotation prediction to obtain signal features and model parameters also includes: The auxiliary data set is input into a neural network model, a feature encoder of the neural network model extracts image coding features of a preset dimension, and divides the image coding features of the preset dimension into image coding features of at least two sub-dimensions; Determine a high-dimensional mean feature according to an average value of the image encoding features of at least two sub-dimensions; Mapping the high-dimensional mean feature to a low-dimensional space to obtain a low-dimensional mean feature; Obtain the image coding features of the neural network model based on preset dimensions, the predicted model prediction value and the associated model weight value.

5. The method according to claim 1, characterized in that The signal features include model encoding features, high-dimensional mean features, and low-dimensional mean features, the model parameters include model prediction values ​​and model weights, and the step of determining the loss values ​​of the signal features and the model parameters includes: Obtaining a classification loss value according to a cross entropy loss function, the model prediction value, and the rotation label value; Obtaining a mean square error loss value according to the mean square error function, the high-dimensional mean feature and the model encoding feature of the sub-dimension; A noise contrast estimation loss value is obtained according to the noise contrast estimation loss function, the low-dimensional mean feature and the instance number label value.

6. The method according to claim 5, characterized in that The step of updating the neural network model according to the loss value to obtain the device identification model comprises: Obtaining an adaptive loss weight, weighting and accumulating the cross entropy loss value, the mean square error loss value, and the noise contrast estimation loss value respectively, to obtain a joint loss value; Perform back propagation based on the joint loss value to obtain the gradient; The parameters of the neural network model are updated according to the gradient until a device recognition model is obtained.

7. A model training device, characterized in that: The device comprises: an acquisition module, configured to acquire a Wi-Fi signal dataset of a communication device, wherein the Wi-Fi signal dataset includes an auxiliary dataset, wherein signal samples in the auxiliary dataset are obtained by performing a signal phase rotation on the Wi-Fi signal, and each of the signal samples has a rotation label value; A prediction module, used for inputting the auxiliary data set into a preset neural network model to perform signal rotation prediction to obtain signal features and model parameters; An updating module is used to determine the loss value of the signal characteristics and the model parameters, and update the neural network model according to the loss value to obtain a device identification model, wherein the device identification model is used to identify the category of the communication device.

8. A model training device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the model training method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the model training method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the steps of the model training method according to any one of claims 1 to 6.

Images