Equipment identity recognition method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides an equipment identity recognition method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a frequency hopping signal, and obtaining a target polarization fingerprint feature from the frequency hopping signal; the target polarization fingerprint feature comprises target real part data and target imaginary part data; based on the target real part data and the target imaginary part data, constructing a dual-channel data set containing the target polarization fingerprint features; determining a target frequency band, and performing data slicing on a target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band to generate slice data; fingerprint classification operation is executed based on the slice data, and a classification result for the target equipment is generated; and the identity information of the target equipment is identified by adopting the classification result, so that the high-dimensional features of the polarized fingerprints are effectively extracted, and the equipment identification efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of device identification technology, and in particular to a device identification method, a device identification apparatus, an electronic device and a computer-readable storage medium. Background Art

[0002] The wireless Internet of Things faces severe security threats due to its openness. Attackers can easily access the network, steal data, and even disguise the identity of the device. Radio frequency fingerprints, which originate from random defects in device hardware, are widely used for device identity authentication. However, with the improvement of manufacturing technology, the difference of radio frequency fingerprints has gradually decreased, resulting in a decrease in recognition accuracy.

[0003] Polarization fingerprints, which use the unique features generated by antenna differences, are considered an effective means to reduce the difference in RF fingerprints. However, most of the current polarization fingerprint-based recognition methods use traditional methods and cannot fully exploit their high-dimensional features, which leads to low recognition accuracy. Summary of the invention

[0004] The embodiments of the present invention provide a device identity recognition method, apparatus, electronic device and computer-readable storage medium to overcome the above problems or at least partially solve the above problems.

[0005] The embodiment of the present invention discloses a device identity recognition method, comprising:

[0006] Acquire a frequency hopping signal, and acquire a target polarization fingerprint feature from the frequency hopping signal; the target polarization fingerprint feature includes target real part data and target imaginary part data;

[0007] Constructing a dual-channel data set including the target polarization fingerprint feature based on the target real data and the target imaginary data;

[0008] Determine a target frequency band, and perform data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points in the target frequency band to generate slice data;

[0009] Performing a fingerprint classification operation based on the slice data to generate a classification result for the target device;

[0010] The classification result is used to identify the identity information of the target device.

[0011] Optionally, the step of determining the target frequency band includes:

[0012] Determine the number and width of frequency bands, and determine the start and end frequencies of each frequency band;

[0013] A target frequency band is determined based on the number of frequency bands, the frequency band width, the start frequency, and the end frequency.

[0014] Optionally, the step of performing data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points in the target frequency band to generate sliced ​​data includes:

[0015] For the target polarization fingerprint feature in the dual-channel data set, data slicing is performed based on the number of sampling points in the target frequency band to generate initial slicing data;

[0016] Performing a spectrum analysis operation on the initial slice data to generate a spectrum analysis result for the initial slice data;

[0017] When it is determined that the cutting is accurate based on the spectrum analysis result, the initial slice data is determined as the slice data.

[0018] Optionally, the step of performing a fingerprint classification operation based on the slice data to generate a classification result for the target device includes:

[0019] Traversing the slice data, and determining the slice features and slice index of the slice data;

[0020] Inputting the slice feature and the slice index into a classification model, and outputting an initial classification result for the slice data through the classification model;

[0021] Determining a classification accuracy probability of the initial classification result, and determining the classification accuracy probability as a slice weight for the slice data;

[0022] A weighted average operation is performed on the initial classification results of all slice data using the slice weights to determine a classification result for the target device.

[0023] Optionally, the step of performing a weighted average operation on the initial classification results of all slice data by using the slice weight to determine the classification result for the target device includes:

[0024] Multiply the classification results of all slice data by the corresponding slice weights, and add all weighted results to generate the target slice data vector;

[0025] The category with the highest probability in the target slice data vector is determined as the classification result for the target device.

[0026] Optionally, it also includes:

[0027] The slice data is divided into a training set, a validation set and a test set;

[0028] Determine a true category label for the slice data, and calculate a loss function for the classification model using the initial classification result and the true category label;

[0029] The classification model is trained based on the training set, the validation set, the test set and the loss function.

[0030] Optionally, the step of using the classification result to identify the identity information of the target device includes:

[0031] determining a confidence score for the classification result;

[0032] When the confidence score is higher than a preset score threshold, the target device is determined to be a valid device;

[0033] The preset device whitelist is traversed, and when the valid device matches a device in the preset device whitelist, the target device is determined to be a legitimate device.

[0034] The embodiment of the present invention further discloses a device identification apparatus, comprising:

[0035] A target polarization fingerprint feature acquisition module is used to acquire a frequency hopping signal and acquire a target polarization fingerprint feature from the frequency hopping signal; the target polarization fingerprint feature includes target real part data and target imaginary part data;

[0036] A dual-channel data set construction module, used to construct a dual-channel data set containing the target polarization fingerprint feature based on the target real data and the target imaginary data;

[0037] A slice data generation module, used to determine a target frequency band, and perform data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band to generate slice data;

[0038] A classification result generating module, used to perform a fingerprint classification operation based on the slice data to generate a classification result for a target device;

[0039] An identity information recognition module is used to use the classification result to recognize the identity information of the target device.

[0040] The embodiment of the present invention further discloses an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0041] The memory is used to store computer programs;

[0042] The processor is used to implement the method described in the embodiment of the present invention when executing the program stored in the memory.

[0043] The embodiment of the present invention further discloses a computer-readable storage medium having instructions stored thereon, which, when executed by one or more processors, enables the processors to execute the method described in the embodiment of the present invention.

[0044] The embodiments of the present invention include the following advantages:

[0045] The embodiment of the present invention obtains a frequency hopping signal and obtains a target polarization fingerprint feature from the frequency hopping signal; the target polarization fingerprint feature includes target real data and target imaginary data; a dual-channel data set including the target polarization fingerprint feature is constructed based on the target real data and the target imaginary data; a target frequency band is determined, and data slicing is performed on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band to generate slice data; a fingerprint classification operation is performed based on the slice data to generate a classification result for a target device; and the identity information of the target device is identified by using the classification result, thereby effectively realizing the extraction of high-dimensional features for polarization fingerprints, thereby improving the efficiency of device identification.

[0046] Specifically,

[0047] Improve device identification accuracy: Through in-depth analysis and classification of polarization fingerprints, devices can be identified more accurately.

[0048] Enhance system robustness: The use of technologies such as dual-channel data sets and data slicing improves the system's robustness to noise and interference.

[0049] Adaptable to different application scenarios: This method can be applied to various wireless communication scenarios and has wide applicability.

[0050] Through in-depth analysis and classification of the polarization fingerprint of the frequency hopping signal, the target device can be accurately identified. This method has high accuracy, robustness and adaptability, and has broad application prospects in the field of wireless communications. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flowchart of a method for identifying a device identity provided in an embodiment of the present invention;

[0052] Figure 2 is an architecture diagram of a classification model provided in an embodiment of the present invention;

[0053] Figure 3 is a flow chart of a device identity recognition method provided in an embodiment of the present invention;

[0054] Figure 4 is a structural block diagram of a device identity recognition apparatus provided in an embodiment of the present invention;

[0055] Figure 5 is a hardware structure block diagram of an electronic device provided in an embodiment of the present invention;

[0056] Figure 6 is a schematic diagram of a computer-readable medium provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Due to the openness and broadcasting of wireless networks, attackers can easily connect malicious devices to the network and access unauthorized data, or even infer the identity of the device. Therefore, it is necessary to accurately determine the identity of devices in the wireless Internet of Things. The radiation source identification method based on RF fingerprint can identify and authenticate the identity of the radiation source and improve the security of the communication system.

[0059] RF fingerprints come from random hardware defects generated in the transmitter, which cause the frequency, project, amplitude and other parameters of the actual RF signal to deviate from the theoretical values, including I / Q imbalance, DAC / ADC converter nonlinearity and carrier frequency offset. The unique features left by the transmitter hardware defects in the transient and steady-state time domains of the signal are defined as transient device fingerprints and steady-state RF fingerprints, respectively. In existing research, radiation source identification based on RF fingerprints is mostly combined with deep learning methods to improve recognition performance. However, with the improvement of manufacturing process level, the transmitter hardware defects contained in RF fingerprints have been reduced. As the distinction between hardware defects of two different devices decreases, the difference in RF fingerprints of devices also decreases. RF fingerprints face the problem of small fingerprint differences. Polarization fingerprints use the differences in antennas to distinguish different devices. At the same time, the directionality of polarization fingerprints further amplifies the difference in fingerprints. Therefore, polarization fingerprints are called one of the possible solutions to the problem of small fingerprint differences.

[0060] At present, the identification of radiation sources based on polarization fingerprints mainly uses traditional methods, such as Euclidean distance. However, this method can only utilize some low-dimensional features of polarization fingerprints, and its high-dimensional features are not fully utilized. Through research in academia and industry, reinforcement learning has been proven to be effectively applied to multiple fields, especially in the field of identification. Compared with manually extracting data features based on experience, the nonlinear logic function in the hidden layer of reinforcement learning can fully mine higher-dimensional information in the data. In addition, the reinforcement learning-based method directly extracts features from wireless signal fingerprints, making its features more scalable.

[0061] In order to ensure the anti-interference of wireless communication, frequency hopping technology and frequency hopping communication equipment have become more and more common. Due to the discrete communication frequency bands of such equipment, their polarization fingerprints are also discrete, which increases the difficulty of device identification using the fingerprint of frequency hopping equipment. It is necessary to study the radiation source identification algorithm suitable for frequency hopping communication equipment.

[0062] As can be seen from the above, the identity authentication of wireless IoT devices is an important part of ensuring system security. Traditional RF fingerprint-based identification methods use random features generated by transmitter hardware defects, but with the improvement of manufacturing processes, the differences in these features are gradually reduced.

[0063] Problems faced by radio frequency fingerprinting:

[0064] Reduced fingerprint differentiation: Due to the advancement of manufacturing technology, the hardware defects of different devices are becoming more and more similar, resulting in reduced differentiation of RF fingerprints.

[0065] Limitations of traditional methods: Currently, the recognition methods based on RF fingerprints mainly use traditional machine learning algorithms, such as Euclidean distance, which can only utilize some low-dimensional features of fingerprints and cannot fully exploit high-dimensional features.

[0066] Advantages of polarization fingerprint:

[0067] Difference amplification: Polarization fingerprint uses the difference of antennas to further amplify the difference of fingerprints through directionality, which can effectively solve the problem of reducing the difference of RF fingerprints.

[0068] Deficiencies of existing research:

[0069] Limitations of the method: Currently, the identification methods based on polarization fingerprints mainly use traditional methods, which cannot fully utilize the high-dimensional characteristics of polarization fingerprints.

[0070] Based on the above analysis, the research directions include:

[0071] How to effectively extract the high-dimensional features of polarization fingerprints; Polarization fingerprints contain rich feature information. How to design effective feature extraction methods to fully tap these features and improve recognition accuracy?

[0072] How to choose a suitable classifier; How to choose or design a suitable classifier for the high-dimensional features of polarization fingerprints to obtain better classification performance;

[0073] How to improve the robustness of polarization fingerprints; Polarization fingerprints are easily affected by environmental noise and multipath effects. How to improve the robustness of polarization fingerprints so that they can still maintain a high recognition accuracy in complex environments?

[0074] How to build an efficient polarization fingerprint database; How to build an efficient polarization fingerprint database to support the rapid identification of large-scale devices;

[0075] How to solve the problem of dynamic changes in polarization fingerprints; due to factors such as equipment aging and environmental changes, polarization fingerprints may change dynamically. How to design an adaptive recognition algorithm to cope with such changes.

[0076] Therefore, the embodiment of the present invention proposes to use a reinforcement learning method to mine the high-dimensional features of polarization fingerprints and improve the accuracy of radiation source identification. In frequency hopping communication equipment, the signal jumps with the frequency, so the extracted polarization fingerprint also jumps, that is, the polarization fingerprint has the characteristics of fragmentation at this time. Therefore, we propose to use a segmented voting network to utilize the fragmented fingerprint, separate the fingerprint samples of each device, and effectively avoid the impact of fingerprint fragmentation.

[0077] Reference Figure 1 , shows a flowchart of a method for identifying a device identity provided in an embodiment of the present invention, which may specifically include the following steps:

[0078] Step 101, acquiring a frequency hopping signal, and acquiring a target polarization fingerprint feature from the frequency hopping signal; the target polarization fingerprint feature includes target real part data and target imaginary part data;

[0079] Step 102, constructing a dual-channel data set including the target polarization fingerprint feature based on the target real data and the target imaginary data;

[0080] Step 103, determining a target frequency band, and performing data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band to generate slice data;

[0081] Step 104, performing a fingerprint classification operation based on the slice data to generate a classification result for the target device;

[0082] Step 105: Use the classification result to identify the identity information of the target device.

[0083] In a specific implementation, the embodiment of the present invention can obtain a frequency hopping signal, and obtain a target polarization fingerprint feature from the frequency hopping signal.

[0084] Frequency hopping signal is a wireless communication technology in which the frequency of the transmitted signal changes continuously.

[0085] The polarization fingerprint feature is the unique polarization characteristic exhibited by a device in wireless communication and can be used as the "fingerprint" of the device for identification.

[0086] The target polarization fingerprint feature may include target real data and target imaginary data. The target real data and the target imaginary data are polarization fingerprints usually represented by complex numbers, and the real part and the imaginary part respectively represent the projection of the signal in two orthogonal directions.

[0087] For example, a complex number can be expressed in the form of a+bi, where:

[0088] a is the real part, which represents the projection of the complex number on the real number axis.

[0089] b is the imaginary part, which represents the projection of the complex number on the imaginary axis.

[0090] i is an imaginary unit, satisfying i2=-1.

[0091] The reason why polarization fingerprints have real and imaginary parts is that:

[0092] Signal complexity: RF signals are essentially complex electromagnetic waves, which have characteristics such as amplitude, frequency, and phase. In order to fully describe an RF signal, it is usually represented by complex numbers.

[0093] Characterization of polarization information: The polarization fingerprint contains the polarization information of the signal, which is described by complex numbers. The amplitude of the complex number represents the strength of the signal, and the phase represents the polarization direction of the signal.

[0094] The purpose of obtaining the target polarization fingerprint feature and determining the target real data and the target imaginary data is to collect the original data containing the target device information to prepare for subsequent feature extraction and classification.

[0095] Beneficial effects: The original, unprocessed polarization fingerprint data is obtained, which provides a basis for subsequent analysis.

[0096] The embodiment of the present invention can also construct a dual-channel data set containing target polarization fingerprint features based on the target real data and the target imaginary data, so as to make full use of the real and imaginary information of the polarization fingerprint, improve the feature expression capability, enhance the description of the polarization fingerprint features, and provide richer features for subsequent classification.

[0097] In practical applications, the imaginary part mainly contains the phase information of the signal. Phase information is very important for distinguishing the polarization characteristics of different devices; the real part mainly contains the amplitude information of the signal. Amplitude information can reflect information such as the power of the device. Processing the real part and the imaginary part into one channel respectively is equivalent to decomposing a complex signal into two real signals, thereby increasing the feature dimension, providing more information for the model, and helping to improve classification accuracy.

[0098] Furthermore, the dual-channel dataset can also achieve the following beneficial effects.

[0099] More comprehensive feature representation: Since the real part and the imaginary part contain different aspects of the signal, processing them into one channel can extract the signal features more comprehensively and improve the performance of the classification model.

[0100] Adapting to deep learning models: Deep learning models usually require input data to be real matrices. By processing the real and imaginary parts of complex signals into one channel respectively, the data can be easily input into deep learning models for training.

[0101] Processing the real and imaginary parts of the polarization fingerprint into one channel respectively is to convert the complex complex signal into a real number matrix that is easier for computers to process, thereby making full use of the phase and amplitude information in the signal and improving the accuracy of device recognition.

[0102] The embodiment of the present invention can also determine a target frequency band, and perform data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points in the target frequency band to generate sliced ​​data.

[0103] The target frequency band may be a frequency band that needs to be focused on in the frequency hopping signal.

[0104] The number of sampling points can be the number of samples collected in each frequency band.

[0105] The purpose of data slicing is to divide the continuous signal into multiple segments of fixed length to facilitate subsequent feature extraction and classification.

[0106] Beneficial effects: It improves data utilization, facilitates model training, and helps process signals of different lengths.

[0107] The embodiment of the present invention can also perform fingerprint classification operations based on the slice data to generate classification results for the target device.

[0108] The purpose of fingerprint classification is to achieve device identity recognition. According to the extracted features, the data can be divided into different categories, that is, the identity of the device can be determined.

[0109] Beneficial effect: Through classification, the target device can be accurately identified and the device identity authentication can be achieved.

[0110] The embodiment of the present invention may also use the classification result to identify the identity information of the target device, thereby determining whether the identity of the target device is legal.

[0111] The embodiment of the present invention obtains a frequency hopping signal and obtains a target polarization fingerprint feature from the frequency hopping signal; the target polarization fingerprint feature includes target real data and target imaginary data; a dual-channel data set including the target polarization fingerprint feature is constructed based on the target real data and the target imaginary data; a target frequency band is determined, and data slicing is performed on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band to generate slice data; a fingerprint classification operation is performed based on the slice data to generate a classification result for a target device; and the identity information of the target device is identified by using the classification result, thereby effectively realizing the extraction of high-dimensional features for polarization fingerprints, thereby improving the efficiency of device identification.

[0112] Specifically,

[0113] Improve device identification accuracy: Through in-depth analysis and classification of polarization fingerprints, devices can be identified more accurately.

[0114] Enhance system robustness: The use of technologies such as dual-channel data sets and data slicing improves the system's robustness to noise and interference.

[0115] Adaptable to different application scenarios: This method can be applied to various wireless communication scenarios and has wide applicability.

[0116] Through in-depth analysis and classification of the polarization fingerprint of the frequency hopping signal, the target device can be accurately identified. This method has high accuracy, robustness and adaptability, and has broad application prospects in the field of wireless communications.

[0117] Based on the above embodiment, a variant embodiment of the above embodiment is proposed. It should be noted that in order to make the description concise, only the differences from the above embodiment are described in the variant embodiment.

[0118] The step of determining the target frequency band, performing data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band, and generating sliced ​​data comprises:

[0119] Determine the number and width of frequency bands, and determine the start and end frequencies of each frequency band;

[0120] Determine a target frequency band based on the number of frequency bands, the frequency band width, the start frequency, and the end frequency;

[0121] For the target polarization fingerprint feature in the dual-channel data set, data slicing is performed based on the number of sampling points in the target frequency band to generate initial slicing data;

[0122] Performing a spectrum analysis operation on the initial slice data to generate a spectrum analysis result for the initial slice data;

[0123] When it is determined that the cutting is accurate based on the spectrum analysis result, the initial slice data is determined as the slice data.

[0124] Exemplarily, the target frequency band may be determined in the following manner, and data slicing may be performed based on the number of sampling points of the target frequency band to generate sliced ​​data.

[0125] In practical applications, the purpose of determining the target frequency band is to determine all frequency bands, that is, the target frequency band may be all frequency bands.

[0126] 1. The principles for determining frequency band division are:

[0127] Determine the frequency band division principle: According to the frequency band division rules of the frequency hopping signal, the starting frequency and ending frequency of each frequency band can be determined.

[0128] Determine the number of frequency bands: The number of frequency bands depends on the specific application scenario and signal characteristics.

[0129] Determine the frequency band width: The frequency band width should be reasonably selected based on the bandwidth and noise level of the signal.

[0130] 2. Perform data slicing:

[0131] Cutting by frequency band: The continuous received signal is divided into certain frequency bands, and then segmented based on the number of samples collected in each frequency band to obtain initial slice data.

[0132] Spectrum analysis: Perform spectrum analysis on the signal of each frequency band to verify the accuracy of the initial slice data segmentation.

[0133] Time alignment: If the sampling times of different frequency bands are different, time alignment is required to ensure the consistency of data in each frequency band.

[0134] 3. Feature extraction:

[0135] Targeted extraction: Extract the corresponding polarization fingerprint features from the signal in each frequency band.

[0136] Feature selection: Select appropriate features such as amplitude, phase, polarization ellipse parameters, etc. according to the subsequent classification task.

[0137] 4. Data normalization:

[0138] Eliminate dimensionality effects: Normalize the extracted features to eliminate dimensional differences between different features and improve model training efficiency.

[0139] Commonly used normalization methods: Min-Max normalization, Z-score normalization, etc.

[0140] 5. Sample construction:

[0141] Fixed length: The feature vectors of each frequency band are concatenated into a vector of fixed length as the input of the sample.

[0142] Sample label: Label each sample with the corresponding device category or other relevant labels.

[0143] After completing the above operations, the initial slice data may be determined as the slice data.

[0144] Optionally, the embodiment of the present invention can also divide the sliced ​​data into a training set, a validation set and a test set. The training set is used to train the classification model, the validation set is used to adjust the model parameters, and the test set is used to evaluate the generalization performance of the model.

[0145] Specifically, the division can be as follows:

[0146] Random division: The processed sample set is randomly divided into training set, validation set and test set.

[0147] Proportional allocation: The training set is used to train the model, the validation set is used to adjust the model hyperparameters, and the test set is used to evaluate the generalization performance of the model.

[0148] Partition ratio: Usually, the training set accounts for 70%, the validation set accounts for 15%, and the test set accounts for 15%.

[0149] In an embodiment of the present invention, the number of frequency bands and the frequency band width are determined, and the starting frequency and the ending frequency of each frequency band are determined; a target frequency band is determined based on the number of frequency bands, the frequency band width, the starting frequency and the ending frequency; data slicing is performed on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band to generate initial slicing data; a spectrum analysis operation is performed on the initial slicing data to generate a spectrum analysis result for the initial slicing data; when it is determined that the cutting is accurate based on the spectrum analysis result, the initial slicing data is determined as slicing data, which can achieve the following beneficial effects:

[0150] Meet the requirements of reinforcement learning: The fixed-length input samples meet the requirements of reinforcement learning algorithms for input data.

[0151] Dealing with missing data: Even if data for some frequency bands is missing, device identification can be performed by analyzing data from other frequency bands.

[0152] Improve the generalization ability of the model: By training data in different frequency bands, the model's ability to adapt to different scenarios can be improved.

[0153] Convenient feature engineering: Features of different frequency bands can be processed separately to improve the flexibility and efficiency of feature engineering.

[0154] Data slicing is to divide the continuous signal into frequency bands and extract the characteristics of each frequency band to form samples of fixed length. This method not only meets the requirements of reinforcement learning for input data, but also improves the robustness and generalization ability of the model.

[0155] In an optional embodiment of the present invention, the step of performing a fingerprint classification operation based on the slice data to generate a classification result for the target device includes:

[0156] Traversing the slice data, and determining the slice features and slice index of the slice data;

[0157] Inputting the slice feature and the slice index into a classification model, and outputting an initial classification result for the slice data through the classification model;

[0158] Determining a classification accuracy probability of the initial classification result, and determining the classification accuracy probability as a slice weight for the slice data;

[0159] By using the slice weights, a weighted average operation is performed on the initial classification results of all slice data to determine a classification result for the target device;

[0160] The step of performing a weighted average operation on the initial classification results of all slice data by using the slice weights to determine the classification result for the target device includes:

[0161] Multiply the classification results of all slice data by the corresponding slice weights, and add all weighted results to generate the target slice data vector;

[0162] The category with the highest probability in the target slice data vector is determined as the classification result for the target device.

[0163] The slice data is divided into a training set, a validation set and a test set;

[0164] Determine a true category label for the slice data, and calculate a loss function for the classification model using the initial classification result and the true category label;

[0165] The classification model is trained based on the training set, the validation set, the test set and the loss function.

[0166] Outer loops and inner loops are common concepts in programming. They are used to describe the nested loop structure in a program. Imagine a Russian nesting doll, each nesting doll has a smaller nesting doll inside it. This is a vivid metaphor for loop nesting.

[0167] Outer loop:

[0168] Definition: The outer loop is the framework that controls the entire loop process. It determines the total number of loops, and at each iteration, the inner loop is executed.

[0169] effect:

[0170] Iterate over all samples in the dataset.

[0171] Controls the flow of the entire algorithm.

[0172] Inner loop:

[0173] Definition: The inner loop is nested in the outer loop. Each time the outer loop iterates, the inner loop will be executed once.

[0174] effect:

[0175] Iterate over a subset or elements of the sample.

[0176] Complete specific subtasks, such as traversing all frequency bands of a sample.

[0177] Exemplarily, the classification results may be generated and the classification model may be trained in the following manner.

[0178] 1. Outer loop:

[0179] Iterate over samples: Iterate over each sample in the dataset.

[0180] Traversing slices: For each sample, you need to nest another layer of loop to traverse all slices corresponding to the sample.

[0181] 2. Inner loop:

[0182] Traverse slices: traverse each slice of the current sample.

[0183] 3. Determine the true category label for the sliced ​​data. Since the data has been sliced, each slice corresponds to a category label. This category label is usually consistent with the category label of the original sample.

[0184] Determine slice features: As before, extract features for each slice to form a feature vector.

[0185] Determine the slice index (optional): The slice index can be used as an additional feature to reflect the position of the slice in the entire signal; specifically, whether the slice index is used as input depends on the specific task and model design. If the model can automatically learn the relationship between slices, then the slice index may not be necessary.

[0186] 4. Classification Model Classification:

[0187] Inputting the slice feature and the slice index into a classification model, and outputting an initial classification result for the slice data through the classification model;

[0188] For example, the input content may include: sample effective frequency band number array effectNum, preprocessed test set testSample, signal-to-noise ratio number snrNum, total number of samples under a single signal-to-noise ratio sampleNum. The classification model output content may include: true value matrix y_tru, classification result matrix y_output, matrix weight weight.

[0189] 5. Calculate loss: Compare the initial classification results with the true labels and calculate the loss function.

[0190] 6. Calculate accuracy: Calculate the classification accuracy of the initial classification result of the current slice data and use it as the slice weight of the slice data.

[0191] 7. Soft voting: The initial classification results of all slice data are weighted averaged according to the slice weights to obtain the final classification result.

[0192] In order to improve the classification performance, we use a voting mechanism to fuse the output results of multiple classifiers. The voting mechanism is mainly divided into hard voting and soft voting.

[0193] Hard voting: Each classifier has a weight of one vote, and the category with the most votes is selected as the final result. This method is simple and intuitive, but it is sensitive to the performance differences of different classifiers.

[0194] Soft voting: Each classifier is assigned a weight, which represents its influence on the final result. By calculating the weighted average of the output results of all classifiers, the category with the highest probability is obtained as the final result. This method is more flexible and different weights can be assigned according to the performance of different classifiers.

[0195] In our frequency hopping signal classification task, since the features of different frequency bands are quite different and complementary to each other, we choose to use soft voting to fuse the results of multiple classifiers. Specifically, we assign weights based on the accuracy of each frequency band classifier, and the classifier with higher accuracy will receive a larger weight. In this way, we can make full use of the feature information of each frequency band and improve the accuracy of classification.

[0196] The purpose of using soft voting is to:

[0197] Frequency band information fusion: The polarization fingerprint of the frequency hopping signal is scattered on multiple frequency bands, and the characteristics of each frequency band may be different. The soft voting mechanism can fuse these scattered information to improve the accuracy of classification.

[0198] Reasonable weight allocation: By using classification accuracy as a weight, frequency bands with high accuracy can be given greater weight, thereby improving the reliability of classification results.

[0199] Adapting to noise and interference: In practical applications, signals may be affected by noise and interference, resulting in inaccurate classification results for certain frequency bands. The soft voting mechanism can reduce the impact of noise by integrating information from multiple frequency bands.

[0200] The specific implementation methods of soft voting include:

[0201] Multiply the classification result of each frequency band by the corresponding weight, and then add all the weighted results to obtain a new vector, namely the target slice data vector, and select the category with the highest probability in the new vector (target slice data vector) as the final classification result.

[0202] Optionally, the embodiment of the present invention may train the classification model in the following manner.

[0203] The sliced ​​data is divided into three parts: training set, validation set and test set.

[0204] Exemplarily, the division method may include:

[0205] Random division: The processed sample set is randomly divided into training set, validation set and test set.

[0206] Proportional allocation: The training set is used to train the model, the validation set is used to adjust the model hyperparameters, and the test set is used to evaluate the generalization performance of the model.

[0207] Partition ratio: Usually, the training set accounts for 70%, the validation set accounts for 15%, and the test set accounts for 15%.

[0208] Training set: used to train the model so that the model can learn the characteristics and patterns in the data.

[0209] Validation set: used to adjust the model's hyperparameters, such as learning rate, network depth, etc., to avoid overfitting.

[0210] Test set: used to evaluate the generalization ability of the trained model on never-seen data.

[0211] By dividing the model into training sets, validation sets, and test sets, the performance of the model can be objectively evaluated to ensure the reliability of the model.

[0212] The purpose of determining the true category label for the slice data and calculating the loss function for the classification model through the initial classification result and the true category label is to calculate the difference between the model prediction result and the true label, thereby guiding the optimization of the model.

[0213] Get the true label: As can be seen from the above, for each slice, its corresponding true category label is already known.

[0214] Calculate loss: Compare the model's prediction results with the true labels, and calculate the loss value based on the selected loss function (such as cross entropy loss, mean square error, etc.). The significance of this is that the loss function is the goal of model optimization. By minimizing the loss function, the model can continuously adjust parameters to improve prediction accuracy.

[0215] The purpose of training the classification model based on the training set, validation set, test set and the loss function is to adjust the parameters of the model through continuous iteration so that the model can accurately classify new data.

[0216] The specific training methods can be as follows:

[0217] method:

[0218] Choose an optimizer: Choose a suitable optimizer, such as stochastic gradient descent (SGD), Adam, etc., to update the parameters of the model.

[0219] Iterative training:

[0220] Randomly extract a batch of data from the training set.

[0221] Input the data into the model and get the predicted results.

[0222] Calculate the loss function.

[0223] Based on the loss function, an optimizer is used to update the parameters of the model.

[0224] Validation: At the end of each epoch, the validation set is used to evaluate the performance of the model and the hyperparameters of the model are adjusted based on the performance on the validation set.

[0225] The goal of the training process is to find a set of model parameters that enables the model to achieve good performance on both the training set and the validation set.

[0226] The advantages of generating classification results and training classification models in the above way are:

[0227] 1. Improve classification accuracy:

[0228] Slicing: Dividing data into smaller segments can better capture local features and improve the classification model's sensitivity to subtle differences.

[0229] Feature extraction: By extracting slice features, the data can be described more comprehensively, thereby improving the accuracy of classification.

[0230] Weighted average: Taking the weighted average of the classification results of different slices can effectively reduce the impact of noise and improve the reliability of the classification results.

[0231] 2. Adapt to non-stationary signals:

[0232] Characteristics of frequency hopping signals: The frequency of frequency hopping signals is constantly changing, and this non-stationarity brings challenges to classification.

[0233] Slicing processing: By slicing the signal, the non-stationary signal can be decomposed into multiple relatively stable segments, thereby reducing the impact of non-stationarity on classification.

[0234] 3. Improve the generalization ability of the model:

[0235] Division of training set, validation set, and test set: By dividing the data into training set, validation set, and test set, overfitting can be effectively avoided and the generalization ability of the model can be improved.

[0236] Optimization of loss function: By minimizing the loss function, the model can better fit the training data and achieve better performance on the test set.

[0237] 4. Enhance the robustness of the model:

[0238] Weighted average: Weighted average can reduce the impact of single slice classification errors on the final result and enhance the model’s robustness to noise and interference.

[0239] 5. Applicable to various scenarios:

[0240] Universality: This method is not only applicable to frequency hopping signal classification, but can also be applied to other types of time series data classification problems.

[0241] The classification method described in this text effectively solves the challenges in the frequency hopping signal classification problem through the steps of slicing processing, feature extraction, classification model training and weighted averaging, and improves the accuracy, robustness and generalization ability of classification. This method has wide applicability in practical applications.

[0242] Further advantages may include:

[0243] Fine-grained analysis: Slicing allows for more fine-grained analysis of signals.

[0244] Multi-angle feature extraction: Combining multiple features can describe the signal more comprehensively.

[0245] Ensemble learning idea: Weighted average is equivalent to a simple ensemble learning method, which can improve classification performance.

[0246] Strong adaptability: It can adapt to different types of signals and classification tasks.

[0247] In an optional embodiment of the present invention, the step of using the classification result to identify the identity information of the target device includes:

[0248] determining a confidence score for the classification result;

[0249] When the confidence score is higher than a preset score threshold, the target device is determined to be a valid device;

[0250] The preset device whitelist is traversed, and when the valid device matches a device in the preset device whitelist, the target device is determined to be a legitimate device.

[0251] Exemplarily, the identity of the target device may be determined in the following manner.

[0252] Threshold judgment: Compare the classification result with the preset threshold. If the confidence of the classification result is higher than the threshold, the classification result is considered valid.

[0253] Whitelist comparison: The device category obtained by classification is compared with the pre-set legal device whitelist. If the classification result is consistent with the category in the whitelist, the device is considered a legal device, otherwise it is an illegal device.

[0254] Manual processing: If the confidence of the classification result is lower than the threshold, the sample is marked as requiring manual processing.

[0255] Beneficial effects:

[0256] Implement device identity authentication: By comparing the classification results with the whitelist, the device identity can be quickly authenticated.

[0257] Reduce the false positive rate: By setting the threshold, the false positive rate can be effectively reduced.

[0258] Improve system security: By manually handling unknown devices, the intrusion of illegal devices can be effectively prevented.

[0259] In order to enable those skilled in the art to better understand the embodiment of the present invention, an example is used below to illustrate the embodiment of the present invention.

[0260] The embodiment of the present invention can be applied to a recognition model for a target device, and the recognition model mainly includes a training module and a recognition module.

[0261] The training module consists of a data acquisition and preprocessing unit, a training unit and a discrimination unit.

[0262] The recognition module is composed of data acquisition and preprocessing unit, classification model and recognition unit.

[0263] Data collection and preprocessing unit: used to complete data collection, device fingerprint characterization, data sample slicing, data set division and other tasks.

[0264] Training unit: used to complete the initial model training and secondary training after the device whitelist is updated.

[0265] Classification model: used to complete the classification of untrained samples and output the results to the recognition unit. The intelligent classification model compares the branches to determine whether the device is a legitimate device.

[0266] Data acquisition and preprocessing unit:

[0267] Assume that there is an orthogonal dual-polarized antenna on the receiver, and its horizontal and vertical linear polarization bases are represented as (H, V). The received signal is r(t) = [r H (t)r V (t)] T , the signal frequency range is (f L ,f H ). Frequency f k The corresponding repolarization ratio is

[0268]

[0269] Where |R H [k]| and |R V [k]|respectively r H (t) and r V (t) at frequency f k The magnitude of argR H [k] and argR V [k] are r H (t) and r V (t) at frequency f k phase.

[0270] P(f k ) is called the receiver at the incident angle The polarization fingerprint value of the transmitter collected at is the elevation angle, and θ is the azimuth angle. k ) is expressed as follows:

[0271]

[0272] in is the frequency dependence of the polarization fingerprint, which includes the structural characteristics of the transmitter antenna and unique hardware defects, is the directional dependence of the polarization fingerprint, which describes the relative position between the transmitter and the receiver.

[0273] Then the polarization fingerprint corresponding to the received signal r(t) is

[0274]

[0275] It can be seen that the polarization fingerprint is a complex array, each element in the array represents a sampling point. Each vector It consists of samples from multiple frequency bands, and the number of sampling points in each frequency band is equal.

[0276] According to the independent characteristics of each frequency band of the frequency hopping signal, the polarization fingerprint is sliced, and the sliced ​​data of each sample is the total number of frequency bands. Data slicing can first meet the requirements of fixed-length input of reinforcement learning, and secondly, it can better deal with the problem that it is always impossible to capture data of each frequency band in the actual sampling process. Finally, the sliced ​​data set is divided into three parts: training set, validation set and test set.

[0277] refer to Figure 2 and Figure 3 , Figure 3 is a flow chart of a device identity recognition method provided in an embodiment of the present invention; Figure 3 is a flow chart of a device identity recognition method provided in an embodiment of the present invention;

[0278] The intelligent agent of reinforcement learning can associate independent features in the frequency hopping signal through self-supervised learning, extract the characteristic signal of a certain frequency band, and thus determine its category. The frequency hopping signal of a certain frequency band is used as the input of the intelligent agent, and the prediction result is used as feedback to motivate the intelligent agent to learn new features. The logical likelihood function of the benchmark value and the prediction result is used as the loss function to train the intelligent agent model, thereby obtaining a classification model.

[0279] The agent classifier consists of two parts: the classification module and the voting module.

[0280] 1) Classification Model:

[0281] This classification module is used to complete the single-band classification of unknown samples. Since frequency-hopping signals use several different frequency bands, their fingerprints are fragmented, and in application scenarios, it is usually difficult to collect fingerprints of all frequency bands. Therefore, the experiment first classifies each frequency band of the sample. The pseudo code of the classification algorithm is as follows:

[0282] Input: effectNum, array of effective frequency bands of samples, testSample, preprocessed test set, snrNum, total number of samples under single signal-to-noise ratio, sampleNum

[0283] Output: true value matrix y_tru, classification result matrix y_output, matrix weight weight

[0284] The classification model can be an Agent model, as shown below:

[0285]

[0286]

[0287] 2) Voting module

[0288] The voting mechanism can fuse multiple classification results to improve their overall performance. Usually, voting methods can be divided into hard voting and soft voting. The principle of hard voting is "the minority obeys the majority". Each classifier gives a classification result independently, and the category with the most votes is output as the final classification result. Hard voting is mostly used in situations where individual classifiers are quite different. In soft voting, each classifier gets a pre-assigned weight. During the result fusion process, the category corresponding to the maximum probability is obtained by calculating the weighted average of all classification results, and is output as the final classification result. Different classification models of the voting classifier can be assigned different weights to improve classification accuracy. The fingerprint features of each frequency band are not similar, and the features of each frequency band complement each other. It is necessary to consider the comprehensive results for classification output. This time we use soft voting to fuse the classification results.

[0289] The structure and hyperparameters of each classifier are set to be consistent, but the classification objects are different. We believe that frequency bands with higher accuracy can provide more valuable features, so the classification weight of each frequency band is proportional to the accuracy of the data of each frequency band in the training set during model training. The voting module calculates the weighted average of each category, and the category with the highest probability is considered to be the final classification result.

[0290] Identification / decision making unit:

[0291] The recognition / decision unit determines the classification result by a preset threshold. If the maximum recognition value corresponding to the sample is greater than the threshold, the classification result is considered valid, and the corresponding category name is compared with the whitelist to determine whether the device is a legitimate device. If the classification result is less than the threshold, the classification result is considered invalid and the sample is transferred to the manual processing unit.

[0292] Step S1: Data collection and preprocessing:

[0293] Purpose: To extract polarization fingerprint features that can be used for device identification from the received frequency hopping signal and preprocess the data to provide high-quality data for subsequent classification.

[0294] Specific process:

[0295] Polarization fingerprint extraction: Use a specific mathematical formula to process the received signal and extract the polarization fingerprint that can reflect the hardware characteristics of the device. Polarization fingerprint is essentially a quantitative description of the signal polarization characteristics, including information about hardware components such as the device antenna and RF front end.

[0296] Data channel processing: The real and imaginary parts of the extracted polarization fingerprint are processed into one channel respectively to form a dual-channel data set. This processing method can more comprehensively characterize the characteristics of the polarization fingerprint.

[0297] Data slicing: Considering the independence of each frequency band of the frequency hopping signal, the original data is sliced ​​according to the number of sampling points in each frequency band. This slicing process can divide the continuous signal into multiple independent segments, which is convenient for subsequent feature extraction and classification.

[0298] Dataset partitioning: The sliced ​​data is divided into training set, validation set and test set. The training set is used to train the classification model, the validation set is used to adjust the model parameters, and the test set is used to evaluate the generalization performance of the model.

[0299] Beneficial effects:

[0300] Improve feature extraction efficiency: By mathematically processing polarization fingerprints, features useful for device identification can be quickly extracted.

[0301] Enhanced feature expression capability: Processing the real and imaginary parts of the polarization fingerprint into one channel respectively can more comprehensively characterize the characteristics of the polarization fingerprint and improve the recognition accuracy of the model.

[0302] Facilitates model training: After slicing the data, it can better adapt to the input requirements of the deep learning model and improve the training efficiency of the model.

[0303] Ensure the reliability of the model: By dividing the model into training sets, validation sets, and test sets, the performance of the model can be objectively evaluated to ensure the reliability of the model.

[0304] Step S2: Fingerprint classification:

[0305] Objective: To classify the extracted polarization fingerprints using the trained agent classification model and to improve the classification accuracy by fusing the results of multiple classifiers through a soft voting mechanism.

[0306] Specific process:

[0307] Model classification: The slice data obtained in step S1 is input into the pre-trained agent classification model. The model will classify each slice data and output the probability distribution of the category to which the slice belongs.

[0308] Soft voting: The output results of multiple classifiers are weighted averaged to obtain the final classification result. The soft voting mechanism can effectively integrate the classification results of multiple models and improve the accuracy of classification.

[0309] Voting module: further processes the results of soft voting, calculates the score of each category, and determines the final classification result based on the size of the score.

[0310] Beneficial effects:

[0311] Improve classification accuracy: The intelligent agent classification model has a strong feature learning ability and can extract features useful for classification from high-dimensional data to improve classification accuracy.

[0312] Enhance model robustness: The soft voting mechanism can effectively reduce the misclassification rate of a single model and improve the robustness of the model.

[0313] Improve classification efficiency: By processing multiple classifiers in parallel, the efficiency of classification can be improved.

[0314] Step S3: Device identification:

[0315] Purpose: To determine whether a device is a legitimate device based on the classification result.

[0316] Specific process:

[0317] Threshold judgment: Compare the classification result with the preset threshold. If the confidence of the classification result is higher than the threshold, the classification result is considered valid.

[0318] Whitelist comparison: The device category obtained by classification is compared with the pre-set legal device whitelist. If the classification result is consistent with the category in the whitelist, the device is considered a legal device, otherwise it is an illegal device.

[0319] Manual processing: If the confidence of the classification result is lower than the threshold, the sample is marked as requiring manual processing.

[0320] Beneficial effects:

[0321] Implement device identity authentication: By comparing the classification results with the whitelist, the device identity can be quickly authenticated.

[0322] Reduce the false positive rate: By setting the threshold, the false positive rate can be effectively reduced.

[0323] Improve system security: By manually handling unknown devices, the intrusion of illegal devices can be effectively prevented.

[0324] Summarize:

[0325] The above three steps together constitute a complete device identification system based on polarization fingerprint. The system achieves accurate identification of device identity by extracting polarization fingerprint features, using intelligent classification models for classification, and combining soft voting and threshold judgment methods. The system has high recognition accuracy, robustness and security, and is suitable for fields such as wireless Internet of Things.

[0326] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0327] Reference Figure 4 , shows a structural block diagram of a device identity recognition apparatus provided in an embodiment of the present invention, which may specifically include the following modules:

[0328] The target polarization fingerprint feature acquisition module 401 is used to acquire a frequency hopping signal and acquire a target polarization fingerprint feature from the frequency hopping signal; the target polarization fingerprint feature includes target real part data and target imaginary part data;

[0329] A dual-channel data set construction module 402 is used to construct a dual-channel data set containing the target polarization fingerprint feature based on the target real data and the target imaginary data;

[0330] The slice data generating module 403 is used to determine a target frequency band, and to perform data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band to generate slice data;

[0331] A classification result generating module 404, configured to perform a fingerprint classification operation based on the slice data to generate a classification result for a target device;

[0332] The identity information recognition module 405 is used to recognize the identity information of the target device using the classification result.

[0333] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0334] In addition, an embodiment of the present invention further provides an electronic device, such as Figure 5As shown, it includes a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504.

[0335] Memory 503, used for storing computer programs;

[0336] The processor 501 is configured to implement the device identification method described in any one of the above embodiments when executing the program stored in the memory 503:

[0337] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0338] The communication interface is used for communication between the above terminal and other devices.

[0339] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0340] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0341] like Figure 6As shown, in another embodiment provided by the present invention, a computer-readable storage medium 601 is also provided, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the device identification method described in the above embodiment.

[0342] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

[0343] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the embodiments of the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0344] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0345] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0346] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0347] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0348] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical disks.

[0349] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A device identification method, characterized in that: include: Acquire a frequency hopping signal, and acquire a target polarization fingerprint feature from the frequency hopping signal; The target polarization fingerprint feature includes target real data and target imaginary data; Constructing a dual-channel data set including the target polarization fingerprint feature based on the target real data and the target imaginary data; Determine a target frequency band, and perform data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points in the target frequency band to generate slice data; Performing a fingerprint classification operation based on the slice data to generate a classification result for the target device; The classification result is used to identify the identity information of the target device.

2. The method according to claim 1, characterized in that The step of determining the target frequency band comprises: Determine the number and width of frequency bands, and determine the start and end frequencies of each frequency band; A target frequency band is determined based on the number of frequency bands, the frequency band width, the start frequency, and the end frequency.

3. The method according to claim 1, characterized in that The step of performing data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points in the target frequency band to generate sliced ​​data comprises: For the target polarization fingerprint feature in the dual-channel data set, data slicing is performed based on the number of sampling points in the target frequency band to generate initial slicing data; Performing a spectrum analysis operation on the initial slice data to generate a spectrum analysis result for the initial slice data; When it is determined that the cutting is accurate based on the spectrum analysis result, the initial slice data is determined as the slice data.

4. The method according to claim 3, characterized in that The step of performing a fingerprint classification operation based on the slice data to generate a classification result for the target device includes: Traversing the slice data, and determining the slice features and slice index of the slice data; Inputting the slice feature and the slice index into a classification model, and outputting an initial classification result for the slice data through the classification model; Determining a classification accuracy probability of the initial classification result, and determining the classification accuracy probability as a slice weight for the slice data; A weighted average operation is performed on the initial classification results of all slice data using the slice weights to determine a classification result for the target device.

5. The method according to claim 4, characterized in that The step of performing a weighted average operation on the initial classification results of all slice data by using the slice weights to determine the classification result for the target device includes: Multiply the classification results of all slice data by the corresponding slice weights, and add all weighted results to generate the target slice data vector; The category with the highest probability in the target slice data vector is determined as the classification result for the target device.

6. The method according to claim 4, characterized in that Also includes: The slice data is divided into a training set, a validation set and a test set; Determine a true category label for the slice data, and calculate a loss function for the classification model using the initial classification result and the true category label; The classification model is trained based on the training set, the validation set, the test set and the loss function.

7. The method according to claim 1 or 4, characterized in that: The step of using the classification result to identify the identity information of the target device includes: determining a confidence score for the classification result; When the confidence score is higher than a preset score threshold, the target device is determined to be a valid device; The preset device whitelist is traversed, and when the valid device matches a device in the preset device whitelist, the target device is determined to be a legitimate device.

8. A device identification device, characterized in that: include: A target polarization fingerprint feature acquisition module, used to acquire a frequency hopping signal and acquire a target polarization fingerprint feature from the frequency hopping signal; The target polarization fingerprint feature includes target real data and target imaginary data; A dual-channel data set construction module, used to construct a dual-channel data set containing the target polarization fingerprint feature based on the target real data and the target imaginary data; A slice data generation module, used to determine a target frequency band, and perform data slicing on the target polarization fingerprint feature in the dual-channel data set based on the number of sampling points of the target frequency band to generate slice data; A classification result generating module, used to perform a fingerprint classification operation based on the slice data to generate a classification result for a target device; An identity information recognition module is used to use the classification result to recognize the identity information of the target device.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; The memory is used to store computer programs; The processor is used to implement the method according to any one of claims 1 to 7 when executing the program stored in the memory.

10. A computer-readable storage medium having instructions stored thereon, which, when executed by one or more processors, cause the processors to perform the method according to any one of claims 1 to 7.

Images