Resource transfer method, device, equipment and storage medium based on facial features

Abstract

The present application discloses a resource transfer method, apparatus, device and storage medium based on facial features, which belongs to the field of payment. The embodiments of the present application can be applied to various scenarios such as cloud technology, artificial intelligence, and smart transportation. Through the historical resource transfer information between the object and the transaction device, the object features and the target device features are determined. Based on the object features and the target device features, the probability of the object and the target transaction device performing a resource transfer operation is predicted, and the facial features of the target object whose transaction probability meets the conditions are sent to the target transaction device in advance. In this way, if the target object uses the face-swiping payment method to conduct a transaction, the target transaction device can directly compare the collected facial features with the received facial features, without having to upload the collected facial features to the server. The server performs a verification process based on the facial features, thereby improving the efficiency and security of face-swiping payment.

Description

Technical Field

[0001] The present application relates to the field of payment, and in particular to a resource transfer method, apparatus, device and storage medium based on facial features. Background Art

[0002] With the development of Internet technology, users have more and more ways to pay when shopping. In addition to cash payment and QR code payment, facial recognition payment is also being used more and more.

[0003] In related technologies, if a user chooses to pay by face recognition, the transaction device needs to collect the user's facial features and send them to the server, which verifies them. Once the verification is successful, the server deducts a certain amount of virtual currency from the user's account, completing the face recognition payment process.

[0004] However, the facial feature verification process is completed by the server. Users who choose face-scanning payment need to wait a long time to complete the payment, and the efficiency of face-scanning payment is low. Summary of the Invention

[0005] The embodiments of the present application provide a resource transfer method, apparatus, device, and storage medium based on facial features, which can improve the efficiency of face-scanning payment. The technical solution is as follows:

[0006] In one aspect, a method for resource transfer based on facial features is provided, the method comprising:

[0007] acquiring, based on historical resource transfer information between a plurality of objects and a plurality of transaction devices in a facial feature database storing facial features of the plurality of objects, object features and target device features of the plurality of objects, the target device features being features of a target transaction device among the plurality of transaction devices;

[0008] Determining a plurality of target parameters between the plurality of objects and the target transaction device based on the object characteristics of the plurality of objects and the target device characteristics, the target parameters being used to indicate a probability that a resource transfer will occur between the corresponding objects based on the target transaction device;

[0009] determining a plurality of target objects from the plurality of objects based on the plurality of target parameters, the target parameters of the target objects satisfying target conditions;

[0010] The facial features of the multiple target objects in the facial feature database are sent to the target transaction device, and the target transaction device is configured to perform a resource transfer operation based on the received facial features.

[0011] In one aspect, a resource transfer device based on facial features is provided, the device comprising:

[0012] a feature acquisition module, configured to acquire object features and target device features of the multiple objects based on historical resource transfer information between the multiple objects and the multiple transaction devices in a facial feature database, wherein the facial feature database stores the facial features of the multiple objects;

[0013] a target parameter acquisition module, configured to determine a plurality of target parameters between the plurality of objects and the target transaction device based on the object characteristics of the plurality of objects and the target device characteristics, the target parameters being used to indicate a probability that a resource transfer will occur between the corresponding objects based on the target transaction device;

[0014] a target object determination module, configured to determine a plurality of target objects from the plurality of objects based on the plurality of target parameters, wherein the target parameters of the target objects meet the target conditions;

[0015] The sending module is used to send the facial features of the multiple target objects in the facial feature database to the target transaction device, and the target transaction device is used to perform a resource transfer operation based on the received facial features.

[0016] In one possible implementation, the feature acquisition module is used to generate a target graph network based on the historical resource transfer information, wherein the target graph network includes multiple object nodes, multiple device nodes, and lines, wherein the object nodes correspond to the objects, the device nodes correspond to the transaction devices, and the lines are used to indicate that resource transfer has occurred between the connected object nodes and device nodes; graph convolution is performed on the target graph network to obtain object features of the multiple objects and the target device features.

[0017] In one possible implementation, the feature acquisition module is used to obtain an adjacency matrix of the target graph network, where the adjacency matrix is ​​used to represent the connection relationship between object nodes and device nodes; obtain a feature matrix of the target graph network, where the feature matrix includes initial object features of each object node and initial device features of each device node; and obtain object features of the multiple objects and target device features based on the feature matrix and the adjacency matrix.

[0018] In one possible implementation, the feature acquisition module is used to input the target graph network into a graph convolution model, perform feature initialization on multiple object nodes and multiple device nodes in the target graph network through the graph convolution model, and obtain the initial object features of each object node and the initial device features of each device node; and splice the initial object features of each object node and the initial device features of each device node to obtain the feature matrix.

[0019] In a possible implementation, the feature acquisition module is configured to, for an initial object feature of any object node in the feature matrix, acquire positive sample device features and negative sample device features from the feature matrix based on the adjacency matrix, where the positive sample device features are initial device features corresponding to device nodes connected to the object node in the feature matrix, and the negative sample device features are initial device features corresponding to device nodes not connected to the object node in the feature matrix; acquire object features of the object node based on the initial object features, the positive sample device features, and the negative sample device features; for an initial device feature of the target device node in the feature matrix, acquire positive sample object features and negative sample object features from the feature matrix based on the adjacency matrix, where the positive sample object features are initial object features corresponding to object nodes connected to the target device node in the feature matrix, and the negative sample object features are initial object features corresponding to object nodes not connected to the target device node in the feature matrix; acquire the target device features based on the initial device features, the positive sample object features, and the negative sample object features.

[0020] In a possible implementation, the feature acquisition module is configured to acquire a first similarity between the initial object feature and the positive sample device feature; and acquire a second similarity between the initial object feature and the negative sample device feature;

[0021] The initial object feature is updated based on a first difference value to obtain the object feature of the object node, where the first difference value is the difference between the first similarity and the second similarity.

[0022] In one possible implementation, the feature acquisition module is configured to acquire a third similarity between the initial device feature and the positive sample object feature; acquire a fourth similarity between the initial device feature and the negative sample object feature; and update the initial device feature based on a second difference to obtain the target device feature, where the second difference is the difference between the third similarity and the fourth similarity.

[0023] In a possible implementation, the feature acquisition module is used to divide the target graph network into multiple sub-graph networks, perform graph convolution on each of the multiple sub-graph networks, and obtain object features of the multiple objects and the target device features.

[0024] In a possible implementation, the target parameter acquisition module is configured to determine a fifth similarity between the target device feature and each of the object features; and determine the target parameter between each of the objects and the target transaction device based on multiple fifth similarities.

[0025] In one possible implementation, the target condition refers to any one of the following:

[0026] The target parameter of the object is in the first N positions of the plurality of objects, where N is a positive integer;

[0027] The target parameter of the object is greater than or equal to the parameter threshold.

[0028] In a possible implementation, the device further includes:

[0029] The coupon sending module is used to send virtual coupons to the target accounts of the multiple target objects, and the virtual coupons are used to deduct virtual resources when resources are transferred based on the target transaction device.

[0030] On the one hand, a computer device is provided, which includes one or more processors and one or more memories, wherein at least one computer program is stored in the one or more memories, and the computer program is loaded and executed by the one or more processors to implement the resource transfer method based on facial features.

[0031] On the one hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, and the computer program is loaded and executed by a processor to implement the resource transfer method based on facial features.

[0032] On the one hand, a computer program product or computer program is provided, which includes a program code, which is stored in a computer-readable storage medium. A processor of a computer device reads the program code from the computer-readable storage medium, and the processor executes the program code, so that the computer device executes the above-mentioned resource transfer method based on facial features.

[0033] In an embodiment of the present application, the object characteristics and the target device characteristics are determined through the historical resource transfer information between the object and the transaction device. Based on the object characteristics and the target device characteristics, the probability of the object and the target transaction device performing a resource transfer operation is predicted, and the facial features of the target object whose transaction probability meets the conditions are sent to the target transaction device in advance. In this way, if the target object uses face-swiping payment to conduct a transaction, the target transaction device can directly compare the collected facial features with the received facial features, and there is no need to upload the collected facial features to the server. The server performs a verification process based on the facial features, thereby improving the efficiency and security of face-swiping payment. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0035] Figure 1 This is a schematic diagram of an implementation environment of a resource transfer method based on facial features provided in an embodiment of the present application;

[0036] Figure 2 This is a flowchart of a resource transfer method based on facial features provided in an embodiment of the present application;

[0037] Figure 3 This is a flowchart of a resource transfer method based on facial features provided in an embodiment of the present application;

[0038] Figure 4 is a schematic diagram of a target graph network provided in an embodiment of the present application;

[0039] Figure 5 This is a flowchart of a resource transfer method based on facial features provided in an embodiment of the present application;

[0040] Figure 6 This is a flowchart of a resource transfer method based on facial features provided in an embodiment of the present application;

[0041] Figure 7 This is a schematic structural diagram of a resource transfer device based on facial features provided in an embodiment of the present application;

[0042] Figure 8 This is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0043] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0044] In this application, the terms "first", "second", etc. are used to distinguish identical or similar items with substantially the same effects and functions. It should be understood that there is no logical or temporal dependency between "first", "second", and "nth", nor is there any limitation on the quantity and execution order.

[0045] In this application, the term "at least one" means one or more, and the term "plurality" means two or more. For example, a plurality of reference facial images refers to two or more reference facial images.

[0046] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to achieve the best results.

[0047] Cloud computing refers to the delivery and usage model of IT (Internet Technology) infrastructure, enabling on-demand, scalable access to resources over the internet. Broadly speaking, cloud computing refers to the delivery and usage model of services, enabling on-demand, scalable access to services over the internet. These services can be IT-related, software-related, internet-related, or other services. Cloud computing is the product of the convergence of traditional computer and network technologies, including grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing. Driven by the growth of the internet, real-time data streams, and the diversification of connected devices, as well as the demand for search services, social networking, mobile commerce, and open collaboration, cloud computing has rapidly developed. Unlike previous parallel and distributed computing approaches, the emergence of cloud computing will fundamentally revolutionize the entire internet model and enterprise management models.

[0048] Intelligent Traffic System (ITS), also known as Intelligent Transportation System, effectively integrates advanced science and technology (information technology, computer technology, data communication technology, sensor technology, electronic control technology, automatic control theory, operations research, artificial intelligence, etc.) into transportation, service control and vehicle manufacturing, strengthens the connection between vehicles, roads and users, and thus forms a comprehensive transportation system that ensures safety, improves efficiency, improves the environment and saves energy.

[0049] A graph network (GN) is a set of functions organized in a topological space according to a graph structure for relational reasoning.

[0050] Graph Neural Network (GNN) is a general term for models that apply neural networks to graphs. Based on the different technologies and classification methods used, it can be divided into different types as shown in the figure below. For example, based on the propagation method, graph neural networks can be divided into graph convolutional neural networks (GCN) and graph attention networks (GAT).

[0051] Face payment: Face payment is a new payment method based on facial recognition, offering advantages such as greater convenience, greater security, and a better user experience. Face payment is enabled by technologies such as artificial intelligence, machine vision, 3D sensing, and big data.

[0052] It should be noted that the user data involved in the embodiments of the present application, such as the user's transaction data, are all data that have been fully authorized by the user.

[0053] Figure 1 This is a schematic diagram of an implementation environment of a resource transfer method based on facial features provided in an embodiment of the present application, see Figure 1 , the implementation environment may include a transaction device 110 and a server 140 .

[0054] The transaction device 110 is connected to the server 140 via a wireless network or a wired network. Optionally, the transaction device 110 is a tablet computer, a laptop computer, a desktop computer, a dedicated payment device, etc., but is not limited thereto, wherein the dedicated payment device includes a device for conducting transactions on a vehicle-mounted terminal. The transaction device 110 is installed and runs an application that supports face-swiping payment. The transaction device 110 includes a data processing device and a shooting device, the shooting device is used to collect facial images during the face-swiping payment process, and the data processing device is used to execute background processing related to face-swiping payment. In some embodiments, the data processing device and the shooting device of the transaction device 110 are two independent devices, and the data processing device and the shooting device are connected via a wireless network or a wired network, that is, the combination of the data processing device and the shooting device is called the transaction device 110. In some embodiments, the data processing device and the shooting device are integrated into the transaction device 110 as functional components of the transaction device 110, and the embodiments of the present application do not limit this.

[0055] The server 140 is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, distribution networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms. The server 140 provides background services for applications that support face payment running on the transaction device 110. In some embodiments, the background server provided by the server 140 for the application is also a cloud computing service, and different applications can call the API (Application Programming Interface) provided by the server 140 to use the cloud computing service.

[0056] Optionally, the transaction device 110 generally refers to one of multiple transaction devices, and the embodiment of the present application only takes the transaction device 110 as an example.

[0057] Those skilled in the art will appreciate that the number of the aforementioned transaction devices may be greater or lesser. For example, there may be only one transaction device, or there may be dozens, hundreds, or even more transaction devices, in which case the aforementioned implementation environment may also include other transaction devices. This embodiment of the present application does not limit the number or type of transaction devices.

[0058] After introducing the implementation environment of the resource transfer method based on facial features provided in the embodiment of the present application, the application scenario of the resource transfer method based on facial features provided in the embodiment of the present application will be described below. In the following description, the transaction device is the transaction device 110 in the above-mentioned implementation environment, the server is the server 140 in the above-mentioned implementation environment, and the object is the user.

[0059] The resource transfer method based on facial features provided in the embodiments of the present application can be applied in various scenarios that support face-scanning payment, such as in stores that support face-scanning payment, or in vending machines that support face-scanning payment, or in self-service checkout machines that support face-scanning payment, or in dedicated transaction equipment based on vehicle-mounted terminals that support face-scanning payment. The embodiments of the present application do not limit this.

[0060] For example, the facial feature-based resource transfer method provided in an embodiment of the present application is applied in a store that supports facial recognition payment. The store is equipped with a transaction device that supports facial recognition payment. After selecting items in the store, a subject hands the merchandise to the salesperson, who then uses the transaction device to obtain the price. The subject is the user conducting the transaction. The subject then uses the transaction device to pay using facial recognition, completing the transaction. Using the facial feature-based resource transfer method provided in an embodiment of the present application, a server can pre-distribute the subject's corresponding target facial features to the transaction device before the subject arrives at the store to make a purchase. When the subject uses facial recognition payment on the transaction device, the subject enters their account number on the transaction device. The transaction device then determines the subject's corresponding target facial features based on the subject's account number. Simultaneously, the transaction device captures a facial image of the subject and performs feature extraction on the facial image to obtain the subject's facial features. The transaction device can then directly match the subject's facial features with pre-stored target facial features to determine whether the facial features correspond to the same person. When it is determined that the facial features and the target facial features correspond to the same face, the transaction device can notify the subject that verification has passed, for example, by displaying the message "Face verification successful, payment in progress, please wait." When the transaction device determines that the facial features and the target facial features correspond to the same face, it simultaneously sends a transaction request to the server, which carries the subject's account number and the purchase amount. After receiving the transaction request, the server obtains the subject's account number and the purchase amount from the transaction request. Based on the subject's account number, the server deducts virtual currency equal to the purchase amount from the account corresponding to the subject's account number. After the deduction is successful, the server sends a payment confirmation message to the transaction device, which notifies the transaction device of the successful payment. The transaction device can display the payment confirmation message.

[0061] For example, consider the application of the facial feature-based resource transfer method provided in an embodiment of the present application to a vending machine that supports facial payment. This vending machine, which has facial payment functionality, is a transaction device. A user can select the items they want to purchase from the vending machine, such as drinks or snacks. After selecting their items, they can choose to pay with facial recognition. After payment is complete, the vending machine "dispenses" the selected items, completing the transaction. Using the facial feature-based resource transfer method provided in an embodiment of the present application, a server can pre-deliver the target facial features corresponding to the user to the vending machine before the user arrives at the vending machine to make a purchase. When the user pays with facial recognition at the vending machine, they need to enter their account number. The vending machine then determines the target facial features corresponding to the user based on their account number. Simultaneously, the vending machine can capture the user's facial image and perform feature extraction on the facial image to obtain the user's facial features. The vending machine can directly match the subject's facial features with pre-stored target facial features to determine whether the facial features correspond to the same person. If the facial features correspond to the target facial features, the vending machine can notify the subject of successful verification, for example by displaying the message "Face verification successful, payment in progress, please wait." Upon confirming that the facial features correspond to the target facial features, the vending machine simultaneously sends a transaction request to the server, which includes the subject's account number and the purchase amount. After receiving the transaction request, the server retrieves the subject's account number and the purchase amount from the transaction request. Based on the subject's account number, the server deducts virtual currency equal to the purchase amount from the account corresponding to the subject's account number. Upon successful deduction, the server sends a payment confirmation message to the vending machine, notifying the vending machine of the successful payment, and the vending machine then dispenses the item selected by the subject.

[0062] For example, the facial feature-based resource transfer method provided in an embodiment of the present application is applied to an automated checkout machine that supports facial payment. This automated checkout machine has facial payment functionality and is, in effect, a transaction device. After selecting items at a store, a person uses the automated checkout machine to scan the item's QR code. The automated checkout machine calculates the item's price. The person selects facial payment on the automated checkout machine, and after completing the payment, the transaction is completed. Using the facial feature-based resource transfer method provided in an embodiment of the present application, a server can pre-deliver the target facial features corresponding to the person to the automated checkout machine before the person arrives at the automated checkout machine to make a purchase. When the person uses facial payment on the automated checkout machine, the person enters their account number. The automated checkout machine then determines the target facial features corresponding to the person based on the account number. Simultaneously, the automated checkout machine can capture the person's facial image and perform feature extraction on the facial image to obtain the person's facial features. The automated checkout machine can directly match the subject's facial features with pre-stored target facial features to determine whether the facial features correspond to the same face. If the facial features correspond to the target facial features, the automated checkout machine can notify the subject of successful verification, for example by displaying the message "Face verification successful, payment in progress, please wait." Upon confirming that the facial features correspond to the target facial features, the automated checkout machine simultaneously sends a transaction request to the server, which includes the subject's account number and the purchase amount. After receiving the transaction request, the server retrieves the subject's account number and the purchase amount from the transaction request. Based on the subject's account number, the server deducts virtual currency equal to the purchase amount from the account corresponding to the subject's account number. After the deduction is successful, the server sends a payment confirmation message to the automated checkout machine, which notifies the automated checkout machine of the successful payment. The automated checkout machine then proceeds with subsequent operations, such as printing a shopping receipt.

[0063] Taking the facial feature-based resource transfer method provided in an embodiment of the present application as an example, applied to a dedicated transaction device based on an in-vehicle terminal that supports facial payment, some stores have vehicle-based quick purchase points, adjacent to which are vehicle-based shopping lanes. The vehicle-based quick purchase points are equipped with dedicated transaction devices based on in-vehicle terminals (hereinafter referred to as dedicated transaction devices). When a person needs to purchase goods, they do not need to find a parking space to park their vehicle. They can simply drive their vehicle into the vehicle-based shopping lane, select goods at the vehicle-based quick purchase point, pay with facial payment, and then drive away with the goods. After using the facial feature-based resource transfer method provided in an embodiment of the present application, the server can pre-deliver the target facial features corresponding to the person to the dedicated transaction device before the person arrives at the vehicle-based quick purchase point to make a purchase. After the person selects goods at the vehicle-based quick purchase point, they can choose to pay with facial payment. In response to the person selecting facial payment, the dedicated transaction device sends an account acquisition request to the in-vehicle terminal. The account acquisition request is used to obtain the person's account number for payment. The in-vehicle terminal sends the bound person's account number to the transaction device. After receiving the target account number, the dedicated transaction device determines the target facial features corresponding to the target account number. Simultaneously, the target transaction device obtains the target's facial image and performs feature extraction on the facial image to obtain the target's facial features. The target transaction device can obtain the target's facial image through either an in-vehicle terminal or a camera. If the target transaction device obtains the target's facial image through an in-vehicle terminal, it can send a facial image acquisition request to the in-vehicle terminal. After receiving the facial feature acquisition request, the in-vehicle terminal uses a camera on the in-vehicle terminal to acquire the facial image and sends the acquired facial image to the target transaction device, thereby obtaining the target transaction device's facial image. If the target transaction device obtains the target's facial image through a camera, the target moves their face into the camera's capture area, and the camera captures the face to obtain the target's facial image.

[0064] The target transaction device then directly matches the subject's facial features with pre-stored target facial features to determine whether the facial features correspond to the same person. If the facial features correspond to the target facial features, the target transaction device can notify the subject of successful verification, for example by displaying the message "Face verification successful, payment in progress, please wait." Upon confirming that the facial features correspond to the target facial features, the target transaction device simultaneously sends a transaction request to the server, which includes the subject's account number and the purchase amount. After receiving the transaction request, the server retrieves the subject's account number and the purchase amount from the transaction request. Based on the subject's account number, the server deducts virtual currency equal to the purchase amount from the account corresponding to the subject's account number. Upon successful deduction, the server sends a payment confirmation message to the target transaction device, notifying the target transaction device of the successful payment. The target transaction device can display the payment confirmation message.

[0065] It should be noted that the above is illustrated by using the facial feature-based resource transfer method provided in the embodiments of the present application as examples of applications in stores that support face payment, vending machines that support face payment, self-service checkout machines that support face payment, and dedicated transaction devices based on vehicle-mounted terminals that support face payment. In other possible implementations, the embodiments of the present application can also be applied to other scenarios that support face payment, such as purchasing concert tickets or movie tickets through face payment, but the embodiments of the present application do not limit this.

[0066] After introducing the implementation environment and application scenarios of the embodiment of the present application, the resource transfer method based on facial features provided by the embodiment of the present application is described below.

[0067] Figure 2 This is a flowchart of a resource transfer method based on facial features provided by an embodiment of the present application, taking the execution subject as a server as an example, see Figure 2 , methods include:

[0068] 201. The server obtains object features and target device features of the multiple objects based on historical resource transfer information between the multiple objects and the multiple transaction devices in the facial feature database, where the facial feature database stores the facial features of the multiple objects. The target device features are features of a target transaction device among the multiple transaction devices.

[0069] Among them, the facial feature database is a database for storing facial features. The facial features in the facial feature database are uploaded by the object using the user device, or are obtained after feature extraction based on the facial image uploaded by the user device. The embodiment of the present application does not limit this. The object is the user who uses the user device. The facial features in the facial feature database are used for identity authentication during the face-swiping payment process. The historical resource transfer information includes multiple historical resource transfer data. Each historical resource transfer data includes the object of the transferred resource, the transaction device that receives the resource, and the amount of resource transfer, etc. The embodiment of the present application does not limit this.

[0070] 202. The server determines a plurality of target parameters between the plurality of objects and the target transaction device based on the object characteristics of the plurality of objects and the target device characteristics, where the target parameters are used to indicate a probability that resource transfer will occur between the corresponding objects based on the target transaction device.

[0071] The multiple target parameters correspond one-to-one to the multiple objects, meaning each target parameter corresponds to an object. For any of the multiple objects, the target parameter corresponding to that object represents the probability that a resource transfer will occur with that object based on the target transaction device, i.e., the probability that a transaction will occur between that object and the target transaction device.

[0072] 203. The server determines a plurality of target objects from the plurality of objects based on the plurality of target parameters, wherein the target parameters of the target objects meet the target conditions.

[0073] The process of the server determining multiple target objects from the multiple objects based on multiple target parameters is, in other words, the process of screening the multiple objects based on the probability of resource transfer between the multiple objects and the target transaction device. In some embodiments, the screened target objects are objects among the multiple objects with a higher probability of resource transfer based on the target transaction device, or objects with a higher probability of transaction with the target transaction device.

[0074] 204. The server sends the facial features of the multiple target objects in the facial feature database to the target transaction device, and the target transaction device is configured to perform a resource transfer operation based on the received facial features.

[0075] Among them, the target transaction device is used to perform a resource transfer operation based on the received facial features, that is, the target transaction device performs a face-swiping payment operation based on the received facial features.

[0076] In an embodiment of the present application, the object characteristics and the target device characteristics are determined through the historical resource transfer information between the object and the transaction device. Based on the object characteristics and the target device characteristics, the probability of the object and the target transaction device performing a resource transfer operation is predicted, and the facial features of the target object whose transaction probability meets the conditions are sent to the target transaction device in advance. In this way, if the target object uses face-swiping payment to conduct a transaction, the target transaction device can directly compare the collected facial features with the received facial features, and there is no need to upload the collected facial features to the server. The server performs a verification process based on the facial features, thereby improving the efficiency and security of face-swiping payment.

[0077] The above steps 201-204 are a brief description of the resource transfer method based on facial features provided in the embodiment of the present application. The resource transfer method based on facial features provided in the embodiment of the present application will be described below with reference to some examples. Figure 3 , methods include:

[0078] 301. The server obtains historical resource transfer information between multiple objects and multiple transaction devices in a facial feature database from an account database, where the facial feature database stores facial features of the multiple objects.

[0079] Among them, the server is a background server that provides resource transfer services. The server correspondingly maintains a facial feature database and an account database. The facial feature database stores facial features of multiple objects. In the facial feature database, each facial feature is bound to the corresponding object account and stored, which facilitates the server to retrieve facial features. For example, the server stores facial features and object accounts in a key-value pair (Key-Value) format, that is, the object account is used as the key and the facial feature is used as the value. When the server searches the facial feature database, the corresponding facial feature (Value) can be retrieved by entering the object account (Key). The account database stores account information of multiple objects, including the number of virtual resources in the object account and the income and expenditure of virtual resources in the object account. In some embodiments, historical resource transfer information between multiple objects and multiple transaction devices is stored in the account database.

[0080] The object in the facial feature database refers to the object whose facial features are stored in the facial feature database. In some embodiments, the object can upload facial features to the server through the user device. After the server receives the facial features, it binds the facial features and the corresponding object account and stores them in the facial feature database. In some embodiments, the user device can perform three-dimensional reconstruction of the object's face based on 3D (3Dimension) structured light technology to obtain a three-dimensional face model, and extract features from the three-dimensional face model to obtain facial features. Alternatively, the facial features are obtained after the server receives the facial image uploaded by the user device and extracts features from the facial image. In some embodiments, the server can use convolutional neural networks (CNN) to extract features from the facial image and obtain facial features corresponding to the facial image. The server binds the extracted facial features and the corresponding object account and stores them in the facial feature database.

[0081] In some embodiments, historical resource transfer information, also referred to as historical transaction information or historical transaction records, records resource transfers between objects and transaction devices. For one of multiple objects, the historical resource transfer information records the transaction devices with which resource transfers occurred. If the transferred resource is virtual currency, the historical resource transfer information also records the transaction devices with which the object conducted transactions. For one of multiple transaction devices, the historical resource transfer information records the objects with which resource transfers occurred. If the transferred resource is virtual currency, the historical resource transfer information also records the objects with which transactions were conducted on the transaction device.

[0082] In some embodiments, the historical resource transfer information includes multiple historical resource transfer data, each of which includes the object of the transferred resource, the transaction device that receives the resource, and the amount of resource transfer, etc. In some embodiments, the object account is used to represent the corresponding object in the historical resource transfer data, and the device identifier is used to represent the corresponding transaction device. For example, a historical resource transfer data is "20XX:XX:XX, User ID:123456 Transfer 100to Device ID:987654", where User ID is the object account and Device ID is the device identifier. In some embodiments, the device identifier is the serial number of the transaction device, or the MAC (Media Access Control Address) of the network card of the transaction device, or other identifiers that can uniquely identify the transaction device. This embodiment of the present application does not limit this. In some embodiments, the historical resource transfer data also includes the method of resource transfer, such as resource transfer through code scanning payment, or resource transfer through face recognition payment.

[0083] In one possible implementation, the server obtains multiple subject accounts from a facial feature database. These multiple subject accounts are also subject accounts whose facial features are stored in the facial feature database. The server queries an account database based on the multiple subject accounts to obtain historical resource transfer information for the multiple subject accounts. The historical account transfer information records resource transfers between the multiple subject accounts and multiple transaction devices.

[0084] Under this implementation, the server can first determine multiple object accounts that store facial features in the facial feature database, and then use the multiple object accounts to query the account database to obtain historical resource transfer information of the multiple object accounts. This can avoid processing object accounts that do not correspond to facial features in subsequent steps, thereby improving computing efficiency.

[0085] In one possible implementation, a server retrieves multiple subject accounts from a facial feature database. These multiple subject accounts are also subject accounts whose facial features are stored in the facial feature database. Based on these multiple subject accounts and a target time period, the server queries the account database to obtain historical resource transfer information for these multiple subject accounts within the target time period. This historical account transfer information records resource transfer records between these multiple subject accounts and multiple transaction devices within the target time period. The time period is set by technical personnel based on actual circumstances, for example, to M months, where M is a positive integer. A larger M indicates a greater amount of historical resource transfer data in the resource transfer record, which in turn indicates a greater number of historical transaction records traceable, covering more subjects and transaction devices. However, the relationship between subjects and transaction devices may change over time. For example, consider subject A, who lived in area B1 a year ago and frequently transacted with transaction device C1. Five months ago, subject A moved to area B2 and began frequently transacting with transaction device C2. Therefore, tracing back historical transaction records that are too far back in time is of limited use. The smaller M is, the smaller the amount of historical resource transfer data in the resource transfer record is, which means that there are fewer historical transaction records to trace, and the fewer objects and transaction devices are covered.

[0086] In this implementation, the server can determine historical resource transfer information based on the target time period to improve computational efficiency and increase the accuracy of subsequent computations.

[0087] In one possible approach, a server retrieves multiple initial target accounts from a facial feature database and filters these initial target accounts based on the location of a target transaction device to obtain multiple target accounts. Based on these multiple target accounts, the server queries an account database to obtain historical resource transfer information for these multiple target accounts. This historical account transfer information records resource transfers between these multiple target accounts and multiple transaction devices, where the target transaction device is the transaction device to which the server sent the facial feature. In some embodiments, the location refers to any of the province, city, or administrative region where the target transaction device is located.

[0088] Under this implementation, the server can screen multiple initial object accounts based on the location of the target transaction device, avoiding adding some object accounts that are far away from the target transaction device into the calculation, thereby saving computing resources and improving computing efficiency.

[0089] For example, the server obtains multiple initial target accounts from a facial feature database and determines the locations of the multiple initial target accounts. For example, if the location is a city, the server determines the cities where the multiple initial target accounts are located. The server compares the cities where the multiple initial target accounts are located with the cities where the target transaction devices are located. From the multiple initial target accounts, the server identifies the initial target accounts whose cities are the same as the cities where the target transaction devices are located as the multiple target accounts. Based on the multiple target accounts, the server queries the account database to obtain historical resource transfer information for the multiple target accounts. The historical account transfer information records the resource transfer records between the multiple target accounts and the multiple transaction devices.

[0090] In one possible implementation, the server obtains historical resource transfer information between multiple objects and multiple transaction devices in the facial feature database from the account database at target time intervals, where the target time is set by technical personnel based on actual conditions, such as 2 days, 3 days, 5 days or 7 days, etc., and this embodiment of the present application does not limit this.

[0091] In this implementation, because resource transfers may occur between multiple objects and multiple transaction devices at any time, the resource transfer information stored in the account database is also updated in real time. The server obtains historical resource transfer information at target intervals, ensuring that the relationships between objects and transaction devices determined subsequently are up to date, thereby improving the accuracy of subsequent facial feature delivery. Furthermore, in combination with the aforementioned implementations, the server may obtain the facial features of new object accounts at any time, meaning that the facial features stored in the facial feature database are also updated in real time. If the server performs screening based on the object accounts in the facial feature database, it can also ensure that the screened object accounts are also up to date.

[0092] In one possible implementation, after the server obtains historical resource transfer information between multiple objects and multiple transaction devices in the facial feature database from the account database, it can also encrypt the object account and device identification recorded in the historical resource transfer information to ensure the privacy of the object, where the encryption process is also called desensitization processing.

[0093] For example, after obtaining historical resource transfer information between multiple objects and multiple transaction devices in a facial feature database from an account database, the server maps the object account recorded in the historical resource transfer information to a first string and the device identifier to a second string. For the same object account in the historical resource transfer information, the first string corresponding to the object account does not change. For different object accounts in the historical resource transfer information, the first strings corresponding to the two object accounts are different, meaning that the first string can uniquely identify the object account. Correspondingly, for the same transaction device in the historical resource transfer information, the second string corresponding to the transaction device does not change. For different transaction devices in the historical resource transfer information, the second strings corresponding to the two transaction devices are different, meaning that the second string can uniquely identify the transaction device. After the server maps the object account to the first string and the transaction device identifier to the second string, it can use the first string to represent the corresponding object account and the second string to represent the corresponding user device. This processing does not involve the actual object account and device identifier, thus preventing privacy leaks.

[0094] For example, the server uses a hash algorithm to map the object account to a first string and the device identifier to a second string. Because the hash algorithm can map characters of different lengths to strings of the same length, the lengths of the mapped first and second strings are the same for object account numbers and device identifiers of different lengths. In some embodiments, both the first and second strings are 32-character strings.

[0095] 302. The server obtains object features and target device features of the multiple objects based on historical resource transfer information between the multiple objects and the multiple transaction devices in the facial feature database. The target device features are features of a target transaction device among the multiple transaction devices.

[0096] Among them, since the historical resource transfer information records the resource transfer situation between the object and the transaction device, the object characteristics determined by the server based on the historical resource transfer information can reflect the resource transfer characteristics of the corresponding object on multiple transaction devices, or in other words, the object characteristics can reflect the characteristics of the object's transactions on multiple transaction devices. If the similarity between the object characteristics of two objects is high, it means that the transaction devices on which the two objects have transferred resources are relatively similar, or that the transaction habits or consumption habits of the two objects are relatively similar. Correspondingly, the target device characteristics determined by the server based on the historical resource transfer information can also reflect the resource transfer characteristics of multiple objects on the target transaction device, or in other words, the target device characteristics can reflect the characteristics of multiple objects' transactions on the target transaction device. If the target device characteristics have a high similarity to any device characteristic, it means that the target device is relatively similar to the object corresponding to the transaction device corresponding to the device characteristic, or that the group of objects that have conducted transactions with the target transaction device has a high similarity to the group of objects that have conducted transactions with the transaction device.

[0097] In one possible implementation, the server generates a target graph network based on the historical resource transfer information. The target graph network includes multiple object nodes, multiple device nodes, and lines, where the object nodes correspond to the objects, the device nodes correspond to the transaction devices, and the lines indicate that resource transfers have occurred between the connected object nodes and device nodes. The server performs graph convolution on the target graph network to obtain object features of the multiple objects and the target device features. Figure 4 The target graph network is a schematic diagram, which includes multiple object nodes and multiple device nodes. Figure 4 In the example, device A and device B are two device nodes, and objects A, B, C, D, and E are five object nodes. Connections between device nodes and object nodes indicate resource transfers between them. For example, a connection between object A and device A indicates a resource transfer between them, or a transaction between object A and device A.

[0098] In order to explain the above embodiment more clearly, the above embodiment will be explained in two sub-steps below.

[0099] 3021. The server generates a target graph network based on the historical resource transfer information.

[0100] In one possible implementation, the server determines the objects and transaction devices that have performed resource transfers from the historical resource transfer information, that is, obtains the object account and the device identifier of the transaction device that has performed resource transfers. The server generates multiple object nodes based on the obtained object account, and each object node is marked with the corresponding object account. It also generates multiple device nodes based on the obtained device identifier, and each device node is marked with the corresponding device identifier. Based on the transaction status between the object account and the device identifier in the historical resource transfer information, the server adds connections between the multiple object nodes and the multiple device nodes. The connection indicates that a resource transfer has occurred between the connected object nodes and the device nodes, and the target graph network is obtained. Of course, if the server has desensitized the object account and the device identifier, then the first character string will be used to represent the object account, and the second character string will be used to represent the device identifier.

[0101] In one possible implementation, historical resource transfer information includes multiple pieces of historical resource transfer data. The server aggregates the multiple pieces of historical resource transfer data based on the dimension of object accounts to obtain historical resource transfer data groups corresponding to the multiple object accounts, each of which includes at least one piece of historical resource transfer data. The server generates object nodes equal to the number of the multiple object accounts, and labels each object node with the corresponding object account. Based on the historical resource transfer data groups corresponding to the multiple object accounts, the server generates multiple device nodes and adds connections between the object nodes and the corresponding device nodes. In some embodiments, the server can deduplicate at least one piece of historical resource transfer data in each historical resource transfer data group. Deduplication here means that if two or more pieces of historical resource transfer data correspond to the same transaction device, that is, to the same device identifier, only one piece of historical resource transfer data is retained in the historical resource transfer data group to reduce computational complexity and improve computational efficiency. If two or more pieces of historical resource transfer data correspond to the same transaction device in a historical resource transfer data group, this indicates that multiple resource transfers have occurred between the object corresponding to the historical resource transfer data group and the transaction device.

[0102] For example, the server converts multiple historical resource transfer data into key-value pairs, where the key of the key-value pair is the object account and the value is the device identifier. A key-value pair represents a group of object accounts and device identifiers in which resource transfers have occurred. The server aggregates multiple key-value pairs based on the dimension of the object account, that is, the dimension of the key, to obtain a key-value pair group corresponding to multiple object accounts, and the key-value pair group includes at least one key-value pair. The server deduplicates the key-value pairs in each key-value pair group. Deduplication means that when the same key-value pair is identified, only one key-value pair is retained. The server generates multiple object nodes based on multiple object accounts, and different object nodes are distinguished by different object accounts. The server generates multiple device nodes based on the deduplicated key-value pair group, and adds connections between the object nodes and the device nodes to obtain the target graph network.

[0103] 3022. The server performs graph convolution on the target graph network to obtain object features of the multiple objects and features of the target device.

[0104] In one possible implementation, the server obtains an adjacency matrix of the target graph network, where the adjacency matrix represents the connection relationships between object nodes and device nodes. The server obtains a feature matrix of the target graph network, where the feature matrix includes initial object features of each object node and initial device features of each device node. Based on the feature matrix and the adjacency matrix, the server obtains object features of the multiple objects and the target device features.

[0105] In order to explain the above embodiment more clearly, the following will be divided into three parts to explain the above embodiment.

[0106] In the first part, the server obtains the adjacency matrix of the target graph network, which is used to represent the connection relationship between the object node and the device node.

[0107] Wherein, the adjacency matrix is ​​a matrix composed of the numbers 0 and 1, the number 0 indicates that there is no connection between the two nodes, and the number 1 indicates that there is a connection between the two nodes, and the node here refers to the object node and the device node. In some embodiments, the adjacency matrix is ​​a square matrix, each row of the adjacency matrix corresponds to a node, and each column also corresponds to a node. If the number in the second row and third column of the adjacency matrix is ​​1, it means that there is a connection between the second node and the third node in the target graph network; if the number in the second row and third column of the adjacency matrix is ​​0, it means that there is no connection between the second node and the third node in the target graph network. In an embodiment of the present application, since resource transfer does not occur between transaction devices and transaction devices, and resource transfer between objects is not considered, the adjacency matrix of the target graph network can also be optimized, that is, the rows and columns of the adjacency matrix are set to correspond to object nodes and device nodes respectively, such as each row of the adjacency matrix is ​​set to correspond to an object node, and each column is set to correspond to a device node. If there is a connection between an object node and a device node, the position corresponding to the object node and the device node is the number 1; if there is no connection between an object node and a device node, the position corresponding to the object node and the device node is the number 0. This method can reduce the size of the adjacency matrix and improve computational efficiency.

[0108] In a possible implementation, the server obtains an adjacency matrix of the target graph network based on object nodes, device nodes, and connections between object nodes and devices in the target graph network.

[0109] For example, the server determines the number of object nodes in the target graph network as the number of rows in the adjacency matrix, and the number of device nodes in the target graph network as the number of columns in the adjacency matrix. The server fills the adjacency matrix with numbers based on the connections between the object nodes and the device nodes.

[0110] In the second part, the server obtains the feature matrix of the target graph network, which includes the initial object features of each object node and the initial device features of each device node.

[0111] In one possible implementation, the server inputs the target graph network into a graph convolutional model, and performs feature initialization on multiple object nodes and multiple device nodes in the target graph network through the graph convolutional model to obtain initial object features of each object node and initial device features of each device node. The server concatenates the initial object features of each object node and the initial device features of each device node to obtain the feature matrix. Feature initialization is to randomly generate initial object features and initial device features, assign the randomly generated initial object features to the object nodes, and assign the randomly generated device features to the device nodes. The graph convolutional model is also a graph convolutional neural network.

[0112] For example, the server inputs the target graph network into a graph convolutional model, and the graph convolutional model's feature initialization unit performs feature initialization on multiple object nodes and multiple device nodes in the target graph network. When initializing the features of the multiple object nodes and multiple device nodes, the feature initialization unit uses a target algorithm to generate multiple initial features. These multiple initial features include multiple initial object features and multiple initial device features. The multiple initial features generated by the target algorithm conform to a normal distribution, which facilitates the convergence of the subsequent graph convolutional model. The multiple initial features generated by the feature initialization unit have a dimension K, where K is a positive integer. A larger dimension K of the initial feature increases the expressive power of the initial feature for the corresponding node, but also increases the computational load. Correspondingly, a smaller dimension K of the initial feature decreases the expressive power of the initial feature for the corresponding node, but reduces the computational load. K is set by technicians based on actual conditions, such as 32 or 64, and is not limited in this embodiment of the present application. During the experiment, K was set to 64 to balance expressive power and computational load.

[0113] In one possible implementation, the server obtains the first embedding features of multiple object nodes and the second embedding features of multiple device nodes in the target graph network, the first embedding features being the embedding features obtained after word embedding of the corresponding object account, and the second embedding features being the features obtained after word embedding of the corresponding device identifier. In some embodiments, if the server maps the object account to a first string and the device identifier to a second string, then the server can also directly determine the first string of the object account as the first embedding feature of the corresponding object node, and determine the second string of the device identifier as the second embedding feature of the corresponding device node. Alternatively, the server can also perform a distribution transformation on the multiple first strings and the multiple second strings, and the multiple first strings and the multiple second strings after the distribution transformation conform to a normal distribution. Alternatively, the server performs embedding coding on the multiple first strings and the multiple second strings to obtain the first embedding features corresponding to each first string and the second embedding features corresponding to each second string.

[0114] In the third part, the server obtains the object features of the multiple objects and the target device features based on the feature matrix and the adjacency matrix.

[0115] The following describes a method for the server to obtain the object features of multiple objects and the target device features based on the feature matrix and the adjacency matrix respectively.

[0116] A. First, a method for the server to obtain object features of multiple objects based on the feature matrix and the adjacency matrix is ​​described.

[0117] In one possible implementation, for the initial object features of any object node in the feature matrix, the server obtains positive sample device features and negative sample device features from the feature matrix based on the adjacency matrix. The positive sample device features are the initial device features corresponding to device nodes connected to the object node in the feature matrix, and the negative sample device features are the initial device features corresponding to device nodes not connected to the object node in the feature matrix. The server obtains the object features of the object node based on the initial object features, the positive sample device features, and the negative sample device features.

[0118] For example, for any object node in the feature matrix, the server randomly determines multiple positive sample device nodes based on the adjacency matrix. The positive sample device nodes are also the device nodes connected to the object node in the target graph network. In other words, the transaction devices indicated by the multiple positive sample device nodes have undergone resource transfer with the object account corresponding to the object node. As described above, the adjacency matrix is ​​a matrix composed of numbers 0 and 1. 0 indicates that there is no connection between the object node and the device node, that is, no resource transfer has occurred between the object account corresponding to the object node and the device identifier corresponding to the device node. 1 indicates that there is a connection between the object node and the device node, that is, a resource transfer has occurred between the object account corresponding to the object node and the device identifier corresponding to the device node. The server determines the row corresponding to the object node in the adjacency matrix, and randomly determines multiple columns with the number 1 from the row. The device nodes corresponding to the columns are also positive sample device nodes. The number of positive sample device nodes determined by the server is set by the technician according to the actual situation, such as 8, 12 or 16, etc., and this embodiment of the application does not limit this. Compared to using all device nodes connected to the object node as positive sample nodes, the random selection method can reduce the number of positive sample device nodes, thereby reducing the amount of computation. During the experiment, the impact of reducing the number of positive sample device nodes on subsequent steps was very subtle, which means that this method can reduce the amount of computation while ensuring accuracy.

[0119] While determining multiple positive sample device nodes, the server can also randomly determine multiple negative sample device nodes based on the adjacency matrix. Negative sample device nodes are device nodes in the target graph network that are not connected to the object node. In other words, the transaction devices indicated by the multiple negative sample device nodes have not undergone resource transfer with the object account corresponding to the object node. Similarly, the adjacency matrix is ​​a matrix composed of numbers 0 and 1. 0 indicates that there is no connection between the object node and the device node, that is, no resource transfer has occurred between the object account corresponding to the object node and the device identifier corresponding to the device node. 1 indicates that there is a connection between the object node and the device node, that is, a resource transfer has occurred between the object account corresponding to the object node and the device identifier corresponding to the device node. The server determines the row corresponding to the object node in the adjacency matrix, and randomly determines multiple columns with the number 0 from the row. The device nodes corresponding to the columns are also negative sample device nodes. The number of negative sample device nodes determined by the server is set by the technician according to the actual situation, such as 8, 12 or 16, etc., and this embodiment of the application does not limit this.

[0120] The server obtains the first similarity between the initial object feature and the positive sample device feature. The server obtains the second similarity between the initial object feature and the negative sample device feature. The server updates the initial object feature based on the first difference to obtain the object feature of the object node, and the first difference is the difference between the first similarity and the second similarity. If there are multiple positive sample device features, the server can obtain the first similarity between the initial object feature and each positive sample device feature respectively; if there are multiple negative sample device features, the server can obtain the second similarity between the initial object feature and each negative sample device feature respectively. At this time, the first difference is the sum of multiple third differences, and each third difference is the difference between a first similarity and a second similarity. The number of third differences is the same as the number of first similarities.

[0121] In some embodiments, the server uses a vector to represent features. For example, an initial object feature vector is used to represent the initial object feature, a positive sample device feature vector is used to represent the positive sample device feature, and a negative sample device feature vector is used to represent the negative sample device feature. Then, the server can use the cosine similarity between the initial object feature vector and the positive sample device feature vector to represent the first similarity between the initial object feature and the positive sample device feature. The cosine similarity between the initial object feature vector and the negative sample device feature vector is used to represent the second similarity between the initial object feature and the negative sample device feature. The server constructs a first loss function based on the first difference between the first similarity and the second similarity, updates the initial object feature based on the first loss function, and obtains the object feature of the object node. It should be noted that the server's update of the initial object feature based on the first loss function includes multiple iterative processes, and the following is an example of an iterative process.

[0122] In some embodiments, the server constructs a first loss function based on the first difference using the following formula (1).

[0123]

[0124] Among them, L1 is the first loss function, M is the number of all nodes in the target graph network, all nodes here include object nodes and device nodes, N u represents all first-order neighbors of the object node u, that is, all device nodes directly connected to the object node u, σ() is a normalization function, is the first similarity, i is the serial number of the positive sample device node, is the second similarity, j is the serial number of the negative sample device node, is the first difference, λ is a hyperparameter, which is set by technicians according to actual conditions, E (0) is the initial object feature.

[0125] The server updates the initial object features based on the first loss function using the following formula (2).

[0126]

[0127] Among them, k is the number of layers of the graph convolutional neural network, which can also be understood as the number of iterations. is the object feature of the object node u. If it is the first iteration process, then That is E (0) . is the device feature of all first-order neighbors of the object node u. If it is the first iteration process, then this device feature is also the initial device feature of all first-order neighbors. AGG() is an aggregation function, such as the average value (mean function) or the sum (sum) function. In formula (2), is the object feature of the object node u The object features are updated after one iteration. The number of layers of the graph convolutional neural network is set by technicians according to actual conditions, and is not limited in this embodiment of the present application.

[0128] The purpose of using formula (2) to update the initial object features based on the first loss function is to make formula (1) converge to the target value as much as possible, or to make the similarity between the object feature of the object node and the device feature of the connected device node as high as possible, and the similarity between the object feature of the object node and the device feature of the unconnected device node as low as possible, or to increase the first similarity as much as possible and reduce the second similarity, so that the connected nodes are more similar in feature expression, and realize modeling based on the connection relationship between the nodes. Among them, the target value is set by the technician according to the actual situation, and the embodiment of the application does not limit this.

[0129] The above process can be summarized as follows: using the loss function provided by formula (1), the gradient for updating the initial features is determined based on the principle of backpropagation, and the initial object features are updated using the gradient descent method, such as using the Adam optimizer to update the initial object features to obtain the object features. In some embodiments, the function provided by the above formula (1) is implemented by the feature update unit of the graph convolution model, and the function provided by the above formula (2) is implemented by the loss function calculation unit of the graph convolution model.

[0130] B. Then, the server is described in detail for obtaining the target device features based on the feature matrix and the adjacency matrix.

[0131] In one possible implementation, for the initial device features of the target device node in the feature matrix, the server obtains positive sample object features and negative sample object features from the feature matrix based on the adjacency matrix. The positive sample object features are the initial object features corresponding to the object nodes in the feature matrix that are connected to the target device node, and the negative sample object features are the initial object features corresponding to the object nodes in the feature matrix that are not connected to the target device node. The server obtains the target device features based on the initial device features, the positive sample object features, and the negative sample object features.

[0132] For example, for the target device node in the feature matrix, the server randomly determines multiple positive sample object nodes based on the adjacency matrix. The positive sample object node is also the object node connected to the target device node in the target graph network. In other words, the object account indicated by the multiple positive sample object nodes has undergone resource transfer with the transaction device corresponding to the target device node. As described above, the adjacency matrix is ​​a matrix composed of numbers 0 and 1. 0 indicates that there is no connection between the target device node and the object node, that is, no resource transfer has occurred between the transaction device corresponding to the target device node and the device identifier corresponding to the object node. 1 indicates that there is a connection between the target device node and the object node, that is, a resource transfer has occurred between the transaction device corresponding to the target device node and the device identifier corresponding to the object node. The server determines the column corresponding to the target device node in the adjacency matrix, and randomly determines multiple rows with the number 1 from the column. The object nodes corresponding to the rows are also positive sample object nodes. The number of positive sample object nodes determined by the server is set by the technician according to the actual situation, such as 8, 12 or 16, etc., and this embodiment of the application does not limit this.

[0133] While determining multiple positive sample object nodes, the server can also randomly determine multiple negative sample object nodes based on the adjacency matrix. Negative sample object nodes are object nodes in the target graph network that are not connected to the target device node. In other words, the object accounts indicated by the multiple negative sample object nodes have not undergone resource transfer with the transaction devices corresponding to the target device nodes. Similarly, the adjacency matrix is ​​a matrix composed of numbers 0 and 1. 0 indicates that there is no connection between the target device node and the object node, that is, no resource transfer has occurred between the transaction device corresponding to the target device node and the device identifier corresponding to the object node. 1 indicates that there is a connection between the target device node and the object node, that is, a resource transfer has occurred between the transaction device corresponding to the target device node and the device identifier corresponding to the object node. The server determines the column corresponding to the target device node in the adjacency matrix, and randomly determines multiple rows with the number 0 from the column. The object nodes corresponding to the rows are also negative sample object nodes. The number of negative sample object nodes determined by the server is set by the technician according to the actual situation, such as 8, 12 or 16, etc., and this embodiment of the application does not limit this.

[0134] The server obtains the third similarity between the initial device feature and the positive sample object feature. The server obtains the fourth similarity between the initial device feature and the negative sample object feature. The server updates the initial device feature based on the second difference to obtain the target device feature, and the second difference is the difference between the third similarity and the fourth similarity. If there are multiple positive sample object features, the server can obtain the third similarity between the initial device feature and each positive sample object feature respectively; if there are multiple negative sample object features, the server can obtain the fourth similarity between the initial device feature and each negative sample object feature respectively. The first difference at this time is the sum of multiple fifth differences, and each fifth difference is the difference between a third similarity and a fourth similarity. The number of fifth differences is the same as the number of third similarities.

[0135] In some embodiments, the server uses vectors to represent features. For example, an initial device feature vector is used to represent the initial device feature, a positive sample object feature vector is used to represent the positive sample object feature, and a negative sample object feature vector is used to represent the negative sample object feature. The server can then use the cosine similarity between the initial device feature vector and the positive sample object feature vector to represent the third similarity between the initial device feature and the positive sample object feature. The cosine similarity between the initial device feature vector and the negative sample object feature vector is used to represent the fourth similarity between the initial device feature and the negative sample object feature. The server constructs a second loss function based on the first difference between the third similarity and the fourth similarity, and updates the initial device feature based on the second loss function to obtain the device feature of the target transaction device.

[0136] It should be noted that the method by which the server updates the initial device features based on the second loss function to obtain the device features of the target transaction device is the same inventive concept as the method by which the server updates the initial object features based on the first loss function to obtain the object features of the object node described above. The implementation process is described in the description of formulas (1) and (2) and will not be repeated here. In addition, the purpose of the server updating the device features of the target transaction device is also to increase the third similarity as much as possible and reduce the fourth similarity as much as possible, so that the connected nodes are more similar in feature expression, thereby achieving modeling based on the connection relationship between nodes.

[0137] In a possible implementation, the server divides the target graph network into multiple sub-graph networks, and performs graph convolution on each of the multiple sub-graph networks to obtain object features of the multiple objects and the target device features.

[0138] Among them, the server performs graph convolution on the multiple subgraph networks respectively, and the method for obtaining the object features of the multiple objects and the target device features belongs to the same inventive concept as the method described in the previous embodiment. For the implementation process, please refer to the previous description and will not be repeated here. When the server divides the target graph network into multiple subgraph networks, it can be divided based on the nodes in the target graph network or based on the edges (connections) in the target graph network. This embodiment of the present application does not limit this. In some embodiments, the server uses a clustering method to cluster multiple nodes, and implements a method for dividing the target graph network into multiple subgraph networks based on the clustering results.

[0139] Since the number of nodes in the target graph network may be large, processing all nodes in the target graph network at the same time may not be possible due to the limitation of server performance. In this case, the server can split the target graph network into multiple sub-graph networks to facilitate distributed computing.

[0140] The following will be combined Figure 5 The above-mentioned step 302 is explained, taking the example of step 302 being implemented by a graph convolution model. The graph convolution model includes a graph network generation unit, a feature initialization unit, a positive and negative sample feature sampling unit, a feature update unit, and a loss function calculation unit. The graph network generation unit generates a target graph network based on the historical resource transfer information. The feature initialization unit obtains a feature matrix of the target graph network, which includes the initial features of each node. The nodes include object nodes and device nodes. The initial features include the initial object features of the object nodes and the initial device features of each device node. The positive and negative sample feature sampling unit randomly collects positive sample features and negative sample features. The positive sample features include positive sample object features and positive sample device features, and the negative sample features include negative sample object features and negative sample object features. The loss function calculation unit determines the update gradient based on the initial features, positive sample features, and negative sample features. The feature update unit updates the initial features based on the update gradient until the loss function converges to the target value.

[0141] 303. The server determines a plurality of target parameters between the plurality of objects and the target transaction device based on the object characteristics of the plurality of objects and the target device characteristics, where the target parameters are used to indicate the probability that resource transfer will occur between the corresponding objects based on the target transaction device.

[0142] The probability that the object will transfer resources based on the target transaction device can be understood as the probability of a transaction occurring between the object and the target transaction device, or as the probability that the object arrives at the store where the target transaction device is located.

[0143] In a possible implementation, the server determines a fifth similarity between the target device feature and each object feature, and based on the plurality of fifth similarities, determines the target parameter between each object and the target transaction device.

[0144] For example, the server uses a first vector to represent the target device feature and a second vector to represent each object feature. The server obtains the cosine similarity between the first vector and each second vector. This cosine similarity is the fifth similarity between the target device feature and each object feature. The server can directly determine the fifth similarity as the target parameter between the object and the target transaction device, or the server can normalize the fifth similarity, such as multiplying the fifth similarity by a fixed multiple to obtain the target parameter.

[0145] 304. The server determines a plurality of target objects from the plurality of objects based on the plurality of target parameters, wherein the target parameters of the target objects meet the target conditions.

[0146] The target condition refers to any of the following: the target parameter of the object is in the first N positions of the multiple objects, where N is a positive integer; the target parameter of the object is greater than or equal to a parameter threshold.

[0147] In one possible implementation, the target condition is that the target parameter of the object is in the top N of the multiple objects. The server sorts the multiple objects in descending order of the target parameters and selects the top N objects from the sorted multiple objects as the target objects. In other words, the server selects the top N objects from the multiple objects in descending order of the target parameters as the target objects. The number N is determined by the hardware and software requirements of the target transaction device, i.e., the number of facial features that the target transaction device can actually store.

[0148] In one possible implementation, the target condition refers to an object's target parameter being greater than or equal to a parameter threshold. Since the target parameter represents the probability that the object will transfer resources based on the target transaction device, a target parameter greater than the parameter threshold also indicates that the probability that the object will transfer resources based on the target transaction device is greater than or equal to the probability threshold, where the probability threshold is also the parameter threshold. Alternatively, in this manner, the target object is determined from multiple objects based on the probability of a transaction with the target transaction device. Specifically, the object with the highest probability of a transaction with the target transaction device is selected from the multiple objects. Furthermore, the server determines the number of target objects based on the relationship between the target parameter and the parameter threshold, which may also be N as in the previous implementation.

[0149] 305. The server sends the facial features belonging to the multiple target objects in the facial feature database to the target transaction device, and the target transaction device is configured to perform a resource transfer operation based on the received facial features.

[0150] In one possible implementation, the server queries the facial feature database based on the object accounts of the multiple target objects to obtain the facial features of the multiple target objects. The server sends the facial features of the multiple target objects to the target transaction device, and the target transaction device stores the facial features of the multiple target objects. When the target object chooses to make a facial payment on the target transaction device, the target transaction device can directly verify the payment based on the pre-stored facial features, that is, compare the facial features collected on-site with the pre-stored facial features to verify the facial payment. There is no need to upload the facial features collected on-site to the server for comparison, thereby improving the efficiency of facial payment.

[0151] Of course, to improve verification accuracy, decisions can also be made based on cloud-based verification results. For example, while the target transaction device is performing local verification, the target person's account number and collected facial features can also be uploaded to the server, and the verification results returned by the server can be combined to improve verification accuracy. However, before the server returns the verification results, the local verification information is displayed in advance, which reduces the time required for the target person to perceive the information, thereby improving the target person's facial payment experience.

[0152] It should be noted that the above steps 301-305 are explained by taking the example of the server sending facial features to the target transaction device once. In other possible implementations, since the transaction situation between the object and the target transaction device may change at any time, the server can send facial features to the target transaction device once at regular intervals to adapt to changes in the transaction situation between the object and the target transaction device.

[0153] The following will be combined Figure 6 The resource transfer method based on facial features provided in the embodiment of the present application is described, see Figure 6 The server obtains transaction records from the past M months, that is, historical resource transfer information. Based on these transaction records, the server generates a target graph network consisting of object nodes and device nodes. The server inputs the target graph network into a graph convolutional neural network to obtain object and device features. Based on these object and device features, the server determines the probability of a transaction between the object and the transaction device. This probability is also called the store visit probability. Based on the store visit probability, the server sends facial features to the transaction device.

[0154] 306. The server sends virtual coupons to the target accounts of the multiple target objects. The virtual coupons are used to deduct virtual resources when resources are transferred based on the target transaction device.

[0155] Among them, the virtual coupon can be a voucher used for face-scanning payment. When the target object makes face-scanning payment on the target transaction device, the virtual coupon can be deducted from a certain amount of virtual currency, thereby attracting the target object to trade with the target transaction device.

[0156] It should be noted that in the above steps 301-306, the server sending facial features to the target transaction device is used as an example for description. When the server sends facial features to other transaction devices, the method described in the above steps 301-306 can also be used.

[0157] All of the above optional technical solutions can be combined in any way to form optional embodiments of the present application, and will not be described in detail here.

[0158] In an embodiment of the present application, the object characteristics and the target device characteristics are determined through the historical resource transfer information between the object and the transaction device. Based on the object characteristics and the target device characteristics, the probability of the object and the target transaction device performing a resource transfer operation is predicted, and the facial features of the target object whose transaction probability meets the conditions are sent to the target transaction device in advance. In this way, if the target object uses face-swiping payment to conduct a transaction, the target transaction device can directly compare the collected facial features with the received facial features, and there is no need to upload the collected facial features to the server. The server performs a verification process based on the facial features, thereby improving the efficiency and security of face-swiping payment.

[0159] In addition, in the embodiment of the present application, the delivery of facial features does not rely on the difficult-to-obtain location information of the user device and the transaction device, but can be achieved based on historical resource transfer information, which is highly robust and practical. Since the object features and device features are obtained based on the target graph network, the transaction status between the object and the transaction device is fully explored, and the probability of the potential object on the transaction device arriving at the store can be predicted, thereby more accurately guiding the delivery of facial features. Analogously to the concept in the recommendation system, for objects A and B, object A has been to store C, while object B has not been to store C. If the similarity between the object features of object A and the object features of object B is high, then it can be considered that object A and object B have similar transaction habits. Although object B has not actually been to store C, the probability of it going to store C in the future should be high, and its facial features can be delivered to the transaction device of store C in advance.

[0160] Figure 7 This is a schematic diagram of a resource transfer device based on facial features provided in an embodiment of the present application, see Figure 7 The device includes: a feature acquisition module 701, a target parameter acquisition module 702, a target object determination module 703 and a sending module 704.

[0161] The feature acquisition module 701 is used to obtain object features and target device features of the multiple objects based on historical resource transfer information between the multiple objects and multiple transaction devices in the facial feature database, where the target device features are features of the target transaction device among the multiple transaction devices. The facial feature database stores the facial features of the multiple objects.

[0162] The target parameter acquisition module 702 is used to determine multiple target parameters between the multiple objects and the target transaction device based on the object characteristics of the multiple objects and the target device characteristics, and the target parameters are used to represent the probability that the corresponding objects will transfer resources based on the target transaction device.

[0163] The target object determination module 703 is configured to determine a plurality of target objects from the plurality of objects based on the plurality of target parameters, wherein the target parameters of the target objects meet the target conditions.

[0164] The sending module 704 is configured to send the facial features of the multiple target objects in the facial feature database to the target transaction device, and the target transaction device is configured to perform a resource transfer operation based on the received facial features.

[0165] In one possible implementation, the feature acquisition module 701 is configured to generate a target graph network based on the historical resource transfer information. The target graph network includes a plurality of object nodes, a plurality of device nodes, and lines, wherein the object nodes correspond to the objects, the device nodes correspond to the transaction devices, and the lines indicate that a resource transfer has occurred between the connected object nodes and device nodes. Graph convolution is performed on the target graph network to obtain object features of the plurality of objects and the target device features.

[0166] In one possible implementation, the feature acquisition module 701 is configured to acquire an adjacency matrix of the target graph network, where the adjacency matrix represents the connection relationships between object nodes and device nodes. The feature matrix of the target graph network is acquired, where the feature matrix includes initial object features of each object node and initial device features of each device node. Based on the feature matrix and the adjacency matrix, object features of the multiple objects and target device features are acquired.

[0167] In one possible implementation, the feature acquisition module 701 is configured to input the target graph network into a graph convolutional model, and perform feature initialization on multiple object nodes and multiple device nodes in the target graph network using the graph convolutional model to obtain initial object features for each object node and initial device features for each device node. The initial object features of each object node and the initial device features of each device node are concatenated to obtain the feature matrix.

[0168] In one possible embodiment, the feature acquisition module 701 is configured to acquire, for the initial object features of any object node in the feature matrix, positive sample device features and negative sample device features from the feature matrix based on the adjacency matrix, wherein the positive sample device features are the initial device features corresponding to the device nodes connected to the object node in the feature matrix, and the negative sample device features are the initial device features corresponding to the device nodes not connected to the object node in the feature matrix. The object features of the object node are acquired based on the initial object features, the positive sample device features, and the negative sample device features. For the initial device features of the target device node in the feature matrix, positive sample object features and negative sample object features are acquired from the feature matrix based on the adjacency matrix, wherein the positive sample object features are the initial object features corresponding to the object nodes connected to the target device node in the feature matrix, and the negative sample object features are the initial object features corresponding to the object nodes not connected to the target device node in the feature matrix. The target device features are acquired based on the initial device features, the positive sample object features, and the negative sample object features.

[0169] In a possible implementation, the feature acquisition module 701 is configured to acquire a first similarity between the initial object feature and the positive sample device feature, and acquire a second similarity between the initial object feature and the negative sample device feature.

[0170] The initial object feature is updated based on a first difference value to obtain an object feature of the object node, where the first difference value is a difference between the first similarity and the second similarity.

[0171] In one possible implementation, the feature acquisition module 701 is configured to acquire a third similarity between the initial device feature and the positive sample object feature, acquire a fourth similarity between the initial device feature and the negative sample object feature, and update the initial device feature based on a second difference value, where the second difference value is the difference between the third similarity value and the fourth similarity value, to obtain the target device feature.

[0172] In a possible implementation, the feature acquisition module 701 is used to divide the target graph network into multiple sub-graph networks, perform graph convolution on each of the multiple sub-graph networks, and obtain object features of the multiple objects and the target device features.

[0173] In a possible implementation, the target parameter acquisition module 702 is configured to determine a fifth similarity between the target device feature and each of the object features, and to determine the target parameter between each of the objects and the target transaction device based on multiple fifth similarities.

[0174] In one possible implementation, the target condition refers to any of the following:

[0175] The target parameter of the object is located in the first N positions of the multiple objects, where N is a positive integer.

[0176] The target parameter of the object is greater than or equal to the parameter threshold.

[0177] In one possible implementation, the device further includes:

[0178] The coupon sending module is used to send virtual coupons to the object accounts of the multiple target objects, and the virtual coupons are used to deduct virtual resources when resources are transferred based on the target transaction device.

[0179] It should be noted that the facial feature-based resource transfer device provided in the above embodiment only uses the division of the above functional modules as an example to illustrate when transferring resources based on facial features. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the facial feature-based resource transfer device provided in the above embodiment and the facial feature-based resource transfer method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0180] In an embodiment of the present application, the object characteristics and the target device characteristics are determined through the historical resource transfer information between the object and the transaction device. Based on the object characteristics and the target device characteristics, the probability of the object and the target transaction device performing a resource transfer operation is predicted, and the facial features of the target object whose transaction probability meets the conditions are sent to the target transaction device in advance. In this way, if the target object uses face-swiping payment to conduct a transaction, the target transaction device can directly compare the collected facial features with the received facial features, and there is no need to upload the collected facial features to the server. The server performs a verification process based on the facial features, thereby improving the efficiency and security of face-swiping payment.

[0181] The above-mentioned computer device can be implemented as a server. The structure of the server is introduced below:

[0182] Figure 8 This is a structural diagram of a server provided in an embodiment of the present application. The server 800 may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 801 and one or more memories 802, wherein the one or more memories 802 store at least one computer program, and the at least one computer program is loaded and executed by the one or more processors 801 to implement the methods provided in the above-mentioned various method embodiments. Of course, the server 800 may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The server 800 may also include other components for implementing device functions, which will not be described in detail here.

[0183] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including a computer program. The computer program can be executed by a processor to implement the resource transfer method based on facial features in the above embodiment. For example, the computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, or an optical data storage device.

[0184] In an exemplary embodiment, a computer program product or computer program is also provided, which includes a program code, which is stored in a computer-readable storage medium. A processor of a computer device reads the program code from the computer-readable storage medium, and the processor executes the program code, so that the computer device executes the above-mentioned resource transfer method based on facial features.

[0185] In some embodiments, the computer program involved in the embodiments of the present application may be deployed and executed on a computer device, or on multiple computer devices located at one location, or on multiple computer devices distributed at multiple locations and interconnected through a communication network. Multiple computer devices distributed at multiple locations and interconnected through a communication network may constitute a blockchain system.

[0186] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0187] The above are only optional embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A resource transfer method based on facial features, characterized in that: The method comprises: Based on historical resource transfer information between multiple objects and multiple transaction devices in a facial feature database, obtaining object features and target device features of the multiple objects, wherein the target device features are features of a target transaction device among the multiple transaction devices, the facial feature database storing the facial features of the multiple objects, the object features being used to represent resource transfer features of the corresponding objects on the multiple transaction devices, and the target device features being used to represent resource transfer features of the multiple objects on the target transaction device; Determining a plurality of target parameters between the plurality of objects and the target transaction device based on the object characteristics of the plurality of objects and the target device characteristics, the target parameters being used to indicate a probability that a resource transfer will occur between the corresponding objects based on the target transaction device; determining a plurality of target objects from the plurality of objects based on the plurality of target parameters, the target parameters of the target objects satisfying target conditions; The facial features of the multiple target objects in the facial feature database are sent to the target transaction device, and the target transaction device is configured to perform a resource transfer operation based on the received facial features.

2. The method according to claim 1, characterized in that The acquiring of the object features and target device features of the plurality of objects based on the historical resource transfer information between the plurality of objects and the plurality of transaction devices in the facial feature database includes: Based on the historical resource transfer information, a target graph network is generated, wherein the target graph network includes a plurality of object nodes, a plurality of device nodes, and lines, wherein the object nodes correspond to the objects, the device nodes correspond to the transaction devices, and the lines are used to indicate that resource transfers have occurred between the connected object nodes and device nodes; Perform graph convolution on the target graph network to obtain object features of the multiple objects and the target device features.

3. The method according to claim 2, characterized in that The performing graph convolution on the target graph network to obtain the object features of the multiple objects and the target device features includes: Obtaining an adjacency matrix of the target graph network, wherein the adjacency matrix is ​​used to represent the connection relationship between object nodes and device nodes; Acquire a feature matrix of the target graph network, wherein the feature matrix includes initial object features of each of the object nodes and initial device features of each of the device nodes; Based on the feature matrix and the adjacency matrix, object features of the plurality of objects and the target device features are acquired.

4. The method according to claim 3, characterized in that The acquiring of the feature matrix of the target graph network comprises: Inputting the target graph network into a graph convolution model, performing feature initialization on a plurality of object nodes and a plurality of device nodes in the target graph network through the graph convolution model, and obtaining an initial object feature of each of the object nodes and an initial device feature of each of the device nodes; The initial object features of each of the object nodes and the initial device features of each of the device nodes are concatenated to obtain the feature matrix.

5. The method according to claim 3, characterized in that The acquiring, based on the feature matrix and the adjacency matrix, the object features of the plurality of objects and the target device features comprises: For the initial object features of any object node in the feature matrix, based on the adjacency matrix, obtain positive sample device features and negative sample device features from the feature matrix, where the positive sample device features are the initial device features corresponding to the device nodes connected to the object node in the feature matrix, and the negative sample device features are the initial device features corresponding to the device nodes not connected to the object node in the feature matrix; obtain the object features of the object node based on the initial object features, the positive sample device features, and the negative sample device features; For the initial device features of the target device node in the feature matrix, based on the adjacency matrix, positive sample object features and negative sample object features are obtained from the feature matrix, where the positive sample object features are the initial object features corresponding to the object nodes connected to the target device node in the feature matrix, and the negative sample object features are the initial object features corresponding to the object nodes not connected to the target device node in the feature matrix; based on the initial device features, the positive sample object features, and the negative sample object features, the target device features are obtained.

6. The method according to claim 5, characterized in that The acquiring the object feature of the object node based on the initial object feature, the positive sample device feature, and the negative sample device feature includes: Obtaining a first similarity between the initial object feature and the positive sample device feature; Obtaining a second similarity between the initial object feature and the negative sample device feature; The initial object feature is updated based on a first difference value to obtain the object feature of the object node, where the first difference value is the difference between the first similarity and the second similarity.

7. The method according to claim 5, characterized in that The acquiring the target device feature based on the initial device feature, the positive sample object feature, and the negative sample object feature includes: Obtaining a third similarity between the initial device feature and the positive sample object feature; Obtaining a fourth similarity between the initial device feature and the negative sample object feature; The initial device feature is updated based on a second difference value to obtain the target device feature, where the second difference value is the difference between the third similarity and the fourth similarity.

8. The method according to claim 2, characterized in that The performing graph convolution on the target graph network to obtain the object features of the multiple objects and the target device features includes: The target graph network is divided into multiple sub-graph networks, and graph convolution is performed on each of the multiple sub-graph networks to obtain object features of the multiple objects and the target device features.

9. The method according to claim 1, characterized in that The determining, based on the object features of the multiple objects and the target device features, multiple target parameters between the multiple objects and the target transaction device includes: determining a fifth similarity between the target device feature and each of the object features; Based on a plurality of fifth similarities, the target parameter between each of the objects and the target transaction device is determined.

10. The method according to claim 1, characterized in that The target condition refers to any of the following: The target parameter of the object is in the first N positions of the plurality of objects, where N is a positive integer; The target parameter of the object is greater than or equal to the parameter threshold.

11. The method according to claim 1, wherein The method further comprises: Sending virtual coupons to the target accounts of the multiple target objects, wherein the virtual coupons are used to deduct virtual resources when resources are transferred based on the target transaction device.

12. A resource transfer device based on facial features, characterized in that: The device comprises: a feature acquisition module, configured to acquire object features and target device features of the multiple objects based on historical resource transfer information between the multiple objects and the multiple transaction devices in a facial feature database, wherein the target device features are features of a target transaction device among the multiple transaction devices, the facial feature database storing the facial features of the multiple objects, the object features representing resource transfer features of the corresponding objects on the multiple transaction devices, and the target device features representing resource transfer features of the multiple objects on the target transaction device; a target parameter acquisition module, configured to determine a plurality of target parameters between the plurality of objects and the target transaction device based on the object characteristics of the plurality of objects and the target device characteristics, the target parameters being used to indicate a probability that a resource transfer will occur between the corresponding objects based on the target transaction device; a target object determination module, configured to determine a plurality of target objects from the plurality of objects based on the plurality of target parameters, wherein the target parameters of the target objects meet the target conditions; The sending module is used to send the facial features of the multiple target objects in the facial feature database to the target transaction device, and the target transaction device is used to perform a resource transfer operation based on the received facial features.

13. The device according to claim 12, characterized in that The feature acquisition module is used to generate a target graph network based on the historical resource transfer information, wherein the target graph network includes multiple object nodes, multiple device nodes and lines, wherein the object nodes correspond to the objects, the device nodes correspond to the transaction devices, and the lines are used to indicate that resource transfer has occurred between the connected object nodes and device nodes; graph convolution is performed on the target graph network to obtain object features of the multiple objects and the target device features.

14. The device according to claim 13, characterized in that The feature acquisition module is configured to acquire an adjacency matrix of the target graph network, wherein the adjacency matrix is ​​used to represent the connection relationship between object nodes and device nodes; and acquire a feature matrix of the target graph network, wherein the feature matrix includes initial object features of each object node and initial device features of each device node. Based on the feature matrix and the adjacency matrix, object features of the plurality of objects and the target device features are acquired.

15. The device according to claim 14, characterized in that The feature acquisition module is used to input the target graph network into a graph convolution model, perform feature initialization on multiple object nodes and multiple device nodes in the target graph network through the graph convolution model, and obtain the initial object features of each object node and the initial device features of each device node; and splice the initial object features of each object node and the initial device features of each device node to obtain the feature matrix.

16. The device according to claim 14, characterized in that The feature acquisition module is configured to acquire, for an initial object feature of any object node in the feature matrix, positive sample device features and negative sample device features from the feature matrix based on the adjacency matrix, wherein the positive sample device features are initial device features corresponding to device nodes connected to the object node in the feature matrix, and the negative sample device features are initial device features corresponding to device nodes not connected to the object node in the feature matrix; Based on the initial object features, the positive sample device features, and the negative sample device features, the object features of the object node are obtained; for the initial device features of the target device node in the feature matrix, based on the adjacency matrix, positive sample object features and negative sample object features are obtained from the feature matrix, wherein the positive sample object features are the initial object features corresponding to the object nodes connected to the target device node in the feature matrix, and the negative sample object features are the initial object features corresponding to the object nodes not connected to the target device node in the feature matrix; The target device feature is acquired based on the initial device feature, the positive sample object feature, and the negative sample object feature.

17. The device according to claim 16, characterized in that The feature acquisition module is configured to acquire a first similarity between the initial object feature and the positive sample device feature; acquire a second similarity between the initial object feature and the negative sample device feature; and update the initial object feature based on a first difference to obtain the object feature of the object node, where the first difference is the difference between the first similarity and the second similarity.

18. The device according to claim 16, characterized in that The feature acquisition module is configured to obtain a third similarity between the initial device feature and the positive sample object feature; obtain a fourth similarity between the initial device feature and the negative sample object feature; and update the initial device feature based on a second difference to obtain the target device feature, where the second difference is the difference between the third similarity and the fourth similarity.

19. The device according to claim 13, characterized in that The feature acquisition module is used to divide the target graph network into multiple sub-graph networks, perform graph convolution on the multiple sub-graph networks respectively, and obtain the object features of the multiple objects and the target device features.

20. The device according to claim 12, characterized in that The target parameter acquisition module is configured to determine a fifth similarity between the target device feature and each of the object features; and determine the target parameter between each of the objects and the target transaction device based on multiple fifth similarities.

21. The device according to claim 12, characterized in that The target condition refers to any of the following: The target parameter of the object is in the first N positions of the plurality of objects, where N is a positive integer; The target parameter of the object is greater than or equal to the parameter threshold.

22. The device according to claim 12, characterized in that The device further comprises: The coupon sending module is used to send virtual coupons to the target accounts of the multiple target objects, and the virtual coupons are used to deduct virtual resources when resources are transferred based on the target transaction device.

23. A computer device, characterized in that: The computer device includes one or more processors and one or more memories, and at least one computer program is stored in the one or more memories. The computer program is loaded and executed by the one or more processors to implement the resource transfer method based on facial features as described in any one of claims 1 to 11.

24. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the resource transfer method based on facial features according to any one of claims 1 to 11.

25. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the resource transfer method based on facial features described in any one of claims 1 to 11 is implemented.

Images