CN110363120B - Intelligent terminal touch authentication method and system based on vibration signal - Google Patents

Abstract

When the intelligent terminal detects finger touch, a specific vibration signal is actively generated, the vibration signal is collected through an IMU sensor, and biological characteristics, behavior characteristics and independent touch behavior characteristics are respectively extracted from the received vibration signal; then, a neural network based on a twin network (simple network) architecture is adopted to classify the biological characteristics, and behavior-independent intelligent terminal touch authentication is realized. The invention utilizes the intelligent terminal to actively send out the vibration signal and capture the biological characteristics of the touch finger so as to identify different users, and the method can not depend on the behavior characteristics of the touch.

Description

Smart terminal touch authentication method and system based on vibration signal

technical field

The invention relates to a technology in the field of user authentication, in particular to a behavior-independent smart terminal touch authentication method and system based on vibration signals.

Background technique

In recent years, with the continuous development of smart mobile terminals, the private and sensitive data stored in smart mobile terminal devices, especially smart terminals, have been increasing, resulting in huge hidden dangers of privacy and data leakage. Therefore, a safe and effective intelligent end-user authentication system has become a current research hotspot.

The existing work is mainly divided into three categories. The first category of work is a password-based intelligent terminal user authentication system, which mainly includes digital passwords, gesture passwords and graphic passwords. The common problem of this type of user authentication system is that it is easy to be peeped, and there are great security risks; the second type of work is based on some external biometric features of users, mainly including fingerprints, face recognition, iris recognition, voiceprint identification etc. The common problems of this type of work are that it often requires additional equipment and is sensitive to the environment (humidity, light, noise, etc.), and is vulnerable to replay attacks; the third type of work is to use the behavioral characteristics of fingers to touch the screen. Authentication mainly includes touch position, touch strength, touch time, etc. The common problem of this kind of work is that it is vulnerable to mimic attack and has poor user experience.

SUMMARY OF THE INVENTION

Aiming at the above-mentioned deficiencies in the prior art, the present invention proposes a method and system for smart terminal touch authentication based on vibration signals, which utilizes the smart terminal to actively send out vibration signals to capture the biometric features of the touching fingers to identify different users, thereby enabling the identification of different users. Behavioural characteristics that do not depend on touch (e.g. touch location, strength, duration, etc.).

The present invention is achieved through the following technical solutions:

The invention relates to a touch authentication method for an intelligent terminal based on a vibration signal. When the intelligent terminal detects a finger touch, it actively generates a specific vibration signal and collects the vibration signal through an IMU sensor, and extracts biological features, behavioral features and Independent touch behavior features; then use a neural network based on a siamese network architecture to classify biometric features to achieve behavior-independent touch authentication for smart terminals.

The specific vibration signal is generated by adjusting the switch signal, and each cycle includes a vibration impulse signal, a motor activation signal and a residual vibration elimination part.

When collecting the vibration signal, preferably, the received vibration signal is firstly divided into a transient vibration stage, a stable vibration stage and a decay stage.

作为触碰力度特征，其中：f_r为振动信号的共振频率，Δf定义了能量计算带宽，f(t)是原始信号，在中对应暂态振动阶段的振动信号。The extraction includes feature extraction based on wavelet change, feature extraction based on cepstral transformation, touch location feature extraction, and touch strength feature extraction, specifically: extracting the time spectrum obtained after wavelet transformation from the transient vibration stage. , Extract the cepstrum from the stable vibration stage, extract the peak value of the audio signal and its corresponding time from the microphone of the mobile terminal as the touch position feature, and extract the energy value of the vibration signal near the resonance frequency

As the touch strength feature, where: f _r is the resonance frequency of the vibration signal, Δf defines the energy calculation bandwidth, and f(t) is the original signal, which corresponds to the vibration signal in the transient vibration stage.

The touch behavior characteristics specifically refer to: when the user uses the intelligent terminal, the contact between the user's finger and the intelligent terminal includes the behavior characteristics of the contact with the front, side and back of the intelligent device, including the touch position and the touch strength. feature.

The neural network based on the twin network architecture includes two parallel sub-networks, each sub-network has the same structure and weights, and respectively according to the eigenvalues X ₁ , X ₂ extracted from the input vibration signal after passing through the sub-networks. Obtain the corresponding feature representations C _W (X ₁ ), C _W (X ₂ ), and then obtain the distance between the two feature representations: D _W (X ₁ , X ₂ )=||C _w (X ₁ )-C _w ( X ₂ )||.

The sub-network is a time-delayed neural network (TDNN) structure, including two convolutional layers and a fully connected layer. Within each convolutional layer is a basic convolutional module as the core, a batch normalization (BN) layer to handle the gradient problem and a ReLU layer as the activation function.

The training samples of the neural network have the same label, that is, come from the same user and the touch behavior characteristics are not similar, that is, the similarity of the touch behavior characteristics is lower than the threshold.

The touch behavior feature similarity is measured by the Pearson correlation coefficient:

当

大于预设阈值h，则认为两个样本具有相似的触碰行为特征，否则认为二者的触碰行为特征不相似。

when

If it is greater than the preset threshold h, it is considered that the two samples have similar touch behavior characteristics; otherwise, the two samples are considered to have dissimilar touch behavior characteristics.

The classification refers to: judging whether the two inputs belong to a unified user according to the distance represented by the features corresponding to the two inputs obtained by the neural network, so as to realize the judgment of whether the touch behavior comes from a registered user.

technical effect

Compared with the prior art, the present invention realizes accurate user authentication without relying on any external device by constructing a behavior-independent touch authentication system based on an intelligent mobile terminal. In the system construction process, the present invention combines the sensor-richness and easy development of the intelligent mobile terminal, utilizes the vibration motor and the IMU sensor on the intelligent mobile terminal, and integrates the relevant technology of vibration signal processing and the relevant knowledge of the twin neural network, and finally A touch authentication system with good robustness and high precision is realized, which can effectively resist replay attack and mimic attack, and has no requirements on the user's touch method, which greatly improves the user's ability to touch. experience.

Description of drawings

Figure 1 is a schematic diagram of different touching methods of human hands and mobile phones;

Fig. 2 is the system structure diagram of the present invention;

Fig. 3 is a schematic diagram of vibration signal design;

4 is a schematic diagram of vibration signal segmentation;

Fig. 5 is a three-dimensional schematic diagram of wavelet transform of vibration signal;

Fig. 6 is a three-dimensional schematic diagram of cepstral transformation of vibration signal;

7 is a schematic diagram of touch position feature extraction;

8 is a schematic diagram of touch force feature extraction;

Fig. 9 is a schematic diagram of the selection strategy of twin network training data;

Fig. 10 is the twin network model structure diagram of the present invention;

Figure 11 is a confusion matrix diagram of user authentication accuracy;

Figure 12 shows the success rate of imitation attacks in different environments;

Figure 13 shows the success rate of replay attacks in different environments.

Detailed ways

As shown in Figure 1, there are different ways for human hands to touch the smart terminal. Therefore, it is hoped that by capturing the biometric features of the finger that is in contact with the smart terminal, it is possible to authenticate and identify the touched user under various touch methods. , which implements behavior-independent touch user authentication. To achieve this, the biometrics of the fingers in contact with the smart terminal are captured by analyzing the effect of finger touch on the actively propagated vibration signal using the vibration motor active and IMU sensors on the smart terminal.

This embodiment includes two stages of registration and login:

A registration stage: When the user touches the smartphone, it first actively generates a specific vibration signal by using the vibration motor built into the smart terminal, and at the same time collects the vibration signal through the IMU sensor. After the signal is collected, it is further processed for calibration and segmentation to obtain signal segments containing user information. Based on these signal fragments, signal features (hybrid biometric and behavioral features) and independent touch behavioral features are extracted, respectively. Then, based on the extracted signal features and behavioral features, the influence of the touch behavioral features on the extracted signal features is weakened through the neural network architecture based on the siamese network and the corresponding training sample selection strategy, and further training has nothing to do with the behavior. The touch user authentication classifier model.

B login stage: first capture the finger touch action and actively send and receive specific vibration signals, then extract the signal features of the vibration signal in the way of the registration stage, and use the extracted signal features and the classifier model obtained in the registration stage to get Authenticate the touch. This login phase has no requirements for finger touch.

As shown in FIG. 2 , this embodiment relates to a system for implementing the above method, including: a vibration signal processing module, a signal feature extraction module, a behavior-independent classifier, and an authentication module, wherein: the vibration signal processing module is connected to the signal feature extraction module And transmit the preprocessed vibration signal information, the signal feature extraction module is connected with the behavior independent classifier and transmits the signal features and behavior feature information extracted from the preprocessed vibration signal, the behavior independent classifier is connected with the authentication module and transmitted through The information of the trained classifier model, the authentication module receives the signal features extracted in the authentication phase, connects and transmits it to the trained classifier model to obtain the authentication result of the registrant.

As shown in Figure 3, the specific vibration signal is generated by adjusting the switch signal, specifically: firstly generate a very short vibration impulse signal (<0.1ms), and pause for about 10ms for signal alignment and subsequent touch Touch the position, then activate the vibration motor for 90ms to make it reach the stable vibration stage through the transient vibration stage, and then stop for 10ms to eliminate the residual vibration, and form a 500ms vibration signal by activating and stopping 5 times. Due to the short duration, such a vibration signal Almost no loss of user experience.

The collection of vibration signals, that is, using the IMU sensor to receive, preferably performs time alignment at the sending end and the receiving end. Since the transmission speed of the vibration signal in the smart terminal is generally greater than 2000m/s, it only takes less than 0.05ms to spread through a normal-sized smart terminal. The sampling frequency of the IMU sensor is often less than 1000Hz, corresponding to a time resolution of 1ms, which is much greater than the transmission time of the vibration signal. Therefore, the time corresponding to the sampling point of the vibration signal received by the IMU sensor is considered as the transmission time of the vibration signal, so Complete signal alignment.

For the collection of vibration signals, in order to utilize the information contained in different stages of the vibration signals, the received vibration signals are divided as shown in Figure 4, including:

1) Based on the information and signal alignment of the transmitted signal, the signal is divided into a 90ms vibration phase (A+B phase in Figure 4) and a 10ms decay phase (C phase in Figure 4).

2) Using the variance of the frequency change of the vibration signal as the threshold, further distinguish the transient vibration stage (stage A in Figure 4) and the stable vibration stage (stage B in Figure 4) of the vibration signal. That is, when the variance of the frequency change of the vibration signal is greater than the threshold h, it is considered to be in the transient vibration stage, and when the variance of the vibration signal frequency change is less than the threshold h, it is considered to be in the stable vibration stage.

The extraction includes feature extraction based on wavelet change, feature extraction based on cepstral transformation, touch location feature, and touch strength feature extraction, and the specific steps include:

1) Extract features from the transient vibration phase of the received signal. Since the frequency of the vibration signal is constantly changing during the transient vibration phase, the original vibration signal needs to be transformed to obtain good time and frequency domain resolutions at the same time. Therefore, the continuous wavelet transform (CWT) is used to transform the vibration signal in the transient vibration stage:

其中：CWT_f(a,τ)是获得的时频谱；f(t)是原始信号，在中对应暂态振动阶段的振动信号；ψ_a,τ(t)是小波基函数，选择Morlet函数作为小波基函数以达到更好的时域和频域解析度，将小波变换后得到的时频谱作为暂态振动阶段的信号特征。

Among them: CWT _f (a,τ) is the obtained time spectrum; f(t) is the original signal, corresponding to the vibration signal in the transient vibration stage; ψ _a,τ (t) is the wavelet basis function, and the Morlet function is selected as The wavelet basis function is used to achieve better resolution in the time and frequency domains, and the time spectrum obtained after wavelet transformation is used as the signal characteristic of the transient vibration stage.

As shown in Figure 5, it is a three-dimensional time-spectrogram of a transient vibration signal after wavelet change. It can be seen that its resolution in the time domain and frequency domain is very high, showing the details of the change of vibration frequency with time, and also implicitly Details of the impact of touching your finger on the vibration signal.

2) Feature extraction based on cepstral transformation: At the same time, features are extracted from the stable vibration stage of the received signal. In the stable vibration stage, the vibration signal is stabilized at the resonant frequency, masking other relatively weak frequency components, that is, the side-band (side-band). band frequencies), while the effect of touching a finger on the stable vibration phase is often reflected in the sidebands. Therefore, in order to show the frequency components of the sidebands, the cepstral transformation is used to obtain various frequency components including the sidebands in the stable vibration phase signal: C _y (q)=F ^-1 (log S _y (f(t)) ), where: C _y (q) is the obtained cepstrum; S _y (f(t)) is the power spectral density (PSD) of the signal; F ^-1 corresponds to the inverse Fourier transform (IFFT), and the obtained inverse Spectrum as a signal characteristic of the stable vibration phase.

As shown in FIG. 6 , for the cepstrum corresponding to the vibration signals touched by two different users, it can be seen that the two users exhibit completely different characteristics on the cepstrum.

3) Touch position feature: In order to obtain the user's touch position information, the time of arrival (ToA) of the signal is used for measurement. Since the transmission speed of the vibration signal is too fast (>200m/s) and the sampling frequency of the IMU is too low (<1000Hz), the microphone on the smart terminal is used to capture the sound wave signal generated at the same time as the vibration signal to extract the touch position feature. As shown in Figure 7, the vibration signal mainly reaches the microphone through three paths, the first is transmitted through the internal vibration of the mobile phone, the second is transmitted in a straight line through the sound wave in the air, and the third is transmitted through the reflection of the touching finger. Due to the different distances of propagation speed of different paths, the impulse phase (<0.1ms) of the transmitted vibration signal will form multiple different peaks on the received signal of the microphone (as shown in Figure 7). The first peak (peak A) corresponds to the path of vibration transmission inside the mobile phone, the second peak (peak B) corresponds to the straight transmission path of sound waves in the air, and the third peak (peak C) corresponds to the vibration transmitted by touching the finger. The path the reflection propagates. Therefore, through the time corresponding to these peaks, the length of the corresponding path can be calculated, reflecting the characteristics of the touch position. Therefore, the peak value and corresponding time in the received signal of the microphone are extracted as the touch position feature.

其中：f_r为振动信号的共振频率，Δf定义了能量计算带宽，在中设为5Hz。4) Touch strength feature extraction: In order to obtain the user's touch strength information, the energy change of the vibration signal is used to measure. To avoid interference, calculate the energy value of the vibration signal around the resonance frequency

Where: f _r is the resonance frequency of the vibration signal, Δf defines the energy calculation bandwidth, and is set to 5Hz in .

As shown in FIG. 8, it is the change of the energy value E under different touch strengths. It can be seen that the greater the touch strength, the smaller the energy value E, showing an obvious corresponding relationship. Therefore, the energy value E of the vibration signal near the resonance frequency is extracted as the touch strength feature.

当

大于预先设定的阈值h，则认为两个样本具有相似的触碰行为特征，否则认为二者的触碰行为特征不相似。从而迫使分类器不依赖于触碰行为特征完成分类，从而使得孪生网络能够从信号特征(混合生物特征和触碰行为特征)中进一步提取与触碰行为无关的生物特征。The behavior-independent classifier is a neural network based on a siamese network architecture, specifically a siamese network that requires paired training samples in the training process. touch behavior feature), using the user label and touch behavior feature as a reference to select training samples, as shown in the table in Figure 9, for a set of training samples with the same label (from the same user), if and only if When their touch behavior characteristics are not similar, they are selected as training samples. Similarly, for a set of training samples with different labels (from two different users), and only if their touch behavior characteristics are similar, they are selected as training samples. The similarity of the touch behavior characteristics of the two samples x ₁ , x ₂ is measured by the Pearson correlation coefficient:

when

If it is greater than the preset threshold h, it is considered that the two samples have similar touch behavior characteristics; otherwise, the two samples are considered to have dissimilar touch behavior characteristics. Therefore, the classifier is forced to complete the classification without relying on the touch behavior features, so that the Siamese network can further extract the biological features unrelated to the touch behavior from the signal features (mixed biological features and touch behavior features).

The described Siamese network is shown in Figure 10, including two parallel sub-networks, each sub-network has the same structure and weights, and respectively according to the eigenvalues X ₁ , X ₂ extracted from the input vibration signal after passing through the sub-networks. Obtain the corresponding feature representations C _W (X ₁ ), C _W (X ₂ ), and then obtain the distance between the two feature representations: D _W (X ₁ , X ₂ )=||C _w (X ₁ )-C _w ( X ₂ )||.

This embodiment involves an application scenario: SAMSUNG Galaxy S6, SAMSUNG Galaxy S7, Google Pixel, HTC U Ultra and Huawei Mate8 are selected as prototypes of the described touch authentication system for 15 different volunteers to use to evaluate the actual effect. The experimental process was carried out in three different environments, namely laboratory environment (lab), shopping mall environment (mall) and bar environment (bar), and different types of touch were studied, including off-hand touch (Off-Hand) and In-Hand.

This embodiment specifically includes the following steps:

Step 1. 10 of the 15 volunteers registered to enter the system to become legitimate users, and the other 5 as attackers tried to enter the system by random touch.

Step 2. Five attackers try to use a mimic attack to enter the system.

Step 3. Five attackers try to use a replay attack to enter the system.

There are three main evaluation indicators:

Accuracy of user authentication, that is, the probability of successful authentication for different users.

Describes the False Reject Rate of the user experience, that is, the probability that a registered user is incorrectly authenticated as a non-registered user and cannot enter the system.

False Accept Rate, which describes the attack success rate, i.e. the probability that an attacker is successfully counted into the system without registering.

The evaluation results are shown in the table below. For 15 volunteers participating in the experiment, the accuracy of authenticating different users is shown. It can be seen from the figure that the identification between registered users can reach 92.3%, and the identification of potential non-registered users can reach 98.7%, indicating that the proposed touch authentication system can complete user authentication safely and effectively.

As shown in Figure 11, for three different environments, the false rejection rate is compared with the existing two common commercial smartphone authentication systems (Alipay's face authentication system and WeChat's voice lock authentication system). It can be seen that Alipay's face authentication system is difficult to correctly authenticate registered users when the lighting conditions are poor (bar environment), while WeChat's voice lock authentication system has higher errors when the ambient noise is high (shopping mall environment and bar environment). rejection rate. The proposed touch authentication system maintains a low false rejection rate in various environments, which greatly improves the user experience.

As shown in Figure 12, for different environments and different touch methods, the false acceptance rate of the system against imitation attacks is shown. It can be seen that under various environments and touch methods, the proposed system can effectively resist imitation attacks, and the false acceptance rate remains below 2%. Figure 13 shows the false acceptance rate of the system against replay attacks under different environments and different touch methods. While the false acceptance rate is slightly higher than imitation attacks overall, it remains within an acceptable range, i.e. under 5%. The results show that the proposed system can effectively resist replay attacks.

The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose. The protection scope is subject to the claims and is not limited by the above-mentioned specific implementation. Implementation plans are subject to constraints.

Claims (8)

1. a kind of intelligent terminal touch authentication method based on vibration signal, it is characterized in that, when intelligent terminal detects finger touch, by actively generating specific vibration signal and collecting vibration signal by IMU sensor, and extract biological feature, Behavior characteristics and independent touch behavior characteristics; then use the neural network based on the twin network architecture to classify biological characteristics to realize behavior-independent smart terminal touch authentication; The specific vibration signal is generated by adjusting the switch signal, and each cycle includes a vibration impulse signal, a motor activation signal and a residual vibration elimination part; The touch behavior characteristics specifically refer to: when the user uses the intelligent terminal, the contact between the user's finger and the intelligent terminal includes the behavior characteristics of the contact with the front, side and back of the intelligent device, including the touch position and the touch strength. feature; The neural network based on the Siamese network architecture is specifically a Siamese network that requires paired training samples in the training process. A set of training samples with different labels, if and only if their touch behavior characteristics are not similar, they are selected as training samples; for a set of training samples with different labels, and only if their touch behavior characteristics are similar , select them as training samples; The similarity of the touch behavior characteristics of the two samples x, y is measured by the Pearson correlation coefficient:

in:

and

The vectors are the average value of the elements in x and y respectively, and each vector contains n elements. When r _{x and y} are greater than the preset threshold h, the two samples are considered to have similar touch behavior characteristics, otherwise, the two samples are considered to have similar touch behavior characteristics. The touch behavior characteristics of the users are not similar, so that the classifier is not dependent on the touch behavior characteristics to complete the classification, so that the Siamese network can further extract the biological features unrelated to the touch behavior from the signal features. 2. The intelligent terminal touch authentication method based on vibration signal according to claim 1, is characterized in that, described collection vibration signal, first divides the vibration signal received into transient vibration stage, stable vibration stage and decay stage. 3. The intelligent terminal touch authentication method based on vibration signal according to claim 1, is characterized in that, described extraction comprises feature extraction based on wavelet change, feature extraction based on cepstral transformation, touch position feature extraction , Touch strength feature extraction, specifically: extracting the time spectrum obtained after wavelet transformation from the transient vibration stage, extracting the cepstrum from the stable vibration stage, extracting the peak value of the audio signal and its corresponding time from the microphone of the mobile terminal as the touch Position feature, extract the energy value of vibration signal near resonance frequency

As the touch strength feature, where: f _r is the resonance frequency of the vibration signal, Δf defines the energy calculation bandwidth, and f(t) is the original signal, that is, the vibration signal in the transient vibration stage. 4. The vibration signal-based smart terminal touch authentication method according to claim 1 or 3, wherein the extraction specifically comprises: 1) Extract features from the transient vibration phase of the received signal. Since the frequency of the vibration signal is constantly changing in the transient vibration phase, the original vibration signal needs to be transformed to obtain good time domain and frequency domain resolution at the same time. Therefore, , using continuous wavelet transform to transform the vibration signal in the transient vibration stage:

where: CWT _f (a,τ) is the obtained time spectrum; f(t) is the original signal, that is, the vibration signal in the transient vibration phase; ψ _a,τ (t) is the wavelet basis function, where a and τ represent the resolution in the time and frequency domains,

Represents the form of wavelet basis function time delay as t, selects Morlet function as wavelet basis function to achieve better resolution in time domain and frequency domain, and takes the time spectrum obtained after wavelet transform as the signal characteristic of transient vibration stage; 2) Feature extraction based on cepstral transformation: Use cepstral transformation to obtain various frequency components including sidebands in the signal in the stable vibration stage: C _y (q)=F ^-1 (logS _y (f(t))) , where: C _y (q) is the obtained cepstrum; S _y (f(t)) is the power spectral density of the signal; F ^-1 corresponds to the inverse Fourier transform (IFFT), and the obtained cepstrum is regarded as the stable vibration Signal characteristics of the stage; 3) Touch position feature: the vibration signal reaches the microphone through three paths, the first peak that appears on the microphone received signal in the impulse phase of sending the vibration signal corresponds to the path of the internal vibration transmission of the mobile phone, and the second peak corresponds to the sound wave passing through the air. The path of straight transmission, the third peak corresponds to the path propagating through the reflection of the touch finger, and the peak value and corresponding time in the received signal of the microphone are extracted as the touch position feature; 4) Touch strength feature extraction: Calculate the energy value of the vibration signal near the resonance frequency

where: f _r is the resonance frequency of the vibration signal, and Δf defines the energy calculation bandwidth. 5. The smart terminal touch authentication method based on vibration signal according to claim 1, wherein the described neural network based on twin network architecture comprises two parallel sub-networks, and each sub-network has the same structure and weights, respectively, according to the feature values X ₁ , X ₂ extracted from the input vibration signal, after passing through the sub-network, the corresponding feature representations C _W (X ₁ ), C _W (X ₂ ) are obtained, and then the two feature representations are obtained. Distance: D _W (X ₁ , X ₂ )=||C _w (X ₁ )-C _w (X ₂ )||. 6. The vibration signal-based touch authentication method for an intelligent terminal according to claim 5, wherein the sub-network is a time-delay neural network structure, comprising two convolutional layers and a fully connected layer, and each Within each convolutional layer, a basic convolutional module is used as the core, a BN layer is used to deal with the gradient problem and a ReLU layer is used as an activation function; The training samples of the neural network have the same label, that is, come from the same user and the touch behavior characteristics are not similar, that is, the similarity of the touch behavior characteristics is lower than the threshold. 7. The smart terminal touch authentication method based on vibration signal according to claim 6, wherein the classification refers to: according to the distance of the corresponding feature representation of the two inputs obtained by the neural network, judging the two inputs Whether it belongs to a unified user, so as to realize the judgment of whether the touch behavior comes from the registered user. 8. A system for implementing the method according to any one of claims 1 to 7, characterized in that, comprising: a vibration signal processing module, a signal feature extraction module, a behavior-independent classifier, and an authentication module, wherein: the vibration signal processing module is associated with The signal feature extraction module is connected to and transmits the preprocessed vibration signal information. The signal feature extraction module is connected to the behavior-independent classifier and transmits the signal features and behavioral feature information extracted from the preprocessed vibration signal. The behavior-independent classifier and authentication The modules are connected and transmit the information of the trained classifier model, the authentication module receives the signal features extracted in the authentication stage, connects and transmits it to the trained classifier model to obtain the authentication result of the lander.

Images