Equipment identity determination method in industrial internet of things and related equipment
By obtaining communication data between devices and servers in the industrial Internet of Things, using the identity determination model and attention mechanism, combined with the central confidence loss function, the accurate identification and distinction of device identity is achieved, and the single-user authentication solution is solved in multi-user scenarios The accuracy of the system is reduced and the security threat issues have improved the security and reliability of the system.
Patent Information
- Application Number
- CN202410877043.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-02
- Publication Date
- 2024-11-19
AI Technical Summary
Due to the increase in the number of devices in the industrial Internet of Things, single-user physical layer authentication schemes are difficult to effectively identify the device identity and are vulnerable to attacks, resulting in an increase in security threats. The accuracy of traditional methods in multi-user scenarios is reduced and cannot meet the needs of further development.
An industrial IoT device identity determination method is proposed. By obtaining communication data between the device and the server, determining the attribute set of the device, and using a pre-trained identity determination model for feature extraction and classification, combining attention mechanism and central confidence loss Function to realize the accurate identification and distinction of device identity.
It improves the accuracy and security of device identity identification, reduces the probability of malicious node attacks, and is suitable for multi-user scenarios, improving the security and reliability of the industrial Internet of Things.
Smart Images

Figure CN118981701A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a method for determining the identity of a device in an industrial Internet of Things and related devices. Background Art
[0002] The Industrial Internet of Things (IIoT) connects a variety of devices to the Internet, which can collect, analyze and process a large amount of data generated by a variety of devices. With the increasing number of devices and the openness of electromagnetic wave propagation, IIoT faces a large number of wireless interference and security threats, especially servers located at the edge of industrial wireless networks are more vulnerable to spoofing attacks. The communication channel between the server and the device has a physical layer channel. The physical layer authentication uses the physical layer properties that are difficult for spoofers to steal and imitate as the only key to authenticate the user identity to meet the needs of wireless communication security. Therefore, Physical Layer Authentication (PLA) can solve the above attack problems.
[0003] Traditional authentication strategies, such as the single-user PLA scheme, rely on hypothesis testing thresholds to identify signals from legitimate transmitters by comparing the difference between the received signal and the reference signal of a single attribute. However, with the sharp increase in the number of terminals, the single-user PLA scheme will no longer be applicable and will hinder the further development of IIoT. Since the single attribute in the single-user PLA scheme is easier to be predicted and imitated by attackers, and there may also be inaccurate estimates of the single attribute used, it is not enough to use only a single attribute to identify the identity of the device. Summary of the invention
[0004] In view of this, the purpose of this application is to propose a device identity determination method and related equipment in the industrial Internet of Things to overcome all or part of the deficiencies in the prior art.
[0005] Based on the above purpose, the present application provides a method for determining the identity of a device in an industrial Internet of Things, including: obtaining communication data between each device and a server, and determining a set of attributes corresponding to all devices based on all communication data; inputting the attribute set into a pre-trained identity determination model, and performing the following operations in the identity determination model: performing feature extraction on the attribute set to obtain an attribute feature set; based on the attribute set, assigning a weight to each attribute feature in the attribute feature set to obtain a target attribute feature set; fusing the target attribute feature set with the attribute set to obtain a fused set; performing feature extraction on the fused set to obtain a fused feature; classifying the fused feature to obtain a device identity; and outputting the device identity corresponding to each device through the identity determination model.
[0006] Optionally, the extracting features from the attribute set to obtain an attribute feature set includes: performing global average pooling on the attribute set to obtain a first attribute feature set; performing maximum pooling on the attribute set to obtain a second attribute feature set; and fusing the first attribute feature set with the second attribute feature set to obtain the attribute feature set.
[0007] Optionally, the fusing the target attribute feature set with the attribute set to obtain a fused set includes: multiplying elements in the target attribute feature set with elements in the attribute set to obtain the fused set.
[0008] Optionally, the loss function of the identity determination model is determined by the following formula:
[0009] in, is the loss function, To minimize the classification loss function calculated by the sparse cross entropy loss function, λ ′ is the dynamic balance quantity, is the center loss function.
[0010] Optionally, the method for determining the classification loss function includes:
[0011] in, is the classification loss function, N s is the number of training samples, l i,k The target input label for the kth device, p i,k represents the probability that the i-th sample is classified as the k-th device, satisfying
[0012] Optionally, the method for determining the center loss function includes:
[0013] in, is the center loss function, N s is the number of training samples, x u ∈R d Indicates that it belongs to y u The u-th deep feature of the category, d is the dimension of the feature, Yes u Class center of deep-like features.
[0014] Optionally, the attribute set includes at least amplitude, phase-frequency distortion and delay; and determining the attribute set corresponding to all devices based on all communication data includes: determining the amplitude by the following formula: in, is the amplitude, amplitude represents the amplitude characteristics, is the zth channel impulse response of the pseudo-random code sequence received by the kth device, imag(·) represents the imaginary part of the complex number, and real(·) represents the real part of the complex number; the phase-frequency distortion is determined by the following formula: in, is the phase-frequency distortion, p represents the phase-frequency characteristic, is the zth channel impulse response of the pseudo-random code sequence received by the kth device, imag(·) represents the imaginary part of the complex number, real(·) represents the real part of the complex number, l PN is the length of the pseudo-random code sequence sent by the transmitter, R s is the oversampling rate of the receiver, PN represents a pseudo-random code; the delay is determined by the following formula: Among them, (x k ,y k ) is the location coordinate of device k, d represents the delay characteristic, c is the propagation speed of radio waves, and (x, y) is the location coordinate of the server.
[0015] Based on the same inventive concept, the present application also provides a device identity determination apparatus in an industrial Internet of Things, comprising: a first determination module, configured to obtain communication data between each device and a server, and determine a set of attributes corresponding to all devices based on all communication data; an operation module, configured to input the attribute set into a pre-trained identity determination model, and perform the following operations in the identity determination model: extract features from the attribute set to obtain an attribute feature set; based on the attribute set, assign weights to each attribute feature in the attribute feature set to obtain a target attribute feature set; fuse the target attribute feature set with the attribute set to obtain a fused set; extract features from the fused set to obtain fused features; classify the fused features to obtain a device identity; and an output module, configured to output the device identity corresponding to each device through the identity determination model.
[0016] Based on the same inventive concept, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method as described above when executing the computer program.
[0017] Based on the same inventive concept, the present application also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the method as described above.
[0018] As can be seen from the above, the device identity determination method and related devices in the industrial Internet of Things provided by the present application include obtaining communication data between each device and the server, and determining the attribute set corresponding to all devices based on all communication data, so that the probability of the server being attacked by malicious nodes is low. The attribute set is input into a pre-trained identity determination model, and the following operations are performed in the identity determination model: feature extraction is performed on the attribute set to obtain an attribute feature set, which accurately reflects the characteristics of the attribute. Based on the attribute set, a weight is assigned to each attribute feature in the attribute feature set to obtain a target attribute feature set, so that the weight coefficient of the relatively important attribute feature in the target attribute feature set is relatively large. The target attribute feature set is fused with the attribute set to obtain a fused set, so that the weight coefficient of the important attribute in the attribute set is relatively large. Feature extraction is performed on the fused set to obtain a fused feature, so that the relatively important attribute feature is highlighted. The fused feature is classified to obtain the device identity, so as to achieve the purpose of accurately determining the device identity. The device identity corresponding to each device is output through the identity determination model, so as to achieve the purpose of accurately determining the device name. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions in the present application or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0020] Figure 1 A schematic diagram of a flow chart of a method for determining device identity in an industrial Internet of Things according to an embodiment of the present application;
[0021] Figure 2 A schematic diagram of the architecture of an identity determination model according to an embodiment of the present application;
[0022] Figure 3 A schematic diagram of the architecture details of the identity determination model of an embodiment of the present application;
[0023] Figure 4 A schematic diagram of an attention module of an identity determination model according to an embodiment of the present application;
[0024] Figure 5 A schematic diagram of an architecture of an identity determination model based on central confidence loss according to an embodiment of the present application;
[0025] Figure 6 A schematic diagram of the locations of servers and devices in an embodiment of the present application;
[0026] FIG. 7( a ) is a schematic diagram of the authentication performance of the CANN of the embodiment of the present application and the comparative scheme in a multi-attribute scenario;
[0027] FIG7( b ) is a schematic diagram of the authentication performance of the CANN of the embodiment of the present application and the comparative solution in a single attribute scenario;
[0028] FIG8( a ) is a schematic diagram of the output characteristics of the CANN central layer when the number of devices is seven according to an embodiment of the present application;
[0029] FIG8( b ) is a schematic diagram of the output characteristics of the CANN central layer when the number of devices is eleven according to an embodiment of the present application;
[0030] FIG8( c ) is a schematic diagram of the output characteristics of the CANN central layer when there are fifteen devices in the embodiment of the present application;
[0031] FIG8(d) is a diagram of the CANN- Schematic diagram of the output characteristics of the center layer;
[0032] FIG8(e) is a diagram of CANN- Schematic diagram of the output characteristics of the center layer;
[0033] FIG8(f) is a diagram of the seven devices in the embodiment of the present application. Schematic diagram of the output characteristics of the center layer;
[0034] Fig. 9 A schematic diagram of the authentication accuracy based on the central confidence loss according to an embodiment of the present application;
[0035] Fig.10 This is a schematic diagram of the structure of a device identity determination apparatus in an industrial Internet of Things according to an embodiment of the present application;
[0036] Fig.11 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0037] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.
[0038] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be the usual meanings understood by people with ordinary skills in the field to which the present application belongs. The "first", "second" and similar words used in the embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing in front of the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0039] As described in the background technology section, the Industrial Internet of Things can collect, analyze and process a large amount of data generated by various devices by connecting them to the Internet, including sensors, communication equipment, mechanical equipment and other equipment. The above data-driven method can achieve real-time decision-making and predictive analysis, thereby improving operational efficiency, reducing costs and improving product quality. However, with the increasing number of devices and the openness of electromagnetic wave propagation, IIoT faces a large number of wireless interference and security threats, especially servers located at the edge of industrial wireless networks are more vulnerable to spoofing attacks by disguised devices. The communication channel between the server and the device has a physical layer channel. The physical layer authentication uses physical layer properties that are difficult for spoofers to steal and imitate as unique keys, such as channel impulse response (CIR), wireless channel phase or carrier frequency offset, etc., to authenticate the user identity to meet the needs of wireless communication security. Therefore, physical layer authentication (PLA) can solve the above attack problems. Compared with the upper layer identity authentication, physical layer authentication has low complexity and lightweight, so it is suitable for unmanned industrial Internet of Things scenarios.
[0040] Traditional authentication strategies, such as the single-user PLA scheme, rely on hypothesis testing thresholds to identify signals from legitimate transmitters by comparing the difference between the received signal and the reference signal of a single attribute. However, with the sharp increase in the number of terminals, the single-user PLA scheme will no longer be applicable and will hinder the further development of IIoT. Since the single attribute in the single-user PLA scheme is easier to be predicted and imitated by attackers, and there may also be inaccurate estimates of the single attribute used, it is not enough to use only a single attribute to identify the identity of the device. In addition, in recent years, due to the characteristics of machine learning (ML), which is lightweight, efficient and low-cost, ML-based PLA schemes have also attracted a lot of attention. However, in the dynamic IIoT environment, the existing Mu-PLA schemes are all based on learning, and the accuracy of Mu-PLA schemes decreases exponentially with the increase in the number of users, which becomes an urgent problem to be solved in IIoT. Recent research on Mu-PLA can be divided into two categories: a small number of studies aim to overcome the difficulties of threshold-based PLA schemes in distinguishing multiple users at the same time, while most of the research focuses on designing ML-based PLA frameworks. Existing ML-based Mu-PLA methods all aim to improve the accuracy of identity authentication in single-user PLA, but ignore the means used to improve the accuracy of identity authentication.
[0041] In view of this, the present application embodiment proposes a method for determining the identity of a device in an industrial Internet of Things, referring to Figure 1 , including the following steps:
[0042] Step 101, obtaining communication data between each device and the server, and determining attribute sets corresponding to all devices based on all communication data.
[0043] In this step, in the industrial Internet, a server can communicate with multiple devices. The communication process between the server and the device requires the use of a computer network. The computer network includes a physical layer, which specifies the creation, maintenance, and removal of physical links required for data transmission, as well as the provision of mechanical, electronic, functional, and standardized characteristics. In the communication process between the server and the device, there is communication data, and the properties of the device at the physical layer can be determined through the communication data. In order to improve the verification performance of the physical layer, the properties of all devices are grouped into a set, where each device has multiple properties. The reason why the properties of all devices are grouped into a set is that the probability that different devices have highly similar properties is relatively low, and it is difficult for malicious nodes to imitate multiple properties of a device at the same time. The property set is secure, so that the probability of the server being attacked by malicious nodes is low.
[0044] For example, an access point (AP) Alice and K devices Bob1, Bob2, ..., Bob K communication, where Alice can represent the server and Bob K Able to represent wireless devices. Once Alice receives a data packet, Alice's goal is to use the channel characteristics associated with the device as a unique security signature to detect whether each received data packet comes from a legitimate device and which legitimate device it comes from. Assume that the access point Alice and all legitimate devices are equipped with one antenna. For the multi-user physical layer authentication scenario in the Industrial Internet, the Channel Impulse Response (CIR) can be expressed as:
[0045]
[0046] Among them, H k (t) is Bob K CIR at time slot t, m1 represents the dimension of the CIR vector, and T represents the transpose operation of the matrix / vector.
[0047] Let N be the total number of physical layer attributes, Bob K The nth attribute at time slot t is:
[0048]
[0049] Where m2 represents the dimension of the attribute vector and T represents the transpose operation of the matrix / vector. K The identification characteristics of the received signal can be modeled as a multidimensional random matrix:
[0050]
[0051] It should be noted that, first, multiple attributes are extracted from the received signal and modeled as a three-dimensional matrix. Then, the multi-attribute matrix is divided into two parts, training and testing, and the training data set is input into CANN. Figure 2 The following is a schematic diagram of the architecture of the identity determination model of the embodiment of the present application. The authentication process is divided into online authentication and offline authentication. In the online authentication stage, the network continuously optimizes the parameters using the input training data set. In the offline authentication stage, the access point can directly use the trained network to authenticate the user.
[0052] Step 102, input the attribute set into a pre-trained identity determination model, and perform the following operations in the identity determination model: perform feature extraction on the attribute set to obtain an attribute feature set; based on the attribute set, assign weights to each attribute feature in the attribute feature set to obtain a target attribute feature set; fuse the target attribute feature set with the attribute set to obtain a fused set; perform feature extraction on the fused set to obtain a fused feature; and classify the fused feature to obtain a device identity.
[0053] In this step, the attribute set is input into the pre-trained identity determination model, and the following operations are performed in the identity determination model: feature extraction is performed on the attribute set to obtain an attribute feature set, which accurately reflects the characteristics of the attributes. However, there are multiple attribute features in the attribute feature set. In order to highlight the relatively important attribute features, it is necessary to assign weights to each attribute feature in the attribute set based on the attribute set to obtain a target attribute feature set. When training the identity determination model, it is necessary to make the weight coefficients of the important attributes in the attribute set relatively large, so that when the identity determination model is applied, the weight coefficients of the relatively important attribute features in the target attribute feature set can be relatively large. The target attribute feature set is fused with the attribute set to obtain a fused set, so that the weight coefficients of the important attributes in the attribute set are relatively large. The important attributes in the fused set account for a relatively large proportion, and feature extraction is performed on the fused set to obtain fused features. The fused features reflect the characteristics of each attribute, so that the relatively important attribute features are highlighted. The fused features are classified to obtain the device identity, which can reflect the name of the device, and achieve the purpose of accurately determining the device identity, wherein the device identity is used to characterize the identification of the device and distinguish different devices.
[0054] It should be noted that the identity determination model proposes an attention mechanism to adjust the appropriate weights of the attribute dimensions in the input dataset for identity authentication, and embeds a CNN (Convolutional Neural Networks) architecture with an attention module, called CANN. Figure 3 The following is a detailed diagram of the architecture of the identity determination model of the embodiment of the present application, which consists of an input layer, two convolutional layers containing attention blocks, a culling layer, a dense connection layer, and an output layer. CANN first receives a size of l PN ×R s×N three-dimensional matrix, which contains multiple physical layer attributes of different users from NIST. The output of CANN is the predicted probability that the current input sample belongs to each user category, and the predicted label of the sample is determined by the user category with the highest probability. The convolution layer uses a 5×5 kernel, and the number of kernels in both convolution layers is set to 32, thereby expanding the attribute channels from 3 to 32. The output of the convolution layer is expressed as:
[0055]
[0056] Among them, X [i] is the output of the i-th convolutional layer, X in is the output of the previous layer, w [i] and [i] Represent the convolution kernel and the bias matrix between the previous layer and the i-th convolution layer, respectively. It is the convolution operation.
[0057] Multiple attributes can reflect the environment information and the corresponding position of the fixed equipment. As long as the following assumptions are met, the position of the equipment can be expressed as a function of multiple attributes.
[0058] Assumption: The separation between different devices is more than half a wavelength, which means that the properties of different devices are uncorrelated.
[0059] Assume P k It's Bob K possible positions, and P k It can be expressed as:
[0060] in, is the mapping function between multiple attributes and positions. However, a simple mathematical formula cannot express this mapping function. Therefore, it is necessary to continuously train the neural network to learn this mapping relationship. It is necessary to find a mapping function f(A k ; w), accurately mapping the input multi-attribute sets of multiple devices to the corresponding device sets, thus achieving multi-user identification. That is, f(A k ;w): Where w is the trained network parameter, For the corresponding device set, A set of multiple attributes for multiple devices.
[0061] The process of continuously updating CANN parameters can be regarded as an online authentication process, where the labels predicted by CANN are compared with the input labels of each training to obtain the authentication results. After obtaining the optimized parameters w, the trained CANN model can be directly used for multi-user authentication without updating the parameters, which is regarded as an offline authentication process.
[0062] Step 103: output the device identity corresponding to each device through the identity determination model.
[0063] In this step, the device identity corresponding to each device is output through the identity determination model. There is a mapping relationship between the device identity and the device name. The device name can be reflected through the device identity, thereby achieving the purpose of accurately determining the device name.
[0064] Through the above scheme, the communication data between each device and the server is obtained, and the attribute set corresponding to all devices is determined based on all the communication data, so that the probability of the server being attacked by malicious nodes is low. The attribute set is input into the pre-trained identity determination model, and the following operations are performed in the identity determination model: feature extraction is performed on the attribute set to obtain an attribute feature set, which accurately reflects the characteristics of the attributes. Based on the attribute set, a weight is assigned to each attribute feature in the attribute feature set to obtain a target attribute feature set, so that the weight coefficient of the relatively important attribute features in the target attribute feature set is relatively large. The target attribute feature set is fused with the attribute set to obtain a fused set, so that the weight coefficient of the important attributes in the attribute set is relatively large. Feature extraction is performed on the fused set to obtain a fused feature, so that the relatively important attribute features are highlighted. The fused feature is classified to obtain the device identity, so as to achieve the purpose of accurately determining the device identity. The device identity corresponding to each device is output through the identity determination model, so as to achieve the purpose of accurately determining the device name.
[0065] In some embodiments, the feature extraction of the attribute set to obtain the attribute feature set includes: performing global average pooling processing on the attribute set to obtain a first attribute feature set; performing maximum pooling processing on the attribute set to obtain a second attribute feature set; and fusing the first attribute feature set with the second attribute feature set to obtain the attribute feature set.
[0066] In this embodiment, the attribute set is processed by global average pooling to obtain a first attribute feature set. Global average pooling extracts all feature information in the attribute set to the next layer. The attribute set is processed by maximum pooling to obtain a second attribute feature set. Maximum pooling extracts the most sensitive part of the attribute set to the next layer and discards other redundant information. The first feature set is fused with the second feature set to obtain an attribute feature set. The attribute feature set accurately reflects the characteristics of the attribute. The above process occurs in the attention module of the identity determination model, such as Figure 4 As shown, given an attribute set X as input, the attention mechanism performs global average pooling F on the attribute dimensions of the input matrix AP (·) and the maximum pooling FMP (·), by compressing X's l PN ×R s dimensions to generate attribute statistics in two different contexts. Among them, global average pooling extracts all feature information in the attribute set to the next layer, while maximum pooling extracts the most sensitive part of the attribute set to the next layer and discards other redundant information. In this way, the output of the attribute set is compressed into two feature maps of dimension 1×1×N, which are then input into the shared network as input. The shared network consists of a multi-layer perceptron (MLP), which consists of two N in / r and N in The MLP is trained with the same training settings as the network to further extract features and add the output of the shared network. The sigmoid function is then executed to generate the attribute attention map, which is learned through a fully connected layer and a nonlinear activation function σ to generate the weight vector of the attribute. The multi-attribute attention operation F M The calculation formula is:
[0067] F M (X) = σ(MLP(F AP (X))+MLP(F MP (X))),
[0068] Among them, σ represents the sigmoid operation, and the dimension of the output attribute attention map is R 1×1×N .
[0069] In some embodiments, fusing the target attribute feature set with the attribute set to obtain a fused set includes: multiplying elements in the target attribute feature set with elements in the attribute set to obtain the fused set.
[0070] In this embodiment, by multiplying the elements in the target attribute feature set with the elements in the attribute set, the target attribute feature set is applied to each attribute channel on the attribute set to weigh the features of different attribute channels, thereby achieving the purpose of highlighting relatively important attribute features. It should be noted that the final output fusion set X′ is obtained by rescaling X, which can extract general information about the importance of the input attributes: X ′ =F scale (X,F M (X)) = XF M . It focuses on enhancing the representativeness of the network by building an attribute relationship model, which can be regarded as a weighted adaptive function of the attribute channel.
[0071] In some embodiments, the loss function of the identity determination model is determined by the following formula: in, is the loss function, To minimize the classification loss function calculated by the sparse cross entropy loss function, λ ′ is the dynamic balance quantity, is the center loss function.
[0072] In this embodiment, the loss function of the identity determination model combines the classification loss and the center loss. On the basis of the original multi-classification loss, the network dynamically balances the classification results to adaptively adjust the proportion of the center loss in the overall loss function. Through the above loss function, the training direction of the identity determination model is more accurate. In addition, in order to prevent the huge interference caused by a small number of incorrectly labeled samples, the scalar α is used to control the learning rate of the center, where α is limited to the range of [0,1]. The joint formula of cross entropy loss and center loss is shown in the following formula:
[0073]
[0074] Here, λ is a scalar that balances the two loss functions.
[0075] Different from using a fixed scalar to maintain network balance, a dynamic confidence score c is innovatively introduced to adaptively adjust the network balance by adding a sigmoid layer branch after the dense layer. This confidence score c reflects the confidence of the training network in the certification result. Figure 5 As can be seen in Figure 2, the confidence branch is parallel to the original softmax prediction branch and the center branch. In the confidence branch, the confidence score c, c∈[0,1] is calculated by the sigmoid function, indicating the confidence that the predicted label is correct or not.
[0076]
[0077] Among them, x out is the input of the sigmoid layer. If CANN is confident in the prediction result, the output confidence score will be close to 1. In order to adaptively balance the above joint formula, λ should be a dynamic balance quantity. Therefore, it is possible to consider increasing the confidence c for penalty to prevent CANN from always choosing λ = 0 to minimize the loss function. λ′ = -log(c), where λ′ is the dynamic balance quantity.
[0078] In some embodiments, the method for determining the classification loss function includes: in, is the classification loss function, N s is the number of training samples, l i,kThe target input label for the kth device, p i,k represents the probability that the i-th sample is classified as the k-th device, satisfying
[0079] In this embodiment, each convolutional layer is activated by a rectified linear unit (ReLU), followed by an attention module and a maximum pooling layer of size 2×1. The reduction rate r of the densely connected layer in the MLP of the attention module is 16. In addition, after the last maximum pooling layer, there is a dropout layer to prevent overfitting, and a densely connected layer consisting of 128 neurons, which is activated by the softmax function. The above network is trained using the Adam optimization algorithm as the gradient descent algorithm to minimize the softmax loss calculated by the sparse cross entropy loss function. Among them, w and ξ are the update parameters of the network. Through the classification loss function, the classification result of the identity determination model is made more accurate.
[0080] In some embodiments, the method for determining the center loss function includes:
[0081] in, is the center loss function, N s is the number of training samples, x u ∈R d Indicates that it belongs to y u The u-th deep feature of the category, d is the dimension of the feature, Yes u Class center of deep-like features.
[0082] In this embodiment, in the industrial Internet scenario, users with close geographical locations have greater correlation between their channels when they are stationary or moving at low speed, which may affect the user authentication results. In order to further improve the network authentication performance and prevent adjacent users from being confused, the center loss is introduced, which can effectively enhance the discrimination ability of deep learning features in neural networks. Specifically, the center loss penalizes the distance between the deep features and the centers of the corresponding categories, while learning a center for each category of the user's deep features. Under the joint supervision of cross entropy loss and center loss, a more effective network can be trained to obtain as many deep features as possible and achieve two important learning goals: intra-class compactness and inter-class dispersion. The calculation formula of the center loss function can effectively describe intra-class changes. Ideally, The update should be done based on the deep features, i.e. the center is calculated by averaging the corresponding category features and updated in each mini-batch:
[0083]
[0084] where δ(condition) = 1 (if condition is satisfied) and δ(condition) = 0 (if condition is not satisfied). This means that when y u The category and c j At different times, the feature center c j Specifically, for each category j, the output feature value of the j-th category minus the value of the j-th feature center is averaged and used as the step value for updating the category center.
[0085] In some embodiments, the attribute set includes at least amplitude, phase-frequency distortion and delay; the determining the attribute set corresponding to all devices based on all communication data includes: determining the amplitude by the following formula: in, is the amplitude, amplitude represents the amplitude characteristics, is the zth channel impulse response of the pseudo-random code sequence received by the kth device, imag(·) represents the imaginary part of the complex number, and real(·) represents the real part of the complex number; the phase-frequency distortion is determined by the following formula: in, is the phase-frequency distortion, p represents the phase-frequency characteristic, is the zth channel impulse response of the pseudo-random code sequence received by the kth device, imag(·) represents the imaginary part of the complex number, real(·) represents the real part of the complex number, l PN is the length of the pseudo-random code sequence sent by the transmitter, R s is the oversampling rate of the receiver, PN represents a pseudo-random code; the delay is determined by the following formula: Among them, (x k ,y k ) is the location coordinate of device k, d represents the delay characteristic, c is the propagation speed of radio waves, and (x, y) is the location coordinate of the server.
[0086] In this embodiment, the determination of the attribute set is related to the data set for training the identity determination model, taking into account the attributes available in the data set, such as amplitude, phase-frequency distortion, and delay. Exemplarily, the data set can come from the National Institute of Standards and Technology (NIST), which provides physical layer channel data in industrial wireless environments. In addition, it can be extended to other data sets with different attributes and different numbers of attributes. In order to determine the attribute set corresponding to all devices, it is necessary to calculate the attributes using communication data and digitize the attributes to achieve the purpose of accurately determining the attribute values. It should be noted that all acquired communication data are instantaneous and change over time.
[0087] The transmitter in NIST repeatedly transmits a positive-negative (PN) sequence modulated by a binary phase shift keying (BPSK) signal. To facilitate the merging of multiple attributes, and are expanded into matrices and and For matrices of the same dimension, the value of each element in the matrix is equal to and Before inputting the data into CANN, the dimension l PN ×R s Combine multiple attributes of PN ×R s ×N three-dimensional matrix, where the last dimension is the number of attributes. K The input data is represented as:
[0088]
[0089] Among them, f c Represents an operation that adds the last dimension to the input matrix and establishes a connection. Bob K Medium H k The corresponding label is k-1. Therefore, the training data set It consists of input data H and input label L:
[0090]
[0091] Where H = [H1; ...; H K ] and L=[0;…;K-1].
[0092] In another embodiment provided in the present application, the performance of the proposed CANN-based authentication is analyzed in an automotive assembly plant (AAP) of NIST. The computer configuration is an Intel(R) Core(TM) i7-8700 U CPU, 3.2-GHz base frequency, and 16GB memory. Since the proposed solution is a lightweight authentication solution using TensorFlow as a framework, it does not rely on GPU for training. AAP is a classic Rayleigh fading channel scenario with a dynamic data set. Figure 6 As shown, the AP (server) is fixed at (0m, 0m), and the mobile node moves along the pre-designed path at an average speed of 1.55m / s and returns to the origin at the loop. The total number of nodes is 109, and 15 of them are selected as different mobile devices. The dataset records data for 300 time periods, so each device has 300 CIR records, and the number of samples imported into the network for each device is 300. In order to train, verify and test the CANN-based authentication model, the dataset is randomly divided into training set, verification set and test set in a ratio of 7:1:2. PN sequence l PN The length of is 2047, and the oversampling rate Rs = 4. The parameters in CANN are as follows. Batch size ζ = 100, decay step T = 10, decay rate β = 1, and the learning rate γ is reduced using the inverse time decay function, starting from 0.001. The training process stops, and if the validation loss does not decrease within 20 epochs, the model with the minimum validation loss is stored and used for testing.
[0093] To evaluate the performance of CANN-based Mu-PLA, Acc is defined as the accuracy of Mu-PLA, which is expressed as follows:
[0094]
[0095] Among them, M represents the sample size in the offline verification process, L m is the true label of the mth sample, is its predicted label. Represents an indicator function, which takes the value 1 if the content in the brackets is true, and takes the value 0 otherwise.
[0096] In Mu-PLA, in addition to Acc as an evaluation indicator, precision P and recall R are also considered.
[0097] For the kth device, the accuracy P k and recall R k Respectively expressed as:
[0098]
[0099] Among them, Nk is the number of samples of the kth device that is correctly classified, D k is the sample size of the kth device being judged, and T k is the number of samples of the kth device being classified.
[0100] To evaluate the performance of the proposed Mu-PLA scheme, it is compared with existing Mu-PLA schemes: LPCNN[5]: A Mu-PLA scheme based on a CNN framework that uses latent space to inject Gaussian noise. DNN[6]: A Mu-PLA scheme based on a DNN that uses data augmented IQ data as input. In addition, control experiments of CANN are considered: CNN: A Mu-PLA scheme based on a CNN without an attention module, where the input layer is followed by two convolutional layers and a maximum pooling layer of size 2×1 is integrated. ACNN: A CNN architecture with an embedded attention module, where the position of the attention module is different from that of CANN, and the input layer is followed by an attention module, a maximum pooling layer, and a convolutional layer. In the first attention block, the reduction rate r of the MLP densely connected layer is 2, while in the second attention block, the reduction rate r is 16.
[0101] Considering the different numbers of devices in AAP, the nodes corresponding to different device situations are shown in Table 1.
[0102] Table 1 Points corresponding to different scenes
[0103]
[0104] Figure 7 shows the authentication performance of CANN and the comparison schemes in different situations, where the amplitude represents the single attribute situation, Figure 7(a) shows the authentication performance of CANN and the comparison scheme in the multi-attribute scenario; Figure 7(b) shows the authentication performance of CANN and the comparison scheme in the single-attribute scenario. It can be seen from these two figures that as the number of devices increases, the authentication accuracy also decreases, and CANN has a higher authentication accuracy than other schemes. In the multi-attribute 3-device scenario, the accuracy of CANN, ACNN, CNN, and LPCNN all reached 1.0, which means that these schemes can verify all devices, while the accuracy of DNN can only reach 77.5%. When the number of devices is 15, the accuracy of CANN is 4.34%, 8.11%, 15.23%, and 40.61% higher than that of ACNN, LPCNN, CNN, and DNN, respectively. This shows that the authentication performance of CANN is far better than that of the other four schemes. In the single-attribute scenario, as shown in Figure 7(b), when the number of devices is 15, CANN has an authentication accuracy that is 1.45% higher than that of LPCNN, which is about 58.67%. This is because the attention module can increase the weight of valid attributes of the input dataset and reduce the weight of invalid attributes. Therefore, CANN based on the attention module performs better in multi-attribute scenarios, especially when the number of devices is large.
[0105] Figure 8(a)-Figure 8(f) 8(a) is a schematic diagram of the output characteristics of the central layer when the number of devices is different according to the embodiment of the present application, wherein FIG8(b) is a schematic diagram of the output characteristics of 11 devices-CANN, FIG8(c) is a schematic diagram of the output characteristics of 15 devices-CANN, and FIG8(d) is a schematic diagram of the output characteristics of 7 devices-CANN- Schematic diagram of the output characteristics of , Figure 8 (e) is 11 devices -CANN- Schematic diagram of the output characteristics of , Figure 8 (f) is 15 devices -CANN- Schematic diagram of the output characteristics of the CANN center layer. The red stars in each sub-graph are the feature centers of each type of device. In addition, the features of different device categories are represented by points of different colors. In Figure 7, when the number of devices increases, the spacing between the feature centers of each type of device is small, and even overlaps when CANN is trained using only cross entropy loss. This indicates that the output features of each device category are mixed, which may cause the network to confuse the features of each device category, resulting in misclassification. In contrast, when using CANN trained with the introduction of center confidence loss, the output features of each device category are significantly more concentrated, and the discrimination of the feature centers is relatively high, indicating that the output features are more separable.
[0106] Fig. 9The schematic diagram of the authentication accuracy based on the central confidence loss of the embodiment of the present application shows the authentication accuracy of different models trained with the central confidence loss. The proposed CANN still achieved the best authentication performance, with an authentication accuracy of 76.03% when the number of devices was 15. At the same time, as shown in Figure 7(a), compared with the network trained only with cross entropy loss, the verification accuracy of CANN, ACNN, LPCN, CNN and DNN increased by 2.36%, 3.34%, 2.22%, 3% and 3.44%, respectively. This result further proves the effectiveness of the central confidence loss, which can prevent the authentication network from confusing the device, thereby improving the authentication performance.
[0107] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the described method.
[0108] It should be noted that the above describes some embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0109] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a device identity determination apparatus in the industrial Internet of Things.
[0110] refer to Fig.10 , the device identity determination device in the industrial Internet of Things comprises:
[0111] The first determination module 10 is configured to obtain communication data between each device and the server, and determine a set of attributes corresponding to all devices based on all communication data.
[0112] The operation module 20 is configured to input the attribute set into a pre-trained identity determination model, and perform the following operations in the identity determination model: perform feature extraction on the attribute set to obtain an attribute feature set; assign weights to each attribute feature in the attribute feature set based on the attribute set to obtain a target attribute feature set; fuse the target attribute feature set with the attribute set to obtain a fused set; perform feature extraction on the fused set to obtain a fused feature; and classify the fused feature to obtain a device identity.
[0113] The output module 30 is configured to output the device identity corresponding to each device through the identity determination model.
[0114] Through the above device, the communication data between each device and the server is obtained, and based on all the communication data, the attribute set corresponding to all the devices is determined, so that the probability of the server being attacked by malicious nodes is low. The attribute set is input into the pre-trained identity determination model, and the following operations are performed in the identity determination model: feature extraction is performed on the attribute set to obtain an attribute feature set, which accurately reflects the characteristics of the attributes. Based on the attribute set, a weight is assigned to each attribute feature in the attribute feature set to obtain a target attribute feature set, so that the weight coefficient of the relatively important attribute features in the target attribute feature set is relatively large. The target attribute feature set is fused with the attribute set to obtain a fused set, so that the weight coefficient of the important attributes in the attribute set is relatively large. Feature extraction is performed on the fused set to obtain a fused feature, so that the relatively important attribute features are highlighted. The fused feature is classified to obtain the device identity, so as to achieve the purpose of accurately determining the device identity. The device identity corresponding to each device is output through the identity determination model, so as to achieve the purpose of accurately determining the device name.
[0115] In some embodiments, the operation module 20 is further configured to perform global average pooling processing on the attribute set to obtain a first attribute feature set; perform maximum pooling processing on the attribute set to obtain a second attribute feature set; and fuse the first attribute feature set with the second attribute feature set to obtain the attribute feature set.
[0116] In some embodiments, the operation module 20 is further configured to multiply the elements in the target attribute feature set with the elements in the attribute set to obtain the fused set.
[0117] In some embodiments, a second determination module is further included, wherein the second determination module is configured to determine the loss function of the identity determination model by the following formula: in, is the loss function, To minimize the classification loss function calculated by the sparse cross entropy loss function, λ ′ is the dynamic balance quantity, is the center loss function.
[0118] In some embodiments, the second determination module is further configured as a method for determining the classification loss function, including: in, is the classification loss function, N s is the number of training samples, l i,k The target input label for the kth device, p i,k represents the probability that the i-th sample is classified as the k-th device, satisfying
[0119] In some embodiments, the second determination module is further configured as a method for determining the center loss function, including: in, is the center loss function, N s is the number of training samples, x u ∈R d Indicates that it belongs to y u The u-th deep feature of the category, d is the dimension of the feature, Yes u Class center of deep-like features.
[0120] In some embodiments, the first determination module 10 is further configured such that the attribute set includes at least amplitude, phase-frequency distortion and delay; the amplitude is determined by the following formula: in, is the amplitude, amplitude represents the amplitude characteristics, is the zth channel impulse response of the pseudo-random code sequence received by the kth device, imag(·) represents the imaginary part of the complex number, and real(·) represents the real part of the complex number; the phase-frequency distortion is determined by the following formula: in, is the phase-frequency distortion, p represents the phase-frequency characteristic, is the zth channel impulse response of the pseudo-random code sequence received by the kth device, imag(·) represents the imaginary part of the complex number, real(·) represents the real part of the complex number, l PN is the length of the pseudo-random code sequence sent by the transmitter, R s is the oversampling rate of the receiver, PN represents a pseudo-random code; the delay is determined by the following formula: Among them, (x k ,y k) is the location coordinate of device k, d represents the delay characteristic, c is the propagation speed of radio waves, and (x, y) is the location coordinate of the server.
[0121] For the convenience of description, the above device is described in terms of functions divided into various modules. Of course, when implementing the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.
[0122] The apparatus of the above embodiment is used to implement the device identity determination method in the corresponding industrial Internet of Things in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0123] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the device identity determination method in the industrial Internet of Things as described in any of the above embodiments is implemented.
[0124] Fig.11 A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.
[0125] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.
[0126] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.
[0127] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.
[0128] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).
[0129] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).
[0130] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.
[0131] The electronic device of the above embodiment is used to implement the device identity determination method in the corresponding industrial Internet of Things in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.
[0132] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the device identity determination method in the industrial Internet of Things as described in any of the above embodiments.
[0133] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.
[0134] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the device identity determination method in the industrial Internet of Things as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0135] Based on the same concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a computer program product, including computer program instructions. When the computer program instructions are run on a computer, the computer executes the device identity determination method in the industrial Internet of Things as described in any of the above embodiments, which has the beneficial effects of the corresponding method embodiments and will not be repeated here.
[0136] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application (including the claims) is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.
[0137] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply / ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform to be implemented in the embodiments of the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present application, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered to be illustrative rather than restrictive.
[0138] Although the present application has been described in conjunction with specific embodiments of the present application, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.
[0139] The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the present application.
Claims
1. A method for determining device identity in an industrial Internet of Things, characterized in that: include: Obtain the communication data between each device and the server, and determine the attribute set corresponding to all devices based on all the communication data; The attribute set is input into a pre-trained identity determination model, where the following operations are performed: Performing feature extraction on the attribute set to obtain an attribute feature set; Based on the attribute set, assigning a weight to each attribute feature in the attribute feature set to obtain a target attribute feature set; Fusing the target attribute feature set with the attribute set to obtain a fused set; Performing feature extraction on the fusion set to obtain fusion features; Classifying the fused features to obtain a device identity; The device identity corresponding to each device is outputted through the identity determination model.
2. The method according to claim 1, characterized in that The feature extraction of the attribute set to obtain an attribute feature set includes: Performing global average pooling processing on the attribute set to obtain a first attribute feature set; Performing maximum pooling processing on the attribute set to obtain a second attribute feature set; The first attribute feature set and the second attribute feature set are merged to obtain the attribute feature set.
3. The method according to claim 1, characterized in that: The step of fusing the target attribute feature set with the attribute set to obtain a fused set includes: The elements in the target attribute feature set are multiplied by the elements in the attribute set to obtain the fused set.
4. The method according to claim 1, characterized in that: The loss function of the identity determination model is determined by the following formula: in, is the loss function, To minimize the classification loss function calculated by the sparse cross entropy loss function, λ ′ is the dynamic balance quantity, is the center loss function.
5. According to the method of claim 4, the method for determining the classification loss function comprises: in, is the classification loss function, N s is the number of training samples, l i,k The target input label for the kth device, p i,k represents the probability that the i-th sample is classified as the k-th device, satisfying 6. According to the method of claim 4, the method for determining the center loss function comprises: in, is the center loss function, N s is the number of training samples, x u ∈R d Indicates that it belongs to y u The u-th deep feature of the category, d is the dimension of the feature, Yes u Class center of deep-like features.
7. The method according to claim 1, characterized in that The attribute set includes at least amplitude, phase-frequency distortion and delay; The determining of the attribute set corresponding to all devices based on all communication data includes: The amplitude is determined by the following formula: in, is the amplitude, amplitude represents the amplitude characteristics, is the zth channel impulse response of the pseudo-random code sequence received by the kth device, imag(·) represents the imaginary part of the complex number, and real(·) represents the real part of the complex number; The phase-frequency distortion is determined by the following formula: in, is the phase-frequency distortion, p represents the phase-frequency characteristic, is the zth channel impulse response of the pseudo-random code sequence received by the kth device, imag(·) represents the imaginary part of the complex number, real(·) represents the real part of the complex number, l PN is the length of the pseudo-random code sequence sent by the transmitter, R s is the oversampling rate of the receiver, PN represents the pseudo-random code; The delay is determined by the following formula: Among them, (x k ,y k ) is the location coordinate of device k, d represents the delay characteristic, c is the propagation speed of radio waves, and (x, y) is the location coordinate of the server.
8. A device for determining the identity of an industrial Internet of Things, characterized in that: include: A first determination module is configured to obtain communication data between each device and the server, and determine a set of attributes corresponding to all devices based on all communication data; An operation module is configured to input the attribute set into a pre-trained identity determination model, and perform the following operations in the identity determination model: Performing feature extraction on the attribute set to obtain an attribute feature set; Based on the attribute set, assigning a weight to each attribute feature in the attribute feature set to obtain a target attribute feature set; Fusing the target attribute feature set with the attribute set to obtain a fused set; Performing feature extraction on the fusion set to obtain fusion features; Classifying the fused features to obtain a device identity; The output module is configured to output the device identity corresponding to each device through the identity determination model.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.
10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.
Citation Information
Patent Citations
Equipment identity identification method, computer equipment and storage medium
CN106998553A
Human body behavior recognition and identity authentication method and device and electronic equipment
CN113449587A
Authentication of physical objects based on reliable parts of copy detection patterns
WO2024126062A1
