RFID tag array authentication method based on phase correlation

Abstract

The invention provides an RFID tag array authentication method based on phase correlation, and the method comprises a model training stage and an array authentication stage, the model training stage comprises the steps: collecting and processing a phase sequence received from a conventional tag array, and selecting an optimal tag combination of each tag in the conventional tag array; calculating a conventional phase difference sequence according to the optimal tag combination, and constructing a phase difference sequence of the attacked tag array according to a replacement mechanism; a neural network is designed, a conventional phase difference sequence and a phase difference sequence of an attacked tag array are used as input of the neural network for training, and an array authentication stage: a phase sequence of a to-be-authenticated tag array is collected and processed, a to-be-authenticated phase difference sequence is calculated according to the optimal tag combination and input into the neural network, and an authentication result is obtained. According to the method, clone attack, analog signal attack and deployment tampering attack can be defended, the authentication precision is high, the deployment cost is low, and the security of sensing based on the tag array is improved.

Description

Technical Field

[0001] The invention belongs to the field of radio frequency identification, and in particular relates to an RFID tag array authentication method based on phase correlation. Background Art

[0002] Passive sensing technology based on radio frequency identification (RFID) has attracted increasing attention due to its ability to monitor and analyze dynamic scenes in real time. RFID sensing has a wide range of applications, including positioning, gesture recognition, user authentication, vital sign detection, etc. The advantages of this technology focus on flexible deployment and low cost. A typical RFID sensing system consists of multiple passive tags and a reader with an antenna, where the tags receive energy through continuous signal waves emitted by the reader. Existing sensing methods usually analyze the inherent characteristics of wireless signals, such as phase and received signal strength. In order to achieve multi-dimensional and comprehensive perception, most studies use tag arrays rather than single tags. Although signal feature-based sensing methods have developed rapidly, the deployment and implementation of these methods in real-world scenarios still have key security challenges. Due to the low cost and simple internal structure of RFID tags, they are vulnerable to various attacks, such as cloning, forgery, and replay. If an attacker randomly deploys attack tags with the same identity (ID) as regular tags in the environment, it is difficult for the system to distinguish whether the signal comes from a regular tag. In addition, RFID readers can read a large range (up to 10 meters), making it difficult to directly find these attack tags hidden in the scene.

[0003] Current related methods mainly focus on the identity authentication of the tag itself, which is applied to logistics, warehousing and other scenarios. The upper-layer protocol or cryptography-based methods solve the RFID security problem through a challenge-response mechanism, requiring the tag to provide correct authentication information in response. However, frequent upper-layer protocol authentication cannot meet the high read rate requirements of RFID sensing tasks. To make up for the shortcomings of upper-layer authentication, recent research has focused on exploiting defects in the tag manufacturing process to generate unclonable fingerprints. However, in RFID sensing systems, the objects that need to be protected are the inherent characteristics of the signal (such as phase, received signal strength) rather than the information stored in the tag. Attackers can use signal transmitters to simulate regular tag signals. These simulated signals may cause the reader to parse incorrect phase data, increasing the complexity and threat of the attack. Since such simulated signals may occur and end at any time, it is challenging to detect such attacks without continuous authentication. Few physical layer authentication methods can cope with such attacks because in RFID sensing, both the signals received by the Universal Software Radio Peripheral (USRP) and the reader contain obvious interference from human motion, and the sensing task requires the use of interference generated by the human body to classify and identify human behavior. This contradiction poses a major challenge to security authentication. In addition, most fingerprint-based methods require obtaining the original signal data, which leads to problems such as high authentication cost, difficult deployment, and long authentication time. Summary of the invention

[0004] Purpose of the invention: In order to solve the problems existing in the above-mentioned prior art, the present invention discloses an RFID tag array authentication method based on phase correlation.

[0005] Technical solution: The present invention provides an RFID tag array authentication method based on phase correlation, specifically:

[0006] Step 1: Construct a set of regular label arrays and labels for attack, where the total number of labels in each regular label array is M;

[0007] Step 2: performing data preprocessing on the phase sequence of any conventional tag array in the conventional tag array set;

[0008] Step 3: Select the optimal label combination for each label;

[0009] Step 4: According to the optimal combination of each tag, for any tag array, obtain the phase difference sequence corresponding to the optimal combination under the tag array.

[0010] Step 5: Generate an integer ξ, select ξ labels from the labels used for attack, replace the ξ labels in the regular label array, and obtain the attacked label array;

[0011] Step 6: Calculate the phase difference sequence corresponding to the optimal combination under the attacked tag array.

[0012] Step 7: Construct a training set, which includes the phase difference sequence corresponding to the regular label array and the phase difference sequence corresponding to the attacked label array;

[0013] Step 8: Use the training set to train the neural network;

[0014] Step 9: Collect the tag array to be authenticated and perform the same data preprocessing as step 2. According to the optimal tag combination in step 3, the optimal tag combination of each tag in the tag array to be authenticated is obtained. Then, the phase difference sequence corresponding to the phase array to be authenticated is calculated and input into the trained neural network to determine whether the tag array to be authenticated is attacked.

[0015] Furthermore, the step 2 is specifically as follows:

[0016] Step 2.1: Calculate the difference between two adjacent phases in the phase sequence corresponding to any tag, and adjust the phase according to the difference using the following formula:

[0017]

[0018] in, represents the phase, j represents the jth phase in the phase sequence corresponding to any label;

[0019] Then Gaussian smoothing is performed on the updated phase sequence;

[0020] Step 2.2: Based on the dynamic time warping algorithm, calculate the alignment sequence of each tag and other tags after dynamic time warping in turn, and use the alignment sequence to eliminate the sequence offset caused by the reading delay to obtain the aligned result X:

[0021]

[0022] Among them, X i,i' represents the alignment result of the ith label relative to the ith' label after applying the dynamic time warping algorithm, i=1,2,…,M,i′=1,2,…,M;

[0023] Furthermore, the step 3 is specifically as follows:

[0024] Calculate the Pearson correlation coefficient between the phase sequences of any two tags in the conventional tag array to obtain the correlation coefficient matrix,

[0025] Calculate the correlation coefficient matrix of each regular label array of the regular label array set;

[0026] Calculate the mean of the elements in the i-th row and i'th column of all correlation coefficient matrices to obtain the mean correlation coefficient matrix; i = 1, 2, ..., M, i' = 1, 2, ..., M;

[0027] For the ith label, in the ith row of the mean correlation coefficient matrix, arrange the average correlation coefficients of the ith label with other labels from large to small, and select the first The labels corresponding to the average correlation coefficients are taken as the optimal label combination of the i-th label; Indicates rounding down;

[0028] Furthermore, the step 4 or step 6 generates the phase difference sequence specifically as follows: according to step 3, for any tag array, the optimal combination of each tag under the tag array is obtained, and then the phase sequence under the optimal combination of any tag in the tag array is obtained, the difference between two adjacent phases in the phase sequence is calculated, and the phase difference sequence corresponding to the tag is obtained, and the phase difference sequence corresponding to each tag in the tag array constitutes the phase difference sequence of the tag array.

[0029] Furthermore, the value range of the integer ξ in step 5 is Indicates rounding down.

[0030] Furthermore, the neural network in step 8 adopts a lightweight one-dimensional convolutional neural network that introduces an attention mechanism.

[0031] Furthermore, the one-dimensional convolutional neural network introducing the attention mechanism includes four convolution blocks and four attention mechanisms, each convolution block is connected to an attention mechanism, and each convolution block includes a one-dimensional convolution layer, a batch normalization layer and an activation function layer; the convolution kernel size of the one-dimensional convolution layer is 3′3, and the step size is 2.

[0032] An electronic device / system for RFID tag array authentication includes a processor and a memory, wherein the memory stores execution instructions of the processor, and the processor is configured to execute the execution instructions to implement the RFID tag array authentication method.

[0033] A computer-readable storage medium is used to store a program, and the program is executed to implement the RFID tag array authentication method.

[0034] Beneficial effects:

[0035] 1. The tag array authentication method proposed in the present invention focuses on the phase information of the authentication tag array, which can improve the security of the RFID sensing task based on the tag array; by analyzing the correlation of the phase sequence in the tag array, the present invention can effectively detect the interference of the attack tag in the tag array, specifically defend against cloning attacks, simulated signal attacks and deployment tampering attacks, and realize the authentication of the phase sequence sample of each tag array, which is more detailed than the existing authentication method.

[0036] 2. The tag array authentication method proposed in the present invention has a simple data processing process and uses a lightweight neural network model, which can achieve high-precision authentication with fewer computing resources. In addition, this method does not require additional data collection and shares the phase information of the tag array with the RFID sensing task, and can complete the authentication in parallel without interfering with the sensing task.

[0037] 3. The tag array authentication method proposed in the present invention fully complies with the current RFID standards and protocols, only uses commercial RFID readers and writers, has low deployment costs, and does not require the use of expensive software radio to collect signal raw data; the single authentication delay is low, and the tag array can be authenticated periodically at a higher frequency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of the present invention;

[0039] Figure 2 The comparison diagram of the phase signal of a single tag before and after data preprocessing; (a) is the phase diagram before step 2.1, and (b) is the phase diagram after processing according to step 2.1;

[0040] Figure 3 This is an example diagram of the mean correlation coefficient matrix of this embodiment;

[0041] Figure 4 Histograms of authentication accuracy, false acceptance rate, and false rejection rate for different arrays;

[0042] Figure 5 The authentication accuracy graph corresponding to selecting different numbers of optimal labels;

[0043] Figure 6 This is the recognition accuracy diagram of the attack tag under different antenna distances. DETAILED DESCRIPTION

[0044] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0045] like Figure 1As shown, the present invention provides an RFID tag array authentication method based on phase correlation, which specifically includes the following steps:

[0046] Step 1: Deploy conventional tag arrays and tags for attack, and collect phase sequences. Specifically, this embodiment deploys multiple 3′3 conventional tag arrays, and sticks the arrays on cardboard. The subjects face the arrays and perform various actions, such as pushing, nodding, turning around, etc. This embodiment uses a circularly polarized antenna with a frequency of 920MHz and an antenna gain of 8dBi to collect the phase information of the tag array. The antenna transmits the phase data to the RFID reader and stores it. In addition, in order to use real tag data for attack, this embodiment deploys additional tags at different positions from the antenna in the experimental scenario as tags for attack, which are used for subsequent execution replacement mechanisms to simulate real attacks.

[0047] Step 2: Process the phase sequence received from the conventional tag array. Specifically, the phase sequence of the conventional tag array received each time is regarded as a conventional sample, and the processing of all conventional samples includes phase unwrapping and sequence alignment algorithm, including the following sub-steps:

[0048] Step 2.1: Phase unwrapping: Eliminate the phase jump at a specific distance between the tag and the reader antenna according to whether the absolute value of the phase difference between each tag before and after exceeds π. Specifically, this embodiment compares the phase difference of adjacent sampling points in the phase sequence of a tag and performs the following processing:

[0049]

[0050] in, represents the phase, j represents the jth phase in the phase sequence corresponding to any label; and are two adjacent phase values ​​in a tag phase sequence; Gaussian smoothing, data padding is performed at both ends of the phase sequence, a window of size 5 is set, and for each phase value in a tag phase sequence, this embodiment will calculate the weighted average of the surrounding phase values, where the weight is a set Gaussian function, which is used to filter the noise in the phase sequence; Figure 2 As shown, the phase sequence before and after processing in step 2.1 is shown.

[0051] Step 2.1 removes abnormal jump values ​​in the data, filters out small disturbances caused by the environment and hardware, and extracts effective features from the data.

[0052] Step 2.2: Sequence alignment. Due to the limited reading rate of the reader, the phase sequences of different tags in the array are asynchronous; in order to eliminate this asynchrony, the correlation coefficient matrix calculated later is closer to the true value. This embodiment is based on the dynamic time warping algorithm, and calculates the alignment sequence of each tag and other tags after dynamic time warping in turn. The alignment sequence is used to eliminate the sequence offset caused by the reading delay. The alignment result is specifically X:

[0053]

[0054] Among them, X i,i' represents the result of subscript alignment of the ith label relative to the ith' label after applying the dynamic time warping algorithm, i=1,2,…,M, i'=1,2,…,M; M represents the total number of labels in the label array. Therefore, X i,i' It is a 2×N dimensional matrix, where N represents the length of the phase sequence. i,i′ The two resulting aligned phase sequences are expressed as and

[0055] Step 3: Select the optimal label combination for each label in the regular label array. Specifically, calculate X i,i′ Pearson correlation coefficient ρ between the phase sequence of two labels i,i' :

[0056]

[0057] Among them, Cov(.) represents the covariance of two aligned phase sequences, and σ(.) represents the standard deviation; the Pearson correlation coefficient is a general method to measure the correlation scale between two sequences, and its value range is [-1,1]. The larger the absolute value of the value, the higher the correlation between the two sequences. Calculate the correlation coefficient between each two labels in the sample, and get the correlation coefficient matrix corresponding to a sample as:

[0058]

[0059] The optimal label combination should have strong universality in the data set, so this embodiment uses a large data sample to calculate the average correlation coefficient and observe this statistical characteristic. Specifically, this embodiment takes 70% of the data set as the training set and randomly selects 50% of the training set samples. For the constructed regular label array set, the average value of each element in the correlation coefficient matrix is ​​calculated for each row and column to obtain the mean correlation coefficient matrix. Each value of the mean correlation coefficient matrix stores the average correlation coefficient of one label relative to another label. Figure 3An example of an average correlation coefficient matrix obtained based on a 3′3 tag array is shown. It can be observed that there are obvious differences in each value in the evaluation correlation coefficient matrix, which indicates that the phase sequence of the backscattered signal between the tags in the phase array has the above-mentioned similarity.

[0060] For each label in the mean correlation matrix, sort the other labels by their mean correlation, choosing The label with the largest average correlation coefficient is taken as the optimal label combination of the label, where The optimal label combination is selected only in the training phase, not in the array verification phase, and is uniquely determined for each label array.

[0061] Step 4: Calculate the phase difference sequence of the conventional label array based on the optimal label combination to construct an effective neural network model input. The details are as follows:

[0062]

[0063] in, It represents the phase difference sequence component corresponding to the i-th label, as follows:

[0064]

[0065] in, is the difference between the phase sequence corresponding to the i-th tag and the phase sequence corresponding to the j-th tag.

[0066] Tags with higher similarity have more consistent phase change trend sequences; through sequence alignment, this similarity change trend will be further amplified. However, the attack tag attempts to destroy the perception task, and its phase sequence is different from that of the regular tag and does not have the above similarity. Therefore, this embodiment can distinguish the regular phase sequence from the attacked phase sequence by calculating the phase difference.

[0067] On the one hand, the phase difference sequence obtained in this way increases the data dimension, enabling the neural network to capture and learn more complex information; on the other hand, the phase difference sequence of conventional labels is relatively stable, while the phase difference sequence of the attacked label array is relatively irregular. This difference helps the model speed up training convergence.

[0068] Step 5: Construct the phase difference sequence corresponding to the attacked tag array according to the replacement mechanism, and use the phase difference sequence corresponding to the conventional tag array and the phase difference sequence corresponding to the attacked tag array as neural network input for training. Specifically, this embodiment uses a random integer ξ (ξ has a value range of ) represents the number of labels used for attack. In this embodiment, ξ conventional labels in the conventional label array are randomly replaced with labels used for attack to obtain the label array under attack. For the remaining conventional labels, the remaining conventional phase difference sequence components are generated according to their corresponding optimal label combinations. For the labels under attack, the phase difference sequence components are generated according to the optimal label combination of the conventional labels before replacement. The two phase difference sequence components are connected according to the corresponding positions of the conventional label array as the phase difference sequence of the label array under attack. The label array under attack obtained through this step has the same data dimension as the conventional label array.

[0069] The neural network is specifically: a lightweight one-dimensional convolutional neural network with an attention mechanism, which contains 4 convolution blocks and 4 attention mechanism blocks, and one convolution block is followed by an attention mechanism block; each convolution block contains a one-dimensional convolution layer, a batch normalization layer and an activation function layer; the convolution kernel size of each one-dimensional convolution layer is , the step size is 2, and the ReLU activation function is used; the attention mechanism block is specifically: an attention mechanism block consists of a channel attention layer spliced ​​with a spatial attention layer. The output of the neural network is specifically: a binary classification is output through the Softmax function, that is, 1 or 0; the neural network output 1 indicates that the input is a regular label array, and the output 0 indicates that the input is an attacked label array. The advantages of the designed neural network: lightweight, low computational overhead, can be deployed in common computing terminals, and is suitable for deployment scenarios with scarce computing resources.

[0070] Step 6: Array authentication. Specifically, the phase sequence of the tag array to be authenticated is collected and the same signal processing as step 2 is performed. The phase difference sequence to be authenticated is calculated based on the optimal tag combination and input into the trained neural network, which outputs the authentication result. The array authentication experiment uses the test set. In the data set, the data ratio of the training set to the test set is 7:3.

[0071] The following is the effect of applying the method of the present invention to a data set to demonstrate the detection accuracy of its three-dimensional sleeping posture. The present invention uses two data sets for experiments. (a) This embodiment mainly uses the XRF55 public data set published by Xi'an Jiaotong University to evaluate the present invention in large-scale data. XRF55 is a radio frequency data set designed for indoor human activity analysis, including 39 subjects and 55 different behavioral activities, with a total of 33,000 samples. (b) In order to evaluate the recognition effect of the present invention for attack tags at different distances from the antenna, this embodiment specifically collects a sub-data set in the same deployment mode as the XRF55 data set. The specific equipment configuration includes a reader, a directional reader antenna with a gain of 8dBi, and multiple 3′3 tag arrays are deployed at 40cm from the reader antenna as regular tag arrays, and a total of 20 tags for attack are placed starting from 1.5m from the antenna (the distance between the two farthest tags is 40cm), where the tags used for attack use two mainstream passive tags, namely H47 tags and 9662 tags, including 9 subjects and 6 different behavioral activities.

[0072] This embodiment uses accuracy, false acceptance rate and false rejection rate to evaluate the present invention, wherein the false acceptance rate indicates the probability that the system accepts an array of tags containing attacks, and the false rejection rate indicates the probability that the system refuses to accept a regular array of tags.

[0073] like Figure 4 As shown, for a tag array composed of different tags, the experimental performance of the present invention is stable, the authentication accuracy exceeds 98%, and the average authentication accuracy reaches 98.5%. In addition, the average time for a single authentication is only 56ms, which can meet the high-frequency authentication scenario of the RFID array perception task. The average false rejection rate is 0.74%, and the average false acceptance rate is 1.68%, indicating that the present invention has a strong ability to correctly identify conventional arrays, which is crucial for RFID perception systems with high real-time requirements. This means that the present invention rarely erroneously destroys the perception task and maintains a satisfactory user experience. The higher average false acceptance rate may be due to the fact that some attack signals are too similar to the real signal. However, such indistinguishable attacks usually do not change the perception results.

[0074] like Figure 5As shown, this embodiment analyzes the impact of the number of labels in different optimal label combinations on the authentication accuracy of this method. This embodiment selects a 3′3 label array as an example, and selects 1 to 6 labels with the highest average correlation coefficient as the optimal label combination. The selection of this variable requires a trade-off: selecting more optimal labels means that the data dimension input to the neural network is higher, which is effective for training. However, this method will also lead to the selection of more labels with lower correlation, which may bring uncertainty to model training. According to the experimental results, when 1 or 2 labels are selected as the optimal group, the accuracy is relatively low, so this embodiment can infer that the data dimension still plays a dominant role in this case. The authentication accuracy of the models selecting 3 to 6 optimal labels is not much different, which are 98.58%, 98.62%, 98.48%, and 98.69% respectively. Too many labels will introduce too much computational overhead. After experimental verification, for a 3′3 label array, that is, when the array contains 9 labels, the recommended number of labels in the optimal label combination is 4. For an array with a total number of labels of M, the number of labels in the optimal label combination is

[0075] like Figure 6 As shown, this embodiment uses a sub-data set to evaluate the recognition performance of the present invention for attack tags at different distances from the antenna; for the recognition of attack tags at different distances from the antenna, the present invention can maintain a high recognition accuracy. According to the experimental results, the present invention maintains good performance at most distances, with an average recognition accuracy of 97.0%. In addition, this embodiment finds that the trained neural network has different recognition performance characteristics for the H47 and 9662 tags used for attack. The H47 tag has a higher recognition accuracy at a closer distance (within 2 meters), but the recognition accuracy begins to fluctuate at a longer distance, while the 9662 tag is the opposite. This may be because when the human body continues to move, even tags at a closer distance will experience different multipath effects.

[0076] In summary, the present invention utilizes RFID devices to realize authentication of tag arrays, which can effectively avoid interference of attack tags on tag arrays, realize tag array authentication with high authentication accuracy, low deployment cost, and low computational complexity, and improve the security of perception methods based on tag arrays.

[0077] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A phase correlation based RFID tag array authentication method, characterized in that: Specifically: Step 1: Construct a set of regular tag arrays and tags for attack, where the total number of tags in each regular tag array is M; Step 2: performing data preprocessing on the phase sequence of any conventional tag array in the conventional tag array set; Step 3: Select the optimal label combination for each label; Step 4: According to the optimal combination of each tag, for any tag array, obtain the phase difference sequence corresponding to the optimal combination under the tag array. Step 5: Generate an integer ξ, select ξ labels from the labels used for attack, replace the ξ labels in the regular label array, and obtain the attacked label array; Step 6: Calculate the phase difference sequence corresponding to the optimal combination under the attacked tag array. Step 7: Construct a training set, which includes the phase difference sequence corresponding to the regular label array and the phase difference sequence corresponding to the attacked label array; Step 8: Use the training set to train the neural network; Step 9: Collect the tag array to be authenticated and perform the same data preprocessing as step 2. According to the optimal tag combination in step 3, the optimal tag combination of each tag in the tag array to be authenticated is obtained. Then, the phase difference sequence corresponding to the phase array to be authenticated is calculated and input into the trained neural network to determine whether the tag array to be authenticated is attacked.

2. The RFID tag array authentication method based on phase correlation according to claim 1, characterized in that: The step 2 is specifically as follows: Step 2.1: Calculate the difference between two adjacent phases in the phase sequence corresponding to any tag, and adjust the phase according to the difference using the following formula: in, represents the phase, j represents the jth phase in the phase sequence corresponding to any label; Then Gaussian smoothing is performed on the updated phase sequence; Step 2.2: Based on the dynamic time warping algorithm, calculate the alignment sequence of each tag and other tags after dynamic time warping in turn, and use the alignment sequence to eliminate the sequence offset caused by the reading delay to obtain the aligned result X: Among them, X i，i′ represents the alignment result of the ith label relative to the ith′ label after applying the dynamic time warping algorithm, i=1,2,…,M,i′=1,2,…,M.

3. The RFID tag array authentication method based on phase correlation according to claim 1, characterized in that: The step 3 is specifically as follows: Calculate the Pearson correlation coefficient between the phase sequences of any two tags in the conventional tag array to obtain the correlation coefficient matrix, Calculate the correlation coefficient matrix of each regular label array of the regular label array set; Calculate the mean of the elements in the i-th row and i′th column of all correlation coefficient matrices to obtain the mean correlation coefficient matrix; i = 1, 2, ..., M, i′ = 1, 2, ..., M; For the ith label, in the ith row of the mean correlation coefficient matrix, arrange the average correlation coefficients of the ith label with other labels from large to small, and select the first The labels corresponding to the average correlation coefficients are taken as the optimal label combination of the i-th label; Indicates rounding down.

4. The RFID tag array authentication method based on phase correlation according to claim 1, characterized in that: The step 4 or step 6 of generating the phase difference sequence is specifically as follows: according to step 3, for any tag array, the optimal combination of each tag under the tag array is obtained, and then the phase sequence under the optimal combination of any tag in the tag array is obtained, the difference between two adjacent phases in the phase sequence is calculated, and the phase difference sequence corresponding to the tag is obtained, and the phase difference sequence corresponding to each tag in the tag array constitutes the phase difference sequence of the tag array.

5. The RFID tag array authentication method based on phase correlation according to claim 1, characterized in that: The value range of integer ξ in step 5 is Indicates rounding down.

6. The RFID tag array authentication method based on phase correlation according to claim 1, characterized in that: The neural network in step 8 adopts a lightweight one-dimensional convolutional neural network with an attention mechanism.

7. The RFID tag array authentication method based on phase correlation according to claim 6, characterized in that: The one-dimensional convolutional neural network introducing the attention mechanism includes four convolution blocks and four attention mechanisms. Each convolution block is connected to an attention mechanism. Each convolution block includes a one-dimensional convolution layer, a batch normalization layer and an activation function layer. The convolution kernel size of the one-dimensional convolution layer is 3×3 and the step size is 2.

8. An electronic device / system for RFID tag array authentication, characterized in that: The invention comprises a processor and a memory, wherein the memory stores execution instructions of the processor, and the processor is configured to execute the execution instructions to implement the RFID tag array authentication method according to any one of claims 1 to 7.

9. A computer-readable storage medium for storing a program, characterized in that: Execute the program to implement the RFID tag array authentication method described in any one of claims 1-7.

Images