Payment verification method identification method, device, electronic device and storage medium

Abstract

The embodiment of the present application discloses a payment verification method identification method, device, electronic device and storage medium; the embodiment of the present application can obtain the facial image features of the payment object and the device identification of the facial payment device, search the target facial features matching the facial image features from the facial database, and determine the target object corresponding to the target facial features, search the payment device graph features corresponding to the device identification and the target object graph features corresponding to the target object from the graph feature database, calculate the initial payment probability of the target object making payment at the target payment place based on the payment device graph features and the target object graph features, and then generate a fused payment probability based on the similarity between the facial image features and the target facial features and the initial payment probability, and identify the payment verification method of the payment object according to the fused payment probability. This scheme can effectively improve the accuracy of payment verification method identification.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a payment verification mode identification method, device, electronic device and storage medium. Background Art

[0002] In recent years, identity authentication technologies, such as fingerprint recognition, eye pattern recognition, iris recognition, and face recognition, have been greatly developed. Among them, face recognition technology is the most prominent, and it has been increasingly widely used in various identity authentication systems. With the vigorous development of face recognition technology, face-swiping payment has also developed rapidly. However, most researchers define face-swiping as an image recognition technology. However, relying solely on image recognition technology cannot solve the problem of face recognition of some similar faces, twins, and other highly similar faces, which reduces the accuracy of face recognition, making the accuracy of the payment verification method for identifying users low, and the face-swiping technology cannot be continuously improved during use, affecting the face-swiping experience. Summary of the invention

[0003] The embodiments of the present application provide a payment verification method identification method, device, electronic device and storage medium, which can effectively improve the accuracy of payment verification method identification.

[0004] The present application embodiment provides a method for identifying a payment verification method, including:

[0005] Acquire facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of a payment object and a device identifier of a facial payment device;

[0006] Searching for target facial features that match the facial image features from a facial database, and determining a target object corresponding to the target facial features;

[0007] Searching a graph feature database for a payment device graph feature corresponding to the device identifier and a target object graph feature corresponding to the target object;

[0008] Calculating an initial payment probability of the target object making payment at the target payment location based on the payment device graph feature and the target object graph feature;

[0009] generating a fused payment probability based on the similarity between the facial image features and the target facial features and the initial payment probability;

[0010] A payment verification method of the payment object is identified according to the fused payment probability.

[0011] Accordingly, the embodiment of the present application further provides a payment verification method identification device, including:

[0012] An acquisition unit, configured to acquire facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of a payment object and a device identifier of a facial payment device;

[0013] A matching unit, used to search a face database for a target face feature that matches the face image feature, and determine a target object corresponding to the target face feature;

[0014] A search unit, configured to search a graph feature database for a payment device graph feature corresponding to the device identifier and a target object graph feature corresponding to the target object;

[0015] A calculation unit, configured to calculate an initial payment probability of a target object making payment at the target payment location based on the payment device graph feature and the target object graph feature;

[0016] A fusion unit, configured to generate a fused payment probability based on the similarity between the facial image feature and the target facial feature and the initial payment probability;

[0017] An identification unit is used to identify the payment verification method of the payment object according to the fused payment probability.

[0018] Optionally, in some embodiments, the matching unit can be specifically used to calculate the similarity between the facial image features and the candidate facial features in the facial database; obtain the target facial features whose similarity meets preset conditions from the facial database according to the calculation results, and determine the target object corresponding to the target facial features.

[0019] Optionally, in some embodiments, the payment verification method recognition device may further include a training unit, and the training unit includes a collection subunit, an initialization subunit, a training subunit and a storage subunit, as follows:

[0020] The collection subunit is used to collect a plurality of training sample pairs, wherein the training sample pairs include at least one positive sample and at least one negative sample;

[0021] The initialization subunit is used to respectively initialize the features of the sample objects and sample devices in the training sample pairs to obtain initial object features and initial device features;

[0022] The training subunit is used to train the initial object features and the initial device features based on the positive samples and the negative samples to obtain object graph features and device graph features;

[0023] The storage subunit is used to store the object graph features and the device graph features in a graph feature database.

[0024] Optionally, in some embodiments, the collection subunit can be specifically used to determine the link relationship between the sample object and the sample device based on the historical payment records, and construct a heterogeneous network diagram of the sample object and the sample device based on the link relationship; in the heterogeneous network diagram, the sample objects and sample devices are used as nodes for path sampling, and the paths with link relationships between the collected nodes are used as positive samples, and the paths with at least one no link relationship between the collected nodes are used as negative samples.

[0025] Optionally, in some embodiments, the acquisition subunit can be specifically used to treat sample objects and sample devices as different types of nodes to obtain a predefined meta-path, where the meta-path includes link relationships between different types of nodes; calculate the transition probability of each step based on the link relationship between different types of nodes in the meta-path, and determine a random walk sampling strategy based on the transition probability of each step; and perform path sampling in the heterogeneous network graph based on the random walk sampling strategy.

[0026] Optionally, in some embodiments, the training subunit can be specifically used to calculate the correlation between the initial object features and the initial device features in the positive sample to obtain the prediction result of the positive sample; calculate the correlation between the initial object features and the initial device features in the negative sample to obtain the prediction result of the negative sample; use the loss function to adjust the initial object features and the initial device features according to the prediction result of the positive sample and the actual result of the positive sample, and according to the prediction result of the negative sample and the actual result of the negative sample, until the loss function converges to obtain the object graph features and the device graph features.

[0027] Optionally, in some embodiments, the target facial feature includes a first facial feature and a second facial feature, the target object includes a first object and a second object, and the matching unit may be specifically configured to sort the similarities between the facial image feature and the candidate facial feature according to the calculation result; based on the sorting result, obtain the first facial feature corresponding to the first similarity and the second facial feature corresponding to the second similarity in the sorting; determine the first facial feature and the second facial feature as the target facial feature, and determine the first object corresponding to the first facial feature and the second object corresponding to the second facial feature;

[0028] The search unit may be specifically configured to search a graph feature database for a payment device graph feature corresponding to the device identifier, a first object graph feature corresponding to the first object, and a second object graph feature corresponding to the second object.

[0029] Optionally, in some embodiments, the initial payment probability includes a first payment probability and a second payment probability, and the calculation unit may be specifically used to calculate a first payment probability of the first object making payment at the target payment place based on the payment device graph feature and the first object graph feature; and calculate a second payment probability of the second object making payment at the target payment place based on the payment device graph feature and the second object graph feature;

[0030] The fusion unit may be specifically configured to generate a first fusion probability based on the first similarity and the first payment probability, and to generate a second fusion probability based on the second similarity and the second payment probability.

[0031] Optionally, in some embodiments, the identification unit can be specifically used to obtain the maximum probability of the first fusion probability and the second fusion probability, and determine the maximum probability as the first measurement value; obtain the minimum probability of the first fusion probability and the second fusion probability, calculate the difference between the maximum probability and the minimum probability, and determine the difference as the second measurement value; if the first measurement value is greater than a first preset threshold and the second measurement value is greater than a second preset threshold, the payment verification method of the payment object is the first verification method; if the first measurement value is greater than the first preset threshold or the second measurement value is greater than the second preset threshold, the payment verification method of the payment object is the second verification method; if the first measurement value is less than or equal to the first preset threshold, and the second measurement value is less than or equal to the second preset threshold, the payment verification method of the payment object is the third verification method.

[0032] Optionally, in some embodiments, the calculation unit may be specifically used to calculate the correlation between the payment device graph feature and the target object graph feature; and determine the initial payment probability of the target object making payment at the target payment location based on the correlation.

[0033] Optionally, in some embodiments, the fusion unit can be specifically used to obtain a first interpolation coefficient of the similarity between the facial image features and the target facial features, and a second interpolation coefficient of the initial payment probability; based on the first interpolation coefficient and the second interpolation coefficient, the similarity between the facial image features and the target facial features, and the initial payment probability are fused to obtain a fused payment probability.

[0034] In addition, an embodiment of the present application also provides a computer-readable storage medium, which stores multiple instructions, and the instructions are suitable for a processor to load to execute the steps in the identification method of any payment verification method provided in the embodiment of the present application.

[0035] In addition, an embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for identifying any payment verification method provided in the embodiment of the present application are implemented.

[0036] According to one aspect of the present application, a computer program product or a computer program is provided, which includes computer instructions, which are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in various optional implementations of the above-mentioned payment verification method identification.

[0037] This embodiment can obtain facial payment information initiated at the target payment place, the facial payment information includes the facial image features of the payment object and the device identification of the facial payment device, then search the facial database for the target facial features that match the facial image features, and determine the target object corresponding to the target facial features, search the image feature database for the payment device image features corresponding to the device identification, and the target object image features corresponding to the target object, then calculate the initial payment probability of the target object making payment at the target payment place based on the payment device image features and the target object image features, then generate a fused payment probability based on the similarity between the facial image features and the target facial features, and the initial payment probability, and then identify the payment verification method of the payment object based on the fused payment probability. This solution can effectively improve the accuracy of payment verification method recognition, and thus improve the accuracy of face-swiping payment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1a It is a scenario diagram of a method for identifying a payment verification method provided in an embodiment of the present application;

[0040] Figure 1b This is a first flow chart of a method for identifying a payment verification method provided in an embodiment of the present application;

[0041] Figure 2a is a first schematic diagram of a heterogeneous network diagram provided in an embodiment of the present application;

[0042] Figure 2b is a second schematic diagram of a heterogeneous network diagram provided in an embodiment of the present application;

[0043] Figure 2c is a second flow chart of the method for identifying the payment verification method provided in an embodiment of the present application;

[0044] Figure 2d is a third flow chart of the method for identifying a payment verification method provided in an embodiment of the present application;

[0045] Figure 2e is a fourth flow chart of the method for identifying a payment verification method provided in an embodiment of the present application;

[0046] Figure 3 It is a structural diagram of an identification device of a payment verification method provided in an embodiment of the present application;

[0047] Figure 4 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0049] The principles of the present application are illustrated by implementation in an appropriate computing environment. In the following description, the specific embodiments of the present application will be described with reference to the steps and symbols executed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be mentioned several times as being executed by a computer, and the computer execution referred to herein includes the operation of a computer processing unit by an electronic signal representing data in a structured form. This operation converts the data or maintains it at a location in the memory system of the computer, which can be reconfigured or otherwise change the operation of the computer in a manner familiar to testers in the field. The data structure maintained by the data is the physical location of the memory, which has specific characteristics defined by the data format. However, the principles of the present application are described in the above text, which does not represent a limitation, and testers in the field will understand that the following various steps and operations can also be implemented in hardware.

[0050] The term "unit" used in this article can be regarded as a software object executed on the computing system. The different components, units, engines and services described in this article can be regarded as implementation objects on the computing system. The devices and methods described in this article can be implemented in software, and of course can also be implemented in hardware, all within the scope of protection of this application.

[0051] The terms "first", "second", and "third" in this application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but some embodiments also include steps or units that are not listed, or some embodiments also include other steps or units inherent to these processes, methods, products, or devices.

[0052] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0053] The embodiment of the present application provides a payment verification method identification method, device, electronic device and storage medium. The payment verification method identification device can be integrated in an electronic device, which can be a server, a terminal or other device.

[0054] The payment verification method identification method provided in the embodiment of the present application involves computer vision technology, natural language processing technology and machine learning technology in the field of artificial intelligence. The computer vision technology of artificial intelligence can be used to extract features of facial images, and machine learning can be used to learn features based on the relationship between users and devices. Natural language processing technology can be used to extract the relationship between payment objects and payment devices, thereby realizing the identification of payment verification methods for payment objects.

[0055] Among them, 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 obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level technology and software-level technology. Among them, artificial intelligence software technology mainly includes computer vision technology, machine learning / deep learning and other directions.

[0056] Among them, Computer Vision (CV) is a science that studies how to make machines "see". To put it more specifically, it refers to machine vision that uses computers to replace human eyes to identify and measure targets, and further processes images so that the images are processed by computers to become more suitable for human observation or transmission to instruments for detection. As a scientific discipline, computer vision studies related theories and technologies, and attempts to establish an artificial intelligence system that can obtain information from images or multidimensional data. Computer vision technology usually includes technologies such as image processing and image recognition, as well as common biometric recognition technologies such as face recognition and human posture recognition.

[0057] Among them, natural language processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language people use in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question answering, knowledge graph and other technologies.

[0058] Among them, machine learning (ML) is a multi-disciplinary interdisciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0059] For example, Figure 1a As shown, first, the electronic device integrated with the recognition device of the payment verification method can obtain the facial payment information initiated at the target payment place, the facial payment information includes the facial image features of the payment object and the device identification of the facial payment device, then, the target facial features matching the facial image features are searched from the facial database, and the target object corresponding to the target facial features are determined, and the payment device graph features corresponding to the device identification and the target object graph features corresponding to the target object are searched from the graph feature database, then, the initial payment probability of the target object making payment at the target payment place is calculated based on the payment device graph features and the target object graph features, and then, based on the similarity between the facial image features and the target facial features, and the initial payment probability, a fused payment probability is generated, and then, the payment verification method of the payment object is identified according to the fused payment probability. Since the scheme can construct a heterogeneous network graph of objects and devices by making full use of the payment information between objects and devices in the historical payment records, and fully mine the structural information of objects and devices in the graph through the heterogeneous network graph, it can achieve a similar collaborative filtering effect, and finally, the graph model constructs the object payment probability using the device or the transfer probability between the object and the device through the measurement of the object features and the device features. Finally, facial recognition technology is combined with the recognition device of the payment verification method to make a comprehensive decision, which effectively improves the accuracy of payment verification method recognition, and then improves the accuracy of face payment, solves some highly similar facial recognition problems, and effectively improves the convenience of payment for the payment object, which in turn feeds back to face payment, forming an effective closed loop and improving user experience.

[0060] It should be noted that the description order of the following embodiments is not intended to limit the preferred order of the embodiments.

[0061] This embodiment will be described from the perspective of an identification device for a payment verification method, which can be specifically integrated into an electronic device, which can be a server or a terminal. Among them, the terminal can include mobile phones, tablet computers, laptop computers, and personal computers (PCs).

[0062] A method for identifying a payment verification method comprises: obtaining facial payment information initiated at a target payment location, the facial payment information comprising facial image features of a payment object and a device identification of a facial payment device, then searching a facial database for target facial features that match the facial image features, and determining a target object corresponding to the target facial features, searching a graphical feature database for payment device graphical features corresponding to the device identification, and a target object graphical feature corresponding to the target object, then calculating an initial payment probability of the target object making payment at the target payment location based on the payment device graphical features and the target object graphical features, then generating a fused payment probability based on the similarity between the facial image features and the target facial features, and the initial payment probability, and then identifying the payment verification method of the payment object according to the fused payment probability.

[0063] like Figure 1b As shown, the specific process of the identification method of the payment verification method can be as follows:

[0064] 101. Obtain facial payment information initiated at a target payment location, where the facial payment information includes facial image features of a payment object and a device identifier of a facial payment device.

[0065] The target payment location may refer to a location where the payment object needs to make a payment transaction, such as a supermarket, a store, a shop, a convenience store, etc. The payment object may refer to an object that needs to be identified by a payment verification method when paying by face recognition, such as a user who purchases goods, etc.

[0066] For example, when a payment request initiated by a target payment location is received, the facial image of the payment object and the device identification of the facial payment device that initiated the payment request can be obtained according to the payment request, and features of the facial image of the payment object can be extracted to obtain facial image features.

[0067] For example, when the payment object (such as XX user) purchases goods at the target payment place (such as YY store) and uses a facial payment device (such as ZZ device) to make payment, a payment request is generated and sent to the recognition device of the payment verification method, so that the recognition device of the payment verification method performs face detection and alignment on the XX user and obtains the device identification sn of the ZZ device, such as taking a face photo of the XX user to obtain a face image of the XX user, and performing feature extraction on the face image of the XX user to obtain the face image features of the XX user.

[0068] Optionally, face detection can use the Multi-task convolutional neural network (MTCNN) method. MTCNN is a face detection and face alignment method based on deep learning. It can complete the tasks of face detection and face alignment at the same time. By adopting the cascaded CNN structure and multi-task learning, two tasks - face detection and face alignment - are completed at the same time, and the bounding box of the face and the positions of the key points of the face (eyes, nose, mouth) are output. MTCNN realizes face detection and alignment. It is a model that realizes face detection and five-point calibration in one network, mainly through the cascade of CNN models to realize multi-task learning network. The whole model can be divided into three stages. The first stage quickly generates a series of candidate windows through a shallow CNN network. The second stage filters out most of the non-face candidate windows through a more capable CNN network. The third stage finds the five markers on the face through a more powerful network.

[0069] Optionally, image registration can use affine transformation. If only translation, rotation, and scaling are considered, the transformation between the floating image and the fixed image contains six parameters, which only requires three sets of point pairs, that is, six equations, to solve. Its purpose is to standardize the faces of any posture captured by the camera. Face scanning is mainly for cooperative face recognition. Relatively speaking, the collected face posture is not very large, so a simple MTCNN can cope with it. Face detection technology will not be described in detail here.

[0070] Optionally, the face image can be an RGB (red, green, blue) image, and the RGB image feature extraction can be performed using a deep learning model. The deep learning model can be trained offline using a large number of RGB images with identity annotations. The training algorithm can be a margin based softmax method. For example, by inputting the standardized image into the deep learning model, the feature F of the corresponding dimension can be obtained. Since the scale of industrial-grade face recognition training is very large and each ID corresponds to a category, the final classification layer is very resource-intensive. Therefore, the classification layer adopts a model parallel method, and the classification layer is mixed with data parallel training.

[0071] 102. Search a face database for a target face feature that matches the face image feature, and determine a target object corresponding to the target face feature.

[0072] For example, the similarity between the facial image feature and the candidate facial feature in the facial database may be calculated; based on the calculation result, a target facial feature whose similarity meets a preset condition is obtained from the facial database, and a target object corresponding to the target facial feature is determined.

[0073] The facial database may be configured before the payment verification method is identified and stored in the payment verification method identification device. For example, facial features of historical payment objects may be used as candidate facial features and stored in the facial database based on historical payment information of all payment devices.

[0074] In order to improve the recognition accuracy of the payment verification method and fully explore the structural information between the payment object and the payment device, multiple target facial features whose similarity meets the preset conditions can be selected from the candidate facial features. For example, in order to improve the recognition accuracy of face-swiping payment while ensuring the efficiency of face-swiping payment, the target facial features can be two facial features whose similarity meets the preset conditions.

[0075] There are many ways to set the preset conditions. For example, they can be flexibly set according to the needs of actual applications, or they can be pre-set and stored in the electronic device. In addition, the preset conditions can be built into the electronic device, or they can be stored in a memory and sent to the electronic device, etc. For example, the preset conditions can be the two with the highest similarity (top2), or the similarity meets a certain threshold, etc.

[0076] For example, the target facial features include a first facial feature and a second facial feature, and the target object includes a first object and a second object. Specifically, the similarity between the facial image features and the candidate facial features in the facial database can be calculated, and the similarities between the facial image features and the candidate facial features can be sorted according to the calculation results; based on the sorting results, the first facial features corresponding to the first similarity and the second facial features corresponding to the second similarity in the sorting are obtained; the first facial features and the second facial features are determined as target facial features, and the first object corresponding to the first facial feature and the second object corresponding to the second facial feature are determined. For example, the first facial feature corresponding to the first similarity with the highest similarity (i.e., the first) and the second facial feature corresponding to the second similarity with the second similarity can be obtained according to the sorting results, that is, the first similarity can refer to the first-ranked similarity, and the second similarity can refer to the second-ranked similarity.

[0077] For example, based on the open source faiss library, you can input the extracted facial feature P, retrieve the top 2 most similar facial features, and obtain the corresponding similarity score and user ID. For example, the top 2 are user A and user B, user A's facial feature is Pa, and user B's facial feature is Pb. The similarity score can be measured by cosine. For example, the specific calculation can be as follows:

[0078] Sim_a=cosine(P,Pa)

[0079] Sim_b=cosine(P,Pb)

[0080] 103. Search a graph feature database for a payment device graph feature corresponding to the device identifier and a target object graph feature corresponding to the target object.

[0081] For example, when the target facial features include a first facial feature and a second facial feature, and the target object includes a first object and a second object, the payment device graph feature corresponding to the device identifier, the first object graph feature corresponding to the first object, and the second object graph feature corresponding to the second object can be specifically searched from the graph feature database.

[0082] In order to improve the efficiency of recognition of payment verification methods, a graph feature database may be established first, and then the graph feature database may be used to search for the payment device graph feature corresponding to the device identifier and the target object graph feature corresponding to the target object. Optionally, the graph features in the graph feature database may be trained by multiple training samples. Specifically, it may be trained by other devices and then provided to the recognition device of the payment verification method, or it may be trained by the recognition device of the payment verification method itself, that is, before searching for the payment device graph feature corresponding to the device identifier and the target object graph feature corresponding to the target object from the graph feature database, the recognition method of the payment verification method may also include:

[0083] Collecting a plurality of training sample pairs, wherein the training sample pairs include at least one positive sample and at least one negative sample;

[0084] Initializing features of sample objects and sample devices in the training sample pairs respectively to obtain initial object features and initial device features;

[0085] Based on the positive samples and the negative samples, the initial object features and the initial device features are trained to obtain object graph features and device graph features;

[0086] The object graph features and the device graph features are stored in a graph feature database.

[0087] There are many ways to collect training sample pairs. For example, the method of heterogeneous network representation learning can be used to capture the structural relationship and semantic relationship between different types of nodes such as sample objects and sample devices. For example, the step "collecting multiple training sample pairs, the training sample pairs include at least one positive sample and at least one negative sample" can specifically determine the link relationship between the sample object and the sample device based on the historical payment record, and construct a heterogeneous network graph of the sample object and the sample device based on the link relationship; in the heterogeneous network graph, the sample object and the sample device are used as nodes for path sampling, and the paths with link relationships between the collected nodes are used as positive samples, and at least one path without link relationship between the collected nodes is used as a negative sample. For example, the graph model can be obtained through offline training. First, the sample objects and sample devices in the historical face-swiping records can be used as two types of nodes, and then the sample objects can add edge relationships to the sample devices that have been paid by the sample objects, so as to form a bipartite graph of the sample objects and sample devices. Since the types of nodes are different, this is a heterogeneous graph.

[0088] Among them, in the heterogeneous network graph, there are many ways to sample sample objects and sample devices. For example, a random walk can be performed in a certain order to obtain training sample pairs of sample objects and sample devices. For example, in the step "in the heterogeneous network graph, sample objects and sample devices are used as nodes for path sampling", the sample objects and sample devices can be used as different types of nodes to obtain a predefined meta-path, and the meta-path includes the link relationship between nodes of different types; the transition probability of each step is calculated based on the link relationship between nodes of different types in the meta-path, and the random walk sampling strategy is determined according to the transition probability of each step; in the heterogeneous network graph, path sampling is performed based on the random walk sampling strategy.

[0089] Among them, the meta-path is a specific path connecting two entities, such as "actor->movie->director->movie->actor". This meta-path can connect two actors, so it can be regarded as a way to mine the potential relationship between actors. The advantage of this type of method is that it fully and intuitively utilizes the network structure of the knowledge graph. The meta-path can search for similarities in heterogeneous networks. Meta-path is a path containing a sequence of relations, and these relations are defined between different types of objects.

[0090] For example, a meta-path-based random walk scheme can be used to collect sequences (i.e., paths). This random walk method can simultaneously capture the semantic and structural relationships between different types of vertices, and promote the conversion of heterogeneous network structures to the Skip-Gram model of metapath2vec. A meta-path scheme can be defined as follows:

[0091]

[0092] Among them, E represents an edge, V represents a node, R represents a random walk, N represents a neighbor node, and R i Represents different types of nodes V i With V i+1 The transition probability of the i-th step is defined as:

[0093]

[0094] in, Represents a vertex V t+1 The domain vertex set of type. Then, the three fractions above represent:

[0095] 1. There is an edge between the two points, and the next point belongs to the next type of node on the metapath we defined.

[0096] 2. There is an edge between the two points, but the next point does not belong to the next type of node on the metapath we defined.

[0097] 3. There is no edge between two points.

[0098] Then, using recursive guidance for random walker, for example, we can do the following:

[0099]

[0100] After a large number of training sample pairs are obtained based on the above formulas, the training sample pairs can be used for training. There are many ways to train. For example, the features of sample objects and sample devices can be randomly initialized first, and then trained through a reasonable loss function and gradient back propagation algorithm to obtain the final effective features of sample objects and sample devices.

[0101] For example, the step of "training the initial object features and the initial device features based on the positive sample and the negative sample to obtain object graph features and device graph features" can specifically calculate the correlation between the initial object features and the initial device features in the positive sample to obtain the prediction result of the positive sample; calculate the correlation between the initial object features and the initial device features in the negative sample to obtain the prediction result of the negative sample; use the loss function to adjust the initial object features and the initial device features according to the prediction result of the positive sample and the actual result of the positive sample, and according to the prediction result of the negative sample and the actual result of the negative sample, until the loss function converges to obtain object graph features and device graph features.

[0102] There are many ways to choose the loss function, for example, you can use the maximum loss function (softmax), you can use the triplet loss function (Triplet loss), you can also use other loss functions, and so on.

[0103] For example, the specific formula of softmax loss function training using skip-gram model can be as follows:

[0104]

[0105] This formula is actually a downsampled softmax. Among them, the training sample pairs obtained by the metapath can be one pair of positive samples corresponding to N pairs of negative samples, and then trained with softmax + cross-entropy loss.

[0106] For another example, in order to further improve the accuracy of identifying payment verification methods and more accurately determine whether it is the same face, skip-gram can be used to train Triplet loss. Triplet Loss is a loss function in deep learning, used to train samples with small differences, such as faces, etc. The input data includes anchor examples, positive examples, and negative examples. By optimizing the distance between the anchor example and the positive example to be less than the distance between the anchor example and the negative example, the similarity calculation of the samples is realized. What Triplet loss learns is a good feature vector (embedding). Similar images are close in the embedding space, and it can be determined whether it is the same face. Specifically, it can be as follows:

[0107] The input is a triplet <a, p, n>, where a: anchor; p: positive, a sample of the same category as a; n: negative, a sample of a different category from a. The formula can be as follows:

[0108] L = max(d(a, p) - d(a, n) + margin, 0)

[0109] So the final optimization goal is to shorten the distance between a and p and lengthen the distance between a and n.

[0110] For easy triplets: L = 0, that is, d(a, p) + margin < d(a, n). In this case, no optimization is required. Naturally, the distance between a and p is very close, and the distance between a and n is far.

[0111] For hard triplets: d(a, n) < d(a, p), that is, the distance between a and p is far.

[0112] For semi-hard triplets: d(a, p) < d(a, n) < d(a, p) + margin, that is, the distance between a and n is very close, but there is a margin.

[0113] 104. Calculate the initial payment probability of the target object for payment at the target payment place based on the payment device graph feature and the target object graph feature.

[0114] For example, the correlation between the payment device graph feature and the target object graph feature can be specifically calculated; and the initial payment probability of the target object making payment at the target payment location can be determined based on the correlation. For example, the correlation between the payment device graph feature and the target object graph feature can be calculated using a cosine similarity function.

[0115] For example, when the initial payment probability includes a first payment probability and a second payment probability, the first payment probability of the first object making payment at the target payment place can be calculated based on the payment device graph feature and the first object graph feature; the second payment probability of the second object making payment at the target payment place can be calculated based on the payment device graph feature and the second object graph feature. For example, the first degree of correlation between the payment device graph feature and the first object graph feature can be calculated; based on the first degree of correlation, the first payment probability of the first object making payment at the target payment place can be determined; the second degree of correlation between the payment device graph feature and the second object graph feature can be calculated; and based on the second degree of correlation, the second payment probability of the second object making payment at the target payment place can be determined. For example, it can be specifically as follows:

[0116] First payment probability: P_a_sn = cosine(Fa, sn)

[0117] Second payment probability: P_b_sn = cosine(Fb, sn)

[0118] 105. Generate a fused payment probability based on the similarity between the facial image features and the target facial features and the initial payment probability.

[0119] There are many ways to generate a fused payment probability, for example, similarity and initial payment probability can be fused by linear interpolation fusion. For example, a first interpolation coefficient of the similarity between the facial image feature and the target facial feature and a second interpolation coefficient of the initial payment probability can be obtained; based on the first interpolation coefficient and the second interpolation coefficient, the similarity between the facial image feature and the target facial feature and the initial payment probability are fused to obtain a fused payment probability. In order to improve the accuracy of the fusion, the sum of the first interpolation coefficient and the second interpolation coefficient can be 1.

[0120] For example, when the initial payment probability includes a first payment probability and a second payment probability, a first fusion probability can be generated based on the first similarity and the first payment probability, and a second fusion probability can be generated based on the second similarity and the second payment probability. For example, the first fusion probability Pa(Sim_a, P_a_sn) can be obtained by fusion calculation based on Sim_a and P_a_sn, and the second fusion probability Pb(Sim_b, P_b_sn) can be obtained by fusion calculation based on Sim_b and P_b_sn. For example, it can be linear interpolation fusion, as follows:

[0121] Pa(Sim_a,P_a_sn)=(1-w)*Sim_a+w*P_a_sn

[0122] Pb(Sim_b,P_b_sn)=(1-w)*Sim_b+w*P_b_sn

[0123] Among them, the calculated fusion payment probability may refer to the probability that the payment object uses the facial payment device to pay at the target payment place. If the target payment place has one facial payment device, then the fusion payment probability may refer to the probability that the payment object pays at the target payment place; if the target payment place has multiple facial payment devices, then the fusion payment probabilities of multiple facial payment devices can be combined (for example, the average of the fusion payment probabilities of multiple facial payment devices can be obtained, etc.) to determine the probability that the payment object pays at the target payment place. The specific process will not be described in detail here.

[0124] 106. Identify the payment verification method of the payment object according to the integrated payment probability.

[0125] For example, when the fused payment probability includes a first fused probability and a second fused probability, the maximum probability of the first fused probability and the second fused probability can be obtained, and the maximum probability can be determined as a first measurement value; the minimum probability of the first fused probability and the second fused probability can be obtained, the difference between the maximum probability and the minimum probability is calculated, and the difference is determined as a second measurement value; if the first measurement value is greater than a first preset threshold and the second measurement value is greater than a second preset threshold, the payment verification method of the payment object is the first verification method; if the first measurement value is greater than the first preset threshold or the second measurement value is greater than the second preset threshold, the payment verification method of the payment object is the second verification method; if the first measurement value is less than or equal to the first preset threshold, and the second measurement value is less than or equal to the second preset threshold, the payment verification method of the payment object is the third verification method.

[0126] For example, according to the formulas max(Pa,Pb)>theshold1 (the first preset threshold), and max(Pa,Pb)–min(Pa,Pb)>threshold2 (the second preset threshold), three mobile phone number verification modes of 0, 4, and 11 digits are decided. Both formulas satisfy the verification of 0 digits, one satisfies the verification of 4 digits, and neither satisfies the verification of 11 digits.

[0127] Among them, 0 digits can be used to directly identify who is who from the database of billions of faces, without any verification by the user; 4 digits can be used to identify several people from the database of billions of faces, and the user needs to enter the last 4 digits of the bound mobile phone number for confirmation; 11 digits can mean that the user cannot be identified, and the user needs to enter the bound 11-digit mobile phone number for identification.

[0128] The first preset threshold and the second preset threshold can be set in many ways, for example, they can be flexibly set according to the needs of actual applications, or they can be pre-set and stored in the electronic device. In addition, the first preset threshold and the second preset threshold can be built into the electronic device, or they can be stored in a memory and sent to the electronic device, etc.

[0129] It should be noted that in order to improve the security of the identification of payment verification methods, the data storage in the above method is all stored in the blockchain. Among them, blockchain is a new application model of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the blockchain underlying platform, platform product service layer and application service layer.

[0130] The underlying blockchain platform can include processing modules such as user management, basic services, smart contracts, and operational detection. Among them, the user management module is responsible for the identity information management of all blockchain participants, including maintaining public and private key generation (account management), key management, and the maintenance of the correspondence between the user's real identity and the blockchain address (authority management), etc., and, under authorization, supervises and audits the transactions of certain real identities and provides risk control rule configuration (risk control audit); the basic service module is deployed on all blockchain node devices to verify the validity of business requests, and records valid requests to storage after consensus is reached. For a new business request, the basic service first performs interface adaptation analysis and authentication processing (interface adaptation), and then encrypts the business information through the consensus algorithm (consensus management). The smart contract module is responsible for the registration and issuance of contracts, as well as contract triggering and contract execution. Developers can define the contract logic in a programming language and publish it to the blockchain (contract registration). According to the logic of the contract terms, the key or other events are called to trigger the execution and complete the contract logic. It also provides the function of contract upgrade and cancellation. The operation detection module is mainly responsible for the deployment, configuration modification, contract setting, cloud adaptation and real-time status visualization output of the product during the product release process, such as alarm, network status detection, node equipment health status detection, etc.

[0131] The platform product service layer provides the basic capabilities and implementation framework of typical applications. Developers can superimpose business features based on these basic capabilities to complete the blockchain implementation of business logic. The application service layer provides application services based on blockchain solutions for business participants to use.

[0132] As can be seen from the above, this embodiment can obtain facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of the payment object and a device identifier of the facial payment device. Then, the target facial features matching the facial image features are searched from the facial database, and the target object corresponding to the target facial features is determined. The payment device graph features corresponding to the device identifier and the target object graph features corresponding to the target object are searched from the graph feature database. Then, based on the payment device graph features and the target object graph features, the initial payment probability of the target object making payment at the target payment location is calculated. Then, based on the similarity between the facial image features and the target facial features, and the initial payment probability, a fused payment probability is generated. Then, the payment verification method of the payment object is identified according to the fused payment probability. Since the scheme can fully utilize the payment information between objects and devices in the historical payment records to construct a heterogeneous network graph of objects and devices, and fully mine the structural information of objects and devices in the graph through the heterogeneous network graph (for example, the unsupervised training framework of metapath2vec and skip2gram), it can achieve a similar collaborative filtering effect (for example, many objects of device A have also swiped their faces on device B, then the remaining people of device A who have not swiped their faces on device B also have a certain probability of swiping their faces on device B). Finally, the graph model constructs the probability of an object using a device to pay or the transfer probability between an object and a device through the measurement of object features and device features. Finally, the facial recognition technology is combined with the recognition device of the payment verification method for comprehensive decision-making, thereby greatly improving the accuracy of face recognition for face-swiping payment, solving some highly similar facial recognition problems, and greatly increasing the proportion of zero-bit face-swiping payment, effectively improving the convenience of payment by the payment object, and then feeding back to face-swiping payment, forming an effective closed loop, improving user experience, and making the face-swiping experience "more and more enjoyable".

[0133] According to the method described in the previous embodiment, an example is given below to further illustrate it in detail.

[0134] In this embodiment, the recognition device of the payment verification method is specifically integrated in an electronic device, the payment object is specifically the paying user, the facial payment device is specifically a facial payment device, the facial image features are specifically facial image features, the target facial features specifically include the first facial features and the second facial features, the target objects specifically include the first user and the second user, and the target payment place is specifically the target store (i.e., the store where the facial payment device that currently initiates the payment request is located) for explanation as an example.

[0135] (I) First, a face database can be established, which can be done as follows:

[0136] According to the historical payment information of all payment devices, facial images of historical payment users are obtained, and features of the facial images of historical payment users are extracted to obtain facial features of historical payment users. The facial features of historical payment users are used as candidate facial features and stored in the facial database.

[0137] (II) Next, a graph feature database may be established, which may be specifically as follows:

[0138] In order to improve the efficiency of recognition of payment verification methods, a graph feature database may be established first. The graph features in the graph feature database may be trained from multiple training samples. Specifically, it may be provided to the recognition device of the payment verification method after training by other devices, or it may be trained by the recognition device of the payment verification method itself. For example, the electronic device may specifically collect multiple training sample pairs, wherein the training sample pairs include at least one positive sample and at least one negative sample; initialize the features of the sample users and the sample devices respectively to obtain initial user features and initial device features; train the initial user features and initial device features based on the positive samples and the negative samples to obtain user graph features and device graph features; store the user graph features and device graph features in the graph feature database.

[0139] For example, the electronic device can specifically determine the link relationship between the sample user and the sample device based on the historical payment record, and construct a heterogeneous network graph of the sample user and the sample device based on the link relationship; in the heterogeneous network graph, the sample user and the sample device are used as nodes for path sampling, and the paths with link relationships between the collected nodes are used as positive samples, and at least one path without link relationship between the collected nodes is used as a negative sample. For example, the graph model can be obtained through offline training. First, the sample users in the historical face scanning records and the sample devices of the sample stores can be regarded as two types of nodes. Then, the sample devices of the sample stores where the sample users have made payments can be added with edges to form a bipartite graph of the sample users and the sample stores. Since the types of nodes are different, this is a heterogeneous graph. For example, Figure 2a After the user and the store establish a relationship, the next step is the initial features of the user and the store. Since it is essentially collaborative filtering, the inherent attributes of the store and the person are not very helpful for predicting the person's store visit, and it also involves a lot of feature engineering work, so here we can randomly initialize the features, and then train them through a reasonable loss function and gradient back propagation algorithm to obtain the final effective features of the person and the store.

[0140] Among them, collaborative filtering is to use the preferences of a group of people with similar interests and common experiences to recommend information that users are interested in. Individuals respond to information to a certain extent (such as ratings) through a cooperative mechanism and record it to achieve the purpose of filtering and help others filter information. Responses are not necessarily limited to those that are of particular interest. Records of information that is particularly uninteresting are also very important. In this solution, it can refer to the use of the store visit probability of a group with common characteristics to determine the store visit probability of the paying user, and then determine the payment verification method of the paying user.

[0141] For example, in a heterogeneous network graph, the metapath2vec method can be used to randomly walk in a certain order to obtain training sample pairs of sample users and sample devices. In order to prevent the problem of homogenization of node features, only second-order neighbors can be obtained, because if the order is too high, the problem of homogenization of node features will occur. For example, the electronic device can specifically treat sample users and sample devices as different types of nodes, obtain a predefined meta-path, and the meta-path includes the link relationship between different types of nodes; calculate the transition probability of each step based on the link relationship between different types of nodes in the meta-path, and determine the random walk sampling strategy based on the transition probability of each step; in the heterogeneous network graph, path sampling is performed based on the random walk sampling strategy.

[0142] For example, a meta-path scheme based on random walks can be used to collect sequences (i.e., paths). This random walk method can simultaneously capture the semantic and structural relationships between different types of vertices, and promote the conversion of heterogeneous network structures to the Skip-Gram model of metapath2vec. A meta-path scheme can be defined as follows:

[0143]

[0144] Among them, E represents an edge, V represents a node, R represents a random walk, N represents a neighbor node, and R i Represents different types of nodes V i With V i+1 The transition probability of the i-th step is defined as:

[0145]

[0146] in, Represents a vertex V t+1 The domain vertex set of type. Then, the three fractions above represent:

[0147] 1. There is an edge between the two points, and the next point belongs to the next type of node on the defined metapath.

[0148] 2. There is an edge between two points, but the next point does not belong to the next type of node on the defined metapath.

[0149] 3. There is no edge between two points.

[0150] Then, using recursive guidance for random walker, for example, we can do the following:

[0151]

[0152] After a large number of training sample pairs are obtained by sampling based on the above formulas, the training sample pairs can be used for training. For example, the electronic device can specifically calculate the correlation between the initial user features and the initial device features in the positive sample to obtain the prediction result of the positive sample; calculate the correlation between the initial user features and the initial device features in the negative sample to obtain the prediction result of the negative sample; use the loss function to adjust the initial user features and the initial device features according to the prediction results of the positive sample and the actual results of the positive sample, and according to the prediction results of the negative sample and the actual results of the negative sample, until the loss function converges to obtain the user graph features and the device graph features.

[0153] For example, skip-gram can be used for training to obtain the final person-store features, namely user graph features and device graph features. The specific formula for the loss function trained using skip-gram can be as follows:

[0154]

[0155] This formula is actually a downsampled softmax, where the training sample pairs obtained by metapath can be one pair of positive samples corresponding to N pairs of negative samples, and then trained using softmax+crossentropy-loss. For example, in order to improve the efficiency and accuracy of payment verification method recognition, one positive sample can correspond to 16 negative samples.

[0156] The biggest feature of using graph methods is that we can mine Figure 2b The dotted line relationship on the right means that the user has never been to a certain store, but based on the topological relationship between people and stores, the possibility of visiting the store by face scanning in the future can be captured. This ability cannot be achieved by general rule-based methods.

[0157] (III) By using the established face database and image feature database, the payment verification method of the payment user can be identified. For details, please refer to Figure 2c , Figure 2d and Figure 2e .

[0158] like Figure 2c As shown, a payment verification method identification method, the specific process can be as follows:

[0159] 201. An electronic device receives a payment request initiated at a target store, where the payment request carries a facial image of the paying user and a device identifier of a facial payment device.

[0160] The target store may refer to a store where the paying user needs to pay for the purchase. The paying user may refer to a user who needs to be identified by a payment verification method when paying by face recognition, such as a user who purchases goods.

[0161] For example, when a payment request initiated by a target store is received, a facial image of the paying user and a device identifier of a facial payment device that initiated the payment request may be obtained according to the payment request.

[0162] For example, when a paying user (such as user X) purchases goods at a target store (such as store Y) and uses a facial payment device (such as device S) to pay, the electronic device can perform face detection on user X to obtain a facial image of user X and a device identifier sn of device S, so as to perform face detection and registration on user X.

[0163] For example, face detection can use MTCNN, and registration can use affine transformation. The purpose is to standardize the faces of any posture captured by the camera. Face scanning is mainly cooperative face recognition. Relatively speaking, the collected face posture is not very large, so the MTCNN detection method can be used.

[0164] 202. The electronic device extracts features from the facial image to obtain facial image features.

[0165] The face image acquired by the electronic device can be an RGB image. The RGB image feature extraction can be performed using a deep learning model. The deep learning model can be trained offline using a large number of RGB images with identity annotations. The training algorithm can be a margin based softmax method. For example, by inputting the standardized image into the deep learning model, the feature F of the corresponding dimension can be obtained. Since the scale of industrial-grade face recognition training is very large and each ID corresponds to a category, the final classification layer is very resource-intensive. Therefore, the classification layer adopts a model parallel method, and the classification layer is mixedly trained in a data parallel manner.

[0166] 203. The electronic device searches a face database for a first face feature corresponding to the first user and a second face feature corresponding to the second user that match the face image feature.

[0167] For example, the electronic device can specifically calculate the similarity between the facial image feature and the candidate facial feature in the facial database; sort the similarity between the facial image feature and the candidate facial feature according to the calculation result; obtain the first facial feature corresponding to the first similarity and the second facial feature corresponding to the second similarity in the sorting based on the sorting result; determine the first facial feature and the second facial feature as target facial features, and determine the first user corresponding to the first facial feature and the second user corresponding to the second facial feature. For example, the first facial feature corresponding to the first similarity with the highest similarity (i.e., the first) and the second facial feature corresponding to the second similarity with the second similarity can be obtained according to the sorting result, that is, the first similarity can refer to the similarity ranked first, and the second similarity can refer to the similarity ranked second.

[0168] For example, based on the open source faiss library, you can input the extracted facial feature P, retrieve the top 2 most similar facial features, and obtain the corresponding similarity score and user ID. For example, the top 2 are user A and user B, user A's facial feature is Pa, and user B's facial feature is Pb. The similarity score can be measured by cosine. For example, the specific calculation can be as follows:

[0169] Sim_a=cosine(P,Pa)

[0170] Sim_b=cosine(P,Pb)

[0171] 204. The electronic device searches a graph feature database for a payment device graph feature corresponding to the device identifier, a first user graph feature corresponding to the first user, and a second user graph feature corresponding to the second user.

[0172] For example, the electronic device can specifically search for the payment device graph feature Fsn corresponding to the device identifier, the first user graph feature Fa corresponding to the first user, and the second user graph feature Fb corresponding to the second user from the established graph feature database, or obtain the device graph feature Fsn corresponding to the device number sn from the cloud graph feature library in the cloud, for example. Figure 2d shown.

[0173] 205. The electronic device calculates a first payment probability of the first user making payment at the target payment location based on the payment device graph feature and the first user graph feature.

[0174] For example, user A, user B and device sn are input into the graph feature database to search for the corresponding graph features Fa, Fb and Fsn. The probability of user A and user B arriving at the store can also be calculated using the cosine function. For example, the electronic device can specifically calculate the first correlation between the payment device graph feature and the first user graph feature; based on the first correlation, the first payment probability of the first user making payment at the target payment location is determined. For example, it can be specifically as follows:

[0175] First payment probability: P_a_sn = cosine(Fa, sn)

[0176] 206. The electronic device calculates a second payment probability of the second user making payment at the target payment location based on the payment device graph feature and the second user graph feature.

[0177] For example, the electronic device may specifically calculate a second correlation between the payment device graph feature and the second user graph feature; and determine a second payment probability of the second user making payment at the target payment location based on the second correlation. For example, the specific steps may be as follows:

[0178] Second payment probability: P_b_sn = cosine(Fb, sn)

[0179] 207. The electronic device generates a first fusion probability based on the first similarity between the facial image feature and the first facial feature, and the first payment probability.

[0180] For example, the electronic device may specifically obtain a first interpolation coefficient of a first similarity between the facial image feature and the first facial feature, and a second interpolation coefficient of the first payment probability; based on the first interpolation coefficient and the second interpolation coefficient, the similarity between the facial image feature and the first facial feature, and the first payment probability are fused to obtain a first fusion probability. In order to improve the accuracy of the fusion, the sum of the first interpolation coefficient and the second interpolation coefficient may be 1.

[0181] For example, the first fusion probability Pa(Sim_a, P_a_sn) can be obtained by fusion calculation based on Sim_a and P_a_sn, which can be specifically as follows:

[0182] Pa(Sim_a,P_a_sn)=(1-w)*Sim_a+w*P_a_sn

[0183] 208. The electronic device generates a second fusion probability based on a second similarity between the facial image feature and the second facial feature, and the second payment probability.

[0184] For example, the electronic device may specifically obtain a first interpolation coefficient of a second similarity between the facial image feature and the second facial feature, and a second interpolation coefficient of the second payment probability; based on the first interpolation coefficient and the second interpolation coefficient, the similarity between the facial image feature and the target facial feature, and the second payment probability are fused to obtain a second fusion probability. In order to improve the accuracy of the fusion, the sum of the first interpolation coefficient and the second interpolation coefficient may be 1.

[0185] For example, the second fusion probability Pb(Sim_b, P_b_sn) can be obtained by fusion calculation based on Sim_b and P_b_sn, which can be specifically as follows:

[0186] Pb(Sim_b,P_b_sn)=(1-w)*Sim_b+w*P_b_sn

[0187] 209. The electronic device identifies the payment verification method of the payment user according to the first fusion probability and the second fusion probability.

[0188] For example, the electronic device can specifically obtain the maximum probability of the first fusion probability and the second fusion probability, and determine the maximum probability as the first measurement value; obtain the minimum probability of the first fusion probability and the second fusion probability, calculate the difference between the maximum probability and the minimum probability, and determine the difference as the second measurement value; if the first measurement value is greater than the first preset threshold and the second measurement value is greater than the second preset threshold, the payment verification method of the payment user is the first verification method; if the first measurement value is greater than the first preset threshold or the second measurement value is greater than the second preset threshold, the payment verification method of the payment user is the second verification method; if the first measurement value is less than or equal to the first preset threshold, and the second measurement value is less than or equal to the second preset threshold, the payment verification method of the payment user is the third verification method.

[0189] For example, according to the formula max(Pa,Pb)>theshold1, and max(Pa,Pb)–min(Pa,Pb)>threshold2, three mobile phone number verification modes of 0, 4, and 11 digits are determined. Both formulas satisfy the verification of 0 digits, one satisfies the verification of 4 digits, and neither satisfies the verification of 11 digits. The specific process can be as follows Figure 2c shown.

[0190] Among them, 0 digits can be used to directly identify who is who from the database of billions of faces, without any verification by the user; 4 digits can be used to identify several people from the database of billions of faces, and the user needs to enter the last 4 digits of the bound mobile phone number for confirmation; 11 digits can mean that the user cannot be identified, and the user needs to enter the bound 11-digit mobile phone number for identification.

[0191] This solution abstracts the historical information of face scanning into the connection between people and stores, constructs an unstructured graph relationship between people and stores, and then uses graph learning technology to vectorize the expression of people and stores through loss function training, so that the probability of people scanning their faces in stores can be measured, combined with face recognition for comprehensive decision-making, thereby making full use of the historical information of face scanning, which can greatly improve the face recognition accuracy of face scanning payment, solve some high-similarity face recognition problems, and then feed back to face scanning payment to form an effective closed loop, thereby greatly increasing the proportion of 0-digit face scanning payment, making the face scanning experience "more and more enjoyable".

[0192] As can be seen from the above, this embodiment can obtain facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of the payment object and a device identifier of the facial payment device. Then, the target facial features matching the facial image features are searched from the facial database, and the target object corresponding to the target facial features is determined. The payment device graph features corresponding to the device identifier and the target object graph features corresponding to the target object are searched from the graph feature database. Then, based on the payment device graph features and the target object graph features, the initial payment probability of the target object making payment at the target payment location is calculated. Then, based on the similarity between the facial image features and the target facial features, and the initial payment probability, a fused payment probability is generated. Then, the payment verification method of the payment object is identified according to the fused payment probability. Since this solution can build a graph model of people and stores by making full use of the face-scanning information of users and stores, and fully mine the structural information of people and stores in the graph through graph models (such as the unsupervised training framework of metapath2vec and skip2gram), it can achieve a similar collaborative filtering effect (for example, many people in store A have also scanned their faces in store B, so the remaining people in store A who have not scanned their faces in store B also have a certain probability of going to store B to scan their faces). Finally, the graph model constructs the person-store probability (regular customers) or the person-store transfer probability (new customers) through the measurement of person-store embedding. Finally, the face recognition technology is combined with the face-scanning payment system for comprehensive decision-making, which ultimately greatly enhances the recognition ability of face-scanning payment.

[0193] In order to better implement the above method, accordingly, the embodiment of the present application also provides a payment verification method identification device, which can be specifically integrated in an electronic device, and the electronic device can be a server, a terminal or other equipment.

[0194] For example, Figure 3As shown, the recognition device of the payment verification method may include an acquisition unit 301, a matching unit 302, a search unit 303, a calculation unit 304, a fusion unit 305 and a recognition unit 306, as follows:

[0195] An acquisition unit 301 is used to acquire facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of a payment object and a device identifier of a facial payment device;

[0196] A matching unit 302 is used to search a target facial feature that matches the facial image feature from a facial database, and determine a target object corresponding to the target facial feature;

[0197] A search unit 303 is used to search for a payment device graph feature corresponding to the device identifier and a target object graph feature corresponding to the target object from a graph feature database;

[0198] A calculation unit 304, configured to calculate an initial payment probability of the target object making payment at the target payment location based on the payment device graph feature and the target object graph feature;

[0199] A fusion unit 305, configured to generate a fused payment probability based on the similarity between the facial image feature and the target facial feature and the initial payment probability;

[0200] The identification unit 306 is used to identify the payment verification method of the payment object according to the integrated payment probability.

[0201] Optionally, in some embodiments, the matching unit 302 can be specifically used to calculate the similarity between the facial image features and the candidate facial features in the facial database; obtain the target facial features whose similarity meets preset conditions from the facial database according to the calculation results, and determine the target object corresponding to the target facial features.

[0202] Optionally, in some embodiments, the payment verification method recognition device may further include a training unit, and the training unit includes a collection subunit, an initialization subunit, a training subunit and a storage subunit, as follows:

[0203] The collection subunit is used to collect a plurality of training sample pairs, wherein the training sample pairs include at least one positive sample and at least one negative sample;

[0204] The initialization subunit is used to respectively initialize the features of the sample objects and sample devices in the training sample pairs to obtain initial object features and initial device features;

[0205] The training subunit is used to train the initial object features and the initial device features based on the positive samples and the negative samples to obtain object graph features and device graph features;

[0206] The storage subunit is used to store the object graph features and the device graph features in a graph feature database.

[0207] Optionally, in some embodiments, the collection subunit can be specifically used to determine the link relationship between the sample object and the sample device based on the historical payment records, and construct a heterogeneous network diagram of the sample object and the sample device based on the link relationship; in the heterogeneous network diagram, the sample objects and sample devices are used as nodes for path sampling, and the paths with link relationships between the collected nodes are used as positive samples, and the paths with at least one no link relationship between the collected nodes are used as negative samples.

[0208] Optionally, in some embodiments, the acquisition subunit can be specifically used to treat sample objects and sample devices as different types of nodes to obtain a predefined meta-path, where the meta-path includes link relationships between different types of nodes; calculate the transition probability of each step based on the link relationship between different types of nodes in the meta-path, and determine a random walk sampling strategy based on the transition probability of each step; and perform path sampling in the heterogeneous network graph based on the random walk sampling strategy.

[0209] Optionally, in some embodiments, the training subunit can be specifically used to calculate the correlation between the initial object features and the initial device features in the positive sample to obtain the prediction result of the positive sample; calculate the correlation between the initial object features and the initial device features in the negative sample to obtain the prediction result of the negative sample; use the loss function to adjust the initial object features and the initial device features according to the prediction result of the positive sample and the actual result of the positive sample, and according to the prediction result of the negative sample and the actual result of the negative sample, until the loss function converges to obtain the object graph features and the device graph features.

[0210] Optionally, in some embodiments, the target facial feature includes a first facial feature and a second facial feature, the target object includes a first object and a second object, and the matching unit 302 may be specifically configured to sort the similarities between the facial image feature and the candidate facial feature according to the calculation result; based on the sorting result, obtain the first facial feature corresponding to the first similarity and the second facial feature corresponding to the second similarity in the sorting; determine the first facial feature and the second facial feature as the target facial feature, and determine the first object corresponding to the first facial feature and the second object corresponding to the second facial feature;

[0211] The searching unit 303 may be specifically configured to search a graph feature database for a payment device graph feature corresponding to the device identifier, a first object graph feature corresponding to the first object, and a second object graph feature corresponding to the second object.

[0212] Optionally, in some embodiments, the initial payment probability includes a first payment probability and a second payment probability, and the calculation unit 304 may be specifically configured to calculate a first payment probability of the first object making payment at the target payment location based on the payment device graph feature and the first object graph feature; and calculate a second payment probability of the second object making payment at the target payment location based on the payment device graph feature and the second object graph feature.

[0213] The fusion unit 305 may be specifically configured to generate a first fusion probability based on the first similarity and the first payment probability, and to generate a second fusion probability based on the second similarity and the second payment probability.

[0214] Optionally, in some embodiments, the identification unit 306 can be specifically used to obtain the maximum probability of the first fusion probability and the second fusion probability, and determine the maximum probability as the first measurement value; obtain the minimum probability of the first fusion probability and the second fusion probability, calculate the difference between the maximum probability and the minimum probability, and determine the difference as the second measurement value; if the first measurement value is greater than a first preset threshold and the second measurement value is greater than a second preset threshold, the payment verification method of the payment object is the first verification method; if the first measurement value is greater than the first preset threshold or the second measurement value is greater than the second preset threshold, the payment verification method of the payment object is the second verification method; if the first measurement value is less than or equal to the first preset threshold, and the second measurement value is less than or equal to the second preset threshold, the payment verification method of the payment object is the third verification method.

[0215] Optionally, in some embodiments, the calculation unit 304 can be specifically used to calculate the correlation between the payment device graph feature and the target object graph feature; and determine the initial payment probability of the target object making payment at the target payment location based on the correlation.

[0216] Optionally, in some embodiments, the fusion unit 305 can be specifically used to obtain a first interpolation coefficient of the similarity between the facial image features and the target facial features, and a second interpolation coefficient of the initial payment probability; based on the first interpolation coefficient and the second interpolation coefficient, the similarity between the facial image features and the target facial features, and the initial payment probability are fused to obtain a fused payment probability.

[0217] In specific implementation, the above units can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units can refer to the previous method embodiments, which will not be repeated here.

[0218] As can be seen from the above, in this embodiment, the facial payment information initiated at the target payment place can be obtained by the acquisition unit 301, and the facial payment information includes the facial image features of the payment object and the device identification of the facial payment device. Then, the matching unit 302 searches the facial database for the target facial features that match the facial image features, and determines the target object corresponding to the target facial features. The search unit 303 searches the graph feature database for the payment device graph features corresponding to the device identification and the target object graph features corresponding to the target object. Then, the calculation unit 304 calculates the initial payment probability of the target object making payment at the target payment place based on the payment device graph features and the target object graph features. Then, the fusion unit 305 generates a fused payment probability based on the similarity between the facial image features and the target facial features, and the initial payment probability. Then, the identification unit 306 identifies the payment verification method of the payment object according to the fused payment probability. Since the scheme can fully utilize the payment information between objects and devices in the historical payment records, build a heterogeneous network graph of objects and devices, and fully mine the structural information of objects and devices in the graph through the heterogeneous network graph, it can achieve a similar collaborative filtering effect. Finally, the graph model constructs the probability of an object using a device to pay or the transfer probability between an object and a device through the measurement of object features and device features. Finally, the facial recognition technology is combined with the recognition device of the payment verification method for comprehensive decision-making, effectively improving the accuracy of the payment verification method recognition, thereby improving the accuracy of face-swiping payment, solving some highly similar facial recognition problems, and effectively improving the convenience of payment by the payment object, thereby feeding back to face-swiping payment, forming an effective closed loop, and improving user experience.

[0219] In addition, the present application also provides an electronic device, such as Figure 4 As shown, it shows a schematic diagram of the structure of the electronic device involved in the embodiment of the present application, specifically:

[0220] The electronic device may include components such as a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a power supply 403, and an input unit 404. Those skilled in the art will appreciate that Figure 4 The electronic device structure shown in the figure does not constitute a limitation on the electronic device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0221] The processor 401 is the control center of the electronic device, and uses various interfaces and lines to connect various parts of the entire electronic device. By running or executing software programs and / or modules stored in the memory 402, and calling data stored in the memory 402, the processor 401 performs various functions of the electronic device and processes data, thereby performing overall detection of the electronic device. Optionally, the processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 401.

[0222] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 402 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0223] The electronic device also includes a power supply 403 for supplying power to each component. Preferably, the power supply 403 can be logically connected to the processor 401 through a power management system, so as to manage charging, discharging, power consumption and other functions through the power management system. The power supply 403 can also include one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, power status indicators and other arbitrary components.

[0224] The electronic device may further include an input unit 404, which may be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.

[0225] Although not shown, the electronic device may further include a display unit, etc., which will not be described in detail herein. Specifically in this embodiment, the processor 401 in the electronic device will load the executable files corresponding to the processes of one or more application programs into the memory 402 according to the following instructions, and the processor 401 will run the application programs stored in the memory 402, thereby realizing various functions, as follows:

[0226] Acquire facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of a payment object and a device identifier of a facial payment device; then, search a facial database for target facial features that match the facial image features, and determine a target object corresponding to the target facial features; search a graphic feature database for payment device graphic features corresponding to the device identifier and a target object graphic feature corresponding to the target object; then, calculate an initial payment probability of the target object making payment at the target payment location based on the payment device graphic features and the target object graphic features; then, generate a fused payment probability based on the similarity between the facial image features and the target facial features, and the initial payment probability; and then, identify a payment verification method for the payment object based on the fused payment probability.

[0227] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0228] As can be seen from the above, this embodiment can obtain the facial payment information initiated at the target payment place, the facial payment information includes the facial image features of the payment object and the device identification of the facial payment device, then, search the target facial features matching the facial image features from the facial database, and determine the target object corresponding to the target facial features, search the payment device graph features corresponding to the device identification and the target object graph features corresponding to the target object from the graph feature database, then, calculate the initial payment probability of the target object making payment at the target payment place based on the payment device graph features and the target object graph features, then generate the fused payment probability based on the similarity between the facial image features and the target facial features, and the initial payment probability, and then, identify the payment verification method of the payment object according to the fused payment probability. Since the scheme can construct a heterogeneous network graph of objects and devices by making full use of the payment information between objects and devices in the historical payment records, and fully mine the structural information of objects and devices in the graph through the heterogeneous network graph, it can achieve a similar collaborative filtering effect, and finally the graph model constructs the object payment probability using the device or the transfer probability between the object and the device through the measurement of the object features and the device features. Finally, facial recognition technology is combined with the recognition device of the payment verification method to make a comprehensive decision, which effectively improves the accuracy of payment verification method recognition, and then improves the accuracy of face payment, solves some highly similar facial recognition problems, and effectively improves the convenience of payment for the payment object, which in turn feeds back to face payment, forming an effective closed loop and improving user experience.

[0229] A person of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0230] To this end, the embodiment of the present application further provides a storage medium in which a plurality of instructions are stored, and the instructions can be loaded by a processor to execute the steps in any payment verification method identification method provided in the embodiment of the present application. For example, the instructions can execute the following steps:

[0231] Acquire facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of a payment object and a device identifier of a facial payment device; then, search a facial database for target facial features that match the facial image features, and determine a target object corresponding to the target facial features; search a graphic feature database for payment device graphic features corresponding to the device identifier and a target object graphic feature corresponding to the target object; then, calculate an initial payment probability of the target object making payment at the target payment location based on the payment device graphic features and the target object graphic features; then, generate a fused payment probability based on the similarity between the facial image features and the target facial features, and the initial payment probability; and then, identify a payment verification method for the payment object based on the fused payment probability.

[0232] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0233] The storage medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0234] Since the instructions stored in the storage medium can execute the steps in the method for identifying any payment verification method provided in the embodiments of the present application, the beneficial effects that can be achieved by the method for identifying any payment verification method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0235] The above is a detailed introduction to the identification method, device, electronic device and storage medium of a payment verification method provided in an embodiment of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for identifying a payment verification method, characterized in that: include: Acquire facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of a payment object and a device identifier of a facial payment device; Searching for target facial features that match the facial image features from a facial database, and determining a target object corresponding to the target facial features; Searching a graph feature database for a payment device graph feature corresponding to the device identifier and a target object graph feature corresponding to the target object; Calculating an initial payment probability of the target object making payment at the target payment location based on the payment device graph feature and the target object graph feature; generating a fused payment probability based on the similarity between the facial image features and the target facial features and the initial payment probability; A payment verification method of the payment object is identified according to the fused payment probability.

2. The method according to claim 1, characterized in that: The step of searching a face database for a target face feature that matches the face image feature and determining a target object corresponding to the target face feature comprises: Calculating the similarity between the facial image features and candidate facial features in a facial database; According to the calculation result, target facial features whose similarity meets a preset condition are obtained from the facial database, and a target object corresponding to the target facial features is determined.

3. The method according to claim 1, characterized in that Before searching the graph feature database for the payment device graph feature corresponding to the device identifier and the target object graph feature corresponding to the target object, the method further includes: Collecting a plurality of training sample pairs, wherein the training sample pairs include at least one positive sample and at least one negative sample; Initializing features of sample objects and sample devices in the training sample pairs respectively to obtain initial object features and initial device features; Based on the positive samples and the negative samples, the initial object features and the initial device features are trained to obtain object graph features and device graph features; The object graph features and the device graph features are stored in a graph feature database.

4. The method according to claim 3, characterized in that The collecting of a plurality of training sample pairs, wherein the training sample pairs include at least one positive sample and at least one negative sample, comprises: Determine the link relationship between the sample object and the sample device according to the historical payment record, and construct a heterogeneous network diagram of the sample object and the sample device based on the link relationship; In the heterogeneous network graph, sample objects and sample devices are used as nodes for path sampling, paths with link relationships between the collected nodes are used as positive samples, and paths with at least one no-link relationship between the collected nodes are used as negative samples.

5. The method according to claim 4, characterized in that In the heterogeneous network graph, taking sample objects and sample devices as nodes for path sampling includes: Taking the sample objects and the sample devices as different types of nodes, obtaining a predefined meta-path, wherein the meta-path includes link relationships between the different types of nodes; Calculating the transition probability of each step based on the link relationship between different types of nodes in the meta-path, and determining the random walk sampling strategy according to the transition probability of each step; In the heterogeneous network graph, path sampling is performed based on the random walk sampling strategy.

6. The method according to claim 4, characterized in that The training of the initial object features and the initial device features based on the positive samples and the negative samples to obtain object graph features and device graph features includes: Calculating the correlation between the initial object features and the initial device features in the positive sample to obtain a prediction result of the positive sample; Calculating the correlation between the initial object features and the initial device features in the negative sample to obtain a prediction result of the negative sample; Using the loss function, the initial object features and the initial device features are adjusted according to the predicted results of the positive samples and the actual results of the positive samples, and according to the predicted results of the negative samples and the actual results of the negative samples, until the loss function converges to obtain object graph features and device graph features.

7. The method according to claim 2, characterized in that The target facial features include a first facial feature and a second facial feature, the target object includes a first object and a second object, and the step of acquiring the target facial features whose similarity satisfies a preset condition from the facial database according to the calculation result, and determining the target object corresponding to the target facial features, comprises: According to the calculation result, the similarities between the facial image features and the candidate facial features are sorted; Based on the sorting result, obtaining a first facial feature corresponding to a first similarity and a second facial feature corresponding to a second similarity in the sorting; Determine the first facial feature and the second facial feature as target facial features, and determine a first object corresponding to the first facial feature and a second object corresponding to the second facial feature; The searching, from a graph feature database, for the payment device graph feature corresponding to the device identifier and the target object graph feature corresponding to the target object comprises: searching, from a graph feature database, for the payment device graph feature corresponding to the device identifier, the first object graph feature corresponding to the first object, and the second object graph feature corresponding to the second object.

8. The method according to claim 7, characterized in that The initial payment probability includes a first payment probability and a second payment probability, and the calculating of the initial payment probability of the target object making payment at the target payment place based on the payment device graph feature and the target object graph feature includes: Calculating a first payment probability of a first object making payment at the target payment location based on the payment device graph feature and the first object graph feature; Calculating a second payment probability of a second object making payment at the target payment location based on the payment device graph feature and the second object graph feature; The generating of the fused payment probability based on the similarity between the facial image features and the target facial features, and the initial payment probability, includes: generating a first fused probability based on the first similarity and the first payment probability, and generating a second fused probability based on the second similarity and the second payment probability.

9. The method according to claim 8, characterized in that The step of identifying the payment verification method of the payment object according to the fused payment probability includes: Obtaining a maximum probability value of the first fusion probability and the second fusion probability, and determining the maximum probability value as a first measurement value; Obtaining a minimum probability value of the first fusion probability and the second fusion probability, calculating a difference between the maximum probability value and the minimum probability value, and determining the difference as a second measurement value; If the first measurement value is greater than the first preset threshold and the second measurement value is greater than the second preset threshold, the payment verification method of the payment object is the first verification method; If the first measurement value is greater than the first preset threshold or the second measurement value is greater than the second preset threshold, the payment verification method of the payment object is the second verification method; If the first measurement value is less than or equal to the first preset threshold, and the second measurement value is less than or equal to the second preset threshold, the payment verification method of the payment object is the third verification method.

10. The method according to any one of claims 1 to 7, characterized in that: The calculating, based on the payment device graph feature and the target object graph feature, the initial payment probability of the target object making payment at the target payment location comprises: Calculating the degree of association between the payment device graph feature and the target object graph feature; An initial payment probability of the target object making payment at the target payment location is determined based on the association degree.

11. The method according to any one of claims 1 to 7, characterized in that: The generating of the fused payment probability based on the similarity between the facial image feature and the target facial feature and the initial payment probability includes: Obtaining a first interpolation coefficient of the similarity between the facial image feature and the target facial feature, and a second interpolation coefficient of the initial payment probability; Based on the first interpolation coefficient and the second interpolation coefficient, the similarity between the facial image feature and the target facial feature and the initial payment probability are fused to obtain a fused payment probability.

12. A payment verification method identification device, characterized in that: include: An acquisition unit, used to acquire facial payment information initiated at a target payment location, wherein the facial payment information includes facial image features of a payment object and a device identifier of a facial payment device; A matching unit, configured to search a face database for a target face feature that matches the face image feature, and determine a target object corresponding to the target face feature; A search unit, configured to search a graph feature database for a payment device graph feature corresponding to the device identifier and a target object graph feature corresponding to the target object; A calculation unit, configured to calculate an initial payment probability of a target object making payment at the target payment location based on the payment device graph feature and the target object graph feature; a fusion unit, configured to generate a fused payment probability based on the similarity between the facial image feature and the target facial feature and the initial payment probability; An identification unit is used to identify the payment verification method of the payment object according to the fused payment probability.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the steps in the method for identifying a payment verification method according to any one of claims 1 to 11.

14. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 11 are implemented.

Images