CN116028908A - Continuous identity authentication method and related device based on incremental learning and meta learning - Google Patents

Abstract

A continuous identity authentication method and related devices based on incremental learning and meta-learning, including: visualizing sensor data to determine whether there is drift in sensor data; constructing meta-learning and incremental learning based on offline registration phase and online authentication phase The authentication framework; according to the offline registration to obtain the characteristics of each sensor time dimension and the correlation characteristics between their different dimensions, according to the online authentication stage, learn from the acquired data, and update the online model; the updated model Experimental verification is carried out to obtain the model capability. The identity authentication technology framework MetaAuth based on touch screen behavior solves the problem in the field of long-term touch screen identity authentication. Secondly, an online update mechanism AMUM based on meta-learning is designed for incremental learning in the scenario of long-term continuous identity authentication, so that to improve the stability of the model.

Description

Continuous Identity Authentication Method and Related Devices Based on Incremental Learning and Meta-Learning

technical field

The invention relates to the technical field of deep learning, in particular to a continuous identity authentication method and related devices based on incremental learning and meta-learning.

Background technique

In today's society, mobile phones and tablet computers have become the most basic tools for social networking, entertainment, communication, and e-commerce, and are an indispensable part of today's society. At present, 83.72% of the people in the world have smart phones, and with the continuous development of technology, the computing power and storage capacity of smart phones are also being continuously developed, so that more and more apps are developed and invested Use, such as social networking, online shopping, types of financial loans, etc. According to statistics, nearly 60% of Internet users between the ages of 16 and 64 have done online shopping, and the current annual online shopping amount is about 3.8 trillion, with an average increase of 18%. It can be seen that More and more payment behaviors are carried out on mobile phones.

People's high reliance on smartphones has also led to widespread concern about everyone's private data in their smartphones. Personal privacy data includes bank accounts, health, employment, etc. in mobile phones, as well as confidential data related to some governments, which are all recorded in mobile phones. Under normal circumstances, these data are protected and only the user can obtain them. However, according to data, nearly one-third of users' mobile phones have been stolen, and one-tenth of them believe that the privacy of their mobile phones has been violated to varying degrees. Therefore, the leakage of mobile phone privacy caused by various reasons has become a major problem that plagues all users and major mobile phone manufacturers.

In order to overcome this problem, in order to solve the mobile phone privacy problem and protect the private data of the mobile phone, a large number of identity authentication problems have been carried out. The mobile phone identity authentication problem mainly refers to that the mobile terminal verifies the identity of the user through a relatively short identity indicator. This allows the user to continue access or deny access. With so many years of research on the problem, there are mainly the following three authentication methods: knowledge-based authentication, biometric-based identity authentication, and behavior-based identity authentication; knowledge-based authentication is a It is a relatively traditional method, and has been widely recognized, among which a series of methods such as digital password and graphic unlocking are more representative. Because these methods are relatively simple and require relatively low learning costs, they have been widely promoted. Some studies have shown that if it is a relatively complex password, the memory cost of the user will be greatly increased, and if some relatively simple passwords such as "abc" or "12345678" are selected, the security performance will be insufficient, and users often use mobile phones. He admitted that the stains on the screen or the oil of the skin are used to decipher, so there are obvious hidden dangers in the traditional method.

Contents of the invention

The purpose of the present invention is to provide a continuous identity authentication method and related devices based on incremental learning and meta-learning, so as to solve the problems of security and high memory cost in traditional methods.

To achieve the above object, the present invention adopts the following technical solutions:

A continuous identity authentication method based on incremental learning and meta-learning, including:

Collect the sensor data of the mobile phone during use, visualize the sensor data, and judge whether the sensor data has drift phenomenon;

In the event of drift, construct MetaAuth, an authentication framework based on meta-learning and incremental learning that includes offline registration and online authentication stages;

Establish a meta-learning model, obtain the characteristics of each sensor's time dimension and the correlation characteristics between different dimensions according to offline registration, learn from the acquired data according to the online authentication stage, and perform online model update;

The updated model is tested and verified to obtain the model capability.

Further, collect the sensor data of the mobile phone during use, visualize the sensor data, and judge whether the sensor data has drift phenomenon:

Randomly extract the sensor data of the user's mobile phone during a certain period of time, and divide the period into early, middle and late periods, and then perform feature extraction on the data. After feature extraction, use the t-SNE method to separate the features in two-dimensional space It is visualized to show the similarities and changes in the characteristics between users. When the characteristic data between users changes and the distance is getting farther and farther away, drifting occurs.

Further, the offline registration phase:

Data collection: collect data by selecting the user's touch screen behavior, when the screen lights up, the data collected by the sensor is used as a positive sample, and the data of other users is used as a negative sample for training;

Data preprocessing stage: First, use a fixed-length sliding window to divide the data into a series of repeated time periods, which are used as the basic unit of authentication, and then continue to divide each time period into disjoint small time periods segment, called the time source, and then extract local features from the time source;

Continuous collector part: build a dual-channel deep learning model 2-GCTN model, extract the features of each sensor's time dimension and the correlation features between their different dimensions.

Further, the online certification stage:

At this stage, the improved meta-learning method AMUM is used for online model update, which consists of two parts: 1. Drift detector; 2. Improved MAML update strategy; in the detection process, the detection time window can be based on time and data changes, and will also output a drift signal according to the prediction results in the window. If the probability of wrong line error exceeds the set threshold, it will give a judgment of drift.

Further, the specific process:

First obtain experimental data, including sensor three-dimensional data from accelerometers, gyroscopes and magnetometers;

Set up evaluation indicators: Use indicators to measure the accuracy of identity authentication scenarios, including:

Equal error rate: The false acceptance rate is equal to the value of the false rejection rate, and the equal error rate can be affected by changing the threshold of the prediction score;

False acceptance rate: defined as the number of falsely accepted illegitimate samples compared to the number of all illegitimate test samples; a high false acceptance rate indicates a high rate of false separation of intruder samples;

False rejection rate: defined as the number of valid samples falsely rejected versus all valid test samples; a high false rejection rate indicates poor recognition of legitimate user samples;

Certification accuracy variance: A variance measure that indicates the degree of dispersion of certification results over a long period of time; it is calculated from the average of the square of the difference between the equal error rate of each certification and the average of the equal error rate; the low variance reflects the authenticity of the certification model. Stable predictive ability;

Memory usage: Indicates the memory size occupied by model training and updating;

Time Consumption: Indicates the time spent on model training and updating.

Further, experimental verification:

Evaluate the authentication framework 2-GCTN through the M users in the data set, and train the model and conduct incremental tests for each user individually, and set the mobile phone corresponding to each user as a legal user, while setting the rest M-1 are set as illegal users; for the positive sample data, select the sensor data from each user in the order of time, and divide the data into 20 parts, corresponding to 20 incremental learning tasks; negative samples, Randomly selected among the remaining M-1 users; in the training process, each training and test is defined as a task, and the data block of the next cycle will first be used to test the model, and then the model training and updating.

Further, specific steps:

Sort all processed data according to time and divide them into the same data block T;

Each data block will be the user's positive sample data set, there are T-1 tasks, and each task contains two periods of data D _i and D _i+1 ;

Use D _i as the training set to update the model, and then use D _i+1 to test its ability;

Repeat the above steps for each user.

Further, a continuous identity authentication system based on incremental learning and meta-learning, including:

The data processing and judging module is used to collect sensor data during the use of the mobile phone, visualize the sensor data, and judge whether the sensor data has drift phenomenon;

The framework building module is used for the drift phenomenon, and then constructs MetaAuth, an authentication framework based on meta-learning and incremental learning, including offline registration phase and online authentication phase;

The model building module is used to establish a meta-learning model, obtain the characteristics of each sensor time dimension and the correlation characteristics between different dimensions according to the offline registration, and learn from the obtained data according to the online certification stage, and perform an online model renew;

The verification module is used to test and verify the updated model to obtain the model capability.

Further, a computer device includes a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the computer program, incremental learning and meta-based Learn the steps of the persistent authentication method.

Further, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the continuous identity authentication method based on incremental learning and meta-learning are realized.

Compared with the prior art, the present invention has the following technical effects:

The present invention is based on deep learning and meta-learning, and does not add any additional restrictive conditions in the collection process, so as to fully reflect the real mobile phone usage of users. Secondly, a dual-channel deep learning model 2-GCTN is designed to facilitate the use of data association information between multiple sensors. This model can obtain data association information between different dimensions and can also significantly improve the ability to represent user behavior under multi-sensor data. Finally, an incremental learning mechanism AMUM based on meta-learning technology is proposed. This mechanism can use the knowledge learned by the model and new data to update in real time, which fully reflects the effectiveness of the model. The identity authentication technology framework MetaAuth based on touch screen behavior solves the problem in the field of long-term touch screen identity authentication. Secondly, an online update mechanism AMUM based on meta-learning is designed for incremental learning in the scenario of long-term continuous identity authentication, so that to improve the stability of the model.

Description of drawings

Fig. 1 is a schematic diagram of visualization of sensor data in the present invention.

Fig. 2 is a schematic diagram of an authentication framework based on meta-learning and incremental learning in the present invention.

Fig. 3 is a schematic diagram of the long-term authentication process of the present invention.

Figure 4 is an experimental diagram of the effect of the meta-learning mechanism on the long-term certification of the model.

Figure 5 is an experimental diagram of using adaptive-MAML as an incremental update mechanism in MetaAuth.

Figure 6 is a comparison chart of the capabilities of different models for identity authentication in sensor data.

Detailed ways

The embodiment of the invention of this paper and the experimental steps will be described in detail below:

Step 1: Visualize the sensor data first, randomly select 5 users, and collect the sensor data in the Taobao app. The total duration is divided into 2 months, and the two months can also be divided into early, middle and late stages .

The second step: After the features are extracted, the features are visualized in two-dimensional space by the (t-SNE) method. It can show the similarities and changes in the characteristics of these five users. It can be clearly seen from Figure 1: In the early stage, the characteristic data of No. 3, No. 4 and No. 5 are relatively similar, and No. 5 is far from No. 1. In the mid-term, the original characteristics did not change much, but the distance between No. 1 and No. 4 changed and became farther and farther away. And in the later stage, the distance between No. 3, No. 4, No. 5 and them has become farther and farther. Therefore, it can be seen that the user's behavior data will definitely change over time, that is, there will be drift.

Step 3: In order to solve the problem of performance degradation in long-term identity authentication, an authentication framework MetaAuth based on meta-learning and incremental learning is proposed. The overall framework is shown in Figure 2, which includes two important components, namely the offline registration part and the online authentication part.

The offline registration stage is mainly divided into three steps:

1. Data collection stage: The main way is to collect data through touch screen behavior. As long as the screen is on, the data collected by the sensor is used as a positive sample, and the data of other users is also used as a negative sample for training.

2. Data preprocessing stage: In order to effectively extract the characteristics of time series data, first use a fixed-length sliding window to divide the data into a series of time periods that will appear repeatedly, and use it as the basic unit of authentication, and then divide each The time segment continues to be divided into disjoint small time segments, which are called time sources, and then local features can be extracted from the time sources.

3. Continuous collector part: It needs to be able to provide some more accurate representations of time series data scenarios. Therefore, a new 2-GCTN model is designed, which can extract the features of each sensor time dimension and the correlation features between their different dimensions.

And in the online authentication stage: In this process, the authenticator can continuously learn from the new data, and improve the previous method to update the online model. The ADWIN method can be used as a drift detector to reduce the resource consumption of the mobile terminal. , to better update the model. Next, a meta-learning model will be designed, which can learn the initialization parameters of the model and the learning rate of model training.

In this stage, the improved meta-learning method AMUM is used for online model update. It mainly consists of two parts: 1. Drift detector. 2. Improved MAML update strategy. It is entirely possible to use drift detection technology to improve the efficiency of model updating and its rationality.

During the detection process, the detection time window can be changed according to the change of time and data, and the drift signal will be output according to the prediction results in the window. If the probability of wrong line error exceeds the set threshold, that is Will give the judgment of drift. The second MAML model can improve the efficiency of the model by continuously optimizing the initial parameters of the basic model in task learning.

Specific experimental process: Step 1: First obtain experimental data, mainly including three-dimensional sensor data from accelerometers, gyroscopes and magnetometers.

Step 2: Set up evaluation indicators: The accuracy of identity authentication scenarios can be measured through the following indicators, mainly including:

1. Equal Error Rate (EER): The false acceptance rate (FAR) is equal to the value of the false rejection rate (FRR). The equal error rate can be affected by changing the threshold of the prediction score.

2. False Acceptance Rate (FAR): Defined as the number of falsely accepted illegal samples and the number of all illegal test samples. A higher FAR indicates a higher rate of missegmentation of invader samples.

3. False Rejection Rate (FRR): Defined as the number of valid samples that are falsely rejected and the number of all valid test samples. Higher FRR indicates poorer recognition of legitimate user samples.

4. Certification accuracy variance: A variance measure that indicates the degree of dispersion of certification results in different time periods over a long period of time. It is calculated as the average of the square of the difference between the EER for each certification and the average EER. The lower variance reflects the stable predictive power of the certified model.

5. Memory usage: Indicates the memory size occupied by model training and updating. Lower memory usage reflects less consumption of smartphone resources.

6. Time consumption: Indicates the time spent on model training and updating. Less training time means faster model updates.

Step 3: Experiment. In the experiment, 40 users in the data set are used to evaluate the authentication framework 2-GCTN, and each user is individually trained and incrementally tested. Set each user's corresponding mobile phone as a legal user, and at the same time set the remaining 39 people as illegal users. For the data of the positive sample, the data of the sensor can be selected from each user in the order of time, and the data is divided into 20 parts, corresponding to 20 incremental learning tasks. As for the negative samples, they are randomly selected among the remaining 39 users. In the training process, each training and test can be defined as a task. Therefore, the data block of each next cycle will first be used to test the model, and then the model will be trained and updated.

Specific steps

1. Sort all processed data according to time and divide them into the same data block T.

Each data block will be a positive sample data set of users. So correspondingly, there will be T-1 tasks, and each task Taski contains two periods of data Di and Di+1.

2. Use Di as the training set to update the model, and then use Di+1 to test its ability.

3. Repeat steps 1 and 2 for each user.

In this experiment, a systematic experiment was carried out to analyze the capability of the model, its validity and long-term authentication capability.

First, analyze the problem of model accuracy decline when different models are used by users for a long time. As shown in Figure 3, in the long-term certification process, LSTM, SVM, one-SVM and 2-GCTN models are used, and the accuracy of these four models has declined. The main reason is that the increase in the complexity of the environment leads to changes in the behavior of users using mobile phones, which in turn leads to changes in the distribution of data. Secondly, the models LSTM and 2-GCTN using temporal characteristics have better characteristics initially, but as the data distribution changes, the accuracy drops by a large proportion compared with the remaining models.

Then the impact of the meta-learning mechanism on the long-term certification of the model is analyzed. The experimental results are shown in Figure 4

MetaAuth represents model updates using the AMUM mechanism proposed in this paper. It can be concluded from the figure that when there is no

When the Adaptive-MAML mechanism is used, the accuracy of the long-term continuous authentication model will decrease, and the time to update the model will increase. Because the meta-learning mechanism can get better initialized parameters, it is better for the model to learn. Therefore, it can be seen that the improved meta-learning mechanism can improve the accuracy and model efficiency of the model in the case of long-term identity authentication.

Then discuss the effectiveness of the drift detection mechanism AMUM. The main method of the new incremental update mechanism AMUM is to use the ADWIN method to detect the drift of the new batch of data, and then choose whether to update the model according to the detected results, which will not only make the MetaAuth structure update at a reasonable frequency model and become more efficient. The traditional AMUM is used in combination with 2-GCTN to compare the model capabilities of AMUM. They are Retrain-2-GCTN and MetaAuth respectively. In MetaAuth, using adaptive-MAML as an incremental update mechanism, only new data is used to update the model. The experimental results are shown in Figure 5. It can be seen through experiments that the EER of MetaAuth is very low, that is, it has a high average authentication accuracy, so the stability of the model is relatively good. Because the model can use the information in the historical data through the mata-leanring. It can make the model better learn the user's behavioral characteristics. Secondly, you can also get that the update time of MetaAuth is relatively short, and the material rate has also been improved. So it can be concluded that MetaAuth is more practical and real for mobile phone scenarios.

Finally, long-term certification capabilities are discussed. In the experiment, the ability of different models to authenticate the identity in the use of sensor data is compared. As shown in Fig. 6, compared with conventional methods, the proposed model has better accuracy, i.e., the smallest EER, while MetaAuth also has the best stability in the case of long-term incremental updates. Secondly, there is also a violin diagram of memory usage. It is not difficult to find that the memory usage of MetaAuth is also the lowest, which can reflect that the resource consumption in the authentication process is the least. Finally, it can also be seen that the training time of MetaAuth is also the shortest among all models, which can reflect the real-time nature of MetaAuth and is the most suitable model for real scenarios.

In yet another embodiment of the present invention, a continuous identity authentication system based on incremental learning and meta-learning is provided, which can be used to implement the above-mentioned continuous identity authentication method based on incremental learning and meta-learning. Specifically, the system include:

The data processing and judging module is used to collect sensor data during the use of the mobile phone, visualize the sensor data, and judge whether the sensor data has drift phenomenon;

The framework building module is used for the drift phenomenon, and then constructs MetaAuth, an authentication framework based on meta-learning and incremental learning, including offline registration phase and online authentication phase;

The model building module is used to establish a meta-learning model, obtain the characteristics of each sensor time dimension and the correlation characteristics between different dimensions according to the offline registration, and learn from the obtained data according to the online certification stage, and perform an online model renew;

The verification module is used to test and verify the updated model to obtain the model capability.

The division of modules in the embodiments of the present invention is schematic, and is only a logical function division. In actual implementation, there may be other division methods. In addition, each functional module in each embodiment of the present invention can be integrated into a processing In the controller, it can also be physically present separately, or two or more modules can be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software function modules.

In yet another embodiment of the present invention, a computer device is provided, the computer device includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the computer The program instructions stored in the storage medium. The processor may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable GateArray, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, specifically for Load and execute one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding functions; the processor described in the embodiment of the present invention can be used for the operation of a continuous identity authentication method based on incremental learning and meta-learning.

In yet another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device for storing programs and data. . It can be understood that the computer-readable storage medium here may include a built-in storage medium in the computer device, and of course may also include an extended storage medium supported by the computer device. The computer-readable storage medium provides storage space, and the storage space stores the operating system of the terminal. Moreover, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions may be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one magnetic disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor, so as to realize the corresponding steps in the above-mentioned embodiment about a continuous identity authentication method based on incremental learning and meta-learning.

Those skilled in the art should understand that the embodiments of the present invention may be provided as methods, systems, or computer program products. Accordingly, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the present invention can still be Any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention shall fall within the protection scope of the claims of the present invention.

Claims (10)

1. The continuous identity authentication method based on incremental learning and meta learning is characterized by comprising the following steps:

collecting sensor data of the mobile phone in the using process, performing visual operation on the sensor data, and judging whether the sensor data has a drifting phenomenon or not;

when drift phenomenon occurs, an authentication framework MetaAuth comprising an offline registration stage and an online authentication stage and based on meta learning and incremental learning is constructed;

establishing a meta learning model, acquiring the characteristics of the time dimension of each sensor and the association characteristics among different dimensions of each sensor according to offline registration, learning from the acquired data according to an online authentication stage, and updating the online model;

and performing experimental verification on the updated model to obtain the model capacity.

2. The continuous identity authentication method based on incremental learning and meta learning according to claim 1, wherein the method is characterized in that sensor data of a mobile phone in a use process is collected, the sensor data is subjected to visual operation, and whether the sensor data has a drift phenomenon is judged:

the method comprises the steps of randomly extracting sensor data when a mobile phone of a user is used in a certain period, dividing the period into a front period, a middle period and a later period, then extracting features of the data, respectively visualizing the features in a two-dimensional space through a t-SNE method after the features are extracted, showing the conditions of similar and changing features among the users, and when the feature data among the users change and the distance is further and further away, drifting occurs.

3. The continuous identity authentication method based on incremental learning and meta learning of claim 1, wherein the offline registration stage:

and (3) data acquisition: data acquisition is carried out by selecting the touch screen behaviors of the user, when the screen is lightened, the data acquired by the sensor is taken as a positive sample, and the data of other users are taken as a negative sample for training;

data preprocessing: firstly, dividing data into a series of time periods with repetition by using a fixed-length sliding window, taking the time periods as a basic unit of authentication, then continuously dividing each time period into disjoint small time periods, namely time sources, and extracting local features from the time sources;

continuous collector part: and establishing a two-channel deep learning model 2-GCTN model, and extracting the characteristics of each sensor time dimension and the associated characteristics among different dimensions.

4. The continuous identity authentication method based on incremental learning and meta learning of claim 1, wherein the online authentication phase:

the on-line model update is carried out at this stage by using an improved meta learning method AMUM, which comprises two parts respectively consisting of: 1. a drift detector; 2. improved MAML update policies; in the detection process, the detected time window can be changed according to the time and the change of the data, a drift signal is output according to a prediction result in the window, and if the probability of detecting the error line exceeds a set threshold value, the judgment of drift is given.

5. The continuous identity authentication method based on incremental learning and meta learning according to claim 4, wherein the specific process is as follows:

firstly, acquiring experimental data comprising three-dimensional data of a sensor from an accelerometer, a gyroscope and a magnetometer;

setting an evaluation index: the accuracy in the scene of identity authentication is measured through indexes, wherein the indexes comprise:

equal error rate: the false acceptance rate is equal to the value of the false rejection rate, and the equal false rate can be affected by changing the threshold of the prediction score;

error acceptance rate: defining the number of illegal samples received by error and the number of all illegal test samples; a high false acceptance rate indicates a high false separation rate of the intruder sample;

false rejection rate: defining the number of valid samples and the number of all valid test samples which are refused by mistake; a high false reject rate indicates poor identification of legitimate user samples;

authentication accuracy variance: a variance measure representing the degree of dispersion of authentication results over a long period of time; it is calculated from the average of the squares of the differences between the equal error rate and the average of the equal error rates for each authentication; the low variance reflects the stable predictive ability of the authentication model;

and (3) using a memory: representing the memory size occupied by model training and updating;

time consumption: representing the time it takes for the model to train and update.

6. A continuous identity authentication method based on incremental learning and meta learning as claimed in claim 3, wherein the test verifies that:

evaluating an authentication framework 2-GCTN through M users in a data set, independently training a model and performing incremental test on each user, setting a mobile phone corresponding to each user as a legal user, and setting the rest M-1 as illegal users; selecting data of a sensor from each user according to the sequence of time for the data of the positive sample, and dividing the data into 20 copies, wherein the 20 copies correspond to the incremental learning tasks respectively; negative samples, randomly extracted among the remaining M-1 users; in the training process, each training and test is defined as a task, and each time the data block of the next period is firstly used for testing the model, and then the training and updating of the model are carried out.

7. The continuous identity authentication method based on incremental learning and meta learning according to claim 6, wherein the specific steps are as follows:

ordering all processed data according to time and dividing the processed data into identical data blocks T;

each data block will be a positive sample data set of the user, there are T-1 tasks, each task containing data D for two periods _i D (D) _i+1 ；

Using D _i Updating the model as a training set, then using D _i+1 Testing its ability;

the above steps are repeated for each user.

8. A continuous identity authentication system based on incremental learning and meta learning, comprising:

the data processing judging module is used for collecting sensor data of the mobile phone in the using process, carrying out visual operation on the sensor data and judging whether the sensor data has a drifting phenomenon or not;

the framework building module is used for building an authentication framework MetaAuth comprising an offline registration stage and an online authentication stage and based on meta learning and incremental learning when a drift phenomenon occurs;

the model building module is used for building a meta learning model, acquiring the characteristics of the time dimension of each sensor and the association characteristics among different dimensions of the sensor according to offline registration, learning from the acquired data according to an online authentication stage, and updating the online model;

and the verification module is used for carrying out experimental verification on the updated model to obtain the model capacity.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the steps of the continuous identity authentication method based on incremental learning and meta learning as claimed in any one of claims 1 to 7.

10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the continuous identity authentication method based on incremental learning and meta learning as claimed in any one of claims 1 to 7.

Images