Equipment fingerprint feature extraction method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to an equipment fingerprint feature extraction method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring a to-be-processed signal data set; each piece of to-be-processed signal data in the to-be-processed signal data set carries a level value; counting the distribution condition of the level values on the to-be-processed signal data set to obtain level distribution characteristics; calculating a gradient change condition of the level value on the to-be-processed signal data set to obtain a signal gradient feature; and splicing the level distribution feature and the signal gradient feature to obtain an equipment fingerprint feature. By adopting the method, the cost of extracting the fingerprint features of the equipment can be reduced.

Description

Technical Field

[0001] This application relates to the field of information security technology, and in particular, to a method and device for extracting device fingerprint features, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] A device fingerprint is a hardware characteristic formed by the minute differences between the electronic components of a device, aiming to provide a reliable means of device identity recognition for computer network security to capture the unique features of the hardware. With the continuous development and improvement of technology, device fingerprint feature extraction technology has been widely used in the security field, such as in intrusion detection, identity recognition, access authentication, etc.

[0003] Currently, existing methods for feature extraction based on spectra and the like have a common challenge, that is, a relatively high sampling frequency is required for signal sampling, which leads to the problem of high fingerprint extraction costs. Summary of the Invention

[0004] Based on this, it is necessary to provide a method and device for extracting device fingerprint features, a computer device, a computer-readable storage medium, and a computer program product that can reduce costs in view of the above technical problems.

[0005] In a first aspect, this application provides a method for extracting device fingerprint features, and the method includes:

[0006] Obtain a dataset of signals to be processed; each signal to be processed in the dataset of signals to be processed carries a level value;

[0007] Statistically analyze the distribution of the level values on the dataset of signals to be processed to obtain a level distribution feature;

[0008] Calculate the gradient change of the level values on the dataset of signals to be processed to obtain a signal gradient feature;

[0009] Concatenate the level distribution feature and the signal gradient feature to obtain a device fingerprint feature.

[0010] In one embodiment, the statistically analyzing the distribution of the level values on the dataset of signals to be processed to obtain a level distribution feature includes:

[0011] Based on each of the level values, obtain a plurality of data segments;

[0012] Statistically analyze the frequency of each level value in each of the data segments to obtain the level distribution feature;

[0013] The calculating the gradient change of the level values on the dataset of signals to be processed includes:

[0014] Perform central difference convolution processing on each of the data segments to obtain the signal gradient feature.

[0015] In one embodiment, the step of statistically analyzing the frequencies of the respective level values in each of the data segments to obtain the level distribution feature includes:

[0016] Divide the amplitude of the level value corresponding to each data segment into a first preset number of first intervals;

[0017] Statistically analyze the frequencies of the level values in each of the first intervals;

[0018] Weight the frequencies of the level values of each of the first intervals in each of the data segments to obtain the level distribution feature.

[0019] In one embodiment, the step of performing central difference convolution processing on each of the data segments to obtain the signal gradient feature includes:

[0020] Perform convolution calculation on each of the data segments through a preset convolution kernel to obtain the convolution value corresponding to each data segment.

[0021] Divide the convolution value corresponding to each data segment into a second preset number of second intervals;

[0022] Statistically analyze the frequencies of the convolution values in each of the second intervals;

[0023] Weight the frequencies of the convolution values of each of the second intervals in each of the data segments to obtain the signal gradient feature.

[0024] In one embodiment, after obtaining the level distribution feature, the following steps are further included:

[0025] Perform filtering processing on the level distribution feature;

[0026] After obtaining the signal gradient feature, the following steps are further included:

[0027] Perform filtering processing on the signal gradient feature;

[0028] The filtering processing is to filter out the null values in the level distribution feature or the signal gradient feature.

[0029] In one embodiment, the convolution value is obtained by performing convolution calculation on each of the data segments through a preset convolution kernel; the generation method of the preset convolution kernel includes:

[0030] Determine a third preset number of initial convolution kernels according to the sampling frequency of the signal dataset to be processed;

[0031] Convolve the sample data using each of the initial convolution kernels to obtain reference convolution features;

[0032] Perform a correlation comparison on each of the reference convolution features to obtain the preset convolution kernel.

[0033] In one embodiment, the method further includes:

[0034] Identify the device fingerprint feature to obtain a device identification result.

[0035] In one embodiment, the identifying the device fingerprint feature to obtain a device identification result is obtained through a fingerprint recognition model; the training process of the fingerprint recognition model includes:

[0036] Obtain training data; the training data carries class labels;

[0037] Input the training data into an initial recognition model to obtain a predicted device identification result; the initial recognition model includes a plurality of initial models;

[0038] Based on the predicted device identification result and the class label, obtain a loss function, and adjust the initial recognition model according to the loss function until the training is completed to obtain the fingerprint recognition model. In one embodiment, the obtaining the dataset of the signal to be processed includes:

[0039] Collect device signals through a low-frequency sampling device;

[0040] Modulate the device signal into an original signal;

[0041] Preprocess the original signal to obtain the dataset of the signal to be processed;

[0042] The preprocessing includes any one of filtering and segmentation;

[0043] The filtering is to filter out the noise signals in the original signal through a preset filtering method;

[0044] The segmentation is to segment the original signal according to a preset length to obtain the dataset of the signal to be processed.

[0045] In a second aspect, the present application further provides a device fingerprint feature extraction device, and the device includes:

[0046] An acquisition module, configured to acquire a dataset of a signal to be processed; the dataset of the signal to be processed carries a level value;

[0047] A level feature extraction module, configured to count the distribution of the level values on the dataset of the signal to be processed to obtain a level distribution feature;

[0048] A convolutional feature extraction module, configured to calculate the gradient change of the level value on the to-be-processed signal dataset to obtain signal gradient features;

[0049] A feature fusion module, configured to splice the level distribution features and the signal gradient features to obtain device fingerprint features.

[0050] In a third aspect, the present application further provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the above method are implemented.

[0051] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0052] In a fifth aspect, the present application further provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the above method are implemented.

[0053] For the above device fingerprint feature extraction method, device, computer device, computer-readable storage medium, and computer program product, after obtaining the to-be-processed signal dataset, the distribution of the level value on the to-be-processed signal dataset is respectively statistically analyzed, and the gradient change of the level value on the to-be-processed signal dataset is calculated to obtain level distribution features and signal gradient features. In this way, the to-be-processed signal dataset can be fully processed. Finally, the level distribution features and the signal gradient features are spliced to obtain the final device fingerprint features, reducing the limitation of the signal sampling frequency, and while reducing the cost of device fingerprint feature extraction, still ensuring effective fingerprint extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for the description of the embodiments of the present application or related technologies. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 It is a schematic flowchart of a device fingerprint feature extraction method in an embodiment;

[0056] Figure 2 It is a schematic diagram of a level statistical histogram in an embodiment;

[0057] Figure 3 It is a schematic diagram of the selection of a central difference convolution kernel in an embodiment;

[0058] Figure 4 Schematic diagram of signal distortion collected by a low-frequency sampling device in an embodiment;

[0059] Figure 5 Schematic flowchart of a device fingerprint recognition method in an embodiment;

[0060] Figure 6 Block diagram of the structure of a device fingerprint feature extraction device in an embodiment;

[0061] Figure 7 Internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0062] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0063] Ethernet technology is a data communication technology widely used in local area network (LAN) environments. Its basic principle is to use a shared transmission medium to transmit data frames to achieve communication between network devices. Since its inception in the 1970s, Ethernet technology has undergone continuous evolution and standardization, continuously improving its transmission rate, bandwidth and performance. After years of development and improvement, it has become one of the most common and important technologies in modern networks.

[0064] From the initial 10BASE-T to 100BASE-T, and then to the IEEE802.3z Gigabit Ethernet standard 1000BASE-X specification based on fiber optic media, and finally evolved to the current Gigabit Ethernet technology. In this series of developments, 1000BASE-T technology occupies an important position in Ethernet technology. It uses Category 5 unshielded twisted pair (UTP) as the transmission medium, achieves a transmission rate of gigabits per second, and supports full-duplex communication, thus greatly improving the network throughput and efficiency. In addition, 1000BASE-T technology also particularly considers the compatibility of the existing cable environment, has adaptive rate and auto-negotiation functions, and can dynamically adjust communication parameters with other devices to ensure the best performance of the network.

[0065] Device fingerprint refers to the hardware characteristics formed by the minute differences between the electronic components of a device. This concept originally stemmed from the computer field and aimed to provide a reliable means of device identity recognition for computer network security to capture the unique characteristics of the hardware. By authenticating and identifying the physical characteristics of a communication device, the true identity of the device can be accurately determined, thereby preventing malicious attacks. Currently, device fingerprint extraction technology has been widely applied in the security field, such as in intrusion detection, identity recognition, access authentication, etc.

[0066] With the continuous development and improvement of technology, significant progress has been made in device fingerprint extraction technology. There are various extraction methods, including extracting fingerprints using features such as signal statistics and transient changes. The development of these technologies has made device fingerprint extraction technology more mature and has become an important part of physical layer security. However, it should be noted that although these fingerprints have a high degree of uniqueness and are difficult to forge, currently most fingerprint extraction methods are only applicable to devices in low transmission rate network environments, and for gigabit Ethernet with high transmission rates, the application effect is not good. In the field of gigabit Ethernet, although some related fingerprint extraction methods have been proposed, their perfection still needs to be further strengthened. In addition, existing methods such as those based on spectrum and other feature extractions have a common challenge, that is, a high sampling frequency is required for signal sampling, which leads to the problem of high fingerprint extraction costs. Therefore, it is necessary to research and propose a new fingerprint extraction method that can adapt to lower sampling frequencies, thereby effectively reducing the cost of fingerprint extraction.

[0067] In an exemplary embodiment, as Figure 1 shown, a device fingerprint feature extraction method is provided. In this embodiment, this method is exemplified by its application to a terminal. It can be understood that this method can also be applied to a server and can also be applied to a system including a terminal and a server and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0068] Step S102, obtain a dataset of signals to be processed; each signal data to be processed in the dataset of signals to be processed carries a level value.

[0069] Among them, the dataset of signals to be processed is a set composed of multiple signal data to be processed, and each signal data to be processed can reflect various characteristics and changes of the device during the sampling period. These signal data to be processed are extracted from the collected device signals through a series of digital processing technologies (such as decoding, demodulation, synchronization, and channel estimation) and converted into digital signals that can be processed by a computer.

[0070] Among them, each signal data to be processed contains a timestamp and a corresponding level value, forming a time series data that reflects the state and changes of the device during the sampling period.

[0071] Step S104, statistically analyze the distribution of the level values in the dataset of the signal to be processed to obtain the level distribution characteristics.

[0072] Among them, the level distribution characteristics can effectively reflect the distribution of the signal energy density in the level interval and are not sensitive to the continuity of the signal. Compared with methods such as spectrum analysis and power spectrum feature extraction, this technology is more suitable for low-frequency signal sampling. Moreover, the level distribution characteristics reflect the occurrence frequency and pattern of the signal at different levels, and these characteristics can be used to identify and distinguish different devices.

[0073] Exemplarily, each device may have different distribution characteristics of level values when transmitting signals. For example, some devices may have a preference for specific level values, or due to differences in hardware design, the distribution of level values will be different.

[0074] Exemplarily, the distribution of level values can reflect the noise and interference characteristics of the device during signal transmission. The noise performance of different devices may be different at different level values, and these subtle differences can be used to distinguish devices. For example, some devices may exhibit higher noise at specific level values, while other devices may perform more stably.

[0075] Therefore, by analyzing the distribution of level values, the unique characteristics of the device can be captured and used to identify the device as part of the device fingerprint.

[0076] Optionally, traverse the signal data and statistically analyze the occurrence frequency of each level value. Then, a histogram or frequency distribution table can be used to represent the distribution of level values. Exemplarily, it can be combined with Figure 2 as shown Figure 2 to be a schematic diagram of the level statistical histogram in an embodiment.

[0077] Step S106, calculate the gradient change of the level values in the dataset of the signal to be processed to obtain the signal gradient characteristics.

[0078] The gradient change of the level values can reflect the characteristics and behaviors of the device, thereby helping to identify and distinguish different devices.

[0079] Exemplarily, due to slight differences in the hardware manufacturing process of each device, different level change patterns may be exhibited during signal transmission. These slight differences can be reflected by the gradient change of the level values, thereby being used to identify specific devices.

[0080] For example, there may be differences in the performance of the transmitters and receivers of different devices, which will result in different speeds and amplitudes of level changes.

[0081] For example, due to different circuit designs and layouts, different devices may exhibit different delay and reflection characteristics, all of which can be captured by the gradient change of the level value.

[0082] Similarly, the gradient change situation can reflect the unique characteristics of the device. Therefore, the signal gradient feature is used as part of the device fingerprint to identify the device.

[0083] Optionally, the gradient change of the level value on the dataset of the signal to be processed can be calculated by methods such as the difference method, convolutional difference, and sliding window method.

[0084] Step S108: Concatenate the level distribution feature and the signal gradient feature to obtain the device fingerprint feature.

[0085] Optionally, the level distribution feature and the signal gradient feature can be concatenated by means of cascade fusion, that is, the level distribution feature and the signal gradient feature are gradually combined. Exemplarily, if the dataset of the signal to be processed includes multiple signals to be processed, then the level distribution feature and the signal gradient feature corresponding to each signal to be processed are concatenated.

[0086] Optionally, the level distribution feature and the signal gradient feature can be concatenated by means of weighted fusion, that is, a weighted sum of the level distribution feature and the signal gradient feature is performed, and the weights can be assigned according to the importance or confidence of the features.

[0087] Optionally, the level distribution feature and the signal gradient feature can be concatenated by means of feature transformation fusion, that is, the level distribution feature and the signal gradient feature are mapped to another space, and then the transformed features are fused.

[0088] It should be noted that in this embodiment, the cascade fusion method is selected to concatenate the level distribution feature and the signal gradient feature to obtain the device fingerprint feature.

[0089] Exemplarily, (f1, f2, f3) is the level distribution feature, (f a , f b , f c ) is the signal gradient feature, and the device fingerprint feature obtained by cascading and fusing the level distribution feature and the signal gradient feature is (f1, f2, f3, f a , f b , f c ).

[0090] In the above method for extracting device fingerprint features, after obtaining the dataset of signals to be processed, the distribution of level values on the dataset of signals to be processed is respectively counted, and the gradient change of the level values on the dataset of signals to be processed is calculated, so as to obtain the level distribution features and signal gradient features. In this way, the dataset of signals to be processed can be fully processed. Finally, the level distribution features and signal gradient features are concatenated to obtain the final device fingerprint features, reducing the limitation of the signal sampling frequency. While reducing the cost of extracting device fingerprint features, effective fingerprint extraction is still ensured.

[0091] In one embodiment, multiple data segments are obtained based on each level value; the distribution of level values on the dataset of signals to be processed is counted to obtain the level distribution features, including: counting the frequency of each level value in each data segment to obtain the level distribution features; calculating the gradient change of the level values on the dataset of signals to be processed, including: performing central difference convolution processing on each data segment to obtain the signal gradient features.

[0092] In this embodiment, a dataset of signals to be processed includes multiple data segments, where the data segments are segmented by the level values of the signals, so as to better analyze, process, and utilize the signal data, and further fully extract the level distribution features and signal gradient features. Exemplarily, the dataset of signals to be processed is divided into T segments, and each segment of data will be used to generate 1 fingerprint feature.

[0093] Optionally, the frequency of occurrence of level values in each data segment is respectively counted, and the frequency of occurrence of level values in each data segment is subjected to a preset process to obtain the level distribution features of the current dataset of signals to be processed.

[0094] Optionally, each data segment is processed respectively through central difference convolution operation to obtain the gradient change of the level values on each data segment, and the gradient change of each data segment is subjected to a preset process to obtain the signal gradient features of the current dataset of signals to be processed.

[0095] Optionally, the preset process can be processing methods such as average, weighted summation, median, and variance, so as to comprehensively consider the distribution of level values and gradient changes in each data segment, and obtain the level distribution features and signal gradient features of the data segments of signals to be processed.

[0096] In the above embodiment, by further dividing the dataset of signals to be processed into multiple data segments, and respectively counting the frequency and gradient change of level values, the signal data can be fully processed. This method realizes effective fingerprint feature extraction under the condition of low-frequency sampling. Compared with the traditional fingerprint extraction method, the sampling frequency can be reduced to 1% of the original frequency, realizing lightweight and low-cost fingerprint authentication.

[0097] In one embodiment, the frequencies of each level value in each data segment are counted to obtain the level distribution characteristics, including: dividing the amplitude of the level value corresponding to each data segment into a first preset number of first intervals; counting the level value frequencies of the level values in each first interval; and weighting the level value frequencies of each first interval in each data segment to obtain the level distribution characteristics.

[0098] Optionally, in this embodiment, the data segment is further divided, which can make the discrimination of the level distribution characteristics better. Each data segment is divided into a first preset number of first intervals. Wherein, the first preset number is preset, and the data segment is divided into multiple first intervals by setting the first preset number.

[0099] Further, the amplitude of the level value corresponding to each data segment is divided into a first preset number of first intervals. Exemplarily, if the range of the level value is [-1.0, 1.0], it can be divided into 10 intervals: [-1.0, -0.8], [-0.8, -0.6],... [0.8, 1.0]. In this embodiment, the first preset number is 10.

[0100] It should be noted that the division of the level into intervals has nothing to do with the specific signal, but is divided according to the first preset number. The first interval is the first fingerprint feature, that is, the interval set divided for statistical analysis of the digital histogram feature. For example, the range of [-2, 2] is fixedly divided into 1000 intervals such as [-2, -1.996]... a total of 1000 intervals, and then the signal level is statistically analyzed according to this fixed interval set.

[0101] Optionally, the histogram method can be used to count the level value frequencies of the level values in each first interval. Exemplarily, by traversing the signal data, each level value is assigned to the corresponding interval, and the number of level values included in each interval is counted to obtain the level value frequency.

[0102] Exemplarily, the amplitude of the level value is uniformly divided into intervals, and the frequency of the level value appearing in each interval is counted to obtain the statistical set , that is, the level value frequency, which can be specifically combined with Figure 2 .

[0103] Optionally, the level value frequencies of each first interval in each data segment are weighted to obtain the level distribution characteristics. Exemplarily, the level value frequencies of each first interval in each data segment are weighted using formula (1), and formula (1) is as follows:

[0104] Formula (1)

[0105] Where n is the length of the feature set, that is, the number of divided level intervals, and the W of the T segments is optimized into one segment by weighted averaging Performing the above processing on each data set will result in N weighted average data sets The data set Y is the level distribution feature

[0106] In the above embodiment, the data segment is divided into smaller intervals, and the frequency of the level values is counted on the smaller intervals, so that the discrimination of the level distribution features generated on different data segments is better

[0107] In one of the embodiments, central difference convolution processing is performed on each data segment to obtain signal gradient features, including: dividing each data segment into a second preset number of second intervals according to the convolution value; counting the convolution value frequencies of the convolution values in each second interval; weighting the convolution value frequencies of each second interval in each data segment to obtain signal gradient features

[0108] Optionally, in this embodiment, the data segment is further divided, which can make the discrimination of the signal gradient features better. Each data segment is divided into a second preset number of second intervals. Among them, the second preset number is preset, and the data segment is divided into multiple second intervals by setting the second preset number

[0109] Furthermore, the convolution values corresponding to each data segment are divided into a second preset number of second intervals

[0110] Similarly, in this embodiment, the histogram method is used to count the convolution value frequencies of the convolution values in each second interval. Exemplarily, the interval range in this embodiment is [-3, 3] divided into 1000 intervals, and the result of the convolution difference (an array of the same length as the signal level sequence) is counted for the frequency, and the numerical frequencies falling in different intervals are counted

[0111] The convolution value is obtained by convolution calculation according to a preset convolution kernel, and the calculation method of the preset convolution kernel can be obtained according to the method for generating the preset convolution kernel in the following embodiment

[0112] Exemplarily, the convolution results are uniformly divided into intervals according to the numerical size, and the frequency of the convolution values appearing in each interval is calculated to obtain a statistical set That is, the convolution value frequency

[0113] Optionally, the convolution value frequencies of each second interval in each data segment are weighted to obtain signal gradient features. Exemplarily, the formula (2) is used to weight the level value frequencies of each first interval in each data segment. The formula (2) is as follows

[0114] Formula (2)

[0115] Where is the length of the feature set, that is, the number of divided level intervals. Divide segments of into one segment by weighted averaging. After performing the above processing on each data set, N weighted average data sets will be obtained . The data set is the signal gradient feature .

[0116] In this way, the data segment is divided into smaller intervals, and the frequency of the convolution value is counted on the smaller intervals, which makes the discrimination of the signal gradient features generated on different data segments better

[0117] Furthermore, in the above embodiment, the convolution value is obtained by performing convolution calculation on each data segment through a preset convolution kernel

[0118] Among them, when performing convolution on each data segment through the preset convolution kernel, the preset convolution kernel is set as a centrosymmetric vector, that is when, that is, performing central difference convolution processing on the signal, and the obtained will reflect the level difference of each signal point from its neighborhood signal with a length of 2m, that is, the overall gradient change of the signal

[0119] Since there are differences in the performance of different convolution kernels at different sampling frequencies, if complete fingerprint extraction is performed on different convolution kernels and the recognition rates are compared through subsequent machine learning steps, this prior determination method will cause a large amount of time and computational costs. Therefore, in another embodiment of the present application, a method for generating a preset convolution kernel is proposed

[0120] Among them, the method for generating a preset convolution kernel includes: determining a third preset number of initial convolution kernels according to the sampling frequency of the signal data set to be processed; respectively performing convolution on the sample data using each initial convolution kernel to obtain reference convolution features; and comparing the correlations of the reference convolution features to obtain a preset convolution kernel

[0121] Among them, the sample data refers to the data used for prior testing of the initial convolution kernel, and its form is the same as that in the central difference convolution of the signal data set to be processed in the above embodiment

[0122] First, determine the sampling frequency of the signal data set to be processed, and then determine a third preset number of initial convolution kernels with different lengths and coefficients according to the sampling frequency. Perform convolution processing on the sample data using the third preset number of initial convolution kernels respectively to obtain the reference convolution features corresponding to the third preset number of initial convolution kernels

[0123] Next, perform a correlation comparison on each reference convolutional feature. Exemplarily, assume that the length of the reference convolutional feature is , and calculate the absolute value sum R of the linear correlation coefficients between each feature variable of the reference convolutional feature and the device label.

[0124] Formula (3)

[0125] where r is the correlation coefficient, x is the feature vector, y is the label vector, n is the vector length, and the sum of the absolute values of the obtained correlation coefficients r is R.

[0126] Furthermore, select the initial convolutional kernel corresponding to the reference convolutional feature with the largest correlation coefficient with the device label among the third preset number of reference convolutional features, and use the initial convolutional kernel corresponding to the reference convolutional feature with the largest correlation coefficient with the device label as the preset convolutional kernel for subsequent fingerprint feature extraction of the signal dataset to be processed.

[0127] Exemplarily, as shown in Figure 3 , Figure 3 is a schematic diagram of the selection of the central difference convolutional kernel in an embodiment. Figure 3 The convolutional kernels 1-n in are the third preset number of initial convolutional kernels in the above embodiment, and the signal data sample is the sample data in the above embodiment. Use the convolutional kernels 1-n to perform central difference convolutional processing on the signal data sample respectively to obtain n reference convolutional features, that is, the feature sets 1-n. Then, calculate the linear correlation coefficients between each feature set and the label, and select the convolutional kernel with the largest correlation coefficient with the device label as the preset convolutional kernel.

[0128] In the above embodiment, according to the sampling frequency of the signal dataset, the optimal convolutional kernel is designed to improve the effect of feature extraction.

[0129] In one of the embodiments, after obtaining the level distribution feature, it further includes: performing a filtering process on the level distribution feature; after obtaining the signal gradient feature, it further includes: performing a filtering process on the signal gradient feature; the filtering process is to filter the null values in the level distribution feature or the signal gradient feature.

[0130] In the process of extracting the level distribution feature and the signal gradient feature above, in order to make the discrimination of the level distribution feature and the signal gradient feature better, the data segment is further divided into multiple intervals. However, this will also lead to a large number of intervals. For example, the frequency count of the level statistics is 0, and at this time, the generated will have data sparsity, that is, the null values are too redundant, thus affecting the subsequent identification of the device using the level distribution feature. Similarly, the same problem also exists in the signal gradient feature.

[0131] Therefore, it is necessary to filter out the null values in the filtered level distribution features or signal gradient features.

[0132] Optionally, remove specific intervals in the entire feature set where the sample level frequencies are all 0 from the feature set, and record the set S of remaining intervals for facilitating subsequent fingerprint identity authentication work. For N data sets, after processing according to the above steps, the level distribution features will be obtained, where l is the length of the set S, that is, the number of remaining statistical intervals.

[0133] Optionally, remove specific intervals in the entire feature set where the sample level frequencies are all 0 from the feature set, and record the set of remaining intervals for facilitating subsequent fingerprint identity authentication work. For N data sets, after processing according to the above steps, the signal gradient features will be obtained, where is the length of the set that is, the number of remaining statistical intervals.

[0134] In the above embodiments, removing the null values in the level distribution features and signal gradient features can avoid the problem of data sparsity, which in turn affects the accuracy of subsequent device fingerprint recognition.

[0135] In one of the embodiments, the method further includes: identifying the device fingerprint features to obtain a device identification result.

[0136] The device identification result refers to the unique device identification information obtained after analyzing and processing the device fingerprint features. Based on the device identification result, further security verification of the device can be performed. When a device attempts to access, by comparing the fingerprint information, the device identity can be identified, and fake devices can be rejected from accessing, thereby achieving low-cost device identity authentication. This is because the device fingerprint features are generated based on the physical characteristics of the bus device signals and have the property of being non-clonable.

[0137] Optionally, extract the device fingerprint features by means of decision trees, random forests, support vector machines, etc.

[0138] Optionally, since device fingerprint recognition is a multi-classification problem, the traditional Support Vector Machine (SVM) can be extended into a one-versus-all SVM model. Exemplarily, during training, the samples of a certain class are classified into one class in turn, a binary classifier is trained for each class, and the remaining samples are classified into another class. In this way, k classifiers are constructed from the samples of k classes. During classification, an unknown sample is classified into the class with the largest classification function value. When classifying new samples, the class to which the classifier with the highest confidence belongs is selected as the final prediction result.

[0139] Further, in one embodiment, the device fingerprint features are recognized, and the device recognition result is obtained through a fingerprint recognition model; the training process of the fingerprint recognition model includes: obtaining training data; the training data carries class labels; inputting the training data into an initial recognition model to obtain a predicted device recognition result; the initial recognition model includes multiple initial models; obtaining a loss function based on the predicted device recognition result and the class label, and adjusting the initial recognition model according to the loss function until the training is completed to obtain the fingerprint recognition model.

[0140] Among them, the training data refers to the data used to train the fingerprint recognition model, and the training data can be obtained by processing the signal dataset to be processed according to the method of extracting fingerprint features in any of the above embodiments; the predicted device recognition result is the recognition result of the initial recognition model during the training process.

[0141] Exemplarily, when the training data is input into the initial recognition model, since the initial recognition model includes multiple initial models, the training data will be input into the corresponding initial model according to the class label. Exemplarily, assuming that there are currently three class labels, all the training data belonging to the first class label will be labeled with "1", and the other training data will be labeled with "0", and then input into the first initial model. Next, all the training data belonging to the second class label will be labeled with "1", and the other training data will be labeled with "0", and then input into the second initial model. Finally, all the training data belonging to the third class label will be labeled with "1", and the other training data will be labeled with "0", and then input into the third initial model. Then, the loss function between each initial model and the corresponding label will be calculated, and the sum of the loss functions of each initial model will be used as the loss function of the initial recognition model. The parameters of the initial recognition model are adjusted according to the loss function of the initial recognition model until the training is completed to obtain the fingerprint recognition model.

[0142] In other embodiments, the traditional SVM can also be extended into a one-versus-all SVM model through the strategies of one-versus-one and direct method.

[0143] In the above embodiments, by training the fingerprint recognition model, the rules and patterns can be automatically learned from the training data to improve the recognition efficiency of the signal data set to be processed.

[0144] In one embodiment, obtaining the signal data set to be processed includes: collecting device signals through a low-frequency sampling device; modulating the device signals into original signals; preprocessing the original signals to obtain the signal data set to be processed; the preprocessing includes any one of filtering and segmentation; filtering is to filter out the noise signals in the original signals through a preset filtering method; segmentation is to segment the original signals according to a preset length to obtain the signal data set to be processed.

[0145] Optionally, low-cost devices such as a low-frequency sampling oscilloscope can be used to collect the network card signals of the terminal devices based on the 1000BASE-T protocol on the gigabit Ethernet bus as the device signals.

[0146] Combined with Figure 4 shown, Figure 4 is a schematic diagram of signal distortion collected by a low-frequency sampling device in one embodiment. The device signals have been distorted, that is Figure 4 restore the signals at the low sampling frequency. Therefore, it is necessary to restore the device signals into original signals through acquisition techniques such as decoding, demodulation, synchronization, and channel estimation for the collected device signals. Among them, decoding, demodulation, synchronization, and channel estimation are common technical means in the art, so they will not be described in detail here.

[0147] Optionally, after obtaining the original signals, preprocessing is also required, where the preprocessing includes filtering and segmentation.

[0148] Among them, filtering is to filter out the noise signals in the original signals through a signal filtering method. In this way, the noise or unnecessary frequency components in the signals can be removed, thereby improving the quality and accuracy of the signals.

[0149] Among them, segmentation is to perform an average segmentation process on the signals according to the length to obtain N signal data sets to be processed for fingerprint extraction. In this way, the long-time signals are segmented into shorter segments, which helps to concentrate on analyzing and processing the signal characteristics within a specific time period and avoid difficulties in analysis caused by the overall signals being too complex and lengthy.

[0150] In an exemplary embodiment, combined with Figure 5 , a flow schematic diagram of a device fingerprint recognition method is provided.

[0151] Step 1, collect the network card signals of the gigabit Ethernet device based on the 1000BASE-T protocol through a low-frequency sampling signal acquisition device, and perform processing such as decoding and demodulation on the collected device signals to obtain the original signals.

[0152] Step 2: Preprocess the original signal obtained in Step 1 to obtain a signal dataset for fingerprint extraction.

[0153] Step 3: Statistically analyze and screen the distribution of signal levels in different intervals to extract the statistical features of signal level intervals.

[0154] Step 4: Based on the central difference convolution operation, extract the statistical features of signal central difference convolution.

[0155] Step 5: Concatenate the statistical features of signal level intervals and the statistical features of signal central difference convolution to obtain fused features.

[0156] Step 6: Build a classification and recognition model and train it. Combine the device fingerprint feature information and the trained classification and recognition model to perform secure access authentication for Gigabit Ethernet devices.

[0157] It should be noted that in this embodiment, the statistical features of the level interval are the level distribution features in the above embodiment, the statistical features of the signal central difference convolution are the signal gradient features in the above embodiment, and the fused fingerprint features are the device fingerprint features in the above embodiment. Among them, the specific description of each step can refer to the specific description in the above embodiment and will not be repeated in this embodiment.

[0158] In the above embodiment, through methods such as central difference convolution, the sampling signal data features extracted for the 1000BASE-T protocol in the Gigabit Ethernet scenario are fully processed. At the same time, through the method of feature fusion, the limitation of the signal sampling frequency is reduced. It is reduced by 20 times under the original signal sampling frequency, and still ensures effective fingerprint extraction, thereby reducing the sampling cost of device fingerprint access authentication.

[0159] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are displayed in sequence according to the arrows, these steps do not necessarily need to be executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages do not necessarily need to be executed at the same time, but can be executed at different times. The execution order of these steps or stages does not necessarily need to be sequential, but can be executed alternately or alternately with at least a part of other steps or steps or stages in other steps.

[0160] Based on the same inventive concept, an embodiment of this application further provides a device fingerprint feature extraction apparatus for implementing the device fingerprint feature extraction method involved above. The implementation solutions provided by this apparatus to solve problems are similar to those recorded in the above method. Therefore, the specific limitations in one or more embodiments of the device fingerprint feature extraction apparatus provided below can refer to the limitations on the device fingerprint feature extraction method in the above text and will not be elaborated here.

[0161] In an exemplary embodiment, as Figure 6 shown, a device fingerprint feature extraction apparatus is provided, including: an acquisition module 610, a level feature extraction module 620, a convolution feature extraction module 630, and a feature fusion module 640, where:

[0162] The acquisition module 610 is configured to acquire a dataset of signals to be processed; the dataset of signals to be processed carries level values.

[0163] The level feature extraction module 620 is configured to count the distribution of level values on the dataset of signals to be processed to obtain a level distribution feature.

[0164] The convolution feature extraction module 630 is configured to calculate the gradient change of level values on the dataset of signals to be processed to obtain a signal gradient feature.

[0165] The feature fusion module 640 is configured to splice the level distribution feature and the signal gradient feature to obtain a device fingerprint feature.

[0166] In an embodiment, the above level feature extraction module includes:

[0167] A level statistics unit configured to count the frequencies of each level value in each data segment to obtain a level distribution feature.

[0168] In an embodiment, the above convolution feature extraction module includes:

[0169] A convolution processing unit configured to perform central difference convolution processing on each data segment to obtain a signal gradient feature.

[0170] In an embodiment, the above level statistics unit includes:

[0171] A first sub - division unit configured to divide the amplitude of the level value corresponding to each data segment into a first preset number of first intervals.

[0172] A first statistics sub - unit configured to count the frequencies of level values in each first interval.

[0173] A first weighting unit configured to weight the frequencies of level values of each first interval in each data segment to obtain a level distribution feature.

[0174] In one embodiment, the above-mentioned convolution processing unit includes:

[0175] A convolution calculation unit, configured to perform convolution calculation on each data segment through a preset convolution kernel to obtain a convolution value corresponding to each data segment.

[0176] A second partitioning unit, configured to partition the convolution value corresponding to each data segment into a second preset number of second intervals.

[0177] A second statistical subunit, configured to count the convolution value frequencies of the convolution values in each of the second intervals.

[0178] A second weighting unit, configured to weight the convolution value frequencies of each second interval in each data segment to obtain a signal gradient feature.

[0179] In one embodiment, the above-mentioned level feature extraction module and the above-mentioned convolution feature extraction module further include a filtering unit;

[0180] The filtering unit is configured to filter null values in the level distribution feature or the signal gradient feature.

[0181] In one embodiment, the above-mentioned convolution processing unit further includes:

[0182] A generation subunit, configured to determine a third preset number of initial convolution kernels according to the sampling frequency of the signal dataset to be processed.

[0183] A convolution subunit, configured to perform convolution on the sample data using each of the initial convolution kernels respectively to obtain a reference convolution feature.

[0184] A selection subunit, configured to perform a correlation comparison on each of the reference convolution features to obtain a preset convolution kernel.

[0185] In one embodiment, the above-mentioned device further includes:

[0186] An identification module, configured to identify the device fingerprint feature to obtain a device identification result.

[0187] In one embodiment, the above-mentioned identification module further includes:

[0188] A sample acquisition unit, configured to acquire sample data; the sample data carries a class label.

[0189] A training unit, configured to input the sample data into an initial identification model to obtain a predicted device identification result, compare the predicted device identification result with each class respectively to obtain a difference from each class, and adjust the initial identification model according to the difference to obtain a fingerprint identification model.

[0190] In one embodiment, the above-mentioned device further includes:

[0191] A sampling module, configured to collect device signals through a low-frequency sampling device.

[0192] A modulation module, configured to modulate the device signals into original signals.

[0193] A preprocessing module, configured to preprocess the original signals to obtain a dataset of signals to be processed. Among them, the preprocessing includes any one of filtering and segmentation; filtering is to filter out the noise signals in the original signals through a preset filtering method; segmentation is to segment the original signals according to a preset length to obtain a dataset of signals to be processed.

[0194] Each module in the above device fingerprint feature extraction device can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or stored in a memory in the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0195] In an exemplary embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 7 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store a dataset of signals to be processed. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for extracting device fingerprint features.

[0196] Those skilled in the art can understand that Figure 7 the structure shown in

[0197] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the following steps are implemented: obtaining a dataset of signals to be processed; each signal data to be processed in the dataset of signals to be processed carries a level value; statistically analyzing the distribution of the level values on the dataset of signals to be processed to obtain a level distribution feature; calculating the gradient change of the level values on the dataset of signals to be processed to obtain a signal gradient feature; splicing the level distribution feature and the signal gradient feature to obtain a device fingerprint feature.

[0198] In one embodiment, when the processor executes the computer program, the following steps are further implemented: obtaining a plurality of data segments based on each level value; statistically analyzing the frequency of each level value in each data segment to obtain a level distribution feature; calculating the gradient change of the level values on the dataset of signals to be processed, including: performing central difference convolution processing on each data segment to obtain a signal gradient feature.

[0199] In one embodiment, when the processor executes the computer program, the following steps are further implemented: dividing the amplitude of the level value corresponding to each data segment into a first preset number of first intervals; statistically analyzing the frequency of the level values in each first interval; weighting the frequency of the level values in each first interval in each data segment to obtain a level distribution feature.

[0200] In one embodiment, when the processor executes the computer program, the following steps are further implemented: performing convolution calculation on each data segment through a preset convolution kernel to obtain a convolution value corresponding to each data segment; dividing the convolution value corresponding to each data segment into a second preset number of second intervals; statistically analyzing the frequency of the convolution values in each second interval; weighting the frequency of the convolution values in each second interval in each data segment to obtain a signal gradient feature.

[0201] In one embodiment, when the processor executes the computer program, the following steps are further implemented: performing filtering processing on the level distribution feature; after obtaining the signal gradient feature, further including: performing filtering processing on the signal gradient feature; the filtering processing is to filter null values in the level distribution feature or the signal gradient feature.

[0202] In one embodiment, when the processor executes the computer program, the following steps are further implemented: determining a third preset number of initial convolution kernels according to the sampling frequency of the dataset of signals to be processed; respectively performing convolution on the sample data through each initial convolution kernel to obtain a reference convolution feature; performing correlation comparison on each reference convolution feature to obtain a preset convolution kernel.

[0203] In one embodiment, when the processor executes the computer program, the following steps are further implemented: identifying the device fingerprint feature to obtain a device identification result.

[0204] In one embodiment, when the processor executes the computer program, the following steps are further implemented: obtaining training data; the training data carries class labels; inputting the training data into an initial recognition model to obtain a predicted device recognition result; the initial recognition model includes a plurality of initial models; obtaining the loss function of each initial model based on the predicted device recognition result and the class label, and adjusting the initial recognition model according to the loss function of each initial model until the training is completed to obtain a fingerprint recognition model.

[0205] In one embodiment, when the processor executes the computer program, the following steps are further implemented: collecting device signals through a low-frequency sampling device; modulating the device signals into original signals; preprocessing the original signals to obtain a dataset of signals to be processed; the preprocessing includes any one of filtering and segmentation; filtering is to filter out noise signals in the original signals through a preset filtering method; segmentation is to segment the original signals according to a preset length to obtain a dataset of signals to be processed.

[0206] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: obtaining a dataset of signals to be processed; each signal to be processed in the dataset of signals to be processed carries a level value; statistically analyzing the distribution of the level values on the dataset of signals to be processed to obtain a level distribution feature; calculating the gradient change of the level values on the dataset of signals to be processed to obtain a signal gradient feature; splicing the level distribution feature and the signal gradient feature to obtain a device fingerprint feature.

[0207] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: obtaining a plurality of data segments based on each level value; statistically analyzing the frequency of each level value in each data segment to obtain a level distribution feature; calculating the gradient change of the level values on the dataset of signals to be processed, including: performing central difference convolution processing on each data segment to obtain a signal gradient feature.

[0208] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: dividing the amplitude of the level value corresponding to each data segment into a first preset number of first intervals; statistically analyzing the frequency of the level values in each first interval; weighting the frequency of the level values in each first interval in each data segment to obtain a level distribution feature.

[0209] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: performing convolution calculation on each data segment through a preset convolution kernel to obtain a convolution value corresponding to each data segment; dividing the convolution value segment corresponding to each data segment into a second preset number of second intervals; statistically analyzing the frequency of the convolution values in each second interval; weighting the frequency of the convolution values in each second interval in each data segment to obtain a signal gradient feature.

[0210] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: performing a filtering process on the level distribution feature; after obtaining the signal gradient feature, further including: performing a filtering process on the signal gradient feature; the filtering process is to filter null values in the level distribution feature or the signal gradient feature.

[0211] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: determining a third preset number of initial convolution kernels according to the sampling frequency of the signal dataset to be processed; respectively performing convolution on the sample data using each initial convolution kernel to obtain reference convolution features; performing a correlation comparison on each reference convolution feature to obtain a preset convolution kernel.

[0212] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: identifying the device fingerprint feature to obtain a device identification result.

[0213] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining training data; the training data carries a class label; inputting the training data into an initial recognition model to obtain a predicted device identification result; the initial recognition model includes a plurality of initial models; obtaining the loss function of each initial model based on the predicted device identification result and the class label, and adjusting the initial recognition model according to the loss function of each initial model until the training is completed to obtain a fingerprint recognition model.

[0214] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: collecting a device signal through a low-frequency sampling device; modulating the device signal into a raw signal; performing preprocessing on the raw signal to obtain a signal dataset to be processed; the preprocessing includes any one of filtering and segmentation; filtering is to filter out noise signals in the raw signal through a preset filtering method; segmentation is to segment the raw signal according to a preset length to obtain a signal dataset to be processed.

[0215] In one embodiment, a computer program product is provided, including a computer program, which when executed by a processor, implements the following steps: obtaining a signal dataset to be processed; each signal data to be processed in the signal dataset to be processed carries a level value; counting the distribution of the level values on the signal dataset to be processed to obtain a level distribution feature; calculating the gradient change of the level values on the signal dataset to be processed to obtain a signal gradient feature; splicing the level distribution feature and the signal gradient feature to obtain a device fingerprint feature.

[0216] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining a plurality of data segments based on each level value; counting the frequency of each level value in each data segment to obtain a level distribution feature; calculating the gradient change condition of the level value in the data set to be processed, including: performing central difference convolution processing on each data segment to obtain a signal gradient feature.

[0217] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: dividing the amplitude of the level value corresponding to each data segment into a first preset number of first intervals; counting the frequency of the level value in each first interval; weighting the frequency of the level value in each first interval in each data segment to obtain a level distribution feature.

[0218] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: dividing the convolution value corresponding to each data segment into a second preset number of second intervals; counting the frequency of the convolution value in each second interval; weighting the frequency of the convolution value in each second interval in each data segment to obtain a signal gradient feature.

[0219] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: performing filtering processing on the level distribution feature; after obtaining the signal gradient feature, further including: performing filtering processing on the signal gradient feature; the filtering processing is to filter null values in the level distribution feature or the signal gradient feature.

[0220] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: determining a third preset number of initial convolution kernels according to the sampling frequency of the data set to be processed; respectively performing convolution on the sample data using each initial convolution kernel to obtain a reference convolution feature; comparing the correlations of the respective reference convolution features to obtain a preset convolution kernel.

[0221] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: identifying the device fingerprint feature to obtain a device identification result.

[0222] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: obtaining training data; the training data carries a class label; inputting the training data into an initial recognition model to obtain a predicted device identification result; the initial recognition model includes a plurality of initial models; obtaining the loss function of each initial model based on the predicted device identification result and the class label, and adjusting the initial recognition model according to the loss function of each initial model until the training is completed to obtain a fingerprint recognition model.

[0223] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: collecting device signals through a low-frequency sampling device; modulating the device signals into original signals; preprocessing the original signals to obtain a dataset of signals to be processed; the preprocessing includes any one of filtering and segmentation; filtering is to filter out noise signals in the original signals through a preset filtering method; segmentation is to segment the original signals according to a preset length to obtain a dataset of signals to be processed.

[0224] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., and are not limited thereto.

[0225] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this application.

[0226] The above-described embodiments merely represent several implementation manners of this application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of this application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application shall be subject to the appended claims.

Claims

1. A method for extracting device fingerprint features, characterized in that, The method includes: Obtaining a dataset of signals to be processed; each signal data to be processed in the dataset of signals to be processed carries a level value; Statistically analyzing the distribution of the level values on the dataset of signals to be processed to obtain a level distribution feature; Calculating the gradient change of the level values on the dataset of signals to be processed to obtain a signal gradient feature; Concatenating the level distribution feature and the signal gradient feature to obtain a device fingerprint feature.

2. The method for extracting device fingerprint features according to claim 1, wherein The statistically analyzing the distribution of the level values on the dataset of signals to be processed to obtain a level distribution feature includes: Obtaining a plurality of data segments based on each of the level values; Statistically analyzing the frequency of each level value in each of the data segments to obtain the level distribution feature; The calculating the gradient change of the level values in the dataset of signals to be processed includes: Performing central difference convolution processing on each of the data segments to obtain the signal gradient feature.

3. The method for extracting device fingerprint features according to claim 2, wherein The statistically analyzing the frequency of each level value in each of the data segments to obtain the level distribution feature includes: Dividing the amplitude of the level value corresponding to each data segment into a first preset number of first intervals; Statistically analyzing the frequency of the level values in each of the first intervals; Weighting the frequency of the level values in each of the first intervals in each of the data segments to obtain the level distribution feature.

4. The method for extracting device fingerprint features according to claim 2, wherein The performing central difference convolution processing on each of the data segments to obtain the signal gradient feature includes: Performing convolution calculation on each of the data segments through a preset convolution kernel to obtain a convolution value corresponding to each data segment; Dividing the convolution value corresponding to each data segment into a second preset number of second intervals; Statistically analyzing the frequency of the convolution values in each of the second intervals; Weighting the frequency of the convolution values in each of the second intervals in each of the data segments to obtain the signal gradient feature.

5. The method for extracting device fingerprint features according to claim 2, wherein After obtaining the level distribution feature, it further includes: Performing filtering processing on the level distribution feature; After obtaining the signal gradient feature, it further includes: Performing filtering processing on the signal gradient feature; The filtering processing is to filter null values in the level distribution feature or the signal gradient feature.

6. The method for extracting device fingerprint features according to claim 4, wherein, The generation method of the preset convolution kernel includes: Determining a third preset number of initial convolution kernels according to the sampling frequency of the dataset of signals to be processed; Respectively performing convolution on sample data using each of the initial convolution kernels to obtain a reference convolution feature; Performing correlation comparison on each of the reference convolution features to obtain the preset convolution kernel.

7. The method for extracting device fingerprint features according to claim 1, wherein, The method further includes: Identifying the device fingerprint feature to obtain a device identification result.

8. The method for extracting device fingerprint features according to claim 7, wherein, The identifying the device fingerprint feature to obtain a device identification result is obtained through a fingerprint identification model; The training process of the fingerprint identification model includes: Obtaining training data; the training data carries a class label; Inputting the training data into an initial identification model to obtain a predicted device identification result; the initial identification model includes a plurality of initial models; Based on the recognition results of the prediction device and the category labels, the loss functions of each of the initial models are obtained, and the initial recognition model is adjusted according to the loss functions of each of the initial models until the training is completed, and the fingerprint recognition model is obtained.

9. The method for extracting device fingerprint features according to claim 1, wherein The obtaining of the dataset of the signal to be processed includes: Collecting device signals through a low-frequency sampling device; Modulating the device signals into original signals; Performing preprocessing on the original signals to obtain the dataset of the signals to be processed; The preprocessing includes any one of filtering and segmentation; The filtering is to filter out the noise signals in the original signals by a preset filtering method; The segmentation is to segment the original signals according to a preset length to obtain the dataset of the signals to be processed.

10. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images