Electromagnetic latent device identification method and system based on federal large model

Abstract

The invention discloses an electromagnetic latent device identification method and system based on a federated large model, relates to the technical field of electromagnetic device detection and identification, and discloses an electromagnetic latent device identification method based on the federated large model. Determining a target latent equipment identification model based on a preset initial model; acquiring electromagnetic equipment parameter data, and determining a risk electromagnetic equipment list and a distance feature and a signal feature of each risk electromagnetic equipment in the risk electromagnetic equipment list based on the electromagnetic equipment parameter data; obtaining a current risk electromagnetic device based on the risk electromagnetic device list, and inputting a distance feature and a signal feature of the current risk electromagnetic device into the target latent device identification model to obtain a latent risk prediction score; and determining a target latent device identification result based on the latent risk prediction score. According to the scheme, the secret photographing behavior of the electromagnetic hidden equipment can be effectively detected, and the privacy of a user is prevented from being invaded and disturbed.

Description

Technical Field

[0001] The present application relates to the technical field of electromagnetic device detection and identification, and in particular to an electromagnetic latent device identification method and system based on a federated large model. Background Art

[0002] With the technological upgrade of mobile electronic devices, electromagnetic concealed devices such as micro cameras and eavesdroppers are becoming increasingly miniaturized and portable, which may bring privacy and security issues that cannot be ignored in some specific places. The security prevention of mobile electronic devices requires the comprehensive use of various technical means including information fusion technology, Internet of Things technology, artificial intelligence, etc. Large model algorithms have powerful representation learning capabilities and can provide strong support for solving such privacy and security issues.

[0003] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0004] The main purpose of this application is to provide an electromagnetic latent device identification method and system based on a federal large model, aiming to solve the technical problem of how to effectively detect the voyeuristic photography behavior of electromagnetic hidden devices to prevent user privacy from being violated.

[0005] To achieve the above purpose, the present application proposes an electromagnetic latent device identification method based on a federated large model, the electromagnetic latent device identification method based on a federated large model is applied to a client, and the method includes:

[0006] When receiving the preset initial model sent by the server, determining the target latent device identification model based on the preset initial model;

[0007] Acquire electromagnetic device parameter data, and determine a risky electromagnetic device list and a distance feature and a signal feature of each risky electromagnetic device in the risky electromagnetic device list based on the electromagnetic device parameter data;

[0008] Acquire the current risk electromagnetic device based on the risk electromagnetic device list, and input the distance feature and the signal feature of the current risk electromagnetic device into the target latent device identification model to obtain a latent risk prediction score;

[0009] A target latent device identification result is determined based on the latent risk prediction score.

[0010] In one embodiment, the step of determining a risk electromagnetic device list and a distance feature and a signal feature of each risk electromagnetic device in the risk electromagnetic device list based on the electromagnetic device parameter data comprises:

[0011] Establishing a risky electromagnetic equipment list based on the electromagnetic equipment parameter data;

[0012] Acquire distance data, signal receiving power data and signal strength data of each risky electromagnetic device in the risky electromagnetic device list;

[0013] The distance feature and the signal feature of each risky electromagnetic device in the risky electromagnetic device list are determined based on the distance data, the signal receiving power data, and the signal strength data.

[0014] In one embodiment, the step of determining the target latent device identification result based on the latent risk prediction score includes:

[0015] Collecting the potential risk prediction score of the current risk electromagnetic equipment within a preset period;

[0016] Determine the median prediction score, the target prediction score number, and the total prediction score number based on the potential risk prediction score within the preset period;

[0017] The target latent device identification result is determined based on the prediction score median, the target prediction score number, and the total prediction score number.

[0018] In one embodiment, after the step of inputting the distance feature and the signal feature into the target latent device identification model to obtain the target latent device identification result, the step further includes:

[0019] Obtaining a local preset threshold, and determining a target latent device from the risky electromagnetic device list based on the local preset threshold and the target latent device identification result, to obtain a target latent device list;

[0020] The device behavior data of the target latent device is acquired, and a latent device list is generated based on the target latent device list and the device behavior data, so that the user can search for electromagnetic latent devices according to the latent device list.

[0021] In one embodiment, when receiving the preset initial model sent by the server, the step of determining the target latent device identification model based on the preset initial model includes:

[0022] When receiving the preset initial model sent by the server, training the preset initial model based on the local data set to obtain local model parameters, and sending the local model parameters to the server;

[0023] When the global model parameters fed back by the server based on the local model parameters are received, local aggregate model parameters are determined based on the global model parameters and the local model parameters to obtain a target latent device identification model.

[0024] In addition, to achieve the above purpose, the present application also proposes an electromagnetic latent device identification method based on a federated large model, the electromagnetic latent device identification method based on a federated large model is applied to a server, and the method includes:

[0025] Sending a preset initial model to multiple clients, and receiving local model parameters fed back by the multiple clients based on the preset initial model;

[0026] Evaluate the attack recognition capabilities of the plurality of local model parameters based on the attack evaluation data set to obtain an attack impact evaluation result;

[0027] Aggregating the plurality of local model parameters based on the attack impact assessment result to obtain a global model parameter;

[0028] The global model parameters are sent to the plurality of clients, so that the plurality of clients determine a target latent device identification model based on the global model parameters, and identify latent devices based on the target latent device identification model.

[0029] In one embodiment, the step of aggregating the plurality of local model parameters based on the attack impact assessment result to obtain a global model parameter includes:

[0030] Acquire attack impact index values ​​of a plurality of the local model parameters based on the attack impact assessment result;

[0031] Acquire a degradation threshold, and determine a candidate model parameter from a plurality of the local model parameters based on the degradation threshold and the attack impact index value;

[0032] A target model parameter is determined from the plurality of candidate model parameters, and the plurality of target model parameters are aggregated to obtain a global model parameter.

[0033] In one embodiment, the step of determining a target model parameter from a plurality of the candidate model parameters and aggregating the plurality of the target model parameters to obtain a global model parameter comprises:

[0034] Sorting the attack impact index values ​​of the plurality of candidate model parameters to obtain an impact score sorting result;

[0035] Determine the candidate model parameter with a first preset ranking in the impact score sorting result as a first target model parameter;

[0036] Determine a second target model parameter from the candidate model parameters of a second preset ranking in the impact score sorting result;

[0037] A target model parameter is determined based on the first target model parameter and the second target model parameter, and a plurality of the target model parameters are aggregated to obtain a global model parameter.

[0038] In one embodiment, the attack assessment data set includes a first preset data set and a second preset data set, the first preset data set is a data set that does not contain an attack, and the second preset data set is a data set that contains an attack;

[0039] The step of evaluating the attack recognition capabilities of the plurality of local model parameters based on the attack evaluation data set to obtain an attack impact evaluation result comprises:

[0040] Based on the first preset data set and the second preset data set, the preset initial model is trained respectively to obtain a first model and a second model, and the test accuracy of the first model and the second model is obtained to obtain a first accuracy rate;

[0041] Based on the multiple local model parameters, the first model and the second model are respectively updated, and the test accuracy of the updated first model and the second model is obtained to obtain a second accuracy;

[0042] The attack recognition capabilities of the plurality of local model parameters are evaluated based on the first accuracy rate and the second accuracy rate to obtain an attack impact evaluation result.

[0043] In addition, to achieve the above objectives, the present application also proposes an electromagnetic latent device identification system based on a federated large model, wherein the electromagnetic latent device identification system based on a federated large model includes a plurality of the above-mentioned clients and the above-mentioned server.

[0044] In addition, to achieve the above purpose, the present application also proposes an electromagnetic latent device identification device based on a federated large model, which is applied to a client. The electromagnetic latent device identification device based on a federated large model includes:

[0045] A target model determination module, configured to determine a target latent device identification model based on the preset initial model when receiving the preset initial model sent by the server;

[0046] A parameter data acquisition module, used to acquire electromagnetic device parameter data, and determine a risky electromagnetic device list and a distance feature and a signal feature of each risky electromagnetic device in the risky electromagnetic device list based on the electromagnetic device parameter data;

[0047] a latent risk prediction module, configured to obtain a current risk electromagnetic device based on the risk electromagnetic device list, and input the distance feature and the signal feature of the current risk electromagnetic device into the target latent device identification model to obtain a latent risk prediction score;

[0048] The latent device identification module is used to determine the target latent device identification result based on the latent risk prediction score.

[0049] In addition, to achieve the above purpose, the present application also proposes an electromagnetic latent device identification device based on a federated large model, which is applied to a server. The electromagnetic latent device identification device based on a federated large model includes:

[0050] A model initialization module, used to send a preset initial model to multiple clients, and receive local model parameters fed back by multiple clients based on the preset initial model;

[0051] A poisoning attack defense module, used to evaluate the attack recognition capabilities of the multiple local model parameters based on the attack evaluation data set to obtain an attack impact evaluation result;

[0052] A global model aggregation module, used for aggregating the plurality of local model parameters based on the attack impact assessment result to obtain a global model parameter;

[0053] The global model distribution module is used to send the global model parameters to the multiple clients, so that the multiple clients determine the target latent device identification model based on the global model parameters and identify the latent device based on the target latent device identification model.

[0054] In addition, to achieve the above-mentioned purpose, the present application also proposes an electromagnetic latent device identification device based on a federal large model, the device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the electromagnetic latent device identification method based on a federal large model as described above.

[0055] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the electromagnetic latent device identification method based on the federal large model as described above are implemented.

[0056] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the electromagnetic latent device identification method based on the federated large model as described above are implemented.

[0057] One or more technical solutions proposed in this application have at least the following technical effects:

[0058] When receiving the preset initial model sent by the server, determine the target latent device identification model based on the preset initial model; obtain electromagnetic device parameter data, and determine the risk electromagnetic device list and the distance characteristics and signal characteristics of each risk electromagnetic device in the risk electromagnetic device list based on the electromagnetic device parameter data; obtain the current risk electromagnetic device based on the risk electromagnetic device list, and input the distance characteristics and signal characteristics of the current risk electromagnetic device into the target latent device identification model to obtain the latent risk prediction score; determine the target latent device identification result based on the latent risk prediction score. This solution can effectively detect the sneak shooting behavior of electromagnetic hidden devices to prevent the privacy of users from being violated. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0061] Figure 1 This is a flow chart of the first embodiment of the electromagnetic latent device identification method based on the federated large model applied to the client of the present application;

[0062] Figure 2 This is a diagram of a federated large model learning architecture of a first embodiment of an electromagnetic latent device identification method based on a federated large model applied to a client of the present application;

[0063] Figure 3 This is a flow chart of a second embodiment of the electromagnetic latent device identification method based on a federated large model applied to a client of the present application;

[0064] Figure 4 This is a flow chart of the first embodiment of the electromagnetic latent device identification method based on the federated large model applied to the server of the present application;

[0065] Figure 5 A schematic diagram of the module structure of an electromagnetic latent device identification device based on a federated large model applied to a client in an embodiment of the present application;

[0066] Figure 6 This is a schematic diagram of the module structure of an electromagnetic latent device identification device based on a federated large model applied to a server in an embodiment of the present application;

[0067] Figure 7Schematic diagram of the device structure of the hardware operating environment involved in the electromagnetic latent device identification method based on the federal large model in the embodiment of the present application.

[0068] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0069] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0070] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0071] The main solution of the embodiment of the present application is: when a preset initial model sent by the server is received, a target latent device identification model is determined based on the preset initial model; electromagnetic device parameter data is obtained, and a risk electromagnetic device list and distance characteristics and signal characteristics of each risk electromagnetic device in the risk electromagnetic device list are determined based on the electromagnetic device parameter data; the current risk electromagnetic device is obtained based on the risk electromagnetic device list, and the distance characteristics and signal characteristics of the current risk electromagnetic device are input into the target latent device identification model to obtain a latent risk prediction score; and the target latent device identification result is determined based on the latent risk prediction score.

[0072] In this embodiment, for the convenience of description, the electromagnetic latent device identification device is used as the execution subject for explanation.

[0073] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, an electromagnetic latent device identification device, etc. The following takes the electromagnetic latent device identification device as an example to illustrate this embodiment and the following embodiments.

[0074] Based on this, the embodiment of the present application provides an electromagnetic latent device identification method based on a federated large model, and the electromagnetic latent device identification method based on a federated large model is applied to a client, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the electromagnetic latent device identification method based on the federated large model applied to the client of the present application.

[0075] In this embodiment, the electromagnetic latent device identification method based on the federated large model includes steps S10 to S40:

[0076] Step S10, upon receiving the preset initial model sent by the server, determining a target latent device identification model based on the preset initial model;

[0077] It should be noted that FL (Federated Learning) is a machine learning framework, also known as federated machine learning, joint learning, and alliance learning, which can effectively help multiple institutions use data and conduct machine learning modeling while meeting the requirements of user privacy protection, data security, and government regulations. The solution of this application can ensure that the participating parties can exchange information and model parameters in an encrypted manner while maintaining independence, and at the same time enhance the algorithmic capabilities with the help of the federated learning framework.

[0078] It should be noted that hidden cameras and hidden cameras are concealed and difficult to detect. The present application provides an electromagnetic latent device identification method based on a federal large model. By performing behavioral analysis on electromagnetic signals within a range, the latent characteristics of abnormal devices can be effectively discovered, and such abnormal devices, namely electromagnetic latent devices, can be detected, thereby helping users counter the intrusion of such devices.

[0079] Please refer to Figure 2 , Figure 2 This is a diagram of the federated large model learning architecture of the first embodiment of the electromagnetic latent device identification method based on the federated large model applied to the client in this application. The electromagnetic latent device identification method proposed in this application adopts a distributed architecture of federated learning, in which the server can be a central server responsible for managing and coordinating the federated learning process of multiple clients, for initializing global model parameters, collecting local updates uploaded by clients, performing screening and aggregation operations, and updating the global model. Each client has its own private data set, which the client can use to train personalized model parameters and send updates back to the server.

[0080] It should be understood that the preset initial model is the initial model provided by the server and is the starting point of the federated learning process. The target latent device identification model is a model that can be used to identify electromagnetic latent devices obtained by training the received preset initial model based on the private data set.

[0081] In a feasible implementation, step S10 may include steps S11 to S12:

[0082] Step S11, when receiving the preset initial model sent by the server, training the preset initial model based on the local data set to obtain local model parameters, and sending the local model parameters to the server;

[0083] It should be understood that when each client receives the preset initial model sent by the server, it will independently train the preset initial model based on the local data set. In order to ensure that the model is fully trained on the local data set, during the training process, the client will perform multiple rounds of local iterations on the model and calculate the update amount of the model parameters such as the adjustment of weights and biases until the local iteration reaches the predetermined rounds or until the model performance meets the preset requirements, and then send the model update (i.e., local model parameters) to the server. Among them, the local data set contains multiple pieces of device behavior feature data of electromagnetic devices and their corresponding labels.

[0084] Step S12: upon receiving the global model parameters fed back by the server based on the local model parameters, determining local aggregate model parameters based on the global model parameters and the local model parameters to obtain a target latent device identification model.

[0085] It should be noted that when the client receives the global model parameters from the server based on the local model parameters, it will download the global model parameters and update them locally. By learning the local aggregation weight for each parameter and performing adaptive local aggregation, each parameter of the model is updated based on the global model parameters and the local model parameters to obtain the local aggregation parameters.

[0086] Exemplarily, the formula for updating the model parameters is as follows:

[0087]

[0088] in, represents the adaptive parameter, e is used to represent the training round, and its value ranges from 1 to n; i is used to identify the client, and for n clients, the value of i ranges from 1 to n; W is a weight vector composed of 0 to 1.

[0089] It should be understood that through adaptive updating, the model of each client can retain the characteristics of local data as much as possible while learning the global model. Each client can adaptively adjust its local model parameters according to its own data characteristics and the update of the global model, which can improve the generalization ability and adaptability of the model, especially when the data is unevenly distributed or there are data islands. A personalized federated learning method designed in this embodiment can adapt to different clients.

[0090] Step S20, acquiring electromagnetic device parameter data, and determining a risky electromagnetic device list and a distance feature and a signal feature of each risky electromagnetic device in the risky electromagnetic device list based on the electromagnetic device parameter data;

[0091] It should be noted that the client includes detection devices distributed in different geographical locations, and the detection device needs to include a receiver for capturing specific electromagnetic signals, a decoder for converting received electromagnetic signals into parameter data, a processor for analyzing and processing parameter data; a memory for storing parameter data analysis and processing results, and a display or output device for pushing information containing latent device identification results to users. Specifically, the detection device can collect IMSI (International Mobile Subscriber Identity) signals of mobile phones within a certain range around it, and record information such as latitude and longitude, time, IMSI number, and number location. Depending on the type of device and the setting of device parameters, the collection range is usually between 200 and 1500 meters. This type of device is the device mainly used in this application to identify electromagnetic risk devices.

[0092] It should be understood that the process of obtaining electromagnetic device parameter data can be as follows: the receiver of the detection device collects electromagnetic signals within a preset range around it, and the decoder of the detection device obtains relevant parameters transmitted in the electromagnetic device communication (such as device identification, network connection status, signal strength, etc.), that is, electromagnetic device parameter data, by analyzing the electromagnetic signals.

[0093] In addition, it should be understood that the risky electromagnetic device list refers to a list of devices derived from the analysis of electromagnetic device parameter data. The electromagnetic devices in the device list are considered to have potential security risks, and each electromagnetic device is represented by device identification information. The distance feature is used to represent the distance from the client to the detected risky electromagnetic device. The signal feature is used to represent the properties of the captured electromagnetic signal of the risky electromagnetic device, such as signal strength, frequency, modulation mode, etc. Based on the distance features and signal features of the risky electromagnetic device, devices with privacy and security risks can be further identified from the risky electromagnetic device list.

[0094] In a feasible implementation, step S20 may include: establishing a risk electromagnetic device list based on the electromagnetic device parameter data; obtaining distance data, signal reception power data and signal strength data for each risk electromagnetic device in the risk electromagnetic device list; and determining the distance characteristics and signal characteristics of each risk electromagnetic device in the risk electromagnetic device list based on the distance data, the signal reception power data and the signal strength data.

[0095] It should be understood that the electromagnetic device parameter data at least includes signal strength data RSSI (Received Signal Strength Indicator) and signal received power data RSRP (Reference Signal Received Power). Since the signal strength decays with increasing distance, the distance data distance (i.e., the distance from the client to the detected risky electromagnetic device) can be estimated based on RSSI.

[0096] It should be noted that the distance feature includes a feature based on distance, and the signal feature includes a feature based on RSRP and a feature based on RSRP. Specifically, the feature based on distance is the third-order differential average of distance; the feature based on RSRP is the first-order differential standard deviation of RSRP; and the feature based on RSRP is the first-order differential average of RSRP. Among them, by calculating the third-order differential average of distance, the acceleration pattern of the movement of risky electromagnetic equipment can be captured and the movement trend of risky electromagnetic equipment can be identified. The first-order differential average of RSRP reflects the average change of the signal receiving power over time, reflecting the stability and consistency of the electromagnetic signal. The first-order differential standard deviation of RSSI is used to measure the degree of discreteness of the signal strength over time, reflecting the volatility of the signal.

[0097] It should be understood that electromagnetic device parameter data is time series data collected periodically. By structuring the periodically collected electromagnetic device parameter data into distance features and signal features, the target latent device identification model can not only capture the static signal strength of risky electromagnetic devices, but also capture the dynamic changes in device behavior. The model for latent device identification based on structured distance features and signal features can maintain stable performance under a variety of different environmental conditions (such as signal interference, physical obstacles, and high device density) and accurately identify latent devices.

[0098] Step S30, obtaining the current risk electromagnetic device based on the risk electromagnetic device list, and inputting the distance feature and the signal feature of the current risk electromagnetic device into the target latent device identification model to obtain a latent risk prediction score;

[0099] It should be noted that by inputting the distance characteristics and signal characteristics of the current risk electromagnetic devices into the target latent device identification model, the current risk electromagnetic devices can be deeply analyzed and evaluated, and the potential risk prediction scores obtained can be quantified and predicted with the potential privacy or security risks of these devices.

[0100] Step S40: determining a target latent device identification result based on the latent risk prediction score.

[0101] It should be understood that for those hidden devices that are not easy to identify directly, such as hidden cameras, it is necessary to continuously observe and analyze their behavior patterns to assess their potential risks. The target latent device identification results include the risk assessment values ​​of all risky electromagnetic devices in the risky electromagnetic device list.

[0102] In a feasible implementation, step S40 may include: collecting the latent risk prediction score of the current risk electromagnetic equipment within a preset period; determining the median prediction score, the target prediction score number and the total prediction score number based on the latent risk prediction score within the preset period; determining the target latent device identification result based on the median prediction score, the target prediction score number and the total prediction score number.

[0103] It should be understood that time series analysis of continuously collected electromagnetic signal data can obtain periodic distance characteristics and signal characteristics. By inputting the periodic distance characteristics and signal characteristics into the target latent device identification model, the latent risk prediction score within a preset period can be obtained.

[0104] In addition, it should be understood that the potential risk prediction score within the preset period includes multiple risk prediction values ​​for the current risk electromagnetic device within the period, and the higher the prediction value, the greater the possibility of privacy infringement of the risk electromagnetic device. The median of the statistical data prediction score, the target prediction score number and the total prediction score number can be obtained by calculating the multiple risk prediction values, and periodic risk scoring based on the statistical data can determine which devices are considered high-risk potential devices within the scoring period.

[0105] It should be noted that the median prediction score refers to the median of the model prediction value within the scoring period, the target prediction score times refer to the times of high prediction values, and the total prediction score times refer to the total prediction times of the model. The target latent device identification result includes the risk assessment value of the current risk latent device based on the scoring formula for all risk electromagnetic devices in the risk electromagnetic device list within the scoring period.

[0106] For example, the risk assessment value calculation formula for the current latent device within the scoring period is as follows:

[0107]

[0108] Among them, m is the median of the model prediction value within the scoring period; k is the number of high prediction values; t is the total number of model predictions; score is the scoring result.

[0109] This embodiment provides an electromagnetic latent device identification method based on a federated large model. The electromagnetic latent device identification method based on a federated large model is applied to a client, and includes: upon receiving a preset initial model sent by a server, determining a target latent device identification model based on the preset initial model; obtaining electromagnetic device parameter data, and determining a risk electromagnetic device list and the distance characteristics and signal characteristics of each risk electromagnetic device in the risk electromagnetic device list based on the electromagnetic device parameter data; obtaining the current risk electromagnetic device based on the risk electromagnetic device list, and inputting the distance characteristics and signal characteristics of the current risk electromagnetic device into the target latent device identification model to obtain a latent risk prediction score; and determining the target latent device identification result based on the latent risk prediction score. This solution can effectively detect the sneak shooting behavior of electromagnetic hidden devices to prevent user privacy from being infringed.

[0110] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 3 After step S40, the electromagnetic latent device identification method based on the federated large model applied to the client further includes steps A10 to A20:

[0111] Step A10, obtaining a local preset threshold, and determining a target latent device from the risky electromagnetic device list based on the local preset threshold and the target latent device identification result, to obtain a target latent device list;

[0112] It should be understood that the local preset threshold is the risk assessment threshold preset by the client. Due to the differences in the client's environmental conditions and local data characteristics, the risk assessment thresholds preset by different clients should be different. Electromagnetic devices that exceed the local preset threshold are considered potential target latent devices and require the user's special attention. Electromagnetic devices whose risk assessment values ​​exceed the local preset threshold within the scoring period in the target latent device identification results will form a target latent device list.

[0113] Step A20, acquiring device behavior data of the target latent device, and generating a latent device list based on the target latent device list and the device behavior data, so that the user can search for electromagnetic latent devices according to the latent device list.

[0114] It should be noted that electromagnetic latent devices refer to concealed devices that use electromagnetic waves (such as radio waves) for communication or data transmission, and are used for unauthorized eavesdropping, surveillance or other privacy attack activities. The device behavior data of the target latent device includes information such as time distribution, distance distribution, location direction, etc. After obtaining the target latent device list, the device behavior data of the target latent device will be obtained based on the target latent device list, and finally a list of latent devices will be generated. The client's display or output device will push the latent device list to the user to prompt the user that there are electromagnetic latent devices around the user that pose a privacy and security risk, and the user needs to search and confirm.

[0115] This embodiment provides an electromagnetic latent device identification method based on a federated large model, obtains a local preset threshold, and determines a target latent device from a risk electromagnetic device list based on the local preset threshold and the target latent device identification result, obtains a target latent device list, obtains device behavior data of the target latent device, and generates a latent device list based on the target latent device list and the device behavior data, so that a user can search for electromagnetic latent devices according to the latent device list. A detailed latent device list is generated based on the target latent device identification result, so that the user can quickly locate and search for potential electromagnetic latent devices according to the list, take timely response measures, improve the overall security protection capability, and protect the privacy and security of users from infringement.

[0116] In addition, the embodiment of the present application provides an electromagnetic latent device identification method based on a federated large model, and the electromagnetic latent device identification method based on a federated large model is applied to the server, please refer to Figure 4 , Figure 4 This is a flow chart of the first embodiment of the electromagnetic latent device identification method based on the federated large model applied to the server of the present application.

[0117] In this embodiment, the electromagnetic latent device identification method based on the federated large model applied to the server includes steps S50 to S80:

[0118] Step S50, sending a preset initial model to multiple clients, and receiving local model parameters fed back by multiple clients based on the preset initial model;

[0119] It should be understood that at the initial stage of training, the server will initialize a global model, namely the preset initial model, and send the preset initial model to multiple clients participating in the federated large model training to receive the local model parameters fed back by all clients based on the preset initial model. Since the training process of each client for the preset initial model is independent and does not require mutual communication, the server can obtain the local model parameters of multiple clients trained based on fresh and diverse data sets while ensuring user privacy.

[0120] Step S60, evaluating the attack recognition capabilities of the plurality of local model parameters based on the attack evaluation data set to obtain an attack impact evaluation result;

[0121] It should be noted that during the training and learning of the federated large model, a central server, namely the server, is responsible for aggregating and synchronizing the model weights, and a group of clients are used to manipulate multi-site data. Since the federated learning system is dynamic and open, new clients may join at any time during model training, and existing clients participating in the training may leave at any time. Therefore, during the training process, the server must always be vigilant against the occurrence of poisoning attacks. The server cannot monitor the training process conducted locally by the participants. Malicious attackers can affect the prediction results of the model by generating deceptive inputs, thereby posing a security risk to the global model of federated learning. The existing federated large model cannot guarantee the quality of the uploaded model updates from the client. Malicious behavior refers to the behavior of deliberately reducing the learning performance (such as accuracy and convergence) of the global model through poisoning attacks. The attacker adds carefully constructed abnormal data to the training data to destroy the original probability distribution of the training data, causing the model to produce classification or clustering errors under certain conditions to destroy its training data set and accuracy.

[0122] It should be understood that the attack assessment dataset includes clean datasets and datasets with backdoors. Clean datasets refer to datasets that do not contain malicious behavior. Backdoor datasets refer to datasets that have been intentionally tampered with or implanted with backdoors. These backdoors may include incorrect labels and abnormal input features in order to implant malicious behavior in the model. By comparing the performance of local model parameters on the attack assessment dataset, the attack recognition capability of each local model parameter can be evaluated to obtain the attack impact assessment results.

[0123] In a feasible implementation manner, the attack assessment data set includes a first preset data set and a second preset data set, the first preset data set is a data set that does not contain attacks, and the second preset data set is a data set that contains attacks, and step S60 may include: training the preset initial model based on the first preset data set and the second preset data set respectively to obtain a first model and a second model, and obtaining the test accuracy of the first model and the second model to obtain a first accuracy; updating the first model and the second model based on multiple local model parameters respectively, and obtaining the test accuracy of the updated first model and the second model to obtain a second accuracy; evaluating the attack recognition capabilities of multiple local model parameters based on the first accuracy and the second accuracy to obtain an attack impact assessment result.

[0124] It should be understood that the first preset data set does not contain attacks and is a clean data set, and the second preset data set contains attacks and is a data set containing backdoors. The first model obtained by training the preset initial model based on the clean data set until convergence and saving the model parameters is a benign model, which represents the ideal state of the model when it is not attacked. The second model obtained by training the preset initial model based on the data set containing the backdoor until convergence and saving the model parameters is a malicious model, which represents the state in which the model learns wrong patterns and features due to the influence of malicious behavior. After obtaining the benign model and the malicious model, the test accuracy of the model, i.e., the first accuracy, will be obtained based on the performance of the benign model and the malicious model on the test set.

[0125] It should be noted that after receiving the local model parameters sent by the client, the benign model and the malicious model will be updated based on the local model parameters, and the accuracy of the benign model and the malicious model after the update will be obtained by testing them on their respective data sets, that is, the second accuracy. The accuracy difference of the benign model before and after the update can be obtained by subtracting the accuracy of the benign model benign By subtracting the accuracy of the malicious model before and after the update, we can get the accuracy difference of the malicious model ΔACC malicious The difference in accuracy of the benign model and the difference in accuracy of the malicious model are subtracted, and the final value can be used as a measure of the degree of influence of the local model parameter on the preset initial model. The attack impact assessment result includes the indicator value of the degree of influence of each local model parameter on the preset initial model.

[0126] For example, the calculation formula of the impact degree measurement indicator ΔI is as follows:

[0127] ΔI=ΔACC benign -ΔACC malicious

[0128] It should be understood that the smaller the value of ΔI is, the worse the influence of the updated local model parameters on the preset initial model is, and the larger the value of ΔI is, the more beneficial the influence of the updated local model parameters on the preset initial model is.

[0129] Step S70, aggregating the plurality of local model parameters based on the attack impact assessment result to obtain a global model parameter;

[0130] It should be understood that since the attack impact assessment result includes the assessment value of the attack identification capability of each local model parameter, multiple local model parameters can be screened according to the attack impact assessment result, and local model parameters that have a degrading effect on the preset initial model can be filtered out. The global model parameters obtained by aggregating the screened local model parameters can reduce the negative impact of the client model parameter update on the model performance.

[0131] In a feasible implementation manner, step S70 may include steps S71 to S73:

[0132] Step S71, obtaining attack impact index values ​​of a plurality of the local model parameters based on the attack impact assessment result;

[0133] It should be noted that the attack impact index values ​​of the multiple local model parameters obtained based on the attack impact assessment result are evaluation values ​​of the attack recognition capability of each local model parameter.

[0134] Step S72, obtaining a degradation threshold, and determining a candidate model parameter from a plurality of the local model parameters based on the degradation threshold and the attack impact index value;

[0135] It should be noted that the degradation threshold is a set standard used to determine the degree of degradation of model performance by local model parameters. If the parameter update of a client causes the model performance to decline (i.e., the change in the accuracy difference) to be lower than the degradation threshold, then the local model parameters uploaded by this client will be considered degraded and need to be filtered out. After filtering out the local model parameters corresponding to the attack impact index values ​​below the degradation threshold, the remaining local model parameters are the candidate model parameters.

[0136] Step S73, determining a target model parameter from the multiple candidate model parameters, and aggregating the multiple target model parameters to obtain a global model parameter.

[0137] It should be understood that further screening of the candidate model parameters can obtain target model parameters that meet the requirements, and aggregating the target model parameters can obtain global model parameters.

[0138] In a feasible implementation, step S73 may include: sorting the attack impact index values ​​of multiple candidate model parameters to obtain an impact score sorting result; determining the candidate model parameter with a first preset ranking in the impact score sorting result as a first target model parameter; determining a second target model parameter from the candidate model parameter with a second preset ranking in the impact score sorting result; determining a target model parameter based on the first target model parameter and the second target model parameter, and aggregating multiple target model parameters to obtain a global model parameter.

[0139] It should be noted that the process of determining the second target model parameter from the candidate model parameters of the second preset ranking in the impact score sorting result is a random process.

[0140] In a specific implementation, the first preset ranking refers to the top 30% of the candidate model parameters in terms of attack impact index value. The second preset ranking refers to the last 70% of the candidate model parameters in terms of attack impact index value. Using the top 30% of the candidate model parameters in terms of attack impact index value as the first target model parameters can ensure that the updated local model parameters with the greatest positive impact on model performance can participate in model aggregation. Parameters can be randomly selected as the second target model parameters from the last 70% of the candidate model parameters in terms of attack impact index value until the proportion of the first target model parameters and the second target model parameters reaches 80% of the candidate model parameters. The selected first target model parameters and the second target model parameters (the first 30% directly selected and the remaining 70% randomly selected) constitute a candidate pool, i.e., a target model parameter, and the target model parameters are weighted averaged to obtain a global model parameter. The global model parameters obtained by the method for defending against poisoning attacks designed in this embodiment can prevent the model performance from declining.

[0141] Step S80: sending the global model parameters to the multiple clients, so that the multiple clients determine a target latent device identification model based on the global model parameters, and identify the latent device based on the target latent device identification model.

[0142] It should be understood that by distributing the global model parameters aggregated by the server for poisoning detection and screening to each client, each client can further personalize and update the global model parameters based on the global model parameters to obtain the target latent device identification model, which is used in the target latent device identification process. During the entire model parameter update process, there is no need to upload sensitive local data to the central server, and a target latent device identification model with the universal characteristics of the global model and the local data characteristics of the client can be obtained.

[0143] It should be noted that the electromagnetic latent device identification method based on the federal large model proposed in this application can be applied to products such as covert monitoring equipment, anti-theft security inspection equipment, anti-sneak photography devices, privacy guards, and wireless signal detection systems.

[0144] This embodiment provides an electromagnetic latent device identification method based on a federated large model, which is applied to a server, sends a preset initial model to multiple clients, receives local model parameters fed back by multiple clients based on the preset initial model, evaluates the attack identification capabilities of multiple local model parameters based on an attack evaluation data set, obtains an attack impact evaluation result, aggregates multiple local model parameters based on the attack impact evaluation result, obtains global model parameters, and sends global model parameters to multiple clients, so that multiple clients determine a target latent device identification model based on the global model parameters, and identify the latent device based on the target latent device identification model. The target latent device identification model obtained through the model training and parameter update process based on the federated large model can learn and identify the signal characteristics of various latent devices, thereby improving the accuracy of electromagnetic latent device identification.

[0145] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the electromagnetic latent device identification method based on the federal large model of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0146] In addition, to achieve the above objectives, the present application also proposes an electromagnetic latent device identification system based on a federated large model, wherein the electromagnetic latent device identification system based on a federated large model includes a plurality of the above-mentioned clients and the above-mentioned server.

[0147] Since the electromagnetic latent device identification system based on the federal large model adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought by the technical solutions of the above embodiments, which will not be described one by one here.

[0148] This application also provides an electromagnetic latent device identification device based on a federated large model, please refer to Figure 5 The electromagnetic latent device identification device based on the federated large model is applied to the client, and the electromagnetic latent device identification device based on the federated large model includes:

[0149] The target model determination module 10 is used to determine the target latent device identification model based on the preset initial model when receiving the preset initial model sent by the server;

[0150] A parameter data acquisition module 20 is used to acquire electromagnetic device parameter data, and determine a risk electromagnetic device list and a distance feature and a signal feature of each risk electromagnetic device in the risk electromagnetic device list based on the electromagnetic device parameter data;

[0151] A latent risk prediction module 30 is used to obtain a current risk electromagnetic device based on the risk electromagnetic device list, and input the distance feature and the signal feature of the current risk electromagnetic device into the target latent device identification model to obtain a latent risk prediction score;

[0152] The latent device identification module 40 is used to determine a target latent device identification result based on the latent risk prediction score.

[0153] This application also provides an electromagnetic latent device identification device based on a federated large model, please refer to Figure 6 The electromagnetic latent device identification device based on the federated large model is applied to the server, and the electromagnetic latent device identification device based on the federated large model includes:

[0154] A model initialization module 50, configured to send a preset initial model to multiple clients, and receive local model parameters fed back by multiple clients based on the preset initial model;

[0155] A poisoning attack defense module 60, configured to evaluate the attack recognition capabilities of the plurality of local model parameters based on an attack evaluation data set to obtain an attack impact evaluation result;

[0156] A global model aggregation module 70, configured to aggregate the plurality of local model parameters based on the attack impact assessment result to obtain a global model parameter;

[0157] The global model distribution module 80 is used to send the global model parameters to the multiple clients, so that the multiple clients determine the target latent device identification model based on the global model parameters and identify the latent device based on the target latent device identification model.

[0158] The electromagnetic latent device identification device based on the federated large model provided by the present application adopts the electromagnetic latent device identification method based on the federated large model in the above embodiment, which can solve the technical problem of how to effectively detect the sneak shooting behavior of electromagnetic hidden devices to prevent the user's privacy from being invaded. Compared with the prior art, the beneficial effects of the electromagnetic latent device identification device based on the federated large model provided by the present application are the same as the beneficial effects of the electromagnetic latent device identification method based on the federated large model provided by the above embodiment, and the other technical features of the electromagnetic latent device identification device based on the federated large model are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0159] The present application provides an electromagnetic latent device identification device based on a federal large model, and the electromagnetic latent device identification device based on a federal large model includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the electromagnetic latent device identification method based on the federal large model in the above-mentioned embodiment one.

[0160] Reference below Figure 7 , which shows a schematic diagram of the structure of an electromagnetic latent device identification device based on a federated large model suitable for implementing the embodiment of the present application. The electromagnetic latent device identification device based on a federated large model in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The electromagnetic latent device identification device based on the federated large model is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0161] like Figure 7As shown, the electromagnetic latent device identification device based on the federated large model may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the electromagnetic latent device identification device based on the federated large model are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the electromagnetic latent device identification device based on the federated large model to communicate wirelessly or wired with other devices to exchange data. Although the electromagnetic latent device identification device based on the federated large model with various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0162] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0163] The electromagnetic latent device identification device based on the federated large model provided by the present application adopts the electromagnetic latent device identification method based on the federated large model in the above embodiment, which can solve the technical problem of how to effectively detect the sneak shooting behavior of electromagnetic hidden devices to prevent the user's privacy from being invaded. Compared with the prior art, the beneficial effects of the electromagnetic latent device identification device based on the federated large model provided by the present application are the same as the beneficial effects of the electromagnetic latent device identification method based on the federated large model provided by the above embodiment, and the other technical features of the electromagnetic latent device identification device based on the federated large model are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0164] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0165] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0166] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the electromagnetic latent device identification method based on the federated large model in the above-mentioned embodiment.

[0167] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0168] The computer-readable storage medium may be included in the electromagnetic latent device identification device based on the federated large model; or may exist independently without being assembled into the electromagnetic latent device identification device based on the federated large model.

[0169] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the electromagnetic latent device identification device based on the federal large model, the electromagnetic latent device identification device based on the federal large model: when receiving the preset initial model sent by the server, determines the target latent device identification model based on the preset initial model; obtains electromagnetic device parameter data, and determines the risk electromagnetic device list and the distance characteristics and signal characteristics of each risk electromagnetic device in the risk electromagnetic device list based on the electromagnetic device parameter data; obtains the current risk electromagnetic device based on the risk electromagnetic device list, and inputs the distance characteristics and the signal characteristics of the current risk electromagnetic device into the target latent device identification model to obtain a latent risk prediction score; determines the target latent device identification result based on the latent risk prediction score.

[0170] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0171] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0172] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0173] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned electromagnetic latent device identification method based on the federated large model, and can solve the technical problem of how to effectively detect the sneak shots of electromagnetic hidden devices to prevent the user's privacy from being invaded. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the electromagnetic latent device identification method based on the federated large model provided by the above-mentioned embodiment, and will not be repeated here.

[0174] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the electromagnetic latent device identification method based on the federated large model as described above.

[0175] The computer program product provided by the present application can solve the technical problem of how to effectively detect the voyeuristic photography of electromagnetic hidden devices to prevent the privacy of users from being violated. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the electromagnetic latent device identification method based on the federated large model provided by the above embodiment, which will not be elaborated here.

[0176] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for identifying electromagnetic latent devices based on a federated large model, characterized in that: The electromagnetic latent device identification method based on the federated large model is applied to the client, and the method includes: When receiving the preset initial model sent by the server, determining the target latent device identification model based on the preset initial model; Acquire electromagnetic device parameter data, and determine a risky electromagnetic device list and a distance feature and a signal feature of each risky electromagnetic device in the risky electromagnetic device list based on the electromagnetic device parameter data; Acquire the current risk electromagnetic device based on the risk electromagnetic device list, and input the distance feature and the signal feature of the current risk electromagnetic device into the target latent device identification model to obtain a latent risk prediction score; A target latent device identification result is determined based on the latent risk prediction score.

2. The method according to claim 1, characterized in that The step of determining a risk electromagnetic device list and a distance feature and a signal feature of each risk electromagnetic device in the risk electromagnetic device list based on the electromagnetic device parameter data comprises: Establishing a risky electromagnetic equipment list based on the electromagnetic equipment parameter data; Acquire distance data, signal receiving power data and signal strength data of each risky electromagnetic device in the risky electromagnetic device list; The distance feature and the signal feature of each risky electromagnetic device in the risky electromagnetic device list are determined based on the distance data, the signal receiving power data, and the signal strength data.

3. The method according to claim 1, characterized in that The step of determining the target latent device identification result based on the latent risk prediction score includes: Collecting the potential risk prediction score of the current risk electromagnetic equipment within a preset period; Determine the median prediction score, the target prediction score number, and the total prediction score number based on the potential risk prediction score within the preset period; The target latent device identification result is determined based on the prediction score median, the target prediction score number, and the total prediction score number.

4. The method according to claim 1, characterized in that After the step of inputting the distance feature and the signal feature into the target latent device identification model to obtain the target latent device identification result, the method further includes: Obtaining a local preset threshold, and determining a target latent device from the risky electromagnetic device list based on the local preset threshold and the target latent device identification result, to obtain a target latent device list; The device behavior data of the target latent device is acquired, and a latent device list is generated based on the target latent device list and the device behavior data, so that the user can search for electromagnetic latent devices according to the latent device list.

5. The method according to claim 1, characterized in that The step of determining the target latent device identification model based on the preset initial model when receiving the preset initial model sent by the server includes: When receiving the preset initial model sent by the server, training the preset initial model based on the local data set to obtain local model parameters, and sending the local model parameters to the server; When the global model parameters fed back by the server based on the local model parameters are received, local aggregate model parameters are determined based on the global model parameters and the local model parameters to obtain a target latent device identification model.

6. A method for identifying electromagnetic latent devices based on a federated large model, characterized in that: The electromagnetic latent device identification method based on the federated large model is applied to the server, and the method includes: Sending a preset initial model to multiple clients, and receiving local model parameters fed back by the multiple clients based on the preset initial model; Evaluate the attack recognition capabilities of the plurality of local model parameters based on the attack evaluation data set to obtain an attack impact evaluation result; Aggregating the plurality of local model parameters based on the attack impact assessment result to obtain a global model parameter; The global model parameters are sent to the plurality of clients, so that the plurality of clients determine a target latent device identification model based on the global model parameters, and identify latent devices based on the target latent device identification model.

7. The method according to claim 6, characterized in that The step of aggregating the plurality of local model parameters based on the attack impact assessment result to obtain a global model parameter comprises: Acquire attack impact index values ​​of a plurality of the local model parameters based on the attack impact assessment result; Acquire a degradation threshold, and determine a candidate model parameter from a plurality of the local model parameters based on the degradation threshold and the attack impact index value; A target model parameter is determined from the plurality of candidate model parameters, and the plurality of target model parameters are aggregated to obtain a global model parameter.

8. The method according to claim 7, characterized in that The step of determining a target model parameter from a plurality of the candidate model parameters and aggregating the plurality of the target model parameters to obtain a global model parameter comprises: Sorting the attack impact index values ​​of the plurality of candidate model parameters to obtain an impact score sorting result; Determine the candidate model parameter with a first preset ranking in the impact score sorting result as a first target model parameter; Determine a second target model parameter from the candidate model parameters of a second preset ranking in the impact score sorting result; A target model parameter is determined based on the first target model parameter and the second target model parameter, and a plurality of the target model parameters are aggregated to obtain a global model parameter.

9. The method according to claim 6, characterized in that The attack assessment data set includes a first preset data set and a second preset data set, the first preset data set is a data set without attack, and the second preset data set is a data set containing attack; The step of evaluating the attack recognition capabilities of the plurality of local model parameters based on the attack evaluation data set to obtain an attack impact evaluation result comprises: Based on the first preset data set and the second preset data set, the preset initial model is trained respectively to obtain a first model and a second model, and the test accuracy of the first model and the second model is obtained to obtain a first accuracy rate; Based on the multiple local model parameters, the first model and the second model are respectively updated, and the test accuracy of the updated first model and the second model is obtained to obtain a second accuracy; The attack recognition capabilities of the plurality of local model parameters are evaluated based on the first accuracy rate and the second accuracy rate to obtain an attack impact evaluation result.

10. An electromagnetic latent device identification system based on a federated large model, characterized in that: The electromagnetic latent device identification system based on the federated large model includes: a server and multiple clients, wherein the server executes the steps of the electromagnetic latent device identification method based on the federated large model as described in any one of claims 1 to 5 above, and the multiple clients execute the steps of the electromagnetic latent device identification method based on the federated large model as described in any one of claims 6 to 9 above.

Images