False data detection method and device

Abstract

The invention provides a false data detection method and device, and the method is applied to a perception function network element, and the method comprises the steps: responding to a received perception task request, and requesting perception data from at least two target perception devices; and based on a multi-modal comparison generative adversarial network model, performing false data detection on the sensing data corresponding to each target sensing device. According to the method, the multi-modal comparison generative adversarial network algorithm is introduced, false data in the sensing data can be accurately detected, and the detection efficiency and accuracy are improved.

Description

Technical Field

[0001] This application relates to the field of core network technology, and in particular to a method and apparatus for detecting fake data. Background Technology

[0002] In integrated communication and sensing networks, numerous nodes exist, and some heterogeneous sensing devices / nodes have limited resources, making them vulnerable to attacks. Therefore, when heterogeneous sensing devices / nodes are networked for sensing, especially non-3GPP (3rd Generation Partnership Project) sensing nodes which may be semi-trusted, there is a possibility of abnormal devices / nodes compromising system security, tampering with data, and / or injecting false information, thus reducing the performance of sensing data processing functions or related applications. Furthermore, some data can be injected into the sensing data to evade attack detection, making false data injection attacks more difficult to detect, thus minimizing the detectability of attacks while reducing the sensing accuracy of the integrated communication and sensing system. Therefore, an effective solution is needed to detect false data injection attacks or filter falsely injected data to protect the security of the data fusion process in integrated communication and sensing networks. Summary of the Invention

[0003] To address the aforementioned problems, this application provides a method and apparatus for detecting false data.

[0004] This application provides a method for detecting fake data, applied to a sensing function network element, including: In response to a received sensing task request, sensing data is requested from at least two target sensing devices; Based on a multimodal contrastive generative adversarial network model, false data detection is performed on the sensing data corresponding to each of the target sensing devices.

[0005] According to the fake data detection method provided in this application, the method for detecting fake data on the sensing data corresponding to each target sensing device based on a multimodal contrastive generative adversarial network model includes: The data features of each of the aforementioned sensing data are fused using multimodal features to obtain fused features; Based on the generative adversarial network model in the multimodal contrastive generative adversarial network model, the fused features are used for data discrimination to obtain the first probability that each of the perceived data is real data; and the data features of each of the perceived data are subjected to multimodal feature consistency comparison verification to obtain feature consistency score. Based on the first probability and the feature consistency score, a second probability is determined to characterize the presence of false data in each of the perceived data. Based on the second probability, it is determined whether the false data exists in each of the perceived data.

[0006] According to the fake data detection method provided in this application, before fusing the data features of each of the perceived data into multimodal features to obtain the fused features, the method further includes: For each of the sensed data, feature processing is performed on the sensed data based on the feature extraction strategy corresponding to the data type of the sensed data to obtain the data features of the sensed data.

[0007] According to the fake data detection method provided in this application, the step of performing feature processing on the perceived data based on the feature extraction strategy corresponding to the data type of the perceived data to obtain the data features of the perceived data includes: When the data type is point cloud data, the sensing data is aligned, mapped, and aggregated to obtain the data features of the sensing data; When the data type is image data, residual learning and identity mapping are performed on the perceived data to obtain the data features of the perceived data.

[0008] According to the fake data detection method provided in this application, the step of performing multimodal feature consistency comparison verification on the data features of each of the perceived data to obtain a feature consistency score includes: For each of the sensed data, the data features of the sensed data are residually concatenated with the first data feature to obtain the residual corresponding to the sensed data. The first data feature is the data feature of other sensed data or the data feature generated by the generator. Multimodal feature consistency comparison verification is performed on the residuals corresponding to each of the aforementioned sensing data to obtain feature consistency scores.

[0009] According to the fake data detection method provided in this application, the step of determining a second probability representing the presence of fake data in each of the perceived data based on the first probability and the feature consistency score includes: The first probability and the feature consistency score are integrated to obtain integrated data; The integrated data is subjected to a nonlinear transformation to obtain the second probability.

[0010] According to the fake data detection method provided in this application, before performing feature processing on the perceived data based on the feature extraction strategy corresponding to the data type of the perceived data to obtain the data features of the perceived data, the method further includes: Each of the sensed data is preprocessed, including data cleaning and / or format conversion.

[0011] According to the fake data detection method provided in this application, after performing fake data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network model, the method further includes: In the event of detected false data, the identification of the sensing device corresponding to the false data and the location identification of the false data in the sensing data are determined. Based on the sensing device identifier and the location identifier, the false data in each of the sensing data is filtered to obtain the real data; The real data are fused together to obtain a fused perception result, and the fused perception result is reported.

[0012] According to the fake data detection method provided in this application, the method further includes: The PCF network element, through its policy control function, requests a perception data policy from the UDR network element, which includes perception data rules and security levels. When the security level meets the conditions for false data detection, the step of performing false data detection on the sensing data corresponding to each of the target sensing devices is executed based on the multimodal contrastive generative adversarial network model.

[0013] According to a method for detecting false data provided in this application, the perceived data rule further includes a threshold for the number of times false data is reported. The step of requesting sensing data from at least two target sensing devices includes: Based on the threshold for the number of false data reports, device detection is performed on each of the target sensing devices. Request sensing data from the target sensing device that has passed the device detection.

[0014] According to a method for detecting false data provided in this application, the threshold for the number of times false data is reported includes a first threshold at the global granularity. The step of performing device detection on each target sensing device based on the threshold for the number of false data reports includes: For each target sensing device, the global false identifier of the target sensing device is compared with the first number threshold, wherein the global false identifier represents the number of times the target sensing device stops providing sensing data; If the number of global false identifiers exceeds the first threshold, it is determined that the target sensing device has failed device detection. If the number of false global identifiers is less than or equal to the first threshold number, the target sensing device is determined to have passed device detection.

[0015] According to a method for detecting fake data provided in this application, the threshold for the number of times fake data is reported includes a second threshold at the task granularity. The multimodal contrastive generative adversarial network model, after detecting false data in the sensing data corresponding to each target sensing device, further includes: For each target sensing device, if false data exists in the sensing data corresponding to the target sensing device, the task false identifier of the target sensing device is updated. The task false identifier represents the number of times the target sensing device provides false data under the current sensing task. If the number of false task identifiers reaches the second threshold, the request for sensing data from the target sensing device corresponding to the false task identifier will be stopped.

[0016] According to the fake data detection method provided in this application, the method further includes: If the request for sensing data from the target sensing device corresponding to the task false identifier is stopped, the global false identifier of the target sensing device corresponding to the task false identifier is updated.

[0017] According to the fake data detection method provided in this application, the step of requesting a perceived data policy from the Unified Data Repository (UDR) network element through the Policy Control Function (PCF) network element includes: Send a sensing data policy request to the PCF network element; The sensing data policy request carries at least one of the following: information of the application function AF network element, sensing range, and sensing service type. The AF network element is the network element that initiates the sensing task request. The perception data policy request is used by the PCF network element to request the perception data policy from the UDR network element and to feed back the perception data policy to the perception function network element.

[0018] This application also provides a method for detecting fake data, applied to SF-C network elements, including: In response to the received sensing task request, the target SF user plane SF-U network element is identified, and sensing data is requested from at least two target sensing devices; The target SF-U network element is used to receive and perform false data detection on the sensing data corresponding to each target sensing device based on a multimodal contrastive generative adversarial network model.

[0019] According to the fake data detection method provided in this application, before requesting sensing data from at least two target sensing devices, the method further includes: Based on the sensing parameters carried in the sensing task request, select at least two target sensing devices.

[0020] According to the fake data detection method provided in this application, before determining the target SF-U network element, the method further includes: The PCF network element requests the perception data policy from the UDR network element. The determination of the target SF-U network element includes: The target SF-U network element is selected based on the sensing parameters carried in the sensing task request and / or the sensing data strategy.

[0021] According to the fake data detection method provided in this application, the method further includes: Send the sensing task request and the sensing data policy to the target SF-U network element; The perception task request and the perception data policy are used by the target SF-U network element to initiate a false data detection operation after configuring and activating the perception data policy.

[0022] According to a method for detecting false data provided in this application, the data perception strategy includes data perception rules, and the data perception rules include a threshold for the number of times false data is reported. The step of requesting sensing data from at least two target sensing devices includes: Based on the threshold for the number of false data reports, device detection is performed on each of the target sensing devices. Request sensing data from the target sensing device that has passed the device detection.

[0023] According to a method for detecting false data provided in this application, the threshold for the number of times false data is reported includes a first threshold at the global granularity. The step of performing device detection on each target sensing device based on the threshold for the number of false data reports includes: For each target sensing device, the global false identifier of the target sensing device is compared with the first number threshold, wherein the global false identifier represents the number of times the target sensing device stops providing sensing data; If the number of global false identifiers exceeds the first threshold, it is determined that the target sensing device has failed device detection. If the number of false global identifiers is less than or equal to the first threshold number, the target sensing device is determined to have passed device detection.

[0024] According to a method for detecting fake data provided in this application, the threshold for the number of times fake data is reported includes a second threshold at the task granularity. The method further includes: The system receives the identification of the sensing device corresponding to the false data reported by the target SF-U network element, wherein the false data is determined by the target SF-U network element through false data detection of the sensing data corresponding to each target sensing device; Update the task false identifier of the target sensing device corresponding to the sensing device identifier, wherein the task false identifier represents the number of times the target sensing device provides false data under the current sensing task; If the number of false task identifiers reaches the second threshold, the request for sensing data from the target sensing device corresponding to the false task identifier will be stopped.

[0025] According to the fake data detection method provided in this application, the method further includes: If the request for sensing data from the target sensing device corresponding to the task false identifier is stopped, the global false identifier of the target sensing device corresponding to the task false identifier is updated.

[0026] According to the fake data detection method provided in this application, after receiving the sensing device identifier corresponding to the fake data reported by the target SF-U network element, the method further includes: Send a data filtering request to the target SF-U network element; The data filtering request is used by the target SF-U network element to filter and fuse the false data in each of the sensing data to obtain the fused sensing result, and then report the fused sensing result.

[0027] This application also provides a method for detecting fake data, applied to SF-U network elements, including: Receive sensing data from at least two target sensing devices; Based on a multimodal contrastive generative adversarial network model, false data detection is performed on the sensing data corresponding to each of the target sensing devices.

[0028] According to the fake data detection method provided in this application, the method for detecting fake data on the sensing data corresponding to each target sensing device based on a multimodal contrastive generative adversarial network model includes: The data features of each of the aforementioned sensing data are fused using multimodal features to obtain fused features; Based on the generative adversarial network model in the multimodal contrastive generative adversarial network model, the fused features are used for data discrimination to obtain the first probability that each of the perceived data is real data; and the data features of each of the perceived data are subjected to multimodal feature consistency comparison verification to obtain feature consistency score. Based on the first probability and the feature consistency score, a second probability is determined to characterize the presence of false data in each of the perceived data. Based on the second probability, it is determined whether the false data exists in each of the perceived data.

[0029] According to the fake data detection method provided in this application, before fusing the data features of each of the perceived data into multimodal features to obtain the fused features, the method further includes: For each of the sensed data, feature processing is performed on the sensed data based on the feature extraction strategy corresponding to the data type of the sensed data to obtain the data features of the sensed data.

[0030] According to the fake data detection method provided in this application, the step of performing feature processing on the perceived data based on the feature extraction strategy corresponding to the data type of the perceived data to obtain the data features of the perceived data includes: When the data type is point cloud data, the sensing data is aligned, mapped, and aggregated to obtain the data features of the sensing data; When the data type is image data, residual learning and identity mapping are performed on the perceived data to obtain the data features of the perceived data.

[0031] According to the fake data detection method provided in this application, the step of performing multimodal feature consistency comparison verification on the data features of each of the perceived data to obtain a feature consistency score includes: For each of the sensed data, the data features of the sensed data are residually concatenated with the first data feature to obtain the residual corresponding to the sensed data. The first data feature is the data feature of other sensed data or the data feature generated by the generator. Multimodal feature consistency comparison verification is performed on the residuals corresponding to each of the aforementioned sensing data to obtain feature consistency scores.

[0032] According to the fake data detection method provided in this application, the step of determining a second probability representing the presence of fake data in each of the perceived data based on the first probability and the feature consistency score includes: The first probability and the feature consistency score are integrated to obtain integrated data; The integrated data is subjected to a nonlinear transformation to obtain the second probability.

[0033] According to the fake data detection method provided in this application, after performing fake data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network model, the method further includes: In the event of detected false data, the identification of the sensing device corresponding to the false data and the location identification of the false data in the sensing data are determined. Based on the sensing device identifier and the location identifier, the false data in each of the sensing data is filtered to obtain the real data; The real data are fused together to obtain a fused perception result, and the fused perception result is reported.

[0034] This application also provides a fake data detection device, applied to a sensing function network element, including: The first request module is configured to request sensing data from at least two target sensing devices in response to a received sensing task request. The first detection module is configured to perform false data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network model.

[0035] This application also provides a fake data detection device, applied to SF-C network elements, including: The second request module is configured to, in response to a received sensing task request, identify the target SF-U network element and request sensing data from at least two target sensing devices. The target SF-U network element is used to receive and perform false data detection on the sensing data corresponding to each target sensing device based on a multimodal contrastive generative adversarial network model.

[0036] This application also provides a Y-type fake data detection device, applied to SF-U network elements, including: The receiving module is configured to receive sensing data provided by at least two target sensing devices; The second detection module is configured to perform false data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network model.

[0037] This 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 the processor executes the computer program to implement any of the above-described methods for detecting fake data.

[0038] This application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the fake data detection method as described above.

[0039] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the fraudulent data detection methods described above.

[0040] The fake data detection method and apparatus provided in this application include: responding to a received sensing task request by requesting sensing data from at least two target sensing devices; and performing fake data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network (MGA) model. By introducing the MGA algorithm, fake data in the sensing data can be accurately detected, improving detection efficiency and accuracy. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is one of the flowcharts of the false data detection method provided in this application.

[0043] Figure 2 This is the second flowchart of the false data detection method provided in this application.

[0044] Figure 3 This is the third flowchart of the false data detection method provided in this application.

[0045] Figure 4 This is the fourth flowchart of the false data detection method provided in this application.

[0046] Figure 5 This is one of the structural schematic diagrams of the false data detection device provided in this application.

[0047] Figure 6 This is the second schematic diagram of the false data detection device provided in this application.

[0048] Figure 7 This is the third schematic diagram of the false data detection device provided in this application.

[0049] Figure 8 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] First, a brief explanation of the relevant content involved in this application will be given.

[0052] In the current industry-proposed sensing service architecture, the SF (Sensing Function) network element responsible for sensing service processing can perform sensing service analysis and provide sensing results based on the received sensing requirements. Its deployment forms are: it can be deployed independently; it can be deployed in conjunction with 5G (5th Generation Mobile Communication Technology) core network elements, such as AMF (Access and Mobility Management Function) network elements or LMF (Location Management Function) network elements; it can also be deployed in conjunction with 6G (6th Generation Mobile Communication Technology) network elements. At the same time, according to the deployment requirements, SF can exist in two forms: SF-C (SF-Control Plane) and SF-U (SF-User Plane).

[0053] The main body for performing the sensing task is the base station and the terminal side in coordination; that is, the base station can transmit and receive signals to achieve the initial acquisition of sensing signals, or the base station and the terminal can cooperate to complete the transmission and reception of signals, thereby achieving the acquisition of sensing signals. The acquired sensing signals are processed by the base station to generate sensing data. The sensing device (base station / terminal) reports the sensing data to the SF. The sensing data is processed by the SF to form the sensing result.

[0054] SF's basic functions include: supporting perception authorization, perception control, data processing, and result output.

[0055] Existing sensing technologies have the following shortcomings: Vulnerable to security: In the integrated communication and sensing network, due to the large number of nodes and the limited resources of some heterogeneous sensing devices / nodes, the system is easily attacked and hijacked, which threatens security. Data tampering and false information injection: Abnormal devices / nodes may compromise system security by tampering with data or injecting false information, thereby reducing the performance of perceived data processing functions or related applications; The attack is difficult to detect: the injected data can be carefully constructed to avoid being detected by traditional detection methods, making it more difficult to detect fake data injection attacks, reducing the system's perception accuracy, and minimizing the detectability of the attack.

[0056] To address the aforementioned problems, this application provides a method and apparatus for detecting false data.

[0057] The following is combined Figures 1-8 This application describes a method and apparatus for detecting false data.

[0058] Figure 1 This is one of the flowcharts illustrating the fake data detection method provided in this application, such as... Figure 1 As shown, applied to sensing function network elements, this method includes the following: Step 101: In response to the received sensing task request, request sensing data from at least two target sensing devices; Step 102: Based on the multimodal contrastive generative adversarial network model, perform fake data detection on the sensing data corresponding to each target sensing device.

[0059] Specifically, the network element / UE (User Equipment) / third-party service provider and other perception requesting party AF (Application Function) network element initiates perception service request and sends it to the NEF (Network Exposure Function) network element.

[0060] Optionally, the parameters carried in the perception service request include: perception task identifier (afTaskId) and perception data type.

[0061] Optionally, the sensing service request may also carry at least one of sensing range, sensing accuracy, and sensing data reporting frequency.

[0062] Specifically, when a NEF network element receives a sensing task request, it selects an SF (SF-C) network element based on the service area corresponding to the sensing task request (sensing range) and sends the sensing task request to the SF (SF-C) network element.

[0063] The SF (SF-C) network element selects a target sensing device to perform a sensing operation based on the sensing requirements (parameters) carried in the sensing request, and sends a sensing operation request to the target sensing device. The number of target sensing devices can be multiple (at least two), including but not limited to base stations, radar, and non-3GPP (3rd Generation Partnership Project) devices (such as cameras).

[0064] Optionally, the sensing operation request may carry sensing operation parameters, such as at least one of afTaskId, sensing range, sensing accuracy, sensing data type, and sensing data reporting frequency.

[0065] The target sensing device acknowledges receipt of the sensing operation request and begins executing the sensing operation. Before performing the sensing operation, the target sensing device has requested the UDM (Unified Data Management) network element to check user licenses and authorization information. Each target sensing device establishes a data transmission channel with the SF (SF-U) network element to report sensing data.

[0066] Furthermore, the sensing data uploaded by multiple sources and multiple devices (each target sensing device) of the SF (SF-U) network element are used to detect whether there is false data in the sensing data uploaded by each target sensing device through a multimodal comparative generative adversarial network model.

[0067] In addition, upon receiving a sensing task request, the SF (SF-C) network element can also send a request to initiate a sensing task and receive a response to the AF network element through the NEF network element. That is, the SF-C network element returns a request to initiate a sensing task and receives a response to the NEF network element, and the NEF network element returns a request to initiate a sensing task and receives a response to the AF network element.

[0068] The fake data detection method provided in this application introduces a multimodal contrastive generative adversarial network algorithm, which can accurately detect fake data in perceived data, thereby improving detection efficiency and accuracy.

[0069] Optionally, the step of detecting fake data in the sensing data corresponding to each of the target sensing devices based on the multimodal contrastive generative adversarial network model includes: The data features of each of the aforementioned sensing data are fused using multimodal features to obtain fused features; Based on the generative adversarial network model in the multimodal contrastive generative adversarial network model, the fused features are used for data discrimination to obtain the first probability that each of the perceived data is real data; and the data features of each of the perceived data are subjected to multimodal feature consistency comparison verification to obtain feature consistency score. Based on the first probability and the feature consistency score, a second probability is determined to characterize the presence of false data in each of the perceived data. Based on the second probability, it is determined whether the false data exists in each of the perceived data.

[0070] Specifically, the multimodal contrastive generative adversarial network model includes a multimodal feature fusion module, a generative adversarial network module, a multimodal cross-validation module, and an output layer.

[0071] Specifically, the multimodal feature fusion module fuses the feature vectors of different modalities, i.e., the data features of each sensing data, to form a comprehensive feature representation, i.e., the fused feature. This can be achieved by concatenating the data features of each sensing data, such as fused feature = [data feature 1, ..., data feature n], where n is the number of target sensing devices; by adding the data features of each sensing data, such as fused feature = data feature 1 + ... + data feature n; or by performing attention fusion (weighted summation) on the data features of each sensing data, such as fused feature = a1 * data feature 1 + ... + a n *Data feature n, where a1 to a n These are the weights corresponding to each data feature.

[0072] For example, consider an embodiment of dual-modal data that aggregates point cloud data and image data: (1) Splicing: ; (2) Addition: ; (3) Attention fusion: .

[0073] Where f is the fusion feature. For the data characteristics of point cloud data, For image data features, for The corresponding weights for The corresponding weights.

[0074] After obtaining the fusion features, the GAN (Generative Adversarial Network) model in the Generative Adversarial Network module can be used to perform data discrimination on the fusion features, thereby obtaining the first probability.

[0075] Specifically, the generative adversarial network model consists of two parts: a generator and a discriminator. The generator uses the DCGAN (Deep Convolution Generative Adversarial Networks) model, and the discriminator uses the CNN (Convolutional Neural Network) model.

[0076] Specifically, generative adversarial network models are trained adversarially using real data feature samples and fake data feature samples, enabling the generator to generate realistic fake data and the discriminator to accurately distinguish between real and fake data.

[0077] Specifically, the generator plays a role in the training process of the multimodal contrastive generative adversarial network model, and its processing is as follows: in, It is a sample of noise input and / or real data features. It is a generator. These are characteristics of generated fake data.

[0078] Specifically, the discriminator's processing procedure is as follows: in, It is a multimodal feature input, used in the training process of a multimodal contrastive generative adversarial network model. The use of fake data features generated by the generator in the application of multimodal contrastive generative adversarial network models Features of fusion; and These are the weights and biases of the discriminator, respectively. It is an activation function, such as the Sigmoid function; It is the output of the discriminator, which represents the probability that the data is real data.

[0079] Specifically, the multimodal cross-validation module includes a residual connection unit and a feature consistency verification unit: the residual connection unit performs residual connection on the data features of each sensing data and inputs them into the feature consistency verification unit, which processes the residual-connected data to obtain the feature consistency score.

[0080] Specifically, the output layer identifies fake data based on the first probability and feature consistency score, and obtains a second probability representing the presence of fake data in each perceived data.

[0081] Furthermore, if the second probability is greater than the probability threshold, it is determined that there is false data in each of the perceived data; if the second probability is less than or equal to the probability threshold, it is determined that there is no false data in each of the perceived data.

[0082] In this embodiment of the application, by performing multimodal feature fusion, data discrimination, consistency comparison verification, and fake data identification on the data features of the perceived data, the efficiency and accuracy of fake data identification can be improved.

[0083] Optionally, before fusing the data features of each of the perceived data to obtain the fused features, the method further includes: For each of the sensed data, feature processing is performed on the sensed data based on the feature extraction strategy corresponding to the data type of the sensed data to obtain the data features of the sensed data.

[0084] Specifically, the multimodal contrastive generative adversarial network model also includes a multimodal feature extraction module.

[0085] For multimodal data, which includes at least two of the various sensing data such as base station signals, radar signals, point cloud data, and visual (image / video) data, these sensing data come from different sensing devices and have different features and representations. Therefore, different feature extraction models can be used as feature extraction layers. Specific models include, but are not limited to, PointNet (a deep learning network for 3D point cloud data), ResNet50 (Residual Network 50 layers), Bi-LSTM (Bidirectional Long Short-Term Memory), etc., to extract feature vectors that can reflect the essential characteristics of the sensing data, i.e., data features.

[0086] In this embodiment of the application, a corresponding feature extraction strategy, i.e. a feature extraction model, can be selected based on the data type of the perceived data, which can improve the efficiency and accuracy of extracting data features.

[0087] Optionally, the feature extraction strategy based on the data type of the perceived data is used to perform feature processing on the perceived data to obtain the data features of the perceived data, including: When the data type is point cloud data, the sensing data is aligned, mapped, and aggregated to obtain the data features of the sensing data; When the data type is image data, residual learning and identity mapping are performed on the perceived data to obtain the data features of the perceived data.

[0088] Specifically, for point cloud data Point cloud data can be used In PointNet, PointNet first learns an affine transformation matrix to align all input point cloud data, then maps it to a new feature space. Simultaneously, to handle the unordered nature of the point cloud data, a symmetric function is used to aggregate the features of all point cloud data to generate global features, i.e., the data features corresponding to the point cloud data. .

[0089] Specifically, the data features corresponding to point cloud data The acquisition process is represented as follows: in, This is an optional transformation matrix used for alignment. It is a shared multilayer perceptron used for mapping processing; It is a max pooling operation used for aggregation processing.

[0090] Specifically, for image data Image data can be used The input is a ResNet, whose overall structure is composed of multiple stacked residual blocks. Assume the network has... If there are residual blocks, then the output of the entire network can be represented as: in, It is the first Input of each residual block; It is the first Learnable parameters for each residual block, used for residual learning; It is the first The residual function of each residual block is used for identity mapping.

[0091] Specifically, residual function The specific form depends on the design of the residual blocks. For example, the residual function of a two-layer residual block (a residual block containing two layers of residual blocks) is: in, For the residual functions of two layers of residual blocks, and These are the weights of the two convolutional layers (layer residual blocks); This is the input for two layers of residual blocks; BN represents batch normalization, and ReLU is the activation function.

[0092] In this embodiment of the application, by aligning, mapping and aggregating point cloud data, and performing residual learning and identity mapping on image data, the accuracy of data features can be further improved.

[0093] Optionally, the step of performing multimodal feature consistency comparison and verification on the data features of each of the perceived data to obtain a feature consistency score includes: For each of the sensed data, the data features of the sensed data are residually concatenated with the first data feature to obtain the residual corresponding to the sensed data. The first data feature is the data feature of other sensed data or the data feature generated by the generator. Multimodal feature consistency comparison verification is performed on the residuals corresponding to each of the aforementioned sensing data to obtain feature consistency scores.

[0094] Specifically, during model training, the residual connection unit performs a residual connection between the output of the multimodal feature extraction layer and the output of the generator. During model application, the residual connection unit performs a residual connection between the output of the multimodal feature extraction layer and the real data features (the data features of the perceived data) to compare the feature consistency between different modalities.

[0095] The process for handling residual connection units is as follows: in, It is the first output of the multimodal feature extraction layer. Data features of each modality; during model training. These are the data features generated by the generator, which are used in the model application process. For data features of another modality; It is the first Each modal residual; It is a small constant used to avoid gradient vanishing.

[0096] Specifically, the feature consistency verification unit uses a fully connected layer or attention mechanism to process the residuals and outputs a score representing feature consistency, namely the feature consistency score.

[0097] The processing procedure of the feature consistency verification unit is as follows: in, These are attention weights; FC stands for fully connected layer; and Attention is the attention mechanism. It is the feature consistency score.

[0098] In this embodiment, the fused feature (residual) representation is used for cross-validation. By comparing the consistency of information from different modalities, it is determined whether the data is fraudulent. If there are contradictions or inconsistencies in the information from different modalities, there is a possibility of fraudulent data injection attacks. This can improve the accuracy and reliability of feature consistency scores.

[0099] Optionally, determining a second probability representing the presence of false data in each of the perceived data based on the first probability and the feature consistency score includes: The first probability and the feature consistency score are integrated to obtain integrated data; The integrated data is subjected to a nonlinear transformation to obtain the second probability.

[0100] Specifically, the output layer fake data identification uses a fully connected layer and a sigmoid activation function. The probability that the output data is fake data is determined based on the discriminator's output and the feature consistency score, i.e., the second probability.

[0101] The processing procedure of the output layer is as follows: in, and These are the weights and biases of the output layer; It is the concatenation (integration) of the discriminator's output (first probability) and the feature consistency score; It is the probability that the data is false, i.e., the second probability; It is the Sigmoid activation function, used for nonlinear transformations.

[0102] In this embodiment of the application, by integrating and nonlinearly transforming the first probability and the feature consistency score, the efficiency and accuracy of obtaining the second probability can be improved.

[0103] Optionally, before performing feature processing on the perceived data based on the feature extraction strategy corresponding to the data type of the perceived data for each of the perceived data to obtain the data features of the perceived data, the method further includes: Each of the sensed data is preprocessed, including data cleaning and / or format conversion.

[0104] Specifically, the multimodal contrastive generative adversarial network model also includes a data preprocessing module. This module preprocesses the perceived data, including at least one operation such as data cleaning and format conversion, to obtain preprocessed perceived data. The preprocessed perceived data is then packaged and transmitted to the multimodal feature extraction module.

[0105] Data cleaning includes removing at least one of noisy data, outliers, and missing values; format conversion can convert the perceived data into a format suitable for processing by the multimodal contrastive generative adversarial network (MGA) model, such as vectors or matrices, to ensure that the MGA model can efficiently process the data and capture the most critical information.

[0106] Optionally, after performing fake data detection on the sensing data corresponding to each of the target sensing devices based on the multimodal contrastive generative adversarial network model, the method further includes: In the event of detected false data, the identification of the sensing device corresponding to the false data and the location identification of the false data in the sensing data are determined. Based on the sensing device identifier and the location identifier, the false data in each of the sensing data is filtered to obtain the real data; The real data are fused together to obtain a fused perception result, and the fused perception result is reported.

[0107] Specifically, the SF (SF-U) network element detects the existence of false data based on a multimodal contrastive generative adversarial network model. If false data is found, it outputs the sensing device ID (Identity) corresponding to the false data and the location identifier of the false data in the sensing data.

[0108] Furthermore, the SF-U network element reports the sensing device ID corresponding to the false data to the SF-C network element. There are two reporting scenarios: the first is periodic reporting, reporting the execution of sensing data policies and / or the detection of false data; the second is reporting only when false data is detected.

[0109] The SF-C network element sends a data filtering request to the SF-U network element. The SF-U network element then filters the sensing data based on the sensing device ID and location identifier corresponding to the detected false data, filtering out the false data. Then, it performs fusion and collaborative decision-making to obtain the final fusion sensing result, and reports the fusion sensing result to the AF network element.

[0110] This application embodiment designs an effective data filtering mechanism to ensure that false data does not enter the data fusion process, thereby guaranteeing the security and reliability of data fusion and improving data filtering capabilities.

[0111] Optionally, the method further includes: The PCF network element requests a perception data policy from the UDR network element. The perception data policy includes perception data rules, and the perception data rules include security levels. When the security level meets the conditions for false data detection, the step of performing false data detection on the sensing data corresponding to each of the target sensing devices is executed based on the multimodal contrastive generative adversarial network model.

[0112] Optionally, the step of requesting awareness data policies from the Unified Data Repository (UDR) network element through the Policy Control Function (PCF) network element includes: Send a sensing data policy request to the PCF network element; The sensing data policy request carries at least one of the following: information of the AF network element, sensing range, and sensing service type. The AF network element is the network element that initiates the sensing task request. The perception data policy request is used by the PCF network element to request the perception data policy from the UDR network element and to feed back the perception data policy to the perception function network element.

[0113] Specifically, upon receiving a sensing task request, the SF (SF-C) network element sends a sensing data policy request to the PCF (Policy Control Function) network element. The sensing data policy request carries at least one of the following information about the AF network element, sensing range, and sensing service type.

[0114] Furthermore, the PCF network element sends a subscription information request to the UDR (Unified Data Repository) network element. The subscription information request carries parameters including at least one of the following: information of the AF network element, sensing range, and sensing service type.

[0115] Then, the UDR network element retrieves the corresponding data contract information based on the parameters carried in the contract information request, and sends the retrieved data contract information to the PCF network element. The data contract information includes the perception data strategy.

[0116] It should be noted that when the AF network element is activated for sensing services, a fine-grained sensing data strategy with classification and gradation is configured in advance according to the contract. This sensing data strategy can be stored in the UDR network element.

[0117] PCF network elements extract perception data policies from data contract information and send complete perception data policies to SF (SF-C) network elements.

[0118] Specifically, a sensing data policy can include one or more SDRs (Sensing Data Rules). A sensing data rule consists of a Rule ID and Rule content, which includes the security level.

[0119] The SF-C network element sends a perception request to the SF-U network element, carrying the perception data policy; the SF-U network element receives the perception request and the perception data policy, completes the configuration and activation of the perception data policy, triggers the start of the perception data false data detection operation, and replies to the SF-C network element to receive the perception request.

[0120] Furthermore, the SF-U network element aggregates the sensing data uploaded by various target sensing devices, checks the security level in the sensing data rules, and generates a security level identifier.

[0121] Before an SF-C network element sends a perception request to an SF-U network element, the SF-C network element can select an SF-U network element based on the perception parameters and / or perception data strategy carried in the perception task request, and then send a perception request to the selected SF-U network element, carrying the perception data strategy.

[0122] Specifically, the conditions for detecting fake data are that the security level is marked as medium or high, and / or the security level is higher than the level threshold.

[0123] When the security level is identified as medium or high, and / or when the security level is higher than the level threshold, indicating that the perceived data contains more sensitive and critical information or is extremely important for the perception decision-making results, the SF-U network element performs false data detection on the perceived data. If the security level does not meet the conditions for false data detection, then it is not necessary to perform false data detection on the perceived data.

[0124] In this embodiment of the application, by checking the security level, detecting false data for sensitive data, and not executing false data for non-sensitive data, the amount of data processing can be reduced and resource consumption can be avoided.

[0125] Optionally, the perceived data rules also include a threshold for the number of false data reports; The step of requesting sensing data from at least two target sensing devices includes: Based on the threshold for the number of false data reports, device detection is performed on each of the target sensing devices. Request sensing data from the target sensing device that has passed the device detection.

[0126] Specifically, the rule may also include a threshold for the number of times false data is reported.

[0127] Specifically, before requesting sensing data from target sensing devices, SF (SF-C) network elements need to determine whether each target sensing device is an abnormal device based on the threshold for the number of false data reports. For abnormal devices, i.e., target sensing devices that have failed the device detection, sensing data will not be requested from them. For normal devices, i.e., target sensing devices that have passed the device detection, sensing data can be requested from them.

[0128] In this embodiment, by using a threshold for the number of false data reports to detect the target sensing device, abnormal devices can be identified quickly and easily. This avoids requesting sensing data from abnormal devices, further preventing abnormal devices from providing false data, reducing the amount of data for false data detection, improving detection efficiency, and also improving the accuracy of sensing results.

[0129] Optionally, the perception data rules may also include at least one of QoS (Quality of Service) parameters, perception mode, and perception period.

[0130] Optionally, the threshold for the number of false data reports includes a first threshold at the global granularity; The step of performing device detection on each target sensing device based on the threshold for the number of false data reports includes: For each target sensing device, the global false identifier of the target sensing device is compared with the first number threshold, wherein the global false identifier represents the number of times the target sensing device stops providing sensing data; If the number of global false identifiers exceeds the first threshold, it is determined that the target sensing device has failed device detection. If the number of false global identifiers is less than or equal to the first threshold number, the target sensing device is determined to have passed device detection.

[0131] Specifically, the number of times the target sensing device was stopped from providing sensing data, i.e., the number of times the sensing operation was stopped, can be compared with the first threshold to identify whether the target sensing device is an abnormal device: if the Globalfaketag is greater than the first threshold, the target sensing device is an abnormal device, i.e., the device detection failed; if the Globalfaketag is less than or equal to the first threshold, the target sensing device is a normal device, i.e., the device detection passed.

[0132] Specifically, the first threshold (the threshold for Globalfaketag) is at the global granularity of awareness. For the global granularity: the Globalfaketag count and the threshold for Globalfaketag, the threshold can be statically configured or dynamically distributed through the awareness data strategy.

[0133] For example, before issuing a sensing operation request, the SF-C network element checks whether the Globalfaketag of the selected sensing device (target sensing device) is greater than 5. For target sensing devices with a Globalfaketag greater than 5, the SF-C network element will not issue a sensing operation request again; for those with a Globalfaketag less than or equal to 5, the SF-C network element will not continue to issue a sensing operation request. The threshold for the first count is 5.

[0134] Optionally, if the target sensing device is performing a sensing task for the first time, the default value of Globalfaketag is 0.

[0135] Optionally, before the first threshold is reached, the global false identifier and / or its corresponding global counter can be reset to 0 at certain time intervals.

[0136] In this embodiment of the application, abnormal devices are detected at a global granularity, that is, at the granularity of single device multi-tasking, which can determine the accuracy and reliability of the detection.

[0137] Optionally, the threshold for the number of false data reports includes a second threshold at the task granularity. The multimodal contrastive generative adversarial network model, after detecting false data in the sensing data corresponding to each target sensing device, further includes: For each target sensing device, if false data exists in the sensing data corresponding to the target sensing device, the task false identifier of the target sensing device is updated. The task false identifier represents the number of times the target sensing device provides false data under the current sensing task. If the number of false task identifiers reaches the second threshold, the request for sensing data from the target sensing device corresponding to the false task identifier will be stopped.

[0138] Specifically, for each target sensing device, when the SF (SF-U) network element detects spoofed data in the sensing data provided by that device, the SF (SF-C) network element increments the number of times the target sensing device has provided spoofed data during the current sensing task by one, i.e., increments the task spoofing flag LocalFaketag by one. If the LocalFaketag reaches the second threshold, it indicates that the target sensing device is behaving abnormally in the current sensing task, and it is necessary to stop the target sensing device from providing further sensing data to prevent spoofed data in the sensing results.

[0139] Specifically, the second threshold (the threshold for Localfaketag) is at the perception task granularity (single device, single task). For task granularity: the Localfaketag count and the Localfaketag threshold, the threshold can be statically configured or dynamically distributed through the perception data strategy.

[0140] For example, each time an SF-U network element reports a sensing device ID containing false data, the LocalFaketag count corresponding to that sensing device ID in SF-C is incremented by one. Once it exceeds 3, a stop-reporting request is sent to the corresponding sensing device, and the GlobalFaketag count is incremented by one. The threshold for LocalFaketag is 3.

[0141] Optionally, if the target sensing device is performing a sensing task for the first time in this sensing request, the default value of LocalFaketag is 0.

[0142] Optionally, before the second threshold is reached, the task false identifier and / or its corresponding counter can be reset to 0 at certain time intervals.

[0143] In this embodiment of the application, abnormal devices in the current sensing task are detected at the task granularity, that is, at the granularity of a single device and a single task, so as to determine the accuracy and reliability of the detection.

[0144] Optionally, the method further includes: If the request for sensing data from the target sensing device corresponding to the task false identifier is stopped, the global false identifier of the target sensing device corresponding to the task false identifier is updated.

[0145] Specifically, when the LocalFaketag reaches the second threshold, it indicates that the target sensing device is performing abnormally in the current sensing task, and it is necessary to stop the target sensing device from providing sensing data. At this time, the GlobalFaketag of the target sensing device is incremented by one. In this way, the accuracy of the global fake tag can be guaranteed.

[0146] For example, in SF-C, the LocalFaketag count corresponding to the sensing device ID is incremented by one. Once it exceeds 3, a stop reporting request is sent to the corresponding sensing device, and the GlobalFaketag count is incremented by one. The threshold for LocalFaketag is 3.

[0147] The fake data detection method provided in this application first configures a fine-grained, classification-based sensing data strategy, distributing relevant settings for fake data detection to the SF-C and SF-U. This sensing data strategy includes identifying data types and thresholds for the number of fake data reports. The rules in the sensing data strategy include a Security Service Data Flow (SDR) template for data detection. The SF-C selects a sensing device based on the sensing request, sets a global fake data report counter (Globalfaketag) for the device, and sends a sensing operation request to the sensing device. After confirming the request, the device begins the sensing operation and establishes a data transmission channel with the sensing service user plane (SF-U) to report data. The SF-U aggregates the data, checks the security level, and performs a multimodal contrastive generative adversarial network algorithm to detect fake data on sensitive data. If false data is detected, the sensing device ID is reported. Simultaneously, the SF-C activates a task-granular false data reporting counter (LocalFaketag) and increments the counter. If the LocalFaketag threshold is exceeded, reporting by that device for the current sensing task stops, and its GlobalFaketag count is incremented. If the GlobalFaketag threshold is exceeded, all data reporting by that sensing device is stopped. The SF-C then sends a data filtering request to the SF-U. The SF-U filters the data based on the false data information, performs fusion and collaborative decision-making, and finally obtains the fused sensing result, which is reported to the AF. This effectively detects and filters false sensing data, ensuring the accuracy and reliability of the sensing data.

[0148] The following describes the false data detection method for SF-C and SF-U network elements provided in this application. The false data detection method for SF-C and SF-U network elements described below can be referred to in correspondence with the false data detection method for sensing function network elements described above.

[0149] Figure 2 This is the second flowchart of the fake data detection method provided in this application, such as... Figure 2 As shown, when applied to SF-C network elements, this method includes the following: Step 201: In response to the received sensing task request, determine the target SF-U network element and request sensing data from at least two target sensing devices; wherein, the target SF-U network element is used to receive and perform false data detection on the sensing data corresponding to each target sensing device based on a multimodal contrastive generative adversarial network model.

[0150] Specifically, the AF network element, which is the perception requesting party, such as the network element / UE / third-party service provider, initiates a perception service request and sends it to the NEF network element.

[0151] Optionally, the parameters carried in the perception service request include: afTaskId and perception data type.

[0152] Optionally, the sensing service request may also carry at least one of sensing range, sensing accuracy, and sensing data reporting frequency.

[0153] Specifically, when a NEF network element receives a sensing task request, it selects an SF-C network element based on the service area corresponding to the sensing task request (sensing range) and sends the sensing task request to the SF-C network element.

[0154] The SF-C network element can select the SF-U network element and send a sensing operation request to the target sensing device. The number of target sensing devices can be multiple, including but not limited to base stations, radar, and non-3GPP devices (such as cameras).

[0155] Optionally, the sensing operation request may carry sensing operation parameters, such as at least one of afTaskId, sensing range, sensing accuracy, sensing data type, and sensing data reporting frequency.

[0156] The target sensing device acknowledges receipt of the sensing operation request and begins executing the sensing operation. Before performing the sensing operation, the target sensing device has requested the UDM (Unified Data Management) network element to check user licenses and authorization information. Each target sensing device establishes a data transmission channel with the SF-U network element to report sensing data.

[0157] Furthermore, the sensing data uploaded by multiple sources and multiple devices (each target sensing device) of the SF-U network element are used to detect whether there is false data in the sensing data uploaded by each target sensing device through a multimodal comparative generative adversarial network model.

[0158] In addition, upon receiving a sensing task request, the SF-C network element can also send a request to initiate a sensing task and receive a response to the AF network element through the NEF network element. That is, the SF-C network element returns a request to initiate a sensing task and receives a response to the NEF network element, and the NEF network element returns a request to initiate a sensing task and receives a response to the AF network element.

[0159] The fake data detection method provided in this application introduces a multimodal contrastive generative adversarial network algorithm, which can accurately detect fake data in perceived data, thereby improving detection efficiency and accuracy.

[0160] Optionally, before requesting sensing data from at least two target sensing devices, the method further includes: Based on the sensing parameters carried in the sensing task request, select at least two target sensing devices.

[0161] Specifically, the SF-C network element selects the target sensing device to perform the sensing operation based on the sensing requirements (parameters) carried in the sensing request, and sends the sensing operation request to the target sensing device.

[0162] In this embodiment of the application, by selecting the target sensing device through sensing parameters, it can be ensured that the target sensing device can improve the sensing data that meets the sensing requirements, thereby improving the reliability and accuracy of the sensing data to a certain extent.

[0163] Optionally, before determining the target SF-U network element, the method further includes: The PCF network element requests the perception data policy from the UDR network element. The determination of the target SF-U network element includes: The target SF-U network element is selected based on the sensing parameters carried in the sensing task request and / or the sensing data strategy.

[0164] Specifically, the SF-C network element sends a sensing data policy request to the PCF network element; wherein, the sensing data policy request carries at least one of the following: information of the AF network element, sensing range, and sensing service type, and the AF network element is the network element that initiated the sensing task request; the sensing data policy request is used by the PCF network element to request the sensing data policy from the UDR network element and to feed back the sensing data policy to the SF-C network element.

[0165] Furthermore, SF-C network elements can select SF-U network elements based on the sensing parameters and / or sensing data strategies carried in the sensing task request.

[0166] In this embodiment, the target SF-U network element is selected by sensing parameters and / or sensing data strategy, so that the target SF-U network element can better perform sensing tasks, that is, improve the efficiency of false data detection.

[0167] Optionally, the method further includes: Send the sensing task request and the sensing data policy to the target SF-U network element; The perception task request and the perception data policy are used by the target SF-U network element to initiate a false data detection operation after configuring and activating the perception data policy.

[0168] Optionally, the SF-C network element sends a sensing request to the SF-U network element, carrying the sensing data policy. The SF-U network element receives the sensing request and the sensing data policy, completes the configuration and activation of the sensing data policy, triggers the start of the false data detection operation, and replies to the SF-C network element that it has received the sensing request. This ensures the accuracy of false data detection.

[0169] Optionally, the sensing data strategy includes sensing data rules, and the sensing data rules include a threshold for the number of false data reports; The step of requesting sensing data from at least two target sensing devices includes: Based on the threshold for the number of false data reports, device detection is performed on each of the target sensing devices. Request sensing data from the target sensing device that has passed the device detection.

[0170] Specifically, the perceived data strategy can include one or more SDRs. The perceived data rule consists of a Rule ID and Rule content, and the Rule content includes a threshold for the number of false data reports.

[0171] Specifically, before requesting sensing data from target sensing devices, SF-C network elements need to determine whether each target sensing device is an abnormal device based on the threshold for the number of false data reports. For abnormal devices, i.e., target sensing devices that have failed the device detection, sensing data will not be requested from them. For normal devices, i.e., target sensing devices that have passed the device detection, sensing data can be requested from them.

[0172] In this embodiment, by using a threshold for the number of false data reports to detect the target sensing device, abnormal devices can be identified quickly and easily. This avoids requesting sensing data from abnormal devices, further preventing abnormal devices from providing false data, reducing the amount of data for false data detection, improving detection efficiency, and also improving the accuracy of sensing results.

[0173] Optionally, the sensing data rules may also include at least one of security level, QoS parameters, sensing mode, and sensing period.

[0174] Optionally, the threshold for the number of false data reports includes a first threshold at the global granularity; The step of performing device detection on each target sensing device based on the threshold for the number of false data reports includes: For each target sensing device, the global false identifier of the target sensing device is compared with the first number threshold, wherein the global false identifier represents the number of times the target sensing device stops providing sensing data; If the number of global false identifiers exceeds the first threshold, it is determined that the target sensing device has failed device detection. If the number of false global identifiers is less than or equal to the first threshold number, the target sensing device is determined to have passed device detection.

[0175] Specifically, the SF-C network element can compare the number of times the target sensing device has stopped providing sensing data (i.e., the number of times sensing operations have stopped) with the first threshold to identify whether the target sensing device is an abnormal device: if the global false identifier is greater than the first threshold, the target sensing device is an abnormal device, i.e., the device detection has failed; if the global false identifier is less than or equal to the first threshold, the target sensing device is a normal device, i.e., the device detection has passed.

[0176] Specifically, the first threshold (the threshold for Globalfaketag) is at the global granularity of awareness. For the global granularity: the Globalfaketag count and the threshold for Globalfaketag, the threshold can be statically configured or dynamically distributed through the awareness data strategy.

[0177] Optionally, if the target sensing device is performing a sensing task for the first time, the default value of Globalfaketag is 0.

[0178] Optionally, before the first threshold is reached, the global false identifier and / or its corresponding global counter can be reset to 0 at certain time intervals.

[0179] In this embodiment of the application, abnormal devices are detected at a global granularity, that is, at the granularity of single device multi-tasking, which can determine the accuracy and reliability of the detection.

[0180] Optionally, the threshold for the number of false data reports includes a second threshold at the task granularity. The method further includes: The system receives the identification of the sensing device corresponding to the false data reported by the target SF-U network element, wherein the false data is determined by the target SF-U network element through false data detection of the sensing data corresponding to each target sensing device; Update the task false identifier of the target sensing device corresponding to the sensing device identifier, wherein the task false identifier represents the number of times the target sensing device provides false data under the current sensing task; If the number of false task identifiers reaches the second threshold, the request for sensing data from the target sensing device corresponding to the false task identifier will be stopped.

[0181] Specifically, for each target sensing device, when the SF-U network element detects that there is false data in the sensing data provided by the target sensing device, the SF-U network element will report the ID of the sensing device with false data.

[0182] Specifically, the SF-U network element reports the sensing device ID corresponding to false data to the SF-C network element. There are two reporting scenarios: the first is periodic reporting, reporting the execution of sensing data policies and / or the detection of false data; the second is reporting only when false data is detected.

[0183] Specifically, each time an SF-U network element reports a sensing device ID containing false data, the SF-C network element increments the number of times that target sensing device has provided false data during the current sensing task by one, i.e., increments the task false tag LocalFaketag by one. If the LocalFaketag reaches the second threshold, it indicates that the target sensing device is behaving abnormally in the current sensing task, and the SF-C network element needs to stop that target sensing device from providing further sensing data to prevent false data from appearing in the sensing results.

[0184] Optionally, if the target sensing device is performing a sensing task for the first time in this sensing request, the default value of LocalFaketag is 0.

[0185] Optionally, before the second threshold is reached, the task false identifier and / or its corresponding counter can be reset to 0 at certain time intervals.

[0186] In this embodiment of the application, abnormal devices in the current sensing task are detected at the task granularity, that is, at the granularity of a single device and a single task, so as to determine the accuracy and reliability of the detection.

[0187] Optionally, the method further includes: If the request for sensing data from the target sensing device corresponding to the task false identifier is stopped, the global false identifier of the target sensing device corresponding to the task false identifier is updated.

[0188] Specifically, when the LocalFaketag reaches the second threshold, it indicates that the target sensing device is performing abnormally in the current sensing task, and it is necessary to stop the target sensing device from providing sensing data. At this time, the GlobalFaketag of the target sensing device is incremented by one. In this way, the accuracy of the global fake tag can be guaranteed.

[0189] Optionally, after receiving the sensing device identifier corresponding to the false data reported by the target SF-U network element, the method further includes: Send a data filtering request to the target SF-U network element; The data filtering request is used by the target SF-U network element to filter and fuse the false data in each of the sensing data to obtain the fused sensing result, and then report the fused sensing result.

[0190] Specifically, after receiving the sensing device ID corresponding to the false data reported by the SF-U network element, the SF-C network element sends a data filtering request to the SF-U network element. Then, the SF-U network element filters the sensing data based on the sensing device ID and location identifier corresponding to the detected false data, filters out the false data, and then performs fusion and collaborative decision-making to obtain the final fusion sensing result, and reports the fusion sensing result to the AF network element.

[0191] This application embodiment designs an effective data filtering mechanism to ensure that false data does not enter the data fusion process, thereby guaranteeing the security and reliability of data fusion and improving data filtering capabilities.

[0192] Figure 3 This is the third flowchart of the fake data detection method provided in this application, such as... Figure 3 As shown, when applied to SF-U network elements, this method includes the following: Step 301: Receive sensing data from at least two target sensing devices; Step 302: Based on the multimodal contrastive generative adversarial network model, perform fake data detection on the sensing data corresponding to each target sensing device.

[0193] Specifically, the AF network element, which is the perception requesting party, such as the network element / UE / third-party service provider, initiates a perception service request and sends it to the NEF network element.

[0194] Optionally, the parameters carried in the perception service request include: afTaskId and perception data type.

[0195] Optionally, the sensing service request may also carry at least one of sensing range, sensing accuracy, and sensing data reporting frequency.

[0196] Specifically, when a NEF network element receives a sensing task request, it selects an SF-C network element based on the service area corresponding to the sensing task request (sensing range) and sends the sensing task request to the SF-C network element.

[0197] The SF-C network element sends a sensing data policy request to the PCF network element. It can select an SF-U network element based on the sensing parameters and / or sensing data policy carried in the sensing task request, and send the sensing task request and the sensing data policy to the target SF-U network element. The sensing task request and the sensing data policy are used by the target SF-U network element to start a false data detection operation after configuring and activating the sensing data policy.

[0198] The SF-C network element also sends sensing operation requests to target sensing devices. These target sensing devices can be multiple, including but not limited to base stations, radar, and non-3GPP devices (such as cameras).

[0199] Optionally, the sensing operation request may carry sensing operation parameters, such as at least one of afTaskId, sensing range, sensing accuracy, sensing data type, and sensing data reporting frequency.

[0200] The target sensing device acknowledges receipt of the sensing operation request and begins executing the sensing operation. Before performing the sensing operation, the target sensing device has requested the UDM (Unified Data Management) network element to check user licenses and authorization information. Each target sensing device establishes a data transmission channel with the SF-U network element to report sensing data.

[0201] Furthermore, the sensing data uploaded by multiple sources and multiple devices (each target sensing device) of the SF-U network element are used to detect whether there is false data in the sensing data uploaded by each target sensing device through a multimodal comparative generative adversarial network model.

[0202] In addition, upon receiving a sensing task request, the SF-C network element can also send a request to initiate a sensing task and receive a response to the AF network element through the NEF network element. That is, the SF-C network element returns a request to initiate a sensing task and receives a response to the NEF network element, and the NEF network element returns a request to initiate a sensing task and receives a response to the AF network element.

[0203] The fake data detection method provided in this application introduces a multimodal contrastive generative adversarial network algorithm, which can accurately detect fake data in perceived data, thereby improving detection efficiency and accuracy.

[0204] Optionally, the step of detecting fake data in the sensing data corresponding to each of the target sensing devices based on the multimodal contrastive generative adversarial network model includes: The data features of each of the aforementioned sensing data are fused using multimodal features to obtain fused features; Based on the generative adversarial network model in the multimodal contrastive generative adversarial network model, the fused features are used for data discrimination to obtain the first probability that each of the perceived data is real data; and the data features of each of the perceived data are subjected to multimodal feature consistency comparison verification to obtain feature consistency score. Based on the first probability and the feature consistency score, a second probability is determined to characterize the presence of false data in each of the perceived data. Based on the second probability, it is determined whether the false data exists in each of the perceived data.

[0205] Specifically, the multimodal contrastive generative adversarial network model includes a multimodal feature fusion module, a generative adversarial network module, a multimodal cross-validation module, and an output layer.

[0206] Specifically, the multimodal feature fusion module fuses the feature vectors of different modalities, i.e., the data features of each sensing data, to form a comprehensive feature representation, i.e., the fused feature. This can be achieved by concatenating the data features of each sensing data, such as fused feature = [data feature 1, ..., data feature n], where n is the number of target sensing devices; by adding the data features of each sensing data, such as fused feature = data feature 1 + ... + data feature n; or by performing attention fusion (weighted summation) on the data features of each sensing data, such as fused feature = a1 * data feature 1 + ... + a n *Data feature n, where a1 to a n These are the weights corresponding to each data feature.

[0207] After obtaining the fusion features, the GAN (Generative Adversarial Network) model in the Generative Adversarial Network module can be used to perform data discrimination on the fusion features, thereby obtaining the first probability.

[0208] Specifically, the generative adversarial network model consists of two parts: a generator and a discriminator. The generator uses the DCGAN (Deep Convolution Generative Adversarial Networks) model, and the discriminator uses the CNN (Convolutional Neural Network) model.

[0209] Specifically, generative adversarial network models are trained adversarially using real data feature samples and fake data feature samples, enabling the generator to generate realistic fake data and the discriminator to accurately distinguish between real and fake data.

[0210] Specifically, the generator plays a role in the training process of the multimodal contrastive generative adversarial network model, and its processing is as follows: in, It is a sample of noise input and / or real data features. It is a generator. These are characteristics of generated fake data.

[0211] Specifically, the discriminator's processing procedure is as follows: in, It is a multimodal feature input, used in the training process of a multimodal contrastive generative adversarial network model. The use of fake data features generated by the generator in the application of multimodal contrastive generative adversarial network models Features of fusion; and These are the weights and biases of the discriminator, respectively. It is an activation function, such as the Sigmoid function; It is the output of the discriminator, which represents the probability that the data is real data.

[0212] Specifically, the multimodal cross-validation module includes a residual connection unit and a feature consistency verification unit: the residual connection unit performs residual connection on the data features of each sensing data and inputs them into the feature consistency verification unit, which processes the residual-connected data to obtain the feature consistency score.

[0213] Specifically, the output layer identifies fake data based on the first probability and feature consistency score, and obtains a second probability representing the presence of fake data in each perceived data.

[0214] Furthermore, if the second probability is greater than the probability threshold, it is determined that there is false data in each of the perceived data; if the second probability is less than or equal to the probability threshold, it is determined that there is no false data in each of the perceived data.

[0215] In this embodiment of the application, by performing multimodal feature fusion, data discrimination, consistency comparison verification, and fake data identification on the data features of the perceived data, the efficiency and accuracy of fake data identification can be improved.

[0216] Optionally, before fusing the data features of each of the perceived data to obtain the fused features, the method further includes: For each of the sensed data, feature processing is performed on the sensed data based on the feature extraction strategy corresponding to the data type of the sensed data to obtain the data features of the sensed data.

[0217] Specifically, the multimodal contrastive generative adversarial network model also includes a multimodal feature extraction module.

[0218] For multimodal data, which includes at least two of the various sensing data such as base station signals, radar signals, point cloud data, and visual (image / video) data, these sensing data come from different sensing devices and have different features and representations. Therefore, different feature extraction models can be used as feature extraction layers. Specific models include, but are not limited to, PointNet (a deep learning network for 3D point cloud data), ResNet50 (Residual Network 50 layers), Bi-LSTM (Bidirectional Long Short-Term Memory), etc., to extract feature vectors that can reflect the essential characteristics of the sensing data, i.e., data features.

[0219] In this embodiment of the application, a corresponding feature extraction strategy, i.e. a feature extraction model, can be selected based on the data type of the perceived data, which can improve the efficiency and accuracy of extracting data features.

[0220] Optionally, the feature extraction strategy based on the data type of the perceived data is used to perform feature processing on the perceived data to obtain the data features of the perceived data, including: When the data type is point cloud data, the sensing data is aligned, mapped, and aggregated to obtain the data features of the sensing data; When the data type is image data, residual learning and identity mapping are performed on the perceived data to obtain the data features of the perceived data.

[0221] Specifically, for point cloud data Point cloud data can be used In PointNet, PointNet first learns an affine transformation matrix to align all input point cloud data, then maps it to a new feature space. Simultaneously, to handle the unordered nature of the point cloud data, a symmetric function is used to aggregate the features of all point cloud data to generate global features, i.e., the data features corresponding to the point cloud data. .

[0222] Specifically, the data features corresponding to point cloud data The acquisition process is represented as follows: in, This is an optional transformation matrix used for alignment. It is a shared multilayer perceptron used for mapping processing; It is a max pooling operation used for aggregation processing.

[0223] Specifically, for image data Image data can be used The input is a ResNet, whose overall structure is composed of multiple stacked residual blocks. Assume the network has... If there are residual blocks, then the output of the entire network can be represented as: in, It is the first Input of each residual block; It is the first Learnable parameters for each residual block, used for residual learning; It is the first The residual function of each residual block is used for identity mapping.

[0224] Specifically, residual function The specific form depends on the design of the residual blocks. For example, the residual function of a two-layer residual block (a residual block containing two layers of residual blocks) is: in, For the residual functions of two layers of residual blocks, and These are the weights of the two convolutional layers (layer residual blocks); This is the input for two layers of residual blocks; BN represents batch normalization, and ReLU is the activation function.

[0225] In this embodiment of the application, by aligning, mapping and aggregating point cloud data, and performing residual learning and identity mapping on image data, the accuracy of data features can be further improved.

[0226] Optionally, the step of performing multimodal feature consistency comparison and verification on the data features of each of the perceived data to obtain a feature consistency score includes: For each of the sensed data, the data features of the sensed data are residually concatenated with the first data feature to obtain the residual corresponding to the sensed data. The first data feature is the data feature of other sensed data or the data feature generated by the generator. Multimodal feature consistency comparison verification is performed on the residuals corresponding to each of the aforementioned sensing data to obtain feature consistency scores.

[0227] Specifically, during model training, the residual connection unit performs a residual connection between the output of the multimodal feature extraction layer and the output of the generator. During model application, the residual connection unit performs a residual connection between the output of the multimodal feature extraction layer and the real data features (the data features of the perceived data) to compare the feature consistency between different modalities.

[0228] The process for handling residual connection units is as follows: in, It is the first output of the multimodal feature extraction layer. Data features of each modality; during model training. These are the data features generated by the generator, which are used in the model application process. For data features of another modality; It is the first Each modal residual; It is a small constant used to avoid gradient vanishing.

[0229] Specifically, the feature consistency verification unit uses a fully connected layer or attention mechanism to process the residuals and outputs a score representing feature consistency, namely the feature consistency score.

[0230] The processing procedure of the feature consistency verification unit is as follows: In this context, FC stands for fully connected layer, and Attention is the attention mechanism. It is the feature consistency score.

[0231] In this embodiment, the fused feature (residual) representation is used for cross-validation. By comparing the consistency of information from different modalities, it is determined whether the data is fraudulent. If there are contradictions or inconsistencies in the information from different modalities, there is a possibility of fraudulent data injection attacks. This can improve the accuracy and reliability of feature consistency scores.

[0232] Optionally, determining a second probability representing the presence of false data in each of the perceived data based on the first probability and the feature consistency score includes: The first probability and the feature consistency score are integrated to obtain integrated data; The integrated data is subjected to a nonlinear transformation to obtain the second probability.

[0233] Specifically, the output layer fake data identification uses a fully connected layer and a sigmoid activation function. The probability that the output data is fake data is determined based on the discriminator's output and the feature consistency score, i.e., the second probability.

[0234] The processing procedure of the output layer is as follows: in, and These are the weights and biases of the output layer; It is the concatenation (integration) of the discriminator's output (first probability) and the feature consistency score; It is the probability that the data is false, i.e., the second probability; It is the Sigmoid activation function, used for nonlinear transformations.

[0235] In this embodiment of the application, by integrating and nonlinearly transforming the first probability and the feature consistency score, the efficiency and accuracy of obtaining the second probability can be improved.

[0236] Optionally, before performing feature processing on the perceived data based on the feature extraction strategy corresponding to the data type of the perceived data for each of the perceived data to obtain the data features of the perceived data, the method further includes: Each of the sensed data is preprocessed, including data cleaning and / or format conversion.

[0237] Specifically, the multimodal contrastive generative adversarial network model also includes a data preprocessing module. This module preprocesses the perceived data, including at least one operation such as data cleaning and format conversion, to obtain preprocessed perceived data. The preprocessed perceived data is then packaged and transmitted to the multimodal feature extraction module.

[0238] Data cleaning includes removing at least one of noisy data, outliers, and missing values; format conversion can convert the perceived data into a format suitable for processing by the multimodal contrastive generative adversarial network (MGA) model, such as vectors or matrices, to ensure that the MGA model can efficiently process the data and capture the most critical information.

[0239] Optionally, after performing fake data detection on the sensing data corresponding to each of the target sensing devices based on the multimodal contrastive generative adversarial network model, the method further includes: In the event of detected false data, the identification of the sensing device corresponding to the false data and the location identification of the false data in the sensing data are determined. Based on the sensing device identifier and the location identifier, the false data in each of the sensing data is filtered to obtain the real data; The real data are fused together to obtain a fused perception result, and the fused perception result is reported.

[0240] Specifically, the SF-U network element detects the existence of false data based on a multimodal contrastive generative adversarial network model. If false data is found, it outputs the sensing device ID corresponding to the false data and the location identifier of the false data in the sensing data.

[0241] Furthermore, the SF-U network element reports the sensing device ID corresponding to the false data to the SF-C network element. There are two reporting scenarios: the first is periodic reporting, reporting the execution of sensing data policies and / or the detection of false data; the second is reporting only when false data is detected.

[0242] The SF-C network element sends a data filtering request to the SF-U network element. The SF-U network element then filters the sensing data based on the sensing device ID and location identifier corresponding to the detected false data, filtering out the false data. Then, it performs fusion and collaborative decision-making to obtain the final fusion sensing result, and reports the fusion sensing result to the AF network element.

[0243] This application embodiment designs an effective data filtering mechanism to ensure that false data does not enter the data fusion process, thereby guaranteeing the security and reliability of data fusion and improving data filtering capabilities.

[0244] The following is combined Figure 4 This application further explains the methods for detecting false data. See [link to relevant documentation]. Figure 4 , Figure 4 This is the fourth flowchart of the false data detection method provided in this application, which includes the following steps.

[0245] Step 0: When AF activates the sensing service, a fine-grained sensing data strategy based on the contract is configured in advance. This strategy can be stored in UDR, that is, the fine-grained sensing data strategy is configured when the service is activated.

[0246] Step 1: The network element / UE / third-party service provider, i.e., AF, initiates a perception task request. The perception parameters carried in the perception task request include at least one of the following: afTaskId, perception range, perception accuracy, perception data type, and perception data reporting frequency.

[0247] Step 2: NEF receives the request to initiate a perception task, selects SF-C based on the service area corresponding to the perception range, and sends the request to initiate a perception task to SF-C.

[0248] Step 3: SF-C returns a response to NEF to initiate the sensing task request, that is, NEF receives the sensing task request response.

[0249] Step 4: NEF returns a response to AF to initiate the sensing task request, that is, AF receives the sensing task request response.

[0250] Step 5: SF-C sends a sensing data policy request to PCF, which carries the following AF information and / or sensing range and / or sensing service type.

[0251] Step 6a: The PCF requests the UDR for the contract information. The UDR retrieves the corresponding data contract information based on the AF information and / or the sensing range and / or the sensing service type. The data contract information includes the sensing data policy.

[0252] Step 6b: UDR sends data contract information to PCF, i.e., responds with contract information.

[0253] Step 7: The PCF sends the perception data policy to the SF-C. At this time, it can carry one or more perception data policies. The perception data policy can contain one or more SDRs. The SDR consists of a Rule ID and Rule content. The Rule content includes at least one of the following: QoS parameters, perception mode, perception period, security level related parameters, and threshold for the number of false data reports.

[0254] Steps 5-7 can be triggered when a service is perceived to change or be cancelled. For example, when AF initiates a service perception change / cancellation request, it will trigger a change / cancellation of the perception data policy.

[0255] It should be noted that there are two counters and two corresponding thresholds for detecting false perception data: one for the perception task granularity (single device) and the other for the global perception granularity (single device, multi-task). Details are as follows: Global granularity: Globalfaketag count and Globalfaketag threshold. The threshold can be configured statically or dynamically distributed through the awareness data strategy.

[0256] Task granularity: Localfaketag count and Localfaketag threshold. The threshold can be statically configured or dynamically issued through the data awareness strategy.

[0257] Step 8: SF-C selects SF-U based on the sensing parameters and / or sensing data policy carried in the sensing request, and sends a sensing request to SF-U, carrying the sensing data policy. In other words, SF-C selects SF-U and sends a sensing task request and sensing data policy. SF-U receives the sensing request and sensing data policy, completes the configuration and activation of the sensing data policy, triggers the start of the false data detection operation, and replies to SF-C that it has received the sensing request.

[0258] Step 9: SF-C selects sensing devices to perform sensing operations based on the sensing requirements carried in the sensing request, and sends sensing operation requests to multiple sensing devices (including but not limited to base stations, radar, and non-3GPP devices such as cameras). Before sending the request, it checks whether the Globalfaketag of the selected sensing device is greater than 5. If the value is greater than 5, the request will not be sent again; if it is less than 5, the request will continue to be sent. The sensing operation requirements carried in the sensing operation request include at least one of the following: afTaskId, sensing range, sensing accuracy, sensing data type, and sensing data reporting frequency.

[0259] In this scenario, the threshold for obtaining the Globalfaketag is assumed to be set to 5. If the sensing device is performing a sensing task for the first time, the default value for the Globalfaketag is 0. Before reaching the Globalfaketag threshold, this Globalfaketag counter can be reset to 0 at regular intervals. Step 10: The sensing device confirms receipt of the sensing operation request and begins to execute the sensing operation.

[0260] Prior to this, the sensing device had requested the UDM to check the user's license and authorization information.

[0261] Step 11: Multiple sensing devices establish a data transmission channel with SF-U to report sensing data.

[0262] Step 12: SF-U aggregates sensing data uploaded from multiple sources and devices, checks the sensing data security level corresponding to the sensing data security level policy in the sensing data rules, and generates a security level identifier. For sensing data with a medium or high security level, which contains more sensitive and critical information or is extremely important for sensing decision results, SF-U executes a multimodal contrastive generative adversarial network algorithm to detect whether there is fake data. If so, it outputs the corresponding sensing device ID and the location identifier of the fake data in that sensing data.

[0263] If it does not exist, skip steps 13, 14 and 15.

[0264] Step 13: SF-U reports the device ID containing false data to SF-C.

[0265] The reporting situation is divided into two types: the first is periodic reporting, which reports the execution of the perception data strategy and the detection of false data; the second is reporting when false data is detected.

[0266] Step 14: Every time SF-U reports a device ID containing false data, the LocalFaketag in SF-C (corresponding to the sensing device ID) is incremented (count is incremented by one). Once it exceeds 3, a stop reporting request is sent to the corresponding sensing device, and the Globalfaketag count is incremented by one.

[0267] At this point, the threshold for LocalFaketag is 3. If the sensing device is performing a sensing task for the first time in this sensing request, the default value of LocalFaketag is 0. Before the threshold is reached, the LocalFaketag counter can be reset to 0 at certain time intervals. Step 15: SF-C sends a data filtering request to SF-U.

[0268] Step 16: SF-U filters the sensing data based on the sensing device ID and location identifier (of the detected false data), filters out the false data, and then integrates and makes collaborative decisions.

[0269] Step 17: SF-U obtains the final fusion perception result and reports the perception result to AF.

[0270] To address the security vulnerabilities, data tampering and misinformation injection, and difficulty in attack detection inherent in existing technologies, this application's embodiments detect or filter misinformation injection attacks in integrated communication and sensing networks, offering the following advantages: Improve detection accuracy: It can accurately identify fake data, and even carefully constructed fake data is difficult to escape detection. By combining multimodal information, including but not limited to video, radar, and base station information, and through cross-verification, the accuracy of detecting counterfeit content can be improved.

[0271] Enhance data filtering capabilities: Through effective data filtering mechanisms, ensure that false data does not enter the data fusion process, thus guaranteeing the security and reliability of data fusion.

[0272] Enhance overall system security: By providing a comprehensive solution and process design, enhance the security of the integrated communication and sensing network, prevent destructive behavior by abnormal devices / nodes, and improve the system's sensing accuracy and application performance.

[0273] Reduce resource utilization: Detecting false data only for high-security sensor data can effectively reduce resource utilization, improve system efficiency, and ensure the accuracy and reliability of critical data.

[0274] Highly efficient data detection: Only high-security synesthetic data is detected for false data to reduce resource consumption and improve detection efficiency.

[0275] Accurate fake data detection: A multimodal contrastive generative adversarial network algorithm is introduced, which can detect fake data for perceived data with medium or high security levels and accurately output the location identifier of fake data, providing strong support for subsequent filtering processing.

[0276] Precise sensing data filtering and result reporting: SF-U filters sensing data based on data detection ID and false data location identifier, effectively removing false data and obtaining more accurate fused sensing results, which are then reported to the sensing requester, improving the reliability and availability of the data.

[0277] A complete fake data detection and filtering process: By combining perceived data security levels, multimodal contrastive generative adversarial network algorithms, and filtering processes, a complete fake data detection and filtering process is ensured to guarantee the comprehensiveness and systematic nature of data processing.

[0278] Optimized perception data fusion and result reporting mechanism: After filtering out false data, the final fused perception result is obtained through the optimized perception data fusion and result reporting mechanism, and is reported to the perception requester in a timely manner, which improves the efficiency and response speed of data processing.

[0279] This application's embodiments address existing problems in sensor data fusion, such as security vulnerabilities, data tampering and false information injection, difficulty in attack detection, and low resource utilization, by enhancing core network capabilities. It achieves multi-source sensor data false data detection through targeted high-security data, intelligent algorithms, filtering, and shutdown processes, thereby improving detection accuracy, enhancing data filtering capabilities, improving overall system security, reducing resource utilization, and achieving efficient and accurate sensing services. Specifically, through fine-grained sensing data security levels and protection strategies, multimodal comparative generative adversarial network algorithms, and sensing data filtering and result reporting mechanisms, it achieves enhanced data security, high efficiency in false data detection, accurate sensing data filtering and result reporting, a complete false data detection and filtering process, and an optimized sensing data fusion and result reporting mechanism, demonstrating significant technical advantages.

[0280] The false data detection device provided in this application is described below. The false data detection device described below can be referred to in correspondence with the false data detection method described above.

[0281] Figure 5 This is one of the structural schematic diagrams of the fake data detection device provided in this application, such as... Figure 5 As shown, the device, applied to sensing function network elements, includes the following: The first request module 501 is configured to request sensing data from at least two target sensing devices in response to a received sensing task request. The first detection module 502 is configured to perform false data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network model.

[0282] Figure 6 This is the second structural schematic diagram of the false data detection device provided in this application, as shown below. Figure 6 As shown, this device, applied to SF-C network elements, includes the following: The second request module 601 is configured to, in response to a received sensing task request, identify a target SF-U network element and request sensing data from at least two target sensing devices. The target SF-U network element is used to receive and perform false data detection on the sensing data corresponding to each target sensing device based on a multimodal contrastive generative adversarial network model.

[0283] Figure 7 This is the third structural schematic diagram of the false data detection device provided in this application, such as... Figure 7 As shown, this device, applied to SF-U network elements, includes the following: The receiving module 701 is configured to receive sensing data provided by at least two target sensing devices; The second detection module 702 is configured to perform false data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network model.

[0284] Figure 8 This is a schematic diagram of the structure of the electronic device provided in this application, such as... Figure 8 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a spoofing detection method.

[0285] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0286] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to perform the fake data detection methods provided by the above methods.

[0287] In another aspect, this application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the fake data detection methods provided by the above methods.

[0288] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0289] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0290] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting fake data, characterized in that, Applied to sensing function network elements, including: In response to a received sensing task request, sensing data is requested from at least two target sensing devices; Based on a multimodal contrastive generative adversarial network model, false data detection is performed on the sensing data corresponding to each of the target sensing devices.

2. The method for detecting fake data according to claim 1, characterized in that, The multimodal contrastive generative adversarial network model performs fake data detection on the sensing data corresponding to each target sensing device, including: The data features of each of the aforementioned sensing data are fused using multimodal features to obtain fused features; Based on the generative adversarial network model in the multimodal contrastive generative adversarial network model, the fused features are used for data discrimination to obtain the first probability that each of the perceived data is real data; and the data features of each of the perceived data are subjected to multimodal feature consistency comparison verification to obtain feature consistency score. Based on the first probability and the feature consistency score, a second probability is determined to characterize the presence of false data in each of the perceived data. Based on the second probability, it is determined whether the false data exists in each of the perceived data.

3. The method for detecting fake data according to claim 2, characterized in that, The step of performing multimodal feature consistency comparison and verification on the data features of each of the perceived data to obtain a feature consistency score includes: For each of the sensed data, the data features of the sensed data are residually concatenated with the first data feature to obtain the residual corresponding to the sensed data. The first data feature is the data feature of other sensed data or the data feature generated by the generator. Multimodal feature consistency comparison verification is performed on the residuals corresponding to each of the aforementioned sensing data to obtain feature consistency scores.

4. The method for detecting fake data according to claim 2, characterized in that, The step of determining a second probability, based on the first probability and the feature consistency score, representing the presence of false data in each of the perceived data includes: The first probability and the feature consistency score are integrated to obtain integrated data; The integrated data is subjected to a nonlinear transformation to obtain the second probability.

5. The method for detecting fake data according to any one of claims 1-4, characterized in that, The multimodal contrastive generative adversarial network model, after detecting false data in the sensing data corresponding to each target sensing device, further includes: In the event of detected false data, the identification of the sensing device corresponding to the false data and the location identification of the false data in the sensing data are determined. Based on the sensing device identifier and the location identifier, the false data in each of the sensing data is filtered to obtain the real data; The real data are fused together to obtain a fused perception result, and the fused perception result is reported.

6. The method for detecting fake data according to claim 1, characterized in that, The method further includes: The PCF network element, through its policy control function, requests a perception data policy from the UDR network element, which includes perception data rules and security levels. When the security level meets the conditions for false data detection, the step of performing false data detection on the sensing data corresponding to each of the target sensing devices is executed based on the multimodal contrastive generative adversarial network model.

7. The method for detecting fake data according to claim 6, characterized in that, The sensing data rules also include a threshold for the number of false data reports; The step of requesting sensing data from at least two target sensing devices includes: Based on the threshold for the number of false data reports, device detection is performed on each of the target sensing devices. Request sensing data from the target sensing device that has passed the device detection.

8. The method for detecting fake data according to claim 7, characterized in that, The threshold for the number of false data reports includes a first threshold at the global granularity; The step of performing device detection on each target sensing device based on the threshold for the number of false data reports includes: For each target sensing device, the global false identifier of the target sensing device is compared with the first number threshold, wherein the global false identifier represents the number of times the target sensing device stops providing sensing data; If the number of global false identifiers exceeds the first threshold, it is determined that the target sensing device has failed device detection. If the number of false global identifiers is less than or equal to the first threshold number, the target sensing device is determined to have passed device detection.

9. The method for detecting fake data according to claim 7 or 8, characterized in that, The threshold for the number of false data reports includes a second threshold at the task granularity; The multimodal contrastive generative adversarial network model, after detecting false data in the sensing data corresponding to each target sensing device, further includes: For each target sensing device, if false data exists in the sensing data corresponding to the target sensing device, the task false identifier of the target sensing device is updated. The task false identifier represents the number of times the target sensing device provides false data under the current sensing task. If the number of false task identifiers reaches the second threshold, the request for sensing data from the target sensing device corresponding to the false task identifier will be stopped.

10. The method for detecting fake data according to claim 9, characterized in that, The method further includes: If the request for sensing data from the target sensing device corresponding to the task false identifier is stopped, the global false identifier of the target sensing device corresponding to the task false identifier is updated.

11. The method for detecting fake data according to claim 6, characterized in that, The process of requesting awareness data policies from the Unified Data Repository (UDR) network element via the Policy Control Function (PCF) network element includes: Send a sensing data policy request to the PCF network element; The sensing data policy request carries at least one of the following: information of the application function AF network element, sensing range, and sensing service type. The AF network element is the network element that initiates the sensing task request. The perception data policy request is used by the PCF network element to request the perception data policy from the UDR network element and to feed back the perception data policy to the perception function network element.

12. A method for detecting fake data, characterized in that, Applied to SF-C network elements in the sensing function control plane, including: In response to the received sensing task request, the target sensing function user plane SF-U network element is identified, and sensing data is requested from at least two target sensing devices; The target SF-U network element is used to receive and perform false data detection on the sensing data corresponding to each target sensing device based on a multimodal contrastive generative adversarial network model.

13. The method for detecting fake data according to claim 12, characterized in that, Before requesting sensing data from at least two target sensing devices, the process also includes: Based on the sensing parameters carried in the sensing task request, select at least two target sensing devices.

14. The method for detecting fake data according to claim 12 or 13, characterized in that, Before determining the target SF-U network element, the process also includes: The PCF network element requests the perception data policy from the UDR network element. The determination of the target SF-U network element includes: The target SF-U network element is selected based on the sensing parameters carried in the sensing task request and / or the sensing data strategy.

15. The method for detecting fake data according to claim 14, characterized in that, The method further includes: Send the sensing task request and the sensing data policy to the target SF-U network element; The perception task request and the perception data policy are used by the target SF-U network element to initiate a false data detection operation after configuring and activating the perception data policy.

16. The method for detecting fake data according to claim 12 or 13, characterized in that, The method further includes: After receiving the sensing device identifier corresponding to the false data reported by the target SF-U network element, a data filtering request is sent to the target SF-U network element; The data filtering request is used by the target SF-U network element to filter and fuse the false data in each of the sensing data to obtain the fused sensing result, and then report the fused sensing result.

17. A method for detecting fake data, characterized in that, Applied to SF-U network elements, including: Receive sensing data from at least two target sensing devices; Based on a multimodal contrastive generative adversarial network model, false data detection is performed on the sensing data corresponding to each of the target sensing devices.

18. A device for detecting fake data, characterized in that, Applied to sensing function network elements, including: The first request module is configured to request sensing data from at least two target sensing devices in response to a received sensing task request. The first detection module is configured to perform false data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network model.

19. A device for detecting fake data, characterized in that, Applied to SF-C network elements, including: The second request module is configured to, in response to a received sensing task request, identify the target SF-U network element and request sensing data from at least two target sensing devices. The target SF-U network element is used to receive and perform false data detection on the sensing data corresponding to each target sensing device based on a multimodal contrastive generative adversarial network model.

20. A device for detecting fake data, characterized in that, Applied to SF-U network elements, including: The receiving module is configured to receive sensing data provided by at least two target sensing devices; The second detection module is configured to perform false data detection on the sensing data corresponding to each of the target sensing devices based on a multimodal contrastive generative adversarial network model.

21. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the fake data detection method as described in any one of claims 1 to 11, 12 to 16 or 17.

22. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the fake data detection method as described in any one of claims 1 to 11, 12 to 16 or 17.

Images