Non-member customer loss prevention method and system based on payment opening and face recognition

Abstract

The invention relates to a non-member customer loss prevention method and system based on payment opening and face recognition, and the method comprises the steps: obtaining the payment opening, a face image and dynamic behavior data of a non-member customer in real time, and enabling the dynamic behavior data to comprise a shopping path, standing time and moving speed; performing preliminary analysis on the dynamic behavior data by using an embedded edge computing device, identifying abnormal customer behaviors, and generating a preliminary risk assessment result; based on the preliminary risk assessment result, the payment openid, the face image and the dynamic behavior data are sent to a data center for secondary verification, and a verification result is generated; and if the verification result shows that the customer is matched with the blacklist or has an abnormal behavior, triggering an alarm and refusing the transaction, recording transaction information, customer characteristics and identification time, and generating and storing a transaction report. The method has the effect of improving the accuracy of the loss prevention system.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and in particular to a method and system for preventing loss of non-member customers based on payment openid and face recognition. Background Art

[0002] With the rapid development of e-commerce and physical retail, payment methods have gradually diversified, and the transaction frequency and traffic of non-member customers have increased significantly. However, the convenience of transaction methods has also brought increasingly severe security challenges to merchants. Especially in the cashier link, how to effectively prevent blacklisted personnel from using non-member identities to complete malicious transactions has become a major problem in merchant operations. Traditional loss prevention methods mainly rely on manual monitoring or simple member information comparison. This method is not only inefficient, but also prone to inaccurate identification due to human errors. In addition, during the transaction process of non-member customers, it is difficult for merchants to effectively track the identity of customers, further increasing the difficulty of loss prevention. Although some merchants have begun to adopt facial recognition technology, a single identification method lacks flexibility, especially in the context of diversified payment methods, and cannot fully cover all transaction risk scenarios.

[0003] The above-mentioned existing technical solutions have the following defects: the existing technologies mostly rely on payment openid or a single face recognition technology, the recognition accuracy is low, and it is difficult to achieve full coverage in some specific scenarios, so there is room for improvement. Summary of the invention

[0004] In order to improve the accuracy of the loss prevention system, the present application provides a method, system, device and storage medium for frequency conversion switching of industrial frequency.

[0005] The above-mentioned invention objective of the present application is achieved through the following technical solutions: A non-member customer loss prevention method based on payment openid and face recognition, the non-member customer loss prevention method based on payment openid and face recognition comprising: Real-time acquisition of payment openid, face image and dynamic behavior data of non-member customers, wherein the dynamic behavior data includes shopping path, standing time and moving speed; Using embedded edge computing devices to conduct preliminary analysis of the dynamic behavior data, identify abnormal customer behavior, and generate preliminary risk assessment results; Based on the preliminary risk assessment result, the payment openid, the face image and the dynamic behavior data are sent to a data center for secondary verification to generate a verification result; If the verification result shows that the customer matches the blacklist or has abnormal behavior, an alarm is triggered and the transaction is rejected, and the transaction information, customer characteristics and identification time are recorded, and a transaction report is generated for storage; Use machine learning algorithms to dynamically analyze and predict historical transaction data, customer behavior data, and blacklist information, and update blacklist data in real time.

[0006] By adopting the above technical solution, the system can capture the customer's identity information and behavioral characteristics at the first time of the transaction by obtaining the customer's payment openid, facial image and dynamic behavior data in real time. This provides basic data for subsequent real-time risk assessment and fraud detection. At the same time, the customer's payment information, facial image and behavior data are collected to ensure a comprehensive analysis of the customer's identity and behavior, helping the loss prevention system to make accurate decisions and avoid misjudgments that may be caused by a single data source; the embedded edge computing device can process the customer's payment openid and dynamic behavior data locally, reducing dependence on the central server, thereby reducing latency and improving the system's real-time response capability. By analyzing the customer's behavior data on the edge computing device, the system can identify whether the customer has abnormal behavior during the transaction, and can promptly detect potential fraud risks and reduce the burden of subsequent processing; after the initial analysis is completed, the data is transmitted to the data center for secondary verification. During this process, the system will conduct detailed analysis based on more information, further improving the recognition accuracy; if the customer is judged to be high risk, the system will promptly trigger an alarm and reject the transaction, thereby preventing potential fraudulent transactions. This rapid response mechanism can significantly reduce merchants’ losses and protect the security of the platform. By using machine learning algorithms to dynamically analyze historical data and behavioral patterns, the system can intelligently identify and predict potential fraudulent behaviors and update the blacklist in real time to ensure that the system can adapt to new fraud methods.

[0007] In a preferred example, the present application can be further configured as follows: the dynamic behavior data also includes the length of time the non-member customer stays during the payment process, the number of times the product is touched, and whether the customer frequently changes the shopping location in a short period of time.

[0008] By adopting the above technical solution and conducting detailed analysis of customer behavior patterns, potential fraud risks can be identified in advance. Behaviors such as staying too long, frequently touching products or frequently changing shopping locations are usually inconsistent with the purchasing behavior of normal customers, which can effectively help identify customers who try to circumvent payment.

[0009] In a preferred example, the present application may be further configured as follows: the use of the embedded edge computing device to perform preliminary analysis on the dynamic behavior data, identify abnormal customer behavior, and generate preliminary risk assessment results includes: Calculating the stay pattern of the non-member customer based on the shopping path and standing time of the non-member customer, and judging the stay pattern of the non-member customer to determine whether there is potential fraudulent behavior or abnormal shopping behavior, and generating a judgment result; The dynamic behavior data is used to evaluate the movement trajectory of the non-member customer. When the non-member customer frequently changes shopping locations or stays in certain commodity areas for a long time, it is identified whether the behavior pattern meets the requirements of escaping the order or committing fraud, and an identification result is generated; Based on the determination result and the identification result, a preliminary assessment of the customer risk is performed, and the preliminary risk assessment result is generated.

[0010] By adopting the above technical solution, by calculating the customer's shopping path and standing time, the system can determine whether the customer's behavior conforms to the normal shopping pattern. By dynamically calculating the customer's behavior pattern, the manual monitoring and inefficient judgment of customer behavior in traditional methods are reduced, and the automation and accuracy of the system are increased; by evaluating the customer's movement trajectory, the system can determine whether the customer has escaped the bill or committed fraud, and behavioral analysis can help identify these abnormal activities that do not conform to the normal shopping pattern, thereby effectively warning; through real-time monitoring and analysis of customer behavior, the system can automatically make risk assessments based on behavioral patterns, provide data support for merchants, help merchants optimize anti-fraud decisions, reduce manual intervention, and improve decision-making efficiency.

[0011] In a preferred example, the present application may be further configured as follows: based on the preliminary risk assessment result, the payment openid, the face image and the dynamic behavior data are sent to the data center for secondary verification, and the generated verification result includes: When the preliminary risk assessment result is abnormal shopping behavior, the payment openid of the non-member customer is compared with the blacklist data to generate a list comparison result; Comparing the real-time facial image of the non-member customer with the facial features stored in the data center, using liveness detection technology to verify the authenticity of the non-member customer's identity, and generating a facial comparison result; Perform a comprehensive analysis based on the dynamic behavior data and historical risk behavior patterns to determine whether the customer has engaged in fraudulent behavior and generate behavior analysis results; The decision tree algorithm is used to summarize and analyze the list comparison results, the face comparison results and the behavior analysis results to generate the verification result.

[0012] By adopting the above technical solutions, through automated list comparison, the system can quickly screen out customers matching the blacklist without human intervention, significantly improving loss prevention efficiency and reducing human omissions; through liveness detection, the system can more effectively identify and prevent identity theft such as photo fraud and face disguise, enhancing the security and accuracy of identity authentication and reducing the risk of false identity use; by combining historical risk behavior patterns with real-time behavior data, the system can analyze customer behavior from multiple dimensions, improve the accuracy of identifying potential fraudulent behaviors, reduce misjudgments and missed judgments, and help the system dynamically identify some potential fraudulent behaviors that are not on the blacklist, thereby improving risk control capabilities; through the decision tree algorithm, the system can make intelligent decisions between multiple verification results and automatically determine whether a customer belongs to a high-risk group without human intervention, thereby improving the efficiency and automation level of the system.

[0013] In a preferred example, the present application may be further configured as follows: triggering an alarm and rejecting a transaction includes: When the behavior of the non-member customer matches the risk pattern but does not match the blacklist data, a low risk alert is issued and the merchant is triggered to pay attention to the information.

[0014] When the non-member customer matches the blacklist data or the behavior pattern of the non-member customer is abnormal, a high-risk alert is generated and a transaction rejection message is triggered.

[0015] By adopting the above technical solution and distinguishing between low-risk alarms and high-risk alarms, the system can flexibly respond to various risk scenarios, avoiding excessive interference with low-risk customers and ensuring that high-risk customers are intercepted in a timely manner. The triggering mechanism of low-risk alarms leaves merchants room for flexible response. Merchants can decide whether to further verify customer information based on the information provided by the system. The automatic transaction rejection mechanism of high-risk alarms ensures a quick response, immediately blocks potential fraud, and avoids the expansion of losses.

[0016] In a preferred example, the present application may be further configured as follows: the use of machine learning algorithms to dynamically analyze and predict historical transaction data, customer behavior data, and blacklist information, and real-time update of blacklist data includes: Based on the historical transaction data, a behavior analysis model is trained using a k-means algorithm to generate a trained behavior analysis model; According to the real-time transaction data and customer behavior data, the potential risk behavior is predicted by the trained behavior analysis model to obtain a prediction result, and the blacklist data is dynamically adjusted according to the prediction result; Through the data feedback mechanism, new high-risk behavior patterns are automatically identified and added to the blacklist data in real time.

[0017] By adopting the above technical solution and analyzing historical transaction data through the k-means algorithm, the system can build an accurate behavior analysis model to identify normal and abnormal customer behavior patterns; the trained behavior analysis model can analyze customer behavior in real time and dynamically adjust the blacklist based on the prediction results to prevent potential fraudulent transactions and ensure the efficiency of the loss prevention system. The system automatically identifies new high-risk behaviors through a data feedback mechanism and updates the blacklist data in real time to ensure that the system can adapt to changing fraud methods and continuously optimize risk identification capabilities. The automation and intelligence of the entire process greatly reduces manual intervention and improves the efficiency and accuracy of risk identification and blacklist management.

[0018] In a preferred example, the present application can be further configured as follows: the non-member customer loss prevention method based on payment openid and face recognition also includes: During the payment process of the non-member customer, further determine whether the non-member customer has a fraud risk by analyzing the non-member customer's facial expression and eye movement trajectory. If a fraud risk is detected, trigger an alarm and reject the transaction; A customer credit scoring model is constructed through transaction history data, and each non-member customer is scored. In the next payment process, different verification measures are taken according to the score.

[0019] By adopting the above technical solution, through the combined analysis of facial expressions and eye movement trajectories, compared with traditional identity authentication methods, emotional risk identification is added. This method can capture more subtle and difficult-to-disguise fraud behaviors, and improve the accuracy and depth of the loss prevention system; through the credit scoring model, the system can assess the customer's potential risks in advance and make dynamic adjustments during the payment process. This can avoid excessive verification of all customers while ensuring transaction security, thereby improving transaction efficiency.

[0020] The second object of the invention is achieved by the following technical solutions: A non-member customer loss prevention system based on payment openid and face recognition, the non-member customer loss prevention system based on payment openid and face recognition comprising: An information acquisition module is used to acquire the payment openid, face image and dynamic behavior data of non-member customers in real time, wherein the dynamic behavior data includes shopping path, standing time and moving speed; A preliminary assessment module, for performing preliminary analysis on the dynamic behavior data using an embedded edge computing device, identifying abnormal customer behavior, and generating preliminary risk assessment results; A secondary verification module, for sending the payment openid, the face image and the dynamic behavior data to a data center for secondary verification based on the preliminary risk assessment result, and generating a verification result; An alarm module is used to trigger an alarm and reject the transaction if the verification result shows that the customer matches the blacklist or has abnormal behavior, and record the transaction information, customer characteristics and identification time, and generate a transaction report for storage; The update module is used to use machine learning algorithms to dynamically analyze and predict historical transaction data, customer behavior data, and blacklist information, and update blacklist data in real time.

[0021] By adopting the above technical solution, the system can capture the customer's identity information and behavioral characteristics at the first time of the transaction by obtaining the customer's payment openid, facial image and dynamic behavior data in real time. This provides basic data for subsequent real-time risk assessment and fraud detection. At the same time, the customer's payment information, facial image and behavior data are collected to ensure a comprehensive analysis of the customer's identity and behavior, helping the loss prevention system to make accurate decisions and avoid misjudgments that may be caused by a single data source; the embedded edge computing device can process the customer's payment openid and dynamic behavior data locally, reducing dependence on the central server, thereby reducing latency and improving the system's real-time response capability. By analyzing the customer's behavior data on the edge computing device, the system can identify whether the customer has abnormal behavior during the transaction, and can promptly detect potential fraud risks and reduce the burden of subsequent processing; after the initial analysis is completed, the data is transmitted to the data center for secondary verification. During this process, the system will conduct detailed analysis based on more information, further improving the recognition accuracy; if the customer is judged to be high risk, the system will promptly trigger an alarm and reject the transaction, thereby preventing potential fraudulent transactions. This rapid response mechanism can significantly reduce merchants’ losses and protect the security of the platform. By using machine learning algorithms to dynamically analyze historical data and behavioral patterns, the system can intelligently identify and predict potential fraudulent behaviors and update the blacklist in real time to ensure that the system can adapt to new fraud methods.

[0022] In summary, the present application includes at least one of the following beneficial technical effects: 1. By acquiring the customer's payment openid, facial image and dynamic behavior data in real time, the system can capture the customer's identity information and behavioral characteristics at the first moment of the transaction. This provides basic data for subsequent real-time risk assessment and fraud detection. At the same time, the customer's payment information, facial image and behavior data are collected to ensure a comprehensive analysis of the customer's identity and behavior, helping the loss prevention system to make accurate decisions and avoid misjudgments that may be caused by a single data source; embedded edge computing devices can process the customer's payment openid and dynamic behavior data locally, reducing dependence on central servers, thereby reducing latency and improving the system's real-time response capabilities. By analyzing the customer's behavior data on the edge computing device, the system can identify whether the customer has abnormal behavior during the transaction, and can promptly detect potential fraud risks and reduce the burden of subsequent processing; 2. After the initial analysis is completed, the data is transmitted to the data center for secondary verification. During this process, the system will integrate more information for detailed analysis, further improving the accuracy of identification; if the customer is judged to be high risk, the system will promptly trigger an alarm and reject the transaction, thereby preventing potential fraudulent transactions. This rapid response mechanism can significantly reduce the merchant's losses and protect the security of the platform; using machine learning algorithms to dynamically analyze historical data and behavior patterns, the system can intelligently identify and predict potential fraudulent behavior, and update the blacklist in real time to ensure that the system can adapt to new fraud methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of a non-member customer loss prevention method based on payment openid and face recognition in one embodiment of the present application; Figure 2 This is a flowchart for implementing step S10 in a non-member customer loss prevention method based on payment openid and face recognition in one embodiment of the present application; Figure 3 This is a flowchart for implementing step S20 in a non-member customer loss prevention method based on payment openid and face recognition in one embodiment of the present application; Figure 4 This is a flowchart for implementing step S30 in a non-member customer loss prevention method based on payment openid and face recognition in one embodiment of the present application; Figure 5 This is a flowchart for implementing step S40 in a non-member customer loss prevention method based on payment openid and face recognition in one embodiment of the present application; Figure 6 This is a flowchart for implementing step S50 in a non-member customer loss prevention method based on payment openid and face recognition in one embodiment of the present application; Figure 7This is a principle block diagram of a non-member customer loss prevention system based on payment openid and face recognition in one embodiment of the present application; Figure 8 It is a schematic diagram of a device in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The present application is further described in detail below in conjunction with the accompanying drawings.

[0025] In one embodiment, if Figure 1 As shown, the present application discloses a non-member customer loss prevention method based on payment openid and face recognition, which specifically includes the following steps: S10: Acquire the payment openid, face image and dynamic behavior data of non-member customers in real time, wherein the dynamic behavior data includes shopping path, standing time and moving speed.

[0026] Specifically, the payment openid is a unique identifier that is automatically generated when a customer conducts a transaction on a payment platform. This identifier is usually bound to the customer's account and transmitted in real time during the payment process. The system will obtain the customer's payment openid through the interface with the payment platform; the customer's facial image is collected in real time by face recognition cameras installed at the checkout counter or various locations in the store, and the customer's facial features are extracted using image processing technology and face recognition algorithms and uploaded to the background for comparison; dynamic behavior data is obtained in real time through sensors and video surveillance equipment installed in the store. The sensors calculate the customer's shopping path based on the customer's movement trajectory in the mall, and the monitoring equipment tracks the customer's movement speed, length of stay, etc. in the mall area, and determines the customer's standing time and movement speed based on the customer's walking trajectory. These data are transmitted to the system background in real time to help evaluate the customer's behavior patterns in the mall.

[0027] S20: Use embedded edge computing devices to conduct preliminary analysis of dynamic behavior data, identify abnormal customer behavior, and generate preliminary risk assessment results.

[0028] Specifically, edge computing devices are embedded in checkout counters or store entrances, and use local computing power to process customers' dynamic behavior data in real time. The processor of the computing device performs preliminary screening and analysis of the customer's behavior data, and calculates various parameters of customer behavior. For example, whether the customer's movement speed and standing time exceed the normal range, or whether the customer frequently changes product areas in a short period of time. The edge device performs preliminary abnormal behavior detection based on these indicators. For example, if a customer stays in the same area for a long time, or frequently wanders between multiple product areas, the edge computing device will mark the customer's behavior as suspicious, and preliminarily infer that the customer may be at risk of escaping the bill or fraud, and ultimately generate a preliminary risk assessment result.

[0029] S30: Based on the preliminary risk assessment results, the payment openid, face image and dynamic behavior data are sent to the data center for secondary verification to generate a verification result.

[0030] Specifically, when the preliminary risk assessment results show that the customer's behavior is abnormal, the edge device will upload the customer's payment openid, facial image and dynamic behavior data to the central data center through an encrypted channel for further processing and verification. The data center has more powerful computing power and more data sources, and can make more detailed risk judgments through a large amount of historical data and customer behavior models. The data center verifies whether the customer has ever engaged in suspicious behavior by comparing it with historical transaction data, or by matching the facial image uploaded by the customer with the facial feature data stored in the database, and using liveness detection technology to further confirm the authenticity of the customer's identity. If the customer's identity does not match or the behavior pattern is abnormal, the data center will generate a final verification result to decide whether to mark the customer as high risk and take further measures.

[0031] S40: If the verification result shows that the customer matches the blacklist or has abnormal behavior, an alarm is triggered and the transaction is rejected. At the same time, the transaction information, customer characteristics and identification time are recorded, and a transaction report is generated for storage.

[0032] Specifically, if the verification result shows that the customer matches the blacklist or has abnormal behavior, for example, the customer has been marked for skipping the bill in the past, or his behavior is highly similar to historical fraud, the system will immediately trigger an alarm, and automatically send an alarm to the cashier terminal or the merchant backend through the system, indicating that the customer is a high-risk customer, and the transaction will be rejected immediately. At the same time, all transaction information will be automatically recorded and stored in the backend database by the system, and the transaction report will contain detailed information such as the customer's identity information, risk assessment results and final judgment.

[0033] S50: Use machine learning algorithms to dynamically analyze and predict historical transaction data, customer behavior data, and blacklist information, and update blacklist data in real time.

[0034] Specifically, through machine learning algorithms, the system dynamically analyzes historical transaction data, customer behavior data, and blacklist information, and learns the relationship between customer behavior patterns and fraudulent behavior. When a customer's behavior data is highly similar to a historical fraud pattern, the machine learning algorithm predicts that the customer is a potential fraudulent customer, updates the blacklist data in a timely manner, and automatically adds the potential risk customer to the blacklist. This process is real-time. As more transaction data flows in, the model will automatically adjust the accuracy of the prediction based on the new data, and update the blacklist data in a timely manner to avoid omissions due to changes in fraud patterns.

[0035] By adopting the above technical solution, the system can capture the customer's identity information and behavioral characteristics at the first time of the transaction by obtaining the customer's payment openid, facial image and dynamic behavior data in real time. This provides basic data for subsequent real-time risk assessment and fraud detection. At the same time, the customer's payment information, facial image and behavior data are collected to ensure a comprehensive analysis of the customer's identity and behavior, helping the loss prevention system to make accurate decisions and avoid misjudgments that may be caused by a single data source; the embedded edge computing device can process the customer's payment openid and dynamic behavior data locally, reducing dependence on the central server, thereby reducing latency and improving the system's real-time response capability. By analyzing the customer's behavior data on the edge computing device, the system can identify whether the customer has abnormal behavior during the transaction, and can promptly detect potential fraud risks and reduce the burden of subsequent processing; after the initial analysis is completed, the data is transmitted to the data center for secondary verification. During this process, the system will conduct detailed analysis based on more information, further improving the recognition accuracy; if the customer is judged to be high risk, the system will promptly trigger an alarm and reject the transaction, thereby preventing potential fraudulent transactions. This rapid response mechanism can significantly reduce merchants’ losses and protect the security of the platform. By using machine learning algorithms to dynamically analyze historical data and behavioral patterns, the system can intelligently identify and predict potential fraudulent behaviors and update the blacklist in real time to ensure that the system can adapt to new fraud methods.

[0036] In one embodiment, if Figure 2 As shown, in step S10, the dynamic behavior data also includes: S101: Dynamic behavior data also includes the length of time that non-member customers stay during the payment process, the number of times they touch the product, and whether the customer frequently changes shopping locations within a short period of time.

[0037] Specifically, when a customer enters a certain area of ​​a store, the sensor will record the time when the customer enters and leaves the area. Using this data, the system can calculate the length of time the customer stays in the area; when a customer touches, picks up or puts down a product, the sensor will record the contact information of the product and associate it with the customer's identity; the system will record the customer's position changes in the mall, especially the customer's movement speed in a short period of time and the behavior of frequently changing positions. For example, if a customer stays in a certain product area for more than 30 seconds, the system will consider the customer's stay time in the area to be abnormal, which may mean that the customer's behavior is inconsistent with the normal shopping process, and there is a possibility of escaping the bill or other abnormal behavior. The system can calculate the customer's stay time in different product areas to obtain the customer's stay pattern, and compare this pattern with historical fraud behavior data. If the customer's stay pattern is similar to a known fraud pattern, the customer will be marked as a potential high-risk customer, triggering further risk assessment or alarm.

[0038] In one embodiment, if Figure 3 As shown, in step S20, the dynamic behavior data is preliminarily analyzed using the embedded edge computing device to identify abnormal customer behavior and generate preliminary risk assessment results, which specifically include: S21: Calculate the stay pattern of non-member customers based on their shopping paths and standing time, and judge the stay pattern of non-member customers to determine whether there is potential fraud or abnormal shopping behavior, and generate a judgment result.

[0039] Specifically, by installing multiple sensors or video surveillance equipment in the store, the shopping paths of non-member customers, that is, the walking routes of customers in the store, can be tracked in real time, and their stay patterns can be calculated by combining the customer's standing time, that is, the length of time the customer stays in each area. For example, after a customer enters the store, the system obtains the customer's movement trajectory in different commodity areas in real time through video surveillance, and records the length of time the customer stays in each area. If a customer stays in a high-value commodity area for more than a predetermined time, such as more than 5 minutes, without showing obvious purchase intentions or picking up the goods, the behavior will be considered abnormal. By calculating the customer's stay pattern, customers who stay in certain commodity areas for a long time can be identified, which may mean that they are waiting for a certain opportunity to engage in improper behavior, such as escaping or stealing. If the customer's stay pattern is highly similar to historical fraudulent behavior, for example, staying in the same area for a long time many times but not purchasing goods, a judgment result will be generated, indicating that the customer has a potential fraud risk.

[0040] S22: Use dynamic behavior data to evaluate the movement trajectory of non-member customers. When non-member customers frequently change their shopping locations or stay in certain product areas for a long time, identify whether they meet the behavior patterns of skipping the bill or committing fraud and generate identification results.

[0041] Specifically, through the installation of multiple sensors and video surveillance equipment in the store, the customer's movement trajectory is tracked and recorded in real time. For example, after a customer enters the store, the system records his or her movement route and monitors the customer's behavior. If the customer frequently changes the shopping location or moves quickly in a short period of time, this behavior will be marked as suspicious. If a customer walks back and forth between multiple areas without obvious purchase actions, the system will consider that his or her behavior may not conform to the normal shopping pattern. The system will compare such behavior patterns with historical bill-skipping or fraudulent behaviors. If they match, the system will further evaluate whether the customer has engaged in bill-skipping or fraudulent behaviors. For example, if a customer stays in a high-value commodity area for a long time and moves frequently without paying, it will be identified as a potential fraudulent behavior, and the system will generate an identification result indicating that the customer's behavior may be bill-skipping or other fraudulent behavior. This identification result will be provided to the merchant as a basis for further decision-making, such as whether to trigger an alarm or reject the transaction.

[0042] S23: Based on the determination results and identification results, a preliminary assessment of customer risks is conducted and a preliminary risk assessment result is generated.

[0043] Specifically, based on the customer's stay pattern determination results and behavior recognition results, the system can conduct a comprehensive preliminary risk assessment. By comprehensively calculating the customer's stay pattern and behavior pattern, if the customer's behavior is highly similar to historical high-risk behavior or fraud patterns, and the dynamic behavior data evaluation identifies that the customer's behavior does not conform to normal shopping rules, the customer will be marked as a high-risk customer. This assessment is not only based on the customer's real-time behavior data in the store, but also combined with the customer's historical behavior patterns, and based on this information, a preliminary assessment of the customer's risk level is made. The result of the customer risk assessment is a comprehensive score. The higher the score, the greater the possibility that the customer has fraudulent behavior. If the assessment result shows a high risk, the merchant can take further measures, such as triggering an alarm, manual verification, and refusing transactions.

[0044] In one embodiment, if Figure 4 As shown, in step S30, based on the preliminary risk assessment result, the payment openid, face image and dynamic behavior data are sent to the data center for secondary verification to generate a verification result, which specifically includes: S31: When the preliminary risk assessment result is abnormal shopping behavior, the payment openid of the non-member customer is compared with the blacklist data to generate a list comparison result.

[0045] Specifically, when a customer's behavior is initially assessed as abnormal or inconsistent with normal shopping patterns, the customer's payment openid is obtained in real time through the identifier in the payment process. After acquisition, the payment openid will be compared with the blacklist data in the database. The blacklist data usually contains the payment information, behavior patterns, etc. of the recorded fraudulent customers. By comparing the customer's payment openid with the payment openid in the blacklist, it is determined whether the customer is a known fraudster or a potential risk customer. If there is a match, the customer is marked as high risk and a list comparison result is generated for subsequent risk analysis and processing. During the comparison process, if the customer's payment openid finds a match in the blacklist, it means that the customer has been identified as a fraudster in the past. The system will mark the customer as high risk and record the corresponding information for the merchant's reference.

[0046] S32: Compare the real-time facial image of the non-member customer with the facial features stored in the data center, use liveness detection technology to verify the authenticity of the non-member customer's identity, and generate a facial comparison result.

[0047] Specifically, when a customer makes a payment, real-time facial images are collected in real time through high-definition cameras in the store. The cameras can capture the customer's facial features and use image processing technology to identify their facial features. After obtaining the customer's facial images, these images will be compared with the facial feature data pre-stored in the data center. Liveness detection technology monitors whether the customer's face has live features and identifies whether it is a real face rather than a photo or video. If a customer's facial reaction is detected, such as blinking of the eyes, movement of the mouth, changes in facial expressions, etc., the customer is confirmed to be real and a face comparison result is generated. If the liveness detection fails, it indicates that the customer may have used forged identity information, which in turn triggers an alarm or further identity verification requirements.

[0048] S33: Conduct a comprehensive analysis based on dynamic behavior data and historical risk behavior patterns to determine whether the customer has engaged in fraudulent behavior and generate behavioral analysis results.

[0049] Specifically, by obtaining the customer's dynamic behavior data, such as shopping paths, length of stay, number of product contacts, location changes, etc., the customer's behavior will be compared with historical risk behavior patterns. Historical risk behavior patterns usually include characteristics of past fraudulent behavior, such as whether the customer frequently stays in a specific product area, whether the customer frequently changes location without the intention to pay, etc. These behaviors are often consistent with skipping orders or fraudulent behavior. By comprehensively analyzing the similarities between the customer's dynamic behavior data and historical behavior data, the system will determine whether the customer has potential fraudulent behavior. For example, if the customer frequently changes product areas and stays too long, or lingers in a high-value product area without paying, this will be judged as suspicious behavior, generating a behavioral analysis result to help further identify the customer's fraud risk.

[0050] S34: Use a decision tree algorithm to summarize and analyze the list comparison results, face comparison results, and behavior analysis results to generate verification results.

[0051] Specifically, the customer's list comparison results, face comparison results and behavior analysis results are summarized and analyzed through a decision tree algorithm. The decision tree classifies customers according to the priority of each feature. For example, if the customer's payment openid matches the blacklist and the face comparison passes, it is directly marked as high risk; if the customer's behavior analysis results show consistency with historical fraud patterns and the face verification passes, the customer is judged as high risk, and a verification result is generated. Based on this, it is decided whether to reject the transaction or require further verification. The decision tree model prioritizes different verification results, automatically evaluates the risk level of each verification feature through a tree structure, and finally generates a comprehensive judgment result.

[0052] In one embodiment, if Figure 5 As shown, in step S40, if the verification result shows that the customer matches the blacklist or has abnormal behavior, an alarm is triggered and the transaction is rejected. At the same time, the transaction information, customer characteristics and identification time are recorded, and a transaction report is generated for storage, which specifically includes: S41: When the behavior of a non-member customer matches the risk pattern but does not match the blacklist data, a low-risk alert is issued and the merchant is triggered to pay attention to the information.

[0053] Specifically, when a customer's behavior pattern is similar to a historical fraud behavior pattern, but the customer's payment openID does not match the information in the blacklist, the system will consider the customer's behavior to have a certain risk, but not enough to directly mark it as a high-risk customer, and thus trigger a low-risk alarm. In this case, the system will send an alarm message to the merchant through the background, reminding the merchant that the customer's behavior deserves attention, but there is no need to take immediate compulsory measures, such as refusing the transaction.

[0054] S42: When a non-member customer matches the blacklist data or the behavior pattern of the non-member customer is abnormal, a high-risk alert is generated and a transaction rejection message is triggered.

[0055] Specifically, if the customer's payment openid matches the blacklist data, indicating that the customer may have a history of fraudulent behavior, or the customer's behavior pattern is highly similar to known fraudulent behavior patterns, the customer will be marked as a high-risk customer. At this time, the system will trigger a high-risk alarm, generate a transaction rejection message, and automatically notify the merchant to reject the customer's transaction. The merchant can use this data to determine whether the customer is a malicious customer, and will immediately prevent the customer from continuing the transaction to prevent fraud.

[0056] In one embodiment, if Figure 6 As shown, in step S50, the historical transaction data, customer behavior data and blacklist information are dynamically analyzed and predicted using a machine learning algorithm, and the blacklist data is updated in real time, specifically including: S51: Based on the historical transaction data, the behavior analysis model is trained using the k-means algorithm to generate a trained behavior analysis model.

[0057] Specifically, the behavior analysis model is trained using the customer's historical transaction data, which includes information such as the customer's shopping frequency, payment amount, and shopping path. With this data, the k-means algorithm will divide the customer's behavior patterns into different clusters based on similarity. For example, some customers may make high-frequency large payments in a specific time period, while other customers may prefer to frequently purchase low-priced goods. The k-means algorithm can be used to divide these customers into different groups, so that the behavior characteristics of each group can be analyzed separately. By calculating the degree of belonging of each customer in these groups, the k-means algorithm can automatically identify the customer's behavior pattern and train a "behavior analysis model" that can accurately distinguish between normal customers and potential risk customers. During the training process, the algorithm will continuously optimize the center point of each cluster, making the behavior of customers in the same category more similar and the behavior of customers in different categories more different, and finally generating a trained behavior analysis model.

[0058] S52: Based on the real-time transaction data and customer behavior data, the potential risk behavior is predicted through the trained behavior analysis model to obtain the prediction results, and the blacklist data is dynamically adjusted according to the prediction results.

[0059] Specifically, when a customer makes a payment, real-time transaction data and the customer's dynamic behavior data will be collected and input into the trained behavior analysis model. Through the calculation of the behavior analysis model, the system can identify whether the customer's behavior deviates from the normal shopping pattern. If certain behavioral characteristics match known fraud patterns, the model will predict that the customer may have potential risky behavior. For example, if a customer frequently changes shopping areas in a short period of time and stays in the high-value commodity area for a long time without making a payment decision, the model will predict the possibility of the customer running away from the bill through these behaviors. By calculating the customer's behavior score, the model will generate a prediction result to determine whether the customer's fraud risk is high. If the customer is judged to be high risk, the blacklist data will be dynamically adjusted according to the prediction result, and the customer's payment openid and other identification information will be added to the blacklist for inspection and prevention in the next transaction.

[0060] S53: Automatically identify new high-risk behavior patterns through the data feedback mechanism, and add new high-risk behavior patterns to the blacklist data in real time.

[0061] Specifically, as more transaction data accumulates, customer behavior patterns will continue to change, and some new fraud behavior patterns may emerge. Through the data feedback mechanism, each transaction behavior of the customer can be fed back to the background. Combined with the customer's real-time behavior data and historical behavior patterns, the system will continue to analyze these data and identify possible new high-risk behavior patterns. For example, if more and more customers frequently visit multiple stores in a short period of time and make multiple failed payment attempts, the system will identify this new behavior pattern and judge it to be similar to past fraud. New behavior patterns are added to the behavior analysis model through algorithmic analysis to ensure that the model can identify and handle new fraudulent behaviors. The updated model will immediately add these new behavior patterns to the blacklist data, and mark customers who meet the pattern as high-risk customers. When these customers appear in the future, the blacklist data will actively intercept their transactions to ensure that the system has the ability to respond to new fraudulent behaviors in real time. For example, if a customer tries multiple payment methods in a row in one payment and the payment amount is less than the regular consumption amount, this behavior pattern may be identified as potential fraud, and this pattern will be added to the blacklist in real time.

[0062] In one embodiment, if Figure 7 As shown, the non-member customer loss prevention method based on payment openid and face recognition also includes: S60: During the payment process of the non-member customer, further determine whether the non-member customer has a fraud risk by analyzing the non-member customer's facial expression and eye movement trajectory. If a fraud risk is detected, trigger an alarm and reject the transaction.

[0063] Specifically, when customers make payments, high-definition cameras installed in shopping malls or payment points will capture customers' facial images in real time, and use facial expression recognition technology to analyze customers' facial muscle movements to determine their emotional states, such as anxiety, tension, or hesitation, which are potential signs of fraud. At the same time, the customer's eye movement trajectory will be analyzed in real time by eye tracking equipment. Eye movement data can reveal whether the customer is deliberately avoiding eye contact or moving his or her eyes frequently, which is usually a manifestation of trying to hide a certain behavior, such as the customer may be preparing to run away or commit fraud. By calculating the customer's eye movement pattern during the payment process, it is possible to further assess whether the customer is at risk of fraud. If the customer's facial expression and eye movement trajectory both show obvious fraud behavior patterns, a "fraud risk" judgment result will be generated, which will then trigger an alarm and reject the transaction to prevent potential fraud.

[0064] S70: Build a customer credit scoring model through transaction history data, score each non-member customer, and take different verification measures based on the score during the next payment process.

[0065] Specifically, the customer's transaction history data will be used as input data to build a customer credit scoring model. This model is trained based on multiple historical behavior data, including the customer's payment amount, payment frequency, transaction time, payment method, etc. The transaction history of each customer will be analyzed and summarized to generate a credit score. Customers with high scores usually have stable consumption behavior and have never been involved in fraud or escaping the bill, while customers with low scores may frequently attempt to pay but fail, change payment methods, or linger in the commodity area for a long time. Through the analysis of customer historical transactions, the model will automatically generate a credit score for each non-member customer. For example, if customer A's transaction history shows that his payment frequency is low and he has failed to complete the payment many times, the system will give him a low credit score. Based on this score, when the customer makes the next payment, the customer's credit score will be extracted in real time, and the strictness of the verification measures will be determined based on the score. If the customer's credit score is high, the verification process can be simplified, such as allowing the payment method to be used directly to complete the transaction; if the customer's credit score is low, more stringent verification is required, such as requiring the customer to perform facial recognition or verify the accuracy of the payment information, or even manually review their transactions.

[0066] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0067] In one embodiment, a non-member customer loss prevention system based on payment openid and face recognition is provided, and the non-member customer loss prevention system based on payment openid and face recognition corresponds to the non-member customer loss prevention method based on payment openid and face recognition in the above embodiment. Figure 8 As shown, the non-member customer loss prevention system based on payment openID and face recognition includes an information acquisition module, a preliminary evaluation module, a secondary verification module, an alarm module and an update module. The detailed description of each functional module is as follows: The information acquisition module is used to obtain the payment openid, face image and dynamic behavior data of non-member customers in real time, where the dynamic behavior data includes shopping path, standing time and moving speed; A preliminary assessment module, which uses embedded edge computing devices to perform preliminary analysis on payment openID and dynamic behavior data, identify abnormal customer behavior, and generate preliminary risk assessment results; The secondary verification module is used to send the payment openid, face image and dynamic behavior data to the data center for secondary verification based on the preliminary risk assessment results and generate verification results; The alarm module is used to trigger an alarm and reject the transaction if the verification result shows that the customer matches the blacklist or has abnormal behavior, and record the transaction information, customer characteristics and identification time, and generate a transaction report for storage; The update module is used to use machine learning algorithms to dynamically analyze and predict historical transaction data, customer behavior data, and blacklist information, and update blacklist data in real time.

[0068] Optional, initial assessment modules include: A determination submodule, for calculating the stay pattern of non-member customers based on their shopping paths and standing time, and determining the stay pattern of non-member customers to determine whether there is potential fraud or abnormal shopping behavior, and generating a determination result; The recognition submodule is used to evaluate the movement trajectory of non-member customers using dynamic behavior data. When non-member customers frequently change their shopping locations or stay in certain commodity areas for a long time, it is used to identify whether they meet the behavior pattern of skipping orders or committing fraud and generate recognition results. The assessment submodule is used to conduct a preliminary assessment of customer risks based on the determination results and identification results, and generate preliminary risk assessment results.

[0069] Optional, secondary verification modules include: The list comparison submodule is used to compare the payment openid of non-member customers with the blacklist data to generate the list comparison result when the preliminary risk assessment result is abnormal shopping behavior; The face comparison submodule is used to compare the real-time face image of the non-member customer with the face features stored in the data center, use liveness detection technology to verify the authenticity of the non-member customer's identity, and generate face comparison results; The comprehensive analysis submodule is used to conduct a comprehensive analysis based on dynamic behavior data and historical risk behavior patterns to determine whether the customer has engaged in fraudulent behavior and generate behavior analysis results; The summary analysis submodule is used to summarize and analyze the list comparison results, face comparison results and behavior analysis results using the decision tree algorithm to generate verification results.

[0070] Optionally, the alarm module includes: The low-risk alert submodule is used to issue a low-risk alert and trigger the merchant to pay attention to the information when the behavior of a non-member customer matches the risk pattern but does not match the blacklist data.

[0071] The high-risk alarm submodule is used to generate a high-risk alarm and trigger a transaction rejection message when a non-member customer matches the blacklist data or the behavior pattern of the non-member customer is abnormal.

[0072] Optionally, update modules include: The model training submodule is used to train the behavior analysis model based on historical transaction data using the k-means algorithm to generate a trained behavior analysis model; The prediction submodule is used to predict potential risk behaviors based on real-time transaction data and customer behavior data through the trained behavior analysis model, obtain prediction results, and dynamically adjust the blacklist data based on the prediction results; The feedback adding submodule is used to automatically identify new high-risk behavior patterns through the data feedback mechanism, and add the new high-risk behavior patterns to the blacklist data in real time.

[0073] Optionally, the non-member customer loss prevention system based on payment openID and face recognition also includes: The deep judgment submodule is used to further judge whether the non-member customer has fraud risk by analyzing the non-member customer's facial expression and eye movement during the payment process. If fraud risk is detected, an alarm is triggered and the transaction is rejected; The credit scoring submodule is used to build a customer credit scoring model through transaction history data, score each non-member customer, and take different verification measures based on the score during the next payment process.

[0074] For the specific definition of a non-member customer loss prevention system based on payment openid and face recognition, please refer to the definition of a non-member customer loss prevention method based on payment openid and face recognition above, which will not be repeated here. Each module in the above-mentioned non-member customer loss prevention system based on payment openid and face recognition can be implemented in whole or in part through software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0075] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0076] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A non-member customer loss prevention method based on payment openid and face recognition, characterized in that: The non-member customer loss prevention method based on payment openid and face recognition includes: Real-time acquisition of payment openid, face image and dynamic behavior data of non-member customers, wherein the dynamic behavior data includes shopping path, standing time and moving speed; Using embedded edge computing devices to conduct preliminary analysis of the dynamic behavior data, identify abnormal customer behavior, and generate preliminary risk assessment results; Based on the preliminary risk assessment result, the payment openid, the face image and the dynamic behavior data are sent to a data center for secondary verification to generate a verification result; If the verification result shows that the customer matches the blacklist or has abnormal behavior, an alarm is triggered and the transaction is rejected, and the transaction information, customer characteristics and identification time are recorded, and a transaction report is generated for storage; Use machine learning algorithms to dynamically analyze and predict historical transaction data, customer behavior data, and blacklist information, and update blacklist data in real time.

2. According to claim 1, a non-member customer loss prevention method based on payment openid and face recognition is characterized in that: The dynamic behavior data also includes the length of time the non-member customer stays during the payment process, the number of times the customer touches the product, and whether the customer frequently changes the shopping location in a short period of time.

3. According to claim 1, a non-member customer loss prevention method based on payment openid and face recognition is characterized in that: The use of embedded edge computing devices to perform preliminary analysis on the dynamic behavior data, identify abnormal customer behavior, and generate preliminary risk assessment results includes: Calculating the stay pattern of the non-member customer based on the shopping path and standing time of the non-member customer, and judging the stay pattern of the non-member customer to determine whether there is potential fraudulent behavior or abnormal shopping behavior, and generating a judgment result; The dynamic behavior data is used to evaluate the movement trajectory of the non-member customer. When the non-member customer frequently changes shopping locations or stays in certain commodity areas for a long time, it is identified whether the behavior pattern of escaping the bill or committing fraud is met, and an identification result is generated; Based on the determination result and the identification result, a preliminary assessment of the customer risk is performed, and the preliminary risk assessment result is generated.

4. According to claim 1, a non-member customer loss prevention method based on payment openid and face recognition is characterized in that: Based on the preliminary risk assessment result, the payment openid, the face image and the dynamic behavior data are sent to the data center for secondary verification, and the verification result is generated, including: When the preliminary risk assessment result is abnormal shopping behavior, the payment openid of the non-member customer is compared with the blacklist data to generate a list comparison result; Comparing the real-time facial image of the non-member customer with the facial features stored in the data center, using liveness detection technology to verify the authenticity of the non-member customer's identity, and generating a facial comparison result; Perform a comprehensive analysis based on the dynamic behavior data and historical risk behavior patterns to determine whether the customer has engaged in fraudulent behavior and generate a behavior analysis result; The decision tree algorithm is used to summarize and analyze the list comparison results, the face comparison results and the behavior analysis results to generate the verification result.

5. According to claim 1, a non-member customer loss prevention method based on payment openid and face recognition is characterized in that: The triggering of an alarm and rejection of a transaction include: When the behavior of the non-member customer matches the risk pattern but does not match the blacklist data, a low-risk alert is issued and the merchant is triggered to pay attention to the information; When the non-member customer matches the blacklist data or the behavior pattern of the non-member customer is abnormal, a high-risk alert is generated and a transaction rejection message is triggered.

6. According to claim 1, a non-member customer loss prevention method based on payment openid and face recognition is characterized in that: The use of machine learning algorithms to dynamically analyze and predict historical transaction data, customer behavior data, and blacklist information, and to update blacklist data in real time includes: Based on the historical transaction data, a behavior analysis model is trained using a k-means algorithm to generate a trained behavior analysis model; According to the real-time transaction data and customer behavior data, the potential risk behavior is predicted by the trained behavior analysis model to obtain a prediction result, and the blacklist data is dynamically adjusted according to the prediction result; Through the data feedback mechanism, new high-risk behavior patterns are automatically identified and added to the blacklist data in real time.

7. According to claim 1, a non-member customer loss prevention method based on payment openid and face recognition is characterized in that: The non-member customer loss prevention method based on payment openid and face recognition also includes: During the payment process of the non-member customer, further determine whether the non-member customer has a fraud risk by analyzing the non-member customer's facial expression and eye movement trajectory. If a fraud risk is detected, trigger an alarm and reject the transaction; A customer credit scoring model is constructed through transaction history data, and each non-member customer is scored. In the next payment process, different verification measures are taken according to the score.

8. A non-member customer loss prevention system based on payment openid and face recognition, characterized in that: The non-member customer loss prevention system based on payment openid and face recognition includes: An information acquisition module is used to acquire the payment openid, face image and dynamic behavior data of non-member customers in real time, wherein the dynamic behavior data includes shopping path, standing time and moving speed; A preliminary assessment module, for performing preliminary analysis on the dynamic behavior data using an embedded edge computing device, identifying abnormal customer behavior, and generating preliminary risk assessment results; A secondary verification module, for sending the payment openid, the face image and the dynamic behavior data to a data center for secondary verification based on the preliminary risk assessment result, and generating a verification result; An alarm module is used to trigger an alarm and reject the transaction if the verification result shows that the customer matches the blacklist or has abnormal behavior, and record the transaction information, customer characteristics and identification time, and generate a transaction report for storage; The update module is used to use machine learning algorithms to dynamically analyze and predict historical transaction data, customer behavior data, and blacklist information, and update blacklist data in real time.

Images