An e-commerce platform security authentication method and system

Abstract

The present invention relates to the technical field of platform security authentication, and in particular to a method and system for security authentication of an e-commerce platform. The method comprises the following steps: continuously collecting a facial image sequence of a user through a front camera of a user device to obtain an original facial image data stream; adaptively locating dynamic feature points of the original facial image data stream to obtain a personalized facial feature vector; collecting touch behavior data of the user to obtain an original touch behavior data stream; extracting operation features of the original touch behavior data stream to obtain a standardized touch feature vector; weighted fusion of the personalized facial feature vector and the standardized touch feature vector to obtain a user's biometric behavior feature signature; collecting environmental context data of the user device to obtain an operation environment behavior data stream. The present invention reduces the reliance on users setting strong passwords, thereby avoiding the security risks caused by users choosing simple passwords.

Description

Technical Field

[0001] The present invention relates to the technical field of platform security authentication, and in particular to an e-commerce platform security authentication method and system. Background Art

[0002] In the rapid development of e-commerce, security authentication technology plays a vital role. It is not only related to the security of users' personal information, but also directly affects the security of users' transactions on the platform. However, with the continuous advancement of network technology, traditional security authentication methods have gradually exposed their limitations. In particular, those two-factor authentication methods that rely only on passwords or static authentication methods are powerless in the face of increasingly cunning network attacks. As a widely used security measure, password authentication relies heavily on the strength of the passwords set by users. However, in order to facilitate memory, users often choose simple or habitual password combinations, which makes passwords easy to guess or break through brute force. In addition, even if the password is set relatively complex, it is difficult to resist the threats of keyloggers, phishing attacks, etc., which can steal users' password information silently. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide an e-commerce platform security authentication method and system to solve at least one of the above technical problems.

[0004] To achieve the above purpose, an e-commerce platform security authentication method includes the following steps:

[0005] Step S1: continuously collecting a facial image sequence of the user through the front camera of the user device to obtain an original facial image data stream; adaptively locating dynamic feature points of the original facial image data stream to obtain a personalized facial feature vector;

[0006] Step S2: collecting the user's touch behavior data to obtain an original touch behavior data stream; extracting operation features from the original touch behavior data stream to obtain a standardized touch feature vector; performing weighted fusion on the personalized facial feature vector and the standardized touch feature vector to obtain a user's biometric behavior feature signature;

[0007] Step S3: collecting environmental context data from the user device to obtain an operating environment behavior data stream; performing user behavior pattern analysis on the operating environment behavior data stream to obtain a user behavior feature distribution map; performing real-time credibility evaluation on the user behavior feature distribution map to obtain a user interaction behavior credibility index;

[0008] Step S4: Collect security factors of the user's current login environment to obtain an operating environment security data set; perform risk level assessment on the operating environment security data set to obtain an operating environment risk score; generate an authentication strategy based on the user interaction behavior credibility index and the operating environment risk score to obtain a multi-level security authentication solution;

[0009] Step S5: Fuse the user's biometric behavior feature signature, user interaction behavior credibility index, and operating environment security data set to obtain a user security feature portrait; continuously monitor the user security feature portrait to obtain a comprehensive dynamic credibility index; and perform identity authentication and verification operations on the user based on the multi-level security authentication scheme and the comprehensive dynamic credibility index.

[0010] The present invention provides a more secure authentication method by combining facial recognition and touch behavior analysis, which is more resistant to network attacks such as keyboard loggers and phishing attacks than traditional authentication methods that rely only on passwords or static two-factor authentication methods. By utilizing facial image sequence acquisition and dynamic feature point positioning technology, a personalized facial feature vector can be generated, so that the authentication process of each user is unique, and the accuracy and security of authentication are improved. By combining the user's facial features with the touch behavior features, a comprehensive user bio-behavioral feature signature is formed, which increases the complexity of the authentication system and makes it more difficult to forge an identity. By real-time monitoring of the user's behavior pattern and operating environment, the credibility of the user's interactive behavior can be evaluated, and abnormal behavior can be discovered in time, thereby improving the security of the trading platform. The present invention dynamically generates an authentication strategy by evaluating the risk level of the operating environment and combining the credibility index of the user's interactive behavior, so that the security authentication can adapt to the ever-changing security threats. The generated multi-level security authentication scheme provides users with different levels of security protection, and the authentication strategy can be flexibly adjusted according to the actual security needs and risk assessment results. By continuously monitoring the user's security feature portrait, a comprehensive dynamic credibility index can be obtained, which helps to update the user's risk assessment in real time and ensure the continuity and real-time nature of identity authentication. Compared with traditional password input, the present invention uses biometric behavioral features for authentication, which reduces the burden of users remembering and entering passwords, and can improve user satisfaction and the ease of use of the trading platform. The present invention can adapt to different devices and environments because it not only relies on facial images, but also includes the collection of touch behavior and environmental context data, so that it can work effectively in a variety of usage scenarios. In summary, the present invention reduces the reliance on users to set strong passwords, thereby avoiding the security risks caused by users choosing simple passwords.

[0011] Preferably, the present invention further provides an e-commerce platform security authentication system for executing the e-commerce platform security authentication method as described above, the e-commerce platform security authentication system comprising:

[0012] The feature point positioning module is used to continuously collect a sequence of facial images of the user through the front camera of the user device to obtain an original facial image data stream; adaptively and dynamically locate the feature points of the original facial image data stream to obtain a personalized facial feature vector;

[0013] The feature fusion module is used to collect the user's touch behavior data to obtain the original touch behavior data stream; extract the operation features of the original touch behavior data stream to obtain the standardized touch feature vector; perform weighted fusion on the personalized facial feature vector and the standardized touch feature vector to obtain the user's biometric behavior feature signature;

[0014] The behavior assessment module is used to collect environmental context data from user devices to obtain an operating environment behavior data stream; perform user behavior pattern analysis on the operating environment behavior data stream to obtain a user behavior feature distribution map; perform real-time credibility assessment on the user behavior feature distribution map to obtain a user interaction behavior credibility index;

[0015] The environmental risk assessment module is used to collect security elements of the user's current login environment to obtain an operating environment security data set; conduct risk level assessment on the operating environment security data set to obtain an operating environment risk score; generate authentication strategies based on the user interaction behavior credibility index and the operating environment risk score to obtain a multi-level security authentication solution;

[0016] The user authentication module is used to fuse the user's biometric behavior feature signature, user interaction behavior credibility index and operating environment security data set to obtain a user security feature portrait; continuously monitor the user security feature portrait to obtain a comprehensive dynamic credibility index; and perform identity authentication and verification operations on users based on the multi-level security authentication scheme and the comprehensive dynamic credibility index.

[0017] The present invention significantly improves the security and accuracy of the authentication process by integrating facial recognition and touch behavior analysis, and effectively prevents network attacks and fraud. The behavior assessment module of the system can deeply analyze the user's behavior pattern, thereby providing more personalized services. The real-time credibility assessment and environmental risk assessment functions ensure that the system can respond quickly to various potential security threats, enhancing the response speed and processing capacity of the platform. In addition, the system can dynamically generate multi-level security authentication schemes according to security requirements and risk assessment results to provide users with flexible security protection. By reducing the reliance on traditional passwords, the system not only improves the user experience, but also reduces operating costs and reduces the possibility of human errors. At the same time, the compatibility design of the system enables it to work effectively in different devices and environments, while the continuous monitoring function ensures the real-time update of user risk assessment and provides continuous protection. Finally, the system reduces the reliance on user sensitive information by using biometric behavioral characteristics for authentication, thereby protecting user information security while also strengthening the protection of user privacy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Other features, objects and advantages of the present invention will become more apparent from the detailed description made with reference to the following drawings:

[0019] Figure 1 A schematic flow chart of steps of a method for security authentication of an e-commerce platform according to an embodiment is shown.

[0020] Figure 2 A detailed flow chart of step S26 of an embodiment is shown.

[0021] Figure 3 A detailed flowchart diagram of step S263 of an embodiment is shown. DETAILED DESCRIPTION

[0022] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0023] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0024] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0025] To achieve this, please refer to Figures 1 to 3 The present invention provides an e-commerce platform security authentication method, comprising the following steps:

[0026] Step S1: continuously collecting a facial image sequence of the user through the front camera of the user device to obtain an original facial image data stream; adaptively locating dynamic feature points of the original facial image data stream to obtain a personalized facial feature vector;

[0027] Step S2: collecting the user's touch behavior data to obtain an original touch behavior data stream; extracting operation features from the original touch behavior data stream to obtain a standardized touch feature vector; performing weighted fusion on the personalized facial feature vector and the standardized touch feature vector to obtain a user's biometric behavior feature signature;

[0028] Step S3: collecting environmental context data from the user device to obtain an operating environment behavior data stream; performing user behavior pattern analysis on the operating environment behavior data stream to obtain a user behavior feature distribution map; performing real-time credibility evaluation on the user behavior feature distribution map to obtain a user interaction behavior credibility index;

[0029] Step S4: Collect security factors of the user's current login environment to obtain an operating environment security data set; perform risk level assessment on the operating environment security data set to obtain an operating environment risk score; generate an authentication strategy based on the user interaction behavior credibility index and the operating environment risk score to obtain a multi-level security authentication solution;

[0030] Step S5: Fuse the user's biometric behavior feature signature, user interaction behavior credibility index, and operating environment security data set to obtain a user security feature portrait; continuously monitor the user security feature portrait to obtain a comprehensive dynamic credibility index; and perform identity authentication and verification operations on the user based on the multi-level security authentication scheme and the comprehensive dynamic credibility index.

[0031] In this embodiment, first, the front camera of the user's device is used, such as iPhone X or higher. A facial recognition application is developed, which uses the device's camera to continuously capture the user's facial image at a rate of 30 frames per second to generate a raw facial image data stream. Then, the facial recognition module in the OpenCV library is used, combined with the 68-point facial landmark detection algorithm of dlib, to adaptively locate the dynamic feature points of the raw facial image data stream. Through the algorithm, the user's personalized facial feature vector is extracted, including the key position information of the eyes, nose and mouth. At the same time, the built-in pressure sensor and touch sensor of the device are used to collect the user's touch behavior data. A data collection service is developed, which can read the user's touch pressure and sliding speed in real time to generate a raw touch behavior data stream. Then, feature extraction is performed on these data to obtain a standardized touch feature vector. Next, the personalized facial feature vector is weighted fused with the standardized touch feature vector. The NumPy library of Python is used to process these vectors, and principal component analysis (PCA) is applied to reduce the feature dimension while retaining the most important information. In this way, the user's biometric behavior feature signature can be obtained. The environmental context data of the user's device is collected, such as device information and device location. Use the device's sensors and related APIs to obtain this data and generate an operating environment behavior data stream. Then, use machine learning algorithms to analyze the behavior patterns of these data and obtain a user behavior feature distribution map. Perform real-time credibility assessment on the user behavior feature distribution map, and use a logistic regression model to predict the credibility index of user interaction behavior. At the same time, collect security factors of the user's current login environment, such as the security of the network connection and the integrity of the device, and use these data to evaluate the risk level and obtain the operating environment risk score. Generate a multi-level security authentication scheme based on the user interaction behavior credibility index and the operating environment risk score. Finally, fuse the user's biometric behavior feature signature, the user interaction behavior credibility index, and the operating environment security data set to obtain a user security feature profile. Use an online learning algorithm to continuously monitor the user security feature profile and obtain a comprehensive dynamic credibility index. According to the multi-level security authentication scheme and the comprehensive dynamic credibility index, perform identity authentication and verification operations on users to ensure that only legitimate users can access system resources.

[0032] Preferably, step S1 comprises the following steps:

[0033] Step S11: positioning and tracking the user's face through the front camera of the user device to obtain a real-time face area coordinate data stream, and optimizing the spatial range of the real-time face area coordinate data stream to obtain accurate face area positioning data;

[0034] Specifically, you can start the face recognition application on your phone, which uses advanced computer vision algorithms to detect and track the user's face. When the user points his face to the camera, the application recognizes the face from the real-time video stream captured by the camera and marks the position of the face with a rectangular frame on the screen. Next, the application generates a coordinate data stream for the detected face area, which contains the location information of the face in the image. The application also performs spatial range optimization, which involves image processing techniques such as edge enhancement and contrast adjustment on the images captured by the camera to ensure that the face area can be accurately located under different lighting conditions and background interference. Finally, accurate face area location data is obtained.

[0035] Step S12: sampling the user's facial image continuous frames according to the precise facial region positioning data to obtain an original facial image data stream;

[0036] Specifically, the accurate face region positioning data can be used to guide the continuous frame sampling of the face image. Specifically, the camera is set to capture images at a rate of 30 frames per second, and only the face region determined in step S11 is sampled. In this way, an original face image data stream is obtained, which consists of a series of continuous image frames containing only faces.

[0037] Step S13: performing illumination compensation on the original facial image data stream to obtain a standardized facial image data stream, and performing time sequence integration and synchronization on the standardized facial image data stream to obtain an original facial image frame sequence;

[0038] Specifically, an adaptive histogram equalization algorithm can be used for illumination compensation, which is a commonly used image enhancement technology that can improve the contrast of the image and expand its dynamic range. By applying this algorithm, the brightness and contrast of the image can be adjusted so that facial images captured under different lighting conditions can reach a unified and standardized illumination level, and finally a standardized facial image data stream is processed. Next, these illumination-compensated facial images are time-sequentially integrated and synchronized using timestamp synchronization technology. This technology can ensure that each frame of the image matches its corresponding timestamp. By writing a script program, the program can automatically identify the timestamp of each frame of the image and sort the image frames in chronological order, and finally obtain the original facial image frame sequence.

[0039] Step S14: Analyze the facial tissue features of the user based on the original facial image frame sequence to obtain a facial feature distribution density map, and dynamically locate the feature points of the user's face according to the facial feature distribution density map to obtain a facial feature point mapping solution;

[0040] Specifically, please refer to the sub-steps of step S14 for the detailed implementation process of this embodiment.

[0041] Step S15: extracting feature points from the facial feature point mapping scheme to obtain an original feature point coordinate set, and calculating the spatial geometric relationship of the original feature point coordinate set to obtain a facial feature point association matrix;

[0042] Specifically, image processing software can be used that can read the mapping scheme and identify the key feature points of the face. The software extracts the coordinates of each feature point by comparing the dynamic feature points with the points in the standard model. These coordinates form the original feature point coordinate set, which accurately describes the position of the facial feature points in two-dimensional space. Next, the spatial geometric relationship between these feature points is calculated. Use a computer vision library, such as OpenCV, to perform this step. Specifically, the Euclidean distance between each pair of feature points is calculated, which is achieved by applying the Euclidean distance formula, which calculates the distance based on the difference in the coordinates of two points. In addition, the angular relationship between the feature points is calculated by applying the cosine theorem or the sine theorem, which are calculated based on the relationship between the side length and angle of a triangle. After obtaining the distance and angle information of all feature point pairs, these data are organized into an association matrix. In this matrix, each row and column represents a feature point, and the value of each cell represents the distance or angle between the corresponding point pairs. For example, if the cell in the i-th row and j-th column of the matrix represents the distance between feature point i and feature point j, then this value is the Euclidean distance obtained by the above calculation. Finally, a facial feature point association matrix is ​​obtained, which fully describes the spatial geometric relationship of facial feature points.

[0043] Step S16: vectorize the facial feature point association matrix to obtain a personalized facial feature vector.

[0044] Specifically, the numerical data in the facial feature point association matrix can be normalized to ensure that the contribution of each feature point is balanced. Then, the principal component analysis (PCA) method is used to reduce the dimension of the matrix. Through PCA, the facial feature point association matrix is ​​converted into a vector, and finally a personalized facial feature vector is obtained, which contains the most important feature information. When implementing PCA, professional statistical software such as R or Python's NumPy library is used. These tools can automatically calculate the principal components and combine them into a vector.

[0045] The present invention significantly improves the accuracy of facial recognition through face positioning and tracking and spatial range optimization, and ensures that high-quality facial image data streams can be obtained under different lighting conditions through continuous frame sampling and illumination compensation processing. The standardization and continuity of data are further ensured through time integration and synchronization processing. The system's ability to capture subtle changes in the user's face is enhanced through in-depth facial feature analysis and dynamic positioning of feature points, and the details and depth of facial recognition are improved. The robustness of facial recognition is enhanced through spatial geometric relationship calculation, so that the system can accurately recognize even when facial expressions or postures change. The personalized facial feature vector generated by vectorization processing not only improves the personalization of recognition, but also enhances the security of the system.

[0046] Preferably, step S14 comprises the following steps:

[0047] Step S141: selecting the best frame from the original facial image frame sequence to obtain the best facial image frame, and extracting facial features from the best facial image frame to obtain facial feature descriptor data;

[0048] Specifically, video processing software can be used to observe the original facial image frame sequence and preliminarily screen out frames with high image quality, such as those without blur, occlusion or abnormal lighting conditions. Next, an automated method based on image quality scoring is used to select the best frames. This method involves calculating indicators such as clarity, contrast, and brightness for each frame. Image processing libraries, such as OpenCV, are used to implement this function. Specifically, the sharpness value of each frame is calculated, which can be achieved by applying the Laplacian operator to the image and calculating the variance of the results. During the automated scoring process, thresholds are set, and only those frames that exceed these thresholds in clarity, contrast, and brightness are selected as the best frames. For example, the sharpness variance is set to be greater than 100, the contrast balance is set to be greater than 0.5, and the mean brightness is set to be within a reasonable range. Based on these criteria, 10 optimal frames are selected from the original facial image frame sequence. Then, facial features are extracted from these optimal frames. A deep learning framework, such as TensorFlow or PyTorch, is used, with a pre-trained facial recognition model, such as the facial landmarking model of dlib. This model can detect and extract key facial feature points, such as the corners of the eyes, the tip of the nose, the corners of the mouth, etc., and generate corresponding feature descriptor data. These data include the coordinates of the feature points, local texture information, and geometric relationships.

[0049] Step S142: performing density distribution calculation on the facial feature descriptor data based on the original facial image frame sequence to obtain facial feature density distribution data, and performing Gaussian kernel density estimation on the facial feature density distribution data to obtain a facial tissue feature distribution density map;

[0050] Specifically, the facial feature points in the optimal facial image frame can be marked and recorded according to the facial feature descriptor data, and the coordinates of each feature point are obtained. Then, the density distribution of these feature points in the facial image is calculated. Specifically, the number of times each feature point appears in the image is calculated, and these numbers are used as the density values ​​of the feature points. Next, Gaussian kernel density estimation is performed on these density data. This step is implemented using the density function of the R language or the gaussian_kde function in the scipy.stats module of Python. Through Gaussian kernel density estimation, discrete density data is converted into a continuous density distribution function, which can describe the distribution of facial feature points in facial images. Finally, a facial tissue feature distribution density map is generated based on this density distribution function. This graph shows the density of facial feature points, where higher density areas indicate locations where feature points appear frequently, such as around the eyes, nose, and mouth.

[0051] Step S143: performing local extreme point detection on the facial tissue feature distribution density map to obtain a facial feature key point candidate set, and performing initial feature point positioning on the facial tissue feature distribution density map according to the facial feature key point candidate set to obtain a preliminary facial feature point positioning result;

[0052] Specifically, image processing software, such as MATLAB or Python's image processing library OpenCV, can be used to load and display the density map of facial tissue feature distribution. The local maximum detection algorithm is used to scan the neighborhood of each pixel in the density map and compare whether its value is greater than the values ​​of other points in the neighborhood. A threshold is set, and only when the value of a point is greater than the values ​​of its 8 neighboring points and exceeds the set threshold, it is considered a local maximum point. For example, the threshold is set to the 90th percentile of all point values ​​on the density map. In this way, a candidate set of facial feature key points is obtained, which appear as local peaks in the density map. Then, the feature points are initially located for these key point candidate sets. Use image annotation tools, such as GIMP or Photoshop, to manually mark the positions of these key points. For example, mark key points at the expected positions of the eyes, nose, and mouth. Finally, the preliminary facial feature point location results are obtained, which include the coordinates and density values ​​of the key points.

[0053] Step S144: tracking the spatiotemporal feature sequence of the original facial image frame sequence according to the preliminary facial feature point positioning result to obtain a dynamic facial feature point positioning solution;

[0054] Specifically, video processing software can be used to read and display the original facial image frame sequence. Use the optical flow algorithm in the OpenCV library, such as the Lucas-Kanade method, to track the dynamic changes of facial feature points. Specifically, first mark the facial feature points that are initially located in a frame of image, and then use these points as the starting points for tracking. The algorithm will automatically track the positions of these points in each subsequent frame of image. Adjust the parameters of the algorithm, such as setting the appropriate window size, number of iterations, and threshold. In this way, dynamic positioning schemes for facial feature points can be obtained, which include the position, movement direction, and speed of the feature points in each frame of image.

[0055] Step S145: performing affine transformation registration on the dynamic feature point sequence in the facial feature point dynamic positioning solution and the preset standard face feature model to obtain a facial feature point mapping solution.

[0056] Specifically, a standard facial feature model can be defined, which contains the standard positions of key feature points on the face, such as the corners of the eyes, the tip of the nose, the corners of the mouth, etc. This model can be an average face shape based on a large amount of face data, or a statistical model based on a specific population. For example, a standard model containing 68 feature points is used, which covers the main contours and internal features of the face. Next, a computer vision library, such as OpenCV or dlib, is used to perform affine transformation registration. The corresponding sequence of facial feature points is extracted from the dynamic positioning scheme. These points are then used as input to match the corresponding points in the standard model. The correspondence between the two sets of points is determined. This can be done by manual selection or using an automatic feature matching algorithm. Once the correspondence is determined, the estimateAffinePartial2D function in OpenCV can be used to calculate the best affine transformation matrix. This matrix describes how to transform the dynamic feature point sequence to the position of the standard model through operations such as translation, scaling, and rotation. After obtaining the affine transformation matrix, this matrix is ​​applied to all dynamic feature points to calibrate their positions. This results in a facial landmark mapping scheme that describes the corresponding position of each dynamic landmark in the standard model.

[0057] The present invention can capture the most representative facial features by selecting the optimal frame for feature extraction, thereby significantly improving the quality of feature recognition. The accuracy of facial feature density distribution is further improved by the application of Gaussian kernel density estimation. The reliability of key point detection is enhanced by local extreme point detection, providing an accurate starting point for the initial positioning of feature points. The stability and accuracy of feature point positioning are further improved by spatiotemporal feature sequence tracking, ensuring that the system can dynamically track the changes of facial feature points. Through affine transformation registration with the standard facial feature model, it shows stronger adaptability, can adapt to different facial postures and expression changes, and improves the flexibility and robustness of facial recognition. The dynamic positioning scheme supports real-time updating of the position of facial feature points, enhancing the real-time performance of facial recognition. A higher level of security is provided for user verification through an accurate facial feature point mapping scheme, reducing the risk of fraud and impersonation. The user's waiting time is reduced by fast and accurate facial feature point positioning, enhancing the user experience.

[0058] Preferably, step S2 comprises the following steps:

[0059] Step S21: collecting pressure sensing data from the touch screen of the user device to obtain touch pressure time series data, and numerically normalizing the touch pressure time series data to obtain standardized pressure characteristic data;

[0060] Specifically, a touch screen device with pressure sensing function can be used, such as some models of smart phones or tablets. The touch screens of these devices have built-in pressure sensors that can detect the pressure intensity when the user touches. Develop a data acquisition application that can read the pressure sensor data on the touch screen in real time. When the user performs a touch operation, the application records the pressure value of each touch point and generates touch pressure timing data. This data includes information such as the strength of the touch, the duration of the touch, and the position of the touch point on the screen. First, calculate the range of all collected pressure values, that is, the maximum and minimum values. Then, apply the minimum-maximum normalization method to convert each pressure value into a standardized value between 0 and 1, and finally obtain standardized pressure characteristic data.

[0061] Step S22: collecting the track coordinates of the user's touch operation to obtain original touch track data;

[0062] Specifically, a high-precision touch screen device can be used, and special data acquisition software can be developed, which can record the coordinates of each touch operation of the user. When the user uses the device, the software monitors the touch events on the touch screen in real time, records the coordinates (x, y) of the touch point, and the start and end time of each touch. These data are used to track the path of the user's finger or stylus on the screen and generate raw touch trajectory data. These data include a series of coordinate points, each of which corresponds to a specific position and time of the user on the screen. In this way, detailed trajectory data of the user's touch operation can be obtained.

[0063] Step S23: smoothing the original touch trajectory data to obtain a smoothed touch trajectory coordinate sequence, and extracting features from the smoothed touch trajectory coordinate sequence to obtain touch trajectory feature data;

[0064] Specifically, the original touch trajectory data can be imported into Python and converted into a NumPy array for numerical calculation. Next, the moving average method is used for smoothing. Specifically, a window size is set, such as 5 consecutive touch points, and then the average value of each point is calculated to replace the coordinates of the current point. After the smoothing process is completed, the feature extraction of the smoothed touch trajectory coordinate sequence continues. The features such as the length, curvature, and number of turning points of the trace are extracted, and these features are stored in a new data structure to finally obtain the touch trajectory feature data.

[0065] Step S24: collecting time characteristics of the user's touch operation to obtain touch operation timing data, and performing rhythm pattern recognition on the touch operation timing data to obtain user operation rhythm characteristics; recording the standardized pressure characteristic data, the touch trajectory characteristic data and the user operation rhythm characteristics as the original touch behavior data stream;

[0066] Specifically, the data acquisition software of step S22 can be continued to be used, and the software can record the user's touch operation timing data in real time, including the start time, end time and duration of each touch. Specifically, when the user performs a touch operation, the software records the timestamp of each touch and calculates the duration of each touch. These data are stored in a time series array, and the start and end time of each touch operation are recorded. By calculating the time interval between adjacent touch operations, the user's touch operation timing data can be obtained. Next, the rhythm pattern recognition is performed on these timing data. Fast Fourier transform (FFT) is used to analyze the frequency components of the time series data. Through FFT, the rhythm frequency of the user's touch operation can be identified, such as the frequency of the user's fast clicks on the screen or the rhythm of slow sliding. These frequency information are extracted to form the user's operation rhythm characteristics. Finally, the standardized pressure feature data, the touch trajectory feature data and the user operation rhythm features are combined to form an original touch behavior data stream.

[0067] Step S25: extracting spatiotemporal features from the original touch behavior data stream to obtain a multi-dimensional touch feature matrix, and reducing the dimension of the multi-dimensional touch feature matrix to obtain an initial touch feature vector; performing normalized mapping on the initial touch feature vector to obtain a normalized touch feature vector;

[0068] Specifically, the touch operation timing data includes the user's touch position, touch pressure, touch duration, and the time interval between touch points. Machine learning libraries, such as Python's scikit-learn, are used to process this data. First, multiple spatiotemporal features are defined, such as the average force of touch, the frequency of touch, the speed of sliding, and the curvature of the trajectory. For each feature, corresponding code is written to calculate its value. For example, the average force of touch can be obtained by calculating the average of the pressure values ​​of all touch points; the speed of sliding can be obtained by calculating the distance between consecutive touch points divided by the time difference. Next, these feature values ​​are combined into a multidimensional touch feature matrix, where each row represents a touch event and each column represents a feature. Principal component analysis (PCA) is performed on this matrix to reduce the dimension. Through PCA, the initial touch feature vector is finally obtained, which contains the most important feature information. Finally, the initial touch feature vector is normalized and mapped. The mean and standard deviation of each feature are calculated, and then these parameters are used to transform the feature value into a standard normal distribution, and finally the standardized touch feature vector is obtained.

[0069] Step S26: perform feature importance evaluation on the personalized facial feature vector and the standardized touch feature vector to obtain a comprehensive feature weight coefficient, and perform weighted fusion on the personalized facial feature vector and the standardized touch feature vector according to the comprehensive feature weight coefficient to obtain a user biometric behavior feature signature.

[0070] Specifically, please refer to the sub-steps of step S26 for the detailed implementation process of this embodiment.

[0071] The present invention greatly enhances the recognition capability of user behavior characteristics by comprehensively collecting pressure sensing data and trajectory coordinates of the user device touch screen, and can capture the user's unique touch behavior pattern. The data quality is further improved through numerical standardization and data smoothing processing, ensuring the accuracy and reliability of subsequent feature extraction and analysis. By extracting multi-dimensional features such as touch trajectory and operation rhythm, the user's personalized operation habits are fully captured, thereby improving the discrimination and accuracy of the bio-behavioral feature signature. The touch feature vector is optimized through dimensionality reduction and standardized mapping processing, while the evaluation and weighted fusion of feature importance enhance the system's anti-interference ability and improve the robustness of the bio-behavioral feature signature. By integrating facial and touch feature vectors, the generated user bio-behavioral feature signature provides a higher level of security for security authentication and reduces the risk of fraud and impersonation.

[0072] Preferably, step S26 comprises the following steps:

[0073] Step S261: performing self-information entropy calculation on the personalized facial feature vector to obtain a facial feature information entropy distribution matrix;

[0074] Specifically, the Python programming language and the NumPy library can be used to process personalized facial feature vectors. Assume that the personalized facial feature vector consists of a series of numerical values, each of which represents a specific feature in the facial image, such as the distance between the eyes, the width of the nose, etc. First, the frequency distribution of each feature value is calculated. For example, if the facial feature vector contains multiple measurements of the distance between the eyes, calculate the histogram of these measurements to determine the frequency of each distance value. Then, the Shannon entropy formula is used to calculate the self-information entropy: ;in, is entropy, It is The probability of the feature value (here is the frequency distribution). For each feature of the facial feature vector, its self-information entropy is calculated and these values ​​are stored in a matrix, namely the facial feature information entropy distribution matrix. The rows of this matrix represent different features, and the columns represent the self-information entropy value of each feature.

[0075] Step S262: sorting the personalized facial feature vectors by importance according to the facial feature information entropy distribution matrix to obtain facial feature weight distribution coefficients;

[0076] Specifically, Python and NumPy libraries can be used to process the facial feature information entropy distribution matrix. Assume that the self-information entropy value obtained in step S261 has been stored in a NumPy array. First, the facial feature information entropy distribution matrix is ​​normalized to convert the entropy value of each feature into a weight in the range of [0,1]. The normalization process can be implemented by the following formula: ;in, It is The weight of the feature, It is The self-information entropy value of the feature, All characteristics The maximum value among the values. Then, the facial feature vectors are sorted according to the calculated weights. The higher the weight, the more discriminative the feature is, and therefore the more important it is. You can use NumPy's argsort function to get the descending index of the weights, and then re-sort the facial feature vectors according to these indices. Finally, the facial feature weight distribution coefficients are obtained, which represent the importance of each feature in the facial feature vector.

[0077] Step S263: performing principal component correlation analysis on the standardized touch feature vector to obtain a touch feature contribution matrix;

[0078] Specifically, please refer to the sub-steps of step S263 for the detailed implementation process of this embodiment.

[0079] Step S264: performing collaborative weight calculation on the standardized touch feature vector according to the touch feature contribution matrix to obtain a touch feature weight distribution coefficient, and using the facial feature weight distribution coefficient and the touch feature weight distribution coefficient as a comprehensive feature weight coefficient;

[0080] Specifically, the contribution value of each feature can be extracted from the touch feature contribution matrix obtained in step S263. Then, these contribution values ​​are used to calculate the weight distribution coefficient of the touch feature. Specifically, the contribution value of each feature is normalized to ensure that the sum of all weights is 1. The following formula can be used for normalization: ;in, It is The weight of the touch feature, It is The contribution value of each touch feature, n is the total number of features. After completing the touch feature weight calculation, the facial feature weight distribution coefficient is combined with the touch feature weight distribution coefficient to form a comprehensive feature weight coefficient. This is achieved by a simple weighted average. For example, if the weight of the facial feature is 0.6 and the weight of the touch feature is 0.4, they are combined as follows: ; In this way, a comprehensive feature weight coefficient can be obtained.

[0081] Step S265: performing weighted combination of the personalized facial feature vector and the standardized touch feature vector according to the comprehensive feature weight coefficient to obtain an initial bio-behavior feature fusion matrix;

[0082] Specifically, two vectors can be prepared: one is a personalized facial feature vector, and the other is a standardized touch feature vector. Use Python and NumPy libraries to process this data, and apply the comprehensive feature weight coefficient to each feature. First, multiply each dimension of each feature vector by its corresponding weight coefficient. For example, if a feature weight of the facial feature vector is 0.7 and a feature weight of the touch feature vector is 0.3, multiply these weights by the corresponding feature values ​​respectively. Then, add the weighted feature values ​​to form a row of the initial bio-behavioral feature fusion matrix. This process is performed for all feature dimensions, and finally a matrix that combines facial and touch features is obtained. Each row of this initial bio-behavioral feature fusion matrix represents a user, and each column represents a bio-behavioral feature that combines facial and touch features.

[0083] Step S266: Perform feature hashing and encoding on the initial bio-behavioral feature fusion matrix to obtain the user's bio-behavioral feature signature.

[0084] Specifically, Python hash libraries, such as hashlib, can be used to implement feature hashing. First, each row of the initial bio-behavioral feature fusion matrix is ​​converted into a string that contains all the bio-behavioral feature values ​​of the user. Then, the string is hashed using the SHA-256 hash algorithm to obtain a fixed-length hash value. This hash value is unique. The hash value is also encoded into a QR code or Base64 string. Finally, the user's bio-behavioral feature signature is obtained, which is a hash code value that contains the user's facial and touch behavior features.

[0085] The present invention optimizes feature representation through self-information entropy calculation, quantifies the uncertainty of facial features, and provides an effective sorting method for facial feature vectors, highlighting the most recognizable features. The accuracy of feature fusion is improved through importance sorting and weight distribution, and the most critical features for individual identification can be identified and emphasized. The recognition of feature contribution is enhanced through principal component correlation analysis, revealing the most representative components in touch features, and focusing on the features that contribute most to user behavior identification. The rationality of weight distribution is improved through collaborative weight calculation, ensuring that the combination of facial and touch features takes into account their individual importance and mutual relationship. The security of the system is improved through feature hashing and encoding processes, providing an additional security layer for user bio-behavioral feature signatures, which are difficult to tamper with or forge. In addition, the present invention protects the privacy of user information by fusing and converting bio-behavioral features into signature form.

[0086] Preferably, step S263 includes the following steps:

[0087] Step S2631: normalizing the dimension of the standardized touch feature vector to obtain normalized touch feature data;

[0088] Specifically, dimension normalization can be performed using the Python programming language and the NumPy library. The standardized touch feature vector contains features in multiple dimensions, such as touch pressure, sliding speed, touch duration, etc. First, the mean and standard deviation of each dimension are calculated. For example, if one dimension is touch pressure, the mean and standard deviation of all touch pressure values ​​are calculated. Then, these statistical data are used to normalize the feature values ​​of each dimension, and finally obtained. Normalization can be achieved by the following formula to normalize the touch feature data: ; In this way, the feature values ​​of each dimension are converted to a standard scale, so that features of different dimensions can be directly compared and analyzed.

[0089] Step S2632: performing feature covariance calculation on the normalized touch feature data to obtain a touch feature covariance matrix, and performing inter-dimensional correlation coefficient calculation on the touch feature covariance matrix to obtain a touch feature correlation matrix;

[0090] Specifically, Python and NumPy libraries can be used to process normalized touch feature data. First, the covariance matrix of the normalized touch feature data is calculated. The covariance matrix is ​​a square matrix in which each element represents the covariance value between two dimensions. Use NumPy's cov function to calculate the covariance matrix. Next, the correlation matrix is ​​calculated from the covariance matrix. The correlation matrix is ​​also a square matrix, in which each element represents the correlation coefficient between two dimensions. The NumPy corrcoef function is used to calculate the correlation matrix. The diagonal elements of the correlation matrix are all 1 because they represent the correlation between the dimension and itself. The value range of the non-diagonal elements is between -1 and 1, where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation. In this way, the touch feature correlation matrix can be obtained, which provides correlation information between different touch features.

[0091] Step S2633: performing eigenvalue decomposition on the touch feature correlation matrix to obtain a touch feature spectrum decomposition matrix, and extracting principal eigenvectors from the touch feature spectrum decomposition matrix to obtain touch feature principal component basis vectors;

[0092] Specifically, Python's NumPy library can be used to process the touch feature correlation matrix. The correlation matrix is ​​decomposed by eigenvalue using NumPy's linalg.eig function. This function returns two arrays: one is an eigenvalue array and the other is a corresponding eigenvector array (eigenvector matrix). Each column of the eigenvector matrix represents an eigenvector, and each value in the eigenvalue array corresponds to the eigenvalue of each column in the eigenvector matrix. Next, the eigenvectors are sorted according to the size of the eigenvalues ​​to determine the principal eigenvector. In NumPy, the argsort function is used to sort the eigenvalues ​​and obtain the sorted indexes. Then, the eigenvector matrix is ​​rearranged according to these indexes so that each column is arranged in the order of the corresponding eigenvalues. The eigenvector corresponding to the largest eigenvalue is the principal component, that is, the principal component basis vector of the touch feature. Through this process, the touch feature spectrum decomposition matrix can be obtained, and the principal component basis vectors are extracted from it, which form the basis of the new feature space.

[0093] Step S2634: Calculate the variance contribution of the touch feature principal component basis vectors to obtain a touch feature principal component contribution rate sequence;

[0094] Specifically, we can first review the eigenvalue array obtained in step S2633. These eigenvalues ​​actually represent the variance of each principal component. The variance contribution rate of each principal component is obtained by calculating the ratio of each eigenvalue to the sum of all eigenvalues. This process can be expressed by the following formula: ;in, It is The eigenvalues ​​of the principal components, is the total number of principal components. Then, these variance contribution rates are stored in a sequence, namely the touch feature principal component contribution rate sequence. This sequence shows the contribution ratio of each principal component to the total variance of the data set.

[0095] Step S2635: performing cumulative contribution screening on the touch feature principal component contribution rate sequence to obtain touch principal component importance ranking data;

[0096] Specifically, Python and NumPy libraries can be used to process principal component contribution sequences. Calculate the cumulative contribution of each principal component, which is obtained by adding the variance contribution of each principal component to the cumulative contribution of the previous principal component. For example, if the variance contributions of the first three principal components are 0.3, 0.2, and 0.15, respectively, then their cumulative contributions will be 0.3, 0.5, and 0.65. Next, set a threshold, such as 95%, to determine how many principal components need to be retained. Check the cumulative contribution sequence and find the position where the cumulative contribution exceeds the threshold for the first time. In this example, at least the first three principal components are needed because their cumulative contribution reaches 0.65, which exceeds the 95% threshold. Finally, the touch principal component importance ranking data is obtained, which is a list containing each principal component and its cumulative contribution.

[0097] Step S2636: performing weight mapping on the feature dimensions in the standardized touch feature vector based on the touch principal component importance ranking data to obtain an initial touch feature contribution vector;

[0098] Specifically, the touch principal component importance ranking data, the feature vector array obtained in the step S2633 embodiment, and the principal component basis vector can be used to determine the contribution of each original feature to the important principal component. The coefficient of each feature in the principal component basis vector is squared and accumulated to calculate the initial contribution of each feature. This coefficient represents the projection weight of the feature vector on the principal component. For example, if a feature has a coefficient of 0.5 in the first principal component basis vector and a coefficient of 0.3 in the second principal component basis vector, then its initial contribution will be Next, these initial contributions are normalized to get a weight between 0 and 1. The normalization is done using the following formula: ; Finally, the initial contribution vector of the touch feature is obtained, which is a vector containing the normalized contribution of each feature.

[0099] Step S2637: performing nonlinear normalization transformation on the initial touch feature contribution vector to obtain the initial touch feature contribution matrix, and performing feature contribution distribution adjustment on the initial touch feature contribution matrix to obtain the touch feature contribution matrix.

[0100] Specifically, Python and NumPy libraries can be used to process the initial contribution vector of touch features. Apply nonlinear functions, such as the Sigmoid function, to transform each contribution value. The Sigmoid function can map any real value between 0 and 1, and has the characteristic that the intermediate output value is close to 1. Apply the Sigmoid function to each element of the initial contribution vector to obtain a nonlinear normalized vector. Next, the nonlinear normalized vector is further adjusted to obtain the initial matrix of touch feature contributions. Calculate the difference between each element and its adjacent elements, and adjust each element based on these differences. This process can be achieved by the following formula:

[0101] ;

[0102] in, is an adjustment coefficient used to control the influence of differences between adjacent elements. Finally, the touch feature contribution matrix is ​​obtained, which is a matrix containing the adjusted contribution of each feature.

[0103] The present invention ensures that the touch feature data of different dimensions are compared and processed at the same scale through dimensional normalization, thereby improving the consistency of feature expression. By calculating the feature covariance and correlation coefficient, the statistical relationship between different touch features is revealed, providing a mathematical basis for in-depth understanding of user behavior patterns. The application of eigenvalue decomposition and principal component analysis can identify and extract the key touch features that contribute the most to user behavior identification. The feature contribution of each principal component is quantified through variance contribution calculation and cumulative contribution screening, helping to identify the most representative user behavior features. The generalization ability of the system is further improved through nonlinear normalization transformation and feature contribution distribution adjustment, so that it can adapt to different user behavior patterns and device characteristics. The final touch feature contribution matrix provides accurate weight information for the fusion of facial and touch features, and improves the effect of feature fusion. Scientific feature selection and weight distribution enhance the system's ability to resist noise and outlier interference, and improve the robustness of the authentication process. In general, the present invention improves the expression quality of touch features and the effect of feature fusion.

[0104] Preferably, step S3 comprises the following steps:

[0105] Step S31: collecting attitude data of the user device gyroscope to obtain a device tilt angle sequence, and performing data standardization on the device tilt angle sequence to obtain standard device attitude data;

[0106] Specifically, a smartphone equipped with a gyroscope can be used to perform this embodiment. The gyroscope can measure the rotation angle of the device around three axes in space, and these data can help understand how the user holds and operates the device. First, a data acquisition tool is developed that can access the device's gyroscope sensor and read the tilt angle data at a certain sampling rate (for example, 50 times per second). These data include the roll angle, pitch angle, and yaw angle of the device. Next, the collected tilt angle sequence is standardized. First, the mean and standard deviation of the tilt angle of each axis are calculated. Then, each angle value is standardized using these statistical data so that its mean is 0 and the standard deviation is 1. In this way, standard device posture data can be obtained, which reflects the tilt state of the device under different user operations.

[0107] Step S32: real-time monitoring of the user interface interactive operation to obtain an original operation event sequence, and timestamp the original operation event sequence to obtain a time-series operation data stream; operation speed feature extraction is performed on the time-series operation data stream to obtain operation speed feature data;

[0108] Specifically, this embodiment can be performed using a customized Android application that can monitor the user's touch operation on the screen. First, the developed application can record the user's touch events, including touch start, touch end, and touch movement. The application generates a timestamp for each touch event, which marks the exact time when the event occurs. In this way, a sequence of operation events with timing is obtained, which contains the order and time information of the user's operation. Next, the operation speed feature extraction is performed on the timing operation data stream. The average speed and peak speed of the user sliding on the screen are calculated. The average speed is obtained by calculating the total distance that the user's finger moves on the screen divided by the total time, and the peak speed is the maximum speed recorded in any given time period. First determine the distance between consecutive touch events, and then calculate the time difference between them. Using these data, calculate the speed of each touch slide. For example, if the user's finger moves from point A to point B on the screen in 0.2 seconds, and the distance between point A and point B is 5 cm, then the speed of this slide is 25 cm / second. In this way, operation speed feature data can be obtained, which reflects the speed and rhythm of the user's operation on the interface.

[0109] Step S33: Perform frequency statistics on the user's click behavior to obtain a click event time series, and perform interval analysis on the click event time series to obtain click frequency feature data; integrate the standard device posture data, operation speed feature data and click frequency feature data to obtain an operation environment behavior data stream;

[0110] Specifically, a customized software application can be used to monitor and record the user's click behavior, while collecting the device's posture data and operation speed data. First, the application records the timestamp of each user's click and generates a time series of click events. Then, the frequency statistics of these click events are performed to calculate the number of clicks of the user in a specific time period. For example, the average number of clicks per minute by the user, or the frequency of clicks in a specific application, can be calculated. Next, interval analysis is performed to calculate the time interval between consecutive click events. By analyzing these intervals, the user's click frequency feature data can be obtained, which reflects the user's activity level and operation rhythm when operating the device. Finally, the standard device posture data, operation speed feature data and click frequency feature data are integrated. In this process, the time alignment of the data is ensured so that each data point corresponds to the same timestamp. In this way, a comprehensive operating environment behavior data stream is obtained, which contains information on multiple dimensions of user operations.

[0111] Step S34: performing time-series behavior pattern recognition on the operation environment behavior data stream to obtain an interactive behavior feature sequence, and clustering the interactive behavior feature sequence to obtain a user operation pattern cluster; performing spatial distribution mapping on the user operation pattern cluster to obtain a user behavior feature distribution map;

[0112] Specifically, machine learning and data mining techniques can be used to identify time series behavior patterns. First, a hidden Markov model (HMM) is used to identify the time series patterns of user operations. HMM is a statistical model suitable for processing and analyzing sequence data. HMM is trained to identify typical behavior patterns of users when operating devices, such as quick browsing, careful reading, or frequent interaction. Next, cluster analysis is performed on the identified interactive behavior feature sequences. The K-means clustering algorithm is used to divide the user's operation behavior into several categories, each representing a user operation pattern cluster. The number of clusters is determined based on features such as operation speed, click frequency, and device posture, and the distance from each data point to the cluster center is calculated to assign the data point to the nearest cluster. Finally, the user operation pattern clusters are spatially mapped to obtain a user behavior feature distribution map. Use data visualization tools, such as Python's Matplotlib library, to create a chart that shows the location of different operation modes in the feature space.

[0113] Step S35: performing difference measurement on the user behavior feature distribution map according to the preset user behavior baseline model to obtain interaction behavior deviation data;

[0114] Specifically, a user behavior baseline model can be defined, which contains the statistical characteristics of normal user behavior, such as the average click frequency, average operation speed, and the distribution range of device posture; this baseline model is built based on a large amount of normal user behavior data, and is used to evaluate and compare whether the current user's behavior deviates from the normal pattern. Use Python and the machine learning library scikit-learn to process the user behavior feature distribution map and the baseline model. Next, calculate the difference between the user behavior feature distribution map and the baseline model. Use metrics such as mean square error (MSE) or Kullback-Leibler divergence (KL divergence) to quantify the difference. For example, calculate the mean square error between the user's click frequency and the click frequency in the baseline model: ;in, is the actual click frequency of the user, is the corresponding click frequency in the baseline model, is the number of samples. In this way, we can get the interactive behavior deviation data, which reflects the degree of deviation between the user behavior and the normal behavior baseline. In this way, we can get the interactive behavior deviation data, which reflects the degree of deviation between the user behavior and the normal behavior baseline.

[0115] Step S36: quantifying the abnormality degree of the interaction behavior deviation data to obtain an interaction behavior abnormality scoring matrix, and performing credibility conversion on the interaction behavior abnormality scoring matrix to obtain a user interaction behavior credibility index.

[0116] Specifically, statistical methods can be used to determine the degree of abnormality based on the interaction behavior deviation data. The standardized deviation (Z score) between each user behavior feature and the baseline model is calculated, which can be expressed as: ;in, is the user's actual behavior feature value, is the corresponding behavioral feature mean in the baseline model, is the standard deviation of the corresponding behavior feature in the baseline model. Next, an anomaly score is assigned to each behavior feature based on the size of the Z score. For example, a threshold is set. When the absolute value of the Z score is greater than 3, the behavior feature is considered abnormal and a higher anomaly score is given. Finally, the interaction behavior anomaly score matrix is ​​converted to credibility. Use a logical function or other mapping function to convert the anomaly score into a credibility indicator, which reflects the credibility of the user behavior. For example, use the following logical function: ;in, It is an adjustment parameter used to control the range of credibility; in this way, the credibility index of user interaction behavior can be obtained.

[0117] The present invention enhances the consistency of behavior data by standardizing the device tilt angle sequence. By real-time monitoring of user interface interactive operations and adding timestamps, the system's ability to accurately capture user operation timing is improved. By integrating multi-dimensional information such as device posture, operation speed and click frequency, a comprehensive user behavior feature analysis can be performed, deepening the understanding of user behavior patterns. By clustering and analyzing the interactive behavior feature sequence, the user's operation mode cluster can be identified, and then the user's habitual behavior can be understood. By generating a user behavior feature distribution map, the user operation mode cluster is spatially distributed and mapped, and the user behavior feature is visualized, making it more intuitive and easy to analyze. By using a preset user behavior baseline model to measure the difference of the user behavior feature distribution map, the model's adaptability and comparison ability to different user behavior habits are improved. By accurately identifying abnormal behaviors and converting abnormal scoring matrices to credibility indicators, the system's ability to detect potential security threats is enhanced, and the accuracy and reliability of security authentication are improved. By real-time monitoring and analyzing the user's operation behavior, the present invention can dynamically adjust the security authentication strategy, enhance the dynamic and real-time nature of authentication, and provide users with a safer and more reliable authentication experience.

[0118] Preferably, step S4 comprises the following steps:

[0119] Step S41: collecting identification information of the user login device to obtain device fingerprint data, and verifying the integrity of the device fingerprint data to obtain device trusted status data;

[0120] Specifically, a device fingerprint collection tool can be developed that can run automatically when a user logs in. This tool uses the API provided by the operating system to collect information such as the device's hardware specifications, operating system version, installed software list, system settings, etc. For example, on the Windows operating system, use WMICollection and WMI services to obtain hardware information. Next, the collected information is encoded, for example, using a hash function (such as SHA-256) to generate a unique device fingerprint. Then, the device fingerprint data is integrity verified, which usually involves checking the consistency and integrity of the data to ensure that it has not been tampered with during transmission or storage. When the device fingerprint data is generated, the data is signed, and the signature is checked for validity during verification. If the signature is valid, the device trusted status data is obtained, which indicates that the device fingerprint data is complete and has not been tampered with.

[0121] Step S42: Scan the security elements of the user's login network to obtain network communication feature data, and perform protocol-level threat detection on the network communication feature data to obtain a network environment risk feature vector;

[0122] Specifically, a network monitoring system can be deployed that can capture and analyze data packets transmitted through the device network interface. Use network analysis tools such as Wireshark or tcpdump to capture network traffic, and use custom scripts or intrusion detection systems (IDS) to analyze these packets. Next, deep packet inspection (DPI) is performed on the captured packets to identify information such as the protocols used, the servers communicated, and the types of data transmitted. Pay attention to packets that communicate with known malicious servers or those that exhibit abnormal behavior (such as frequent connection attempts and large amounts of data transmission). Then, protocol-level threat detection is performed on the network communication feature data. This involves checking whether the data packet meets the standards of a specific protocol, whether there is abnormal header information, or whether there is suspicious payload content. For example, check whether the HTTP request contains suspicious URLs, whether there are commands that attempt to perform malicious operations, etc. Through these analyses, a network environment risk feature vector is obtained, which is a data set that contains potential threats and abnormal behaviors in network communications.

[0123] Step S43: performing spatiotemporal collection of the user's geographic location to obtain a user location trajectory sequence, and performing abnormal trajectory detection on the user location trajectory sequence to obtain a geographic location risk feature vector; recording the device trusted state data, the network environment risk feature vector, and the geographic location risk feature vector as an operating environment security data set;

[0124] Specifically, the device's GPS sensor can be used to collect the user's geographic location information. Develop a location tracking application that periodically requests the user's geographic location permission and records the timestamp and corresponding longitude and latitude coordinates. These data form the user's location trajectory sequence. Next, perform abnormal trajectory detection on the location trajectory sequence. Use a machine learning-based anomaly detection algorithm, such as the DBSCAN clustering algorithm, to identify abnormal patterns in location data. For example, if the user's location moves quickly from one place to another in a short period of time, and this movement speed is beyond the range that can be achieved by normal transportation, then this is marked as an abnormal trajectory. The detected abnormal trajectory is used as a geographic location risk feature vector. At the same time, the device trusted status data and the network environment risk feature vector are combined with the geographic location risk feature vector to form an operating environment security dataset.

[0125] Step S44: performing multi-level threat modeling on the operating environment security data set to obtain an operating environment threat feature matrix, and performing risk quantification calculation on the operating environment threat feature matrix to obtain an operating environment risk vector;

[0126] Specifically, a multi-level threat modeling framework can be defined, which includes the device layer, the network layer, and the geographic location layer. For each layer, a series of threat indicators are defined, such as malware presence indicators at the device layer, abnormal traffic indicators at the network layer, abnormal movement indicators at the geographic location layer, etc. Data mining techniques, such as random forests or gradient boosting machines (GBM), are used to model each threat indicator in the operating environment security dataset. These models can learn the association between each indicator and known threat cases and predict the threat level of unknown cases. Next, risk quantification is calculated for the operating environment threat feature matrix. A risk value is assigned to each threat indicator, which is based on the prediction results of the model. For example, if a device is predicted to have a high probability of malware, then the device layer threat indicator of this device will obtain a high risk value. Finally, the risk values ​​of all threat indicators at each layer are aggregated to obtain the operating environment risk vector.

[0127] Step S45: Performing hierarchical clustering evaluation on the operating environment risk vector to obtain an operating environment risk weight matrix, and performing weighted aggregation on the operating environment risk weight matrix to obtain an operating environment risk score;

[0128] Specifically, Python and the scikit-learn library can be used to process the operating environment risk vector. A hierarchical clustering method, such as the WARD clustering algorithm, is used to group the risk factors in the risk vector. This method can identify which risk factors are closely related and group them into the same cluster. Next, a weight is assigned to each cluster, which reflects the contribution of the risk factors in the cluster to the overall operating environment risk. These weights are set based on historical data or expert knowledge. For example, if a cluster contains risk factors related to malware, a higher weight is given to this cluster because the malware poses a great threat to the operating environment. Then, the operating environment risk weight matrix is ​​calculated, and each element in the matrix represents the weight of a risk factor. The operating environment risk score is obtained by weighted aggregation of these weights. In this way, a score that comprehensively reflects the risk of the operating environment can be obtained.

[0129] Step S46: Perform collaborative strategy mapping on the user interaction behavior credibility index and the operating environment risk score to obtain security authentication strategy mapping data;

[0130] Specifically, a collaborative strategy mapping model can be defined that determines the strength of the authentication strategy based on the user interaction behavior credibility index and the operating environment risk score. For example, if the user interaction behavior credibility is high and the operating environment risk score is low, a simpler authentication strategy, such as single-factor authentication, is selected. This mapping model is implemented using machine learning algorithms such as logistic regression or decision trees. The training model uses historical data that includes the user interaction behavior credibility index, the operating environment risk score, and the corresponding authentication strategy. Then, the trained model is used to predict the security authentication strategy for new user data. For example, if a user's interaction behavior credibility index is 0.8 and the operating environment risk score is 0.4, the model will predict that a two-factor authentication strategy should be adopted. Finally, the security authentication strategy mapping data is obtained, which includes the user interaction behavior credibility index, the operating environment risk score, and the corresponding authentication strategy.

[0131] Step S47: Adaptively adjust the security authentication policy mapping data to obtain authentication level mapping rules, and generate a multi-level authentication scheme set according to the authentication level mapping rules to obtain a multi-level security authentication scheme.

[0132] Specifically, a security authentication management system can be used, which is capable of receiving security authentication strategy mapping data, which includes user interaction behavior credibility indicators, operating environment risk scores, and predicted authentication strategies. Machine learning techniques, especially reinforcement learning, are used to analyze these data and adjust the authentication strategy to optimize security performance. A reinforcement learning model is defined, which is capable of selecting the best authentication strategy based on the state of the environment (i.e., user behavior and operating environment). The model learns through trial and error to determine which authentication strategy can maximize security while minimizing user inconvenience in different situations. Next, based on the learning results of the model, authentication level mapping rules are formulated. These rules specify the authentication strategies that should be adopted under different user behavior credibility and operating environment risk scores. For example, the rule stipulates that when the user behavior credibility is lower than 0.5 and the operating environment risk score is higher than 0.6, multi-factor authentication should be triggered. Finally, a set of multi-level authentication schemes is generated based on the authentication level mapping rules. This set includes various schemes from simple password authentication to multi-factor authentication. For example, a scheme set is designed, which includes the following schemes: Scheme 1: Password authentication only; Scheme 2: Password plus device fingerprint authentication; Scheme 3: Password, device fingerprint and geographic location verification; Scheme 4: Password, device fingerprint, geographic location and biometric verification; these schemes can be dynamically selected according to the authentication level mapping rules to adapt to the current security needs. In this way, a flexible multi-level security authentication scheme can be implemented.

[0133] The present invention significantly enhances the device security verification by collecting device fingerprint data and performing integrity verification. At the same time, by scanning the security elements of the user logging into the network and performing threat detection at the protocol level, the network security protection is effectively improved, and the security risks in the network environment can be identified. The security analysis through geographic location is strengthened, and the abnormal behavior of the user location can be identified by collecting the user's geographic location and abnormal trajectory detection in time and space. By integrating the risk feature vectors of equipment, network and geographic location, a comprehensive operating environment security assessment is carried out, and the security threats in the operating environment are deeply analyzed and quantified through multi-level threat modeling, thereby improving the accuracy of risk assessment. Through hierarchical clustering evaluation and weighted aggregation of operating environment risk vectors, security risks can be finely managed, and a quantitative basis can be provided for risk control. Through collaborative strategy mapping of user interaction behavior credibility indicators and operating environment risk scores, an accurate and adaptable security authentication strategy can be generated, and the authentication level can be flexibly adjusted according to actual security requirements, thereby enhancing the adaptability and flexibility of the authentication scheme. Finally, the present invention can provide a multi-level authentication scheme set according to different security requirements and risk levels, realize multi-level security authentication, and enhance the hierarchy and comprehensiveness of security authentication.

[0134] Preferably, step S5 comprises the following steps:

[0135] Step S51: Decomposing the user's bio-behavior feature signature into dimensions to obtain a bio-behavior feature vector;

[0136] Specifically, the user bio-behavioral feature signature contains facial recognition features and touch behavior features. Python and NumPy libraries are used to process these data. First, each feature category in the bio-behavioral feature signature is extracted separately to form an independent subset. For example, facial recognition features (such as the distance between the eyes, the width of the nose, etc.) are stored in one vector, and touch behavior features (such as sliding speed, touch pressure, etc.) are stored in another vector. Next, each subset is further processed to extract the most representative features. Deep learning methods such as autoencoders can be used to extract key features. For example, it is found that among the facial recognition features, the distance between the eyes and the width of the nose are the most useful features for user identification, so these features are retained in the bio-behavioral feature vector. Finally, a set of bio-behavioral feature vectors are obtained, each of which represents an aspect of the user's bio-behavioral feature signature.

[0137] Step S52: performing feature tensor fusion on the user according to the biological behavior feature vector, the user interaction behavior credibility index and the operating environment security data set to obtain a user security feature portrait;

[0138] Specifically, Python and NumPy libraries can be used to process biobehavioral feature vectors, user interaction behavior credibility indicators, and operating environment security datasets, and deep learning frameworks such as TensorFlow or PyTorch can be used to implement feature tensor fusion. First, the biobehavioral feature vectors, user interaction behavior credibility indicators, and operating environment security datasets are converted into tensor form. For example, the biobehavioral feature vector is a two-dimensional tensor, the user interaction behavior credibility indicator is a one-dimensional tensor, and the operating environment security dataset is a three-dimensional tensor. Next, tensor fusion techniques such as tensor decomposition or tensor networks are used to integrate these tensors. For example, CP decomposition is used to decompose the three tensors into a set of common factors and a specific factor, and then these factors are recombined to form a user security feature portrait. Finally, the user security feature portrait is obtained, which is a high-dimensional tensor that integrates the user's biobehavioral characteristics, interactive behavior credibility, and operating environment security information.

[0139] Step S53: Perform real-time feature recognition on the user security feature portrait to obtain a dynamic feature indicator matrix;

[0140] Specifically, for example, suppose that a user's behavior is being monitored while using a smartphone. A custom application is used that collects data such as the user's touch behavior, location changes, and device usage patterns in real time. This data is transmitted to the backend server for building a user security feature profile, where a machine learning model is running that can identify and extract key features of user behavior. This model is implemented using Python and scikit-learn libraries. The model extracts features by analyzing the user's touch patterns (such as sliding speed, click frequency) and device usage patterns (such as application switching frequency, number of screen unlocks). For example, if a user frequently switches applications in a short period of time, this indicates that the user is trying to evade monitoring or is engaging in suspicious activities. The model calculates these features in real time and forms a dynamic feature indicator matrix. Each row represents a feature value within a time window, and each column represents a specific behavioral feature. For example, the first column of the matrix contains the sliding speed in the past 5 minutes, the second column contains the click frequency, and so on. In this way, a dynamic feature indicator matrix that is updated in real time can be obtained, which reflects the user's behavior patterns at different points in time.

[0141] Step S54: performing credibility evaluation on the dynamic feature indicator matrix to obtain a time series credibility vector, and performing exponential conversion on the time series credibility vector to obtain a comprehensive dynamic credibility index;

[0142] Specifically, a credit assessment model can be used, which is able to judge the credibility of user behavior based on the user's behavioral characteristics. This model is implemented using Python and TensorFlow libraries. The model evaluates the credibility of user behavior by analyzing the eigenvalues ​​in the dynamic feature indicator matrix, such as sliding speed and click frequency. For example, if the user's sliding speed and click frequency are significantly different from the normal user behavior pattern, the model will give a lower credibility score. The model calculates a credibility score for each feature and combines these scores into a time series credibility vector. Each score is calculated based on the behavioral features within the current time window. Next, the time series credibility vector is exponentially transformed to obtain a comprehensive dynamic credibility index. Use an exponential function, such as the Sigmoid function, to transform each credibility score. The comprehensive dynamic credibility index is a value between 0 and 1, which reflects the overall credibility of the user's behavior. In this way, a comprehensive dynamic credibility index can be obtained, which integrates the credibility of the user's behavior at different time points.

[0143] Step S55: Perform policy matching on the multi-level security authentication scheme and the comprehensive dynamic credibility index to obtain a user authentication security policy set, and perform identity authentication and verification operations on the user according to the user authentication security policy set.

[0144] The present invention significantly improves the accuracy of individual identification through dimensional decomposition and feature vector extraction, and can accurately capture the unique biological behavior characteristics of individuals. By integrating the user's biological behavior, interactive behavior and operating environment security data through feature tensors, a comprehensive user security feature portrait is constructed, providing a multi-dimensional identity for the user. The real-time monitoring capability of the system is enhanced by real-time feature recognition and the generation of dynamic feature indicator matrix, and the changes in user behavior are responded to in a timely manner. The credibility of user behavior is quantified by converting the dynamic feature indicator matrix into a time series credibility vector and performing exponential conversion. The security authentication strategy is optimized by matching the multi-level security authentication scheme with the strategy of the comprehensive dynamic credibility index, and the adaptability and flexibility of the authentication process are improved. The risk response mechanism is strengthened by continuously monitoring the comprehensive dynamic credibility index of the user, and potential security risks are quickly identified and responded to. The user waiting time is reduced through an accurate and real-time authentication mechanism, the user experience is improved, and excessive disturbance of the user is avoided. Finally, accurate user identity authentication and verification operations provide solid security guarantees for transactions on the e-commerce platform, reduce the risks of fraud and illegal access, and ensure the security of transactions.

[0145] Preferably, the present invention further provides an e-commerce platform security authentication system for executing the e-commerce platform security authentication method as described above, the e-commerce platform security authentication system comprising:

[0146] The feature point positioning module is used to continuously collect a sequence of facial images of the user through the front camera of the user device to obtain an original facial image data stream; adaptively and dynamically locate the feature points of the original facial image data stream to obtain a personalized facial feature vector;

[0147] The feature fusion module is used to collect the user's touch behavior data to obtain the original touch behavior data stream; extract the operation features of the original touch behavior data stream to obtain the standardized touch feature vector; perform weighted fusion on the personalized facial feature vector and the standardized touch feature vector to obtain the user's biometric behavior feature signature;

[0148] The behavior assessment module is used to collect environmental context data from user devices to obtain an operating environment behavior data stream; perform user behavior pattern analysis on the operating environment behavior data stream to obtain a user behavior feature distribution map; perform real-time credibility assessment on the user behavior feature distribution map to obtain a user interaction behavior credibility index;

[0149] The environmental risk assessment module is used to collect security elements of the user's current login environment to obtain an operating environment security data set; conduct risk level assessment on the operating environment security data set to obtain an operating environment risk score; generate authentication strategies based on the user interaction behavior credibility index and the operating environment risk score to obtain a multi-level security authentication solution;

[0150] The user authentication module is used to fuse the user's biometric behavior feature signature, user interaction behavior credibility index and operating environment security data set to obtain a user security feature portrait; continuously monitor the user security feature portrait to obtain a comprehensive dynamic credibility index; and perform identity authentication and verification operations on users based on the multi-level security authentication scheme and the comprehensive dynamic credibility index.

[0151] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0152] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An e-commerce platform security authentication method, characterized in that: The following steps are involved: Step S1: continuously collecting a sequence of facial images of the user through the front camera of the user device to obtain an original facial image data stream; Adaptively locate dynamic feature points on the original facial image data stream to obtain personalized facial feature vectors; Step S2: collecting the user's touch behavior data to obtain an original touch behavior data stream; extracting operation features from the original touch behavior data stream to obtain a standardized touch feature vector; Perform weighted fusion on the personalized facial feature vector and the standardized touch feature vector to obtain the user's bio-behavioral feature signature; Step S3: collecting environmental context data from the user device to obtain an operating environment behavior data stream; performing user behavior pattern analysis on the operating environment behavior data stream to obtain a user behavior feature distribution map; Perform real-time credibility evaluation on the user behavior feature distribution map to obtain a user interaction behavior credibility index; wherein step S3 includes the following steps: Step S31: collecting attitude data of the user device gyroscope to obtain a device tilt angle sequence, and performing data standardization on the device tilt angle sequence to obtain standard device attitude data; Step S32: real-time monitoring of the user interface interactive operation to obtain an original operation event sequence, and timestamp the original operation event sequence to obtain a time-series operation data stream; operation speed feature extraction is performed on the time-series operation data stream to obtain operation speed feature data; Step S33: Perform frequency statistics on the user's click behavior to obtain a click event time series, and perform interval analysis on the click event time series to obtain click frequency feature data; integrate the standard device posture data, operation speed feature data and click frequency feature data to obtain an operation environment behavior data stream; Step S34: performing time-series behavior pattern recognition on the operation environment behavior data stream to obtain an interactive behavior feature sequence, and clustering the interactive behavior feature sequence to obtain a user operation pattern cluster; performing spatial distribution mapping on the user operation pattern cluster to obtain a user behavior feature distribution map; Step S35: performing difference measurement on the user behavior feature distribution map according to the preset user behavior baseline model to obtain interaction behavior deviation data; Step S36: quantify the abnormality of the interaction behavior deviation data to obtain an interaction behavior abnormality scoring matrix, wherein the interaction behavior abnormality scoring matrix includes a number of user behavior feature abnormality scores, and perform credibility conversion on the interaction behavior abnormality scoring matrix to obtain a user interaction behavior credibility index; wherein the specific calculation formula for the quantification of the abnormality is as follows: ； Where Z is the standardized deviation between each user behavior feature and the user behavior baseline model, is the user's actual behavior feature value, is the corresponding behavioral feature mean in the user behavior baseline model, is the standard deviation of the corresponding behavior feature in the user behavior baseline model; The credibility conversion of the interaction behavior abnormality scoring matrix is ​​performed by the following formula: ； Among them, P is the credibility index of user interaction behavior, e is the base of natural logarithm, is an adjustment parameter used to control the range of credibility. It is the sum of all user behavior feature anomaly scores in the interaction behavior anomaly score matrix; Step S4: Collect security factors of the user's current login environment to obtain an operating environment security data set; perform risk level assessment on the operating environment security data set to obtain an operating environment risk score; generate an authentication strategy based on the user interaction behavior credibility index and the operating environment risk score to obtain a multi-level security authentication solution; Step S5: Fuse the user's biometric behavior feature signature, user interaction behavior credibility index, and operating environment security data set to obtain a user security feature portrait; continuously monitor the user security feature portrait to obtain a comprehensive dynamic credibility index; and perform identity authentication and verification operations on the user based on the multi-level security authentication scheme and the comprehensive dynamic credibility index.

2. The electronic commerce platform security authentication method according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: positioning and tracking the user's face through the front camera of the user device to obtain a real-time face area coordinate data stream, and optimizing the spatial range of the real-time face area coordinate data stream to obtain accurate face area positioning data; Step S12: sampling the user's facial image continuous frames according to the precise facial region positioning data to obtain an original facial image data stream; Step S13: performing illumination compensation on the original facial image data stream to obtain a standardized facial image data stream, and performing time sequence integration and synchronization on the standardized facial image data stream to obtain an original facial image frame sequence; Step S14: Analyze the facial tissue features of the user based on the original facial image frame sequence to obtain a facial feature distribution density map, and dynamically locate the feature points of the user's face according to the facial feature distribution density map to obtain a facial feature point mapping solution; Step S15: extracting feature points from the facial feature point mapping scheme to obtain an original feature point coordinate set, and calculating the spatial geometric relationship of the original feature point coordinate set to obtain a facial feature point association matrix; Step S16: vectorizing the facial feature point association matrix to obtain a personalized facial feature vector.

3. The electronic commerce platform security authentication method according to claim 2, characterized in that: Step S14 includes the following steps: Step S141: selecting the best frame from the original facial image frame sequence to obtain the best facial image frame, and extracting facial features from the best facial image frame to obtain facial feature descriptor data; Step S142: performing density distribution calculation on the facial feature descriptor data based on the original facial image frame sequence to obtain facial feature density distribution data, and performing Gaussian kernel density estimation on the facial feature density distribution data to obtain a facial tissue feature distribution density map; Step S143: performing local extreme point detection on the facial tissue feature distribution density map to obtain a facial feature key point candidate set, and performing initial feature point positioning on the facial tissue feature distribution density map according to the facial feature key point candidate set to obtain a preliminary facial feature point positioning result; Step S144: tracking the spatiotemporal feature sequence of the original facial image frame sequence according to the preliminary facial feature point positioning result to obtain a dynamic facial feature point positioning solution; Step S145: performing affine transformation registration on the dynamic feature point sequence in the facial feature point dynamic positioning solution and the preset standard face feature model to obtain a facial feature point mapping solution.

4. The electronic commerce platform security authentication method according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: collecting pressure sensing data from the touch screen of the user device to obtain touch pressure time series data, and numerically normalizing the touch pressure time series data to obtain standardized pressure characteristic data; Step S22: collecting the track coordinates of the user's touch operation to obtain original touch track data; Step S23: smoothing the original touch trajectory data to obtain a smoothed touch trajectory coordinate sequence, and extracting features from the smoothed touch trajectory coordinate sequence to obtain touch trajectory feature data; Step S24: collecting time characteristics of the user's touch operation to obtain touch operation timing data, and performing rhythm pattern recognition on the touch operation timing data to obtain user operation rhythm characteristics; recording the standardized pressure characteristic data, the touch trajectory characteristic data and the user operation rhythm characteristics as the original touch behavior data stream; Step S25: extracting spatiotemporal features from the original touch behavior data stream to obtain a multi-dimensional touch feature matrix, and reducing the dimension of the multi-dimensional touch feature matrix to obtain an initial touch feature vector; performing normalized mapping on the initial touch feature vector to obtain a normalized touch feature vector; Step S26: perform feature importance evaluation on the personalized facial feature vector and the standardized touch feature vector to obtain a comprehensive feature weight coefficient, and perform weighted fusion on the personalized facial feature vector and the standardized touch feature vector according to the comprehensive feature weight coefficient to obtain a user biometric behavior feature signature.

5. The electronic commerce platform security authentication method according to claim 4, characterized in that: Step S26 includes the following steps: Step S261: performing self-information entropy calculation on the personalized facial feature vector to obtain a facial feature information entropy distribution matrix; Step S262: sorting the personalized facial feature vectors by importance according to the facial feature information entropy distribution matrix to obtain facial feature weight distribution coefficients; Step S263: performing principal component correlation analysis on the standardized touch feature vector to obtain a touch feature contribution matrix; Step S264: performing collaborative weight calculation on the standardized touch feature vector according to the touch feature contribution matrix to obtain a touch feature weight distribution coefficient, and using the facial feature weight distribution coefficient and the touch feature weight distribution coefficient as a comprehensive feature weight coefficient; Step S265: performing weighted combination of the personalized facial feature vector and the standardized touch feature vector according to the comprehensive feature weight coefficient to obtain an initial bio-behavior feature fusion matrix; Step S266: Perform feature hashing and encoding on the initial bio-behavioral feature fusion matrix to obtain the user's bio-behavioral feature signature.

6. The electronic commerce platform security authentication method according to claim 5, characterized in that: Step S263 includes the following steps: Step S2631: normalizing the dimension of the standardized touch feature vector to obtain normalized touch feature data; Step S2632: performing feature covariance calculation on the normalized touch feature data to obtain a touch feature covariance matrix, and performing inter-dimensional correlation coefficient calculation on the touch feature covariance matrix to obtain a touch feature correlation matrix; Step S2633: performing eigenvalue decomposition on the touch feature correlation matrix to obtain a touch feature spectrum decomposition matrix, and extracting principal eigenvectors from the touch feature spectrum decomposition matrix to obtain touch feature principal component basis vectors; Step S2634: Calculate the variance contribution of the touch feature principal component basis vectors to obtain a touch feature principal component contribution rate sequence; Step S2635: performing cumulative contribution screening on the touch feature principal component contribution rate sequence to obtain touch principal component importance ranking data; Step S2636: performing weight mapping on the feature dimensions in the standardized touch feature vector based on the touch principal component importance ranking data to obtain an initial touch feature contribution vector; Step S2637: performing nonlinear normalization transformation on the initial touch feature contribution vector to obtain the initial touch feature contribution matrix, and performing feature contribution distribution adjustment on the initial touch feature contribution matrix to obtain the touch feature contribution matrix.

7. The electronic commerce platform security authentication method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: collecting identification information of the user login device to obtain device fingerprint data, and verifying the integrity of the device fingerprint data to obtain device trusted status data; Step S42: Scan the security elements of the user's login network to obtain network communication feature data, and perform protocol-level threat detection on the network communication feature data to obtain a network environment risk feature vector; Step S43: performing spatiotemporal collection of the user's geographic location to obtain a user location trajectory sequence, and performing abnormal trajectory detection on the user location trajectory sequence to obtain a geographic location risk feature vector; recording the device trusted state data, the network environment risk feature vector, and the geographic location risk feature vector as an operating environment security data set; Step S44: performing multi-level threat modeling on the operating environment security data set to obtain an operating environment threat feature matrix, and performing risk quantification calculation on the operating environment threat feature matrix to obtain an operating environment risk vector; Step S45: Performing hierarchical clustering evaluation on the operating environment risk vector to obtain an operating environment risk weight matrix, and performing weighted aggregation on the operating environment risk weight matrix to obtain an operating environment risk score; Step S46: Perform collaborative strategy mapping on the user interaction behavior credibility index and the operating environment risk score to obtain security authentication strategy mapping data; Step S47: Adaptively adjust the security authentication policy mapping data to obtain authentication level mapping rules, and generate a multi-level authentication scheme set according to the authentication level mapping rules to obtain a multi-level security authentication scheme.

8. The electronic commerce platform security authentication method according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: Decomposing the user's bio-behavior feature signature into dimensions to obtain a bio-behavior feature vector; Step S52: performing feature tensor fusion on the user according to the biological behavior feature vector, the user interaction behavior credibility index and the operating environment security data set to obtain a user security feature portrait; Step S53: Perform real-time feature recognition on the user security feature portrait to obtain a dynamic feature indicator matrix; Step S54: performing credibility evaluation on the dynamic feature indicator matrix to obtain a time series credibility vector, and performing exponential conversion on the time series credibility vector to obtain a comprehensive dynamic credibility index; Step S55: Perform policy matching on the multi-level security authentication scheme and the comprehensive dynamic credibility index to obtain a user authentication security policy set, and perform identity authentication and verification operations on the user according to the user authentication security policy set.

9. An e-commerce platform security authentication system, characterized in that: Used to execute the e-commerce platform security authentication method as claimed in claim 1, the e-commerce platform security authentication system comprises: The feature point positioning module is used to continuously collect a sequence of facial images of the user through the front camera of the user device to obtain an original facial image data stream; adaptively and dynamically locate the feature points of the original facial image data stream to obtain a personalized facial feature vector; The feature fusion module is used to collect the user's touch behavior data to obtain the original touch behavior data stream; extract the operation features of the original touch behavior data stream to obtain the standardized touch feature vector; perform weighted fusion on the personalized facial feature vector and the standardized touch feature vector to obtain the user's biometric behavior feature signature; The behavior assessment module is used to collect environmental context data from user devices to obtain an operating environment behavior data stream; perform user behavior pattern analysis on the operating environment behavior data stream to obtain a user behavior feature distribution map; perform real-time credibility assessment on the user behavior feature distribution map to obtain a user interaction behavior credibility index; The environmental risk assessment module is used to collect security elements of the user's current login environment to obtain an operating environment security data set; conduct risk level assessment on the operating environment security data set to obtain an operating environment risk score; generate authentication strategies based on the user interaction behavior credibility index and the operating environment risk score to obtain a multi-level security authentication solution; The user authentication module is used to fuse the user's biometric behavior feature signature, user interaction behavior credibility index and operating environment security data set to obtain a user security feature portrait; continuously monitor the user security feature portrait to obtain a comprehensive dynamic credibility index; and perform identity authentication and verification operations on users based on the multi-level security authentication scheme and the comprehensive dynamic credibility index.

Images