Data fusion processing method and system based on edge computing

Abstract

The invention discloses a data fusion processing method and system based on edge computing, and relates to the field of data processing.The method comprises the steps that a plurality of data sources are accessed to an edge node; performing preprocessing operation on the data of the plurality of data sources to obtain preprocessed data; performing feature extraction operation on the preprocessed data to obtain feature information related to the public accumulation fund business; performing local intelligent judgment operation on the edge node by adopting a local fusion model based on the feature information to obtain a fusion processing result; and sending the fusion processing result and the associated label thereof to a central server for business collaborative decision making. According to the method, the framework limitation of traditional centralized approval is broken through, a three-in-one innovative path of data fusion, intelligent judgment and system collaboration is realized, and the method is particularly suitable for large-scale and high-real-time business processing scenes in a public accumulation fund intelligent service system.

Description

Technical Field

[0001] This specification relates to the field of data processing. More specifically, this application relates to a data fusion processing method and system based on edge computing. Background Art

[0002] With the continuous digitization of the housing provident fund business, more and more users submit withdrawal applications, loan applications, or home purchase qualification certifications through online channels. Such business processes usually involve multiple links such as user identity verification, provident fund deposit record verification, material consistency verification, risk assessment, etc., and have the characteristics of diverse data sources, complex material types, and highly dependence on manual experience for judgment rules.

[0003] The traditional way of handling provident fund business mainly relies on the central server to centrally receive and process all application data, and there are the following obvious problems: First, the processing speed is slow. All image, form, and behavior data need to be uploaded to the central server, which is likely to cause network congestion and review delays. Second, the computing pressure is high. The server has to process a large number of tasks such as image recognition and information comparison, with high resource consumption. Third, risk identification lags. It is impossible to quickly identify risk users at the edge, which is likely to cause a burden on the backend system. Fourth, the labor cost is high. Processes such as identity verification and material review still rely on a large number of manual operations, with low efficiency and easy errors. Finally, the system is not flexible enough. During peak business periods or when the network is unstable, the system processing capacity decreases, affecting the user experience.

[0004] Therefore, there is an urgent need for an intelligent processing method that integrates edge computing capabilities and can complete data extraction, judgment, and collaborative transfer nearby to improve the overall system efficiency and intelligence level. Summary of the Invention

[0005] A series of simplified concepts are introduced in the Summary of the Invention section, which will be further detailed in the Detailed Description section. The Summary of the Invention section of this application does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the protection scope of the claimed technical solution.

[0006] In a first aspect, this application proposes a data fusion processing method based on edge computing, including: Connecting multiple data sources at an edge node, where the multiple data sources include user business processing data, identity authentication data, and image or document material data; Performing preprocessing operations on the data of the multiple data sources to obtain preprocessed data, where the preprocessing operations include data format unification, outlier processing, time series alignment, and structured transformation; Perform feature extraction operations on the preprocessed data above to obtain feature information related to the provident fund business. Among them, the above feature information includes user behavior feature information, identity feature information, and material consistency feature information; Based on the above feature information, use a local fusion model to perform local intelligent judgment operations at the edge node to obtain a fusion processing result. Among them, the above local intelligent judgment operations include preliminary qualification review, loan risk assessment, and abnormal behavior detection; Send the above fusion processing result and its associated tags to the central server for business collaborative decision-making.

[0007] In a feasible implementation manner, the specific extraction operation steps of the above user behavior feature information include: Construct a user behavior sequence based on the preprocessed user business handling data, and use position embedding coding combined with a sliding window mechanism to extract user access path pattern information; Based on the above preprocessed user business handling data, use a time-aware behavior clustering algorithm to extract user operation habit information. Among them, the above operation habit information includes operation frequency and interaction time distribution; Determine the above user behavior feature information according to the above user access path pattern information and the above operation habit information.

[0008] In a feasible implementation manner, the above determining the above user behavior feature information according to the above user access path pattern information and the above operation habit information includes: Model the above user access path pattern information as a weighted directed graph. Among them, the nodes of the above weighted directed graph represent operation pages or function entrances, the edges of the above weighted directed graph represent page jumps, and the edge weights of the above weighted directed graph represent jump frequencies; Calculate structural indicators based on the above weighted directed graph. Among them, the above structural indicators include node in-degree and out-degree distributions, path entropy, and average jump length; Construct an operation rhythm vector based on the above operation habit information, and obtain a time consistency score through cosine similarity calculation with the historical rhythm pattern; Construct the above user behavior feature information with the above structural indicators and the above time consistency score.

[0009] In a feasible implementation manner, the specific extraction operation steps of the above identity feature information include: Standardize user identity authentication fields based on the preprocessed identity authentication data and extract regional attributes; Extract identity compliance features related to housing purchase qualifications based on the above preprocessed identity authentication data. Among them, the above identity compliance features include whether it is the first applicant, marital status, and family co-loan suitability; Extract identity stability features related to provident fund contributions based on the above-mentioned preprocessed identity authentication data. Among them, the above-mentioned identity stability features include contribution types, contribution histories, and abnormal information on contribution amounts; Perform a logical consistency comparison and analysis based on the above-mentioned regional attributes, the above-mentioned identity compliance features, and the above-mentioned identity stability features to generate fused identity feature information.

[0010] In a feasible implementation manner, the specific extraction operation steps of the above-mentioned material consistency feature information include: Perform field-level structured recognition on the preprocessed image or document material data to extract keyword field information related to provident fund services; Based on the above-mentioned keyword field information, user-entered information, and identity authentication data, perform one-to-one field comparisons and calculate field consistency scores; Cross-verify the same fields among multiple materials uploaded in the same business application, construct a field covariance comparison matrix, and output a cross-material consistency score; Use a lightweight image forgery detection algorithm to analyze the authenticity of image materials and obtain an image authenticity score; Fuse the above-mentioned field consistency scores, the above-mentioned cross-material consistency scores, and the above-mentioned image authenticity scores to construct the above-mentioned material consistency feature information.

[0011] In a feasible implementation manner, the above-mentioned local intelligent judgment operation is performed at the edge node using the above-mentioned local fusion model based on the above-mentioned feature information to obtain a fusion processing result, including: Obtain business type information; Generate derived feature information based on the above-mentioned feature information and the above-mentioned business type information; Perform sparse coding operations on the above-mentioned derived feature information and the above-mentioned feature information to obtain edge node feature information; Perform a local intelligent judgment operation at the edge node using the edge node feature information with the local fusion model to obtain a fusion processing result.

[0012] In a feasible implementation manner, the above-mentioned local fusion model includes an input layer, a feature transformation layer, a multi-task output structure, and an output layer. The above-mentioned feature transformation layer includes a dynamic feature cross-coding sub-module, a local attention feature enhancement sub-module, and a multi-channel feature fusion sub-module. The above-mentioned multi-task output structure includes a first output channel, a second output channel, and a third output channel. The first output channel is used to perform the task of preliminary qualification judgment, the second output channel is used to perform the task of loan risk scoring, and the third output channel is used to perform the task of abnormal behavior detection.

[0013] In a feasible implementation manner, the above-mentioned first output channel includes a shallow multi-layer perceptron model, the second output channel includes a lightweight regression network, and the third output channel includes a multi-label classifier.

[0014] In a feasible implementation manner, the specific steps for the central server to perform business collaboration decision-making include: Perform field parsing and rule matching on the above-mentioned fusion processing result to obtain an eligibility judgment result, a risk score value, and an abnormal behavior label; Execute decision path shunting according to the above-mentioned eligibility judgment result, the above-mentioned risk score value, and the above-mentioned abnormal behavior label to obtain a decision structure, where the above-mentioned shunting includes any one of automatic transfer, manual review trigger, risk warning registration, and user correction notice; Generate a feedback information packet according to the above-mentioned decision result and push it to the user terminal.

[0015] In a second aspect, the present application proposes a data fusion processing system based on edge computing, including: A receiving unit, configured to access multiple data sources at an edge node, where the above-mentioned multiple data sources include user business handling data, identity authentication data, and image or document material data; A preprocessing unit, configured to perform preprocessing operations on the data of the above-mentioned multiple data sources to obtain preprocessed data, where the above-mentioned preprocessing operations include data format unification, outlier processing, time series alignment, and structured transformation; A feature extraction unit, configured to perform feature extraction operations on the above-mentioned preprocessed data to obtain feature information related to the provident fund business, where the above-mentioned feature information includes user behavior feature information, identity feature information, and material consistency feature information; An acquisition unit, configured to perform local intelligent judgment operations at the edge node based on the above-mentioned feature information using a local fusion model to obtain a fusion processing result, where the above-mentioned local intelligent judgment operations include extraction of preliminary eligibility review, loan risk assessment, and abnormal behavior detection; A decision-making unit, configured to send the above-mentioned fusion processing result and its associated label to the central server for business collaboration decision-making.

[0016] In summary, the present invention proposes a data fusion processing method based on edge computing. Compared with the prior art, the method proposed in this application preprocesses data, extracts features, and performs preliminary intelligent judgment at the edge node, significantly reducing the load on the central server, bandwidth consumption, and network latency. This method supports the extraction of composite features from user behavior data, identity authentication data, and image / document material data, and constructs highly expressive behavior models, identity compliance models, and material consistency models, effectively improving the risk identification ability and judgment accuracy. Through the local fusion model, parallel outputs of eligibility judgment, risk scoring, and anomaly detection are realized, supporting the simultaneous triggering of multi-dimensional service paths, reducing repeated calculations, and improving the audit throughput. The edge node generates structured label results and pushes them to the central system, enabling the central side to make quick decisions and feedback results, while also supporting the continuous optimization of the edge-side model and the strengthening of result interpretability. This method can be deployed on any edge node with basic computing capabilities (such as provident fund self-service machines, business front-end terminals, e-government cloud edge nodes, etc.), does not rely on high-performance computing resources, and has excellent adaptability and scalability. In summary, the present invention breaks through the architectural limitations of traditional centralized approval, realizes an innovative path that combines data fusion, intelligent judgment, and system collaboration, and is particularly suitable for large-scale and real-time business processing scenarios in the provident fund intelligent service system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of illustrating the preferred embodiments and are not considered to be a limitation of this specification. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings: Figure 1 It is a schematic flowchart of a data fusion processing method based on edge computing provided by an embodiment of the present application; Figure 2 It is a schematic flowchart of the specific extraction operation steps of a user behavior feature information provided by an embodiment of the present application; Figure 3 It is a schematic flowchart of the operation steps for determining user behavior feature information provided by an embodiment of the present application; Figure 4 It is a schematic flowchart of the specific extraction operation steps of an identity feature information provided by an embodiment of the present application; Figure 5 It is a schematic flowchart of the specific extraction operation steps of a material consistency feature information provided by an embodiment of the present application; Figure 6 It is a schematic flowchart of the specific extraction operation steps for obtaining the fusion processing result provided by an embodiment of the present application; Figure 7Schematic structural diagram of a data fusion processing system based on edge computing provided by an embodiment of the present application. Detailed implementation manners

[0018] Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and the above-mentioned drawings of the present application are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described here can be implemented in an order different from that illustrated or described here. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily need to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices. The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0019] Figure 1 Schematic flow diagram of a data fusion processing method based on edge computing provided by an embodiment of the present application. The method may specifically include: S110. Access multiple data sources at the edge node, where the multiple data sources include user service handling data, identity authentication data, and image or document material data; S120. Perform preprocessing operations on the data of the multiple data sources to obtain preprocessed data, where the preprocessing operations include data format unification, outlier processing, time series alignment, and structured transformation; S130. Perform feature extraction operations on the preprocessed data to obtain feature information related to the provident fund business, where the feature information includes user behavior feature information, identity feature information, and material consistency feature information; S140. Perform local intelligent judgment operations at the edge node based on the feature information using a local fusion model to obtain a fusion processing result, where the local intelligent judgment operations include initial qualification review extraction, loan risk assessment, and abnormal behavior detection; S150. Send the fusion processing result and its associated tags to the central server for business collaborative decision-making.

[0020] Exemplarily, in step S110, multiple data sources are accessed at the edge node. The multiple data sources include: (1) User business handling data, which is used to describe information such as the interaction behavior, operation track, and click path of the user during the business handling process; (2) Identity authentication data, including structured fields such as the user's real-name registration information, historical provident fund payment records, and unit-reported information; (3) Image or document material data, referring to the image files or PDF documents of materials such as the purchase contract, real estate certificate, provident fund payment certificate, and income certificate uploaded by the user.

[0021] In step S120, the system performs preprocessing operations on the above multiple data sources to obtain a consistent data set with a standardized structure. This preprocessing includes the following aspects: performing a unified data format operation on data from different sources, such as performing OCR structured extraction on PDF documents and mapping standardization on identity fields of different systems; performing outlier processing on missing items, extreme values, etc. in the original data to improve data credibility; aligning data generated across sources and at different times in chronological order according to the user dimension to form a business event stream; performing structured conversion on semi-structured data such as image materials and extracting them into key-value pairs or table fields for subsequent analysis.

[0022] In step S130, the system performs feature extraction operations based on the preprocessed data above to construct a feature vector suitable for intelligent judgment of provident fund business. Specifically, it includes three types of features: 1. User behavior feature information: Based on the user access path, a jump graph and an operation rhythm vector are constructed, and behavior features such as access stability and behavior deviation index are extracted; 2. Identity feature information: Identity compliance and credibility indicators such as geographical attributes, whether the user is the first-time applicant, marriage status and co-loan matching, and payment record stability are extracted from the user identity authentication data; 3. Material consistency feature information: By means of field-level comparison, image forgery detection, cross-material field comparison matrix, etc., the consistency of the materials submitted by the user is evaluated in terms of semantics, image, and logic dimensions.

[0023] In step S140, the system inputs the above features into the local fusion model deployed at the edge node to perform intelligent judgment operations. This fusion model supports a multi-task output structure and can parallelly complete the following tasks: 1. Preliminary qualification review: Determine whether the user meets the basic access conditions, such as the payment month, first-home identification, etc.; 2. Loan risk assessment: Generate a loan risk score based on the cross-features of behavior and identity; 3. Abnormal behavior detection: Identify potential high-risk behaviors such as material forgery, field inconsistency, and abnormal access.

[0024] In step S150, the system packages the fusion processing result and its associated tags into a structured data packet (such as in JSON format), including information such as the determination result, risk score, anomaly identifier, model version, etc., and uploads it to the central server through a secure communication channel. After receiving this information, the central server can perform operations such as automatic approval, manual review, risk registration, or user notification in combination with business rules and platform policies to construct a complete edge-center collaborative decision-making chain.

[0025] In summary, this embodiment proposes a data fusion processing method based on edge computing. Compared with the prior art, the method proposed in this application deploys data preprocessing, feature extraction, and preliminary intelligent judgment to edge nodes, significantly reducing the load on the central server, reducing bandwidth consumption and network latency. This method supports the extraction of composite features from user behavior data, identity authentication data, and image / document material data, and constructs highly expressive behavior models, identity compliance models, and material consistency models, effectively improving the risk identification ability and judgment accuracy. Through the local fusion model, parallel output of eligibility judgment, risk scoring, and anomaly detection is achieved, supporting the simultaneous triggering of multi-dimensional business paths, reducing duplicate calculations, and improving the audit throughput. The edge node generates a structured label result and pushes it to the central system, enabling the central side to make quick decisions and feedback results, and at the same time supporting the continuous optimization of the edge-side model and the enhancement of result interpretability. This method can be deployed on any edge node with basic computing capabilities (such as provident fund self-service machines, business front-end terminals, government cloud edge nodes, etc.), does not rely on high-performance computing resources, and has excellent adaptability and scalability. In summary, this embodiment breaks through the architectural limitations of traditional centralized approval, realizes an innovative path of the trinity of data fusion, intelligent judgment, and system collaboration, and is particularly suitable for large-scale and real-time business processing scenarios in the provident fund intelligent service system.

[0026] In a feasible implementation manner, as Figure 2 shown, Figure 2 This is a flow chart of the specific extraction operation steps of a user behavior feature information provided by an embodiment of this application. The specific extraction operation steps of the above user behavior feature information include: S210. Construct a user behavior sequence based on the preprocessed user business handling data, and extract user access path pattern information by using position embedding coding combined with a sliding window mechanism; S220. Extract the operation habit information of the user by using a time-aware behavior clustering algorithm based on the above preprocessed user business handling data, where the above operation habit information includes operation frequency and interaction time distribution; S230. Determine the above user behavior feature information according to the above user access path pattern information and the above operation habit information.

[0027] Exemplarily, the extraction process of user behavior feature information mainly focuses on the operation path and behavior rhythm of users during the process of handling provident fund business, and realizes the automatic construction and expression of behavior features by combining sequence modeling and time analysis methods. Specifically, this process includes the following consecutive steps: First, the system constructs a user behavior sequence based on the preprocessed user business handling data. This behavior sequence is arranged in chronological order and is denoted as , where each represents the page or operation node accessed by the user at the -th time point, and represents the total number of steps in this handling process. To effectively encode the user's access path, the system uses position embedding encoding (Position Embedding) to vectorize each operation node, and combines a sliding window mechanism to extract local sub-path patterns, thereby forming the access path features of the user during this business process.

[0028] Second, the system performs weighted aggregation on the operation node vectors in the behavior sequence to generate an access path pattern vector , and the specific calculation formula is:

[0029] where represents the embedding vector of the operation node data processing field in the data processing field, the data processing field is the weight function related to the time step data processing field in the data processing field, which can take sine-cosine encoding or linear position bias, and after fusion, the sequence-level representation data processing field in the data processing field is obtained, which is used to characterize the structure and sequential features of the user's operation path.

[0030] Subsequently, the system models the user operation frequency in the time dimension, counts the operation frequencies of the user in each time period within 24 hours of a day, and constructs an operation rhythm vector , where represents the number of operations of the user at the -th hour. To further identify the user's behavior pattern, the system introduces a time-aware clustering algorithm (such as time-weighted K-Means or T-BIRCH method) to perform clustering analysis on the rhythm vector, and maps the user to a specific operation behavior group to generate a behavior category label .

[0031] Then, to quantify the consistency between the current behavior and the user's regular behavior, the system calculates the current rhythm vector and its belonging cluster center rhythm The cosine similarity is used as the time consistency score , and the calculation formula is:

[0032] where • represents the dot product of vectors, is the 2-norm. The closer the score is to 1, the more consistent the user's current operation behavior is with their historical rhythm pattern, and the more credible the behavior is.

[0033] Finally, the system combines the path pattern vector , the time consistency score , the operation behavior category label path length and the path selection entropy entropy to construct the final user behavior feature vector :

[0034] This feature vector can be used as the input in the subsequent edge judgment module to support multiple tasks such as eligibility judgment, risk scoring, and abnormal behavior recognition. The above processing flow has high stability and generality, adapts to complex user behavior scenarios and dynamic business path changes, and is especially suitable for the automated review and behavior anomaly detection scenarios in the housing provident fund business.

[0035] In a feasible implementation manner, as Figure 3 shown, Figure 3 is a schematic flowchart of the operation steps for determining user behavior feature information provided by an embodiment of the present application. Determining the above user behavior feature information according to the above user access path pattern information and the above operation habit information includes: S310. Model the above user access path pattern information as a weighted directed graph, where the nodes of the weighted directed graph represent operation pages or function entrances, the edges of the weighted directed graph represent page jumps, and the edge weights of the weighted directed graph represent jump frequencies; S320. Calculate structural indicators based on the above weighted directed graph, where the structural indicators include node in-degree and out-degree distributions, path entropy, and average jump length; S330. Construct an operation rhythm vector based on the above operation habit information, and obtain a time consistency score by calculating the cosine similarity with the historical rhythm pattern; S340. Construct the above user behavior feature information from the above structural indicators and the above time consistency score.

[0036] Exemplarily, in a feasible implementation, in order to extract behavior feature information strongly related to provident fund business from the user operation process, the system performs a fusion analysis on the user's access path pattern and operation habit information, and constructs multi-dimensional behavior features through graph modeling and time series mining. The specific process is as follows: First, the system constructs the operation records during the business handling process of the user into an access path sequence, where each element in the sequence represents a page access or function call. Based on this sequence, the system establishes a weighted directed graph , which is used to represent the access path structure of the user. Among them, is the node set, and each node represents an operation page or function entry; is the edge set, and each directed edge represents that the user jumps from page to page ; is the edge weight set, represents the frequency of this jump, reflecting the activity degree of this path.

[0037] Specifically, the jump frequency is calculated as follows:

[0038] Among them: is the number of times the user jumps from page to page in the behavior data; is the sum of all jump times issued by the user from page ; is the normalized jump weight, which is used to represent the importance or usage frequency of the jump.

[0039] Based on this graph, the system extracts the following structural indicators as the discriminant basis for the access path: 1. Node in-degree and out-degree distribution: The out-degree is the number of all out-edges of node ; the in-degree is the number of all in-edges of node . The system can count the average in-degree and standard deviation of all nodes as the node in-degree and out-degree distribution to reflect the complexity and dispersion of the operation process.

[0040] 2. Path Entropy: To measure the concentration of the user path, the information entropy can be calculated for the jump distribution of each page through the following formula:

[0041] The total path entropy is the average value of the entropies of all nodes:

[0042] Among them, represents the jump probability from node to node ; the lower the path entropy, the more concentrated and predictable the user behavior is, usually corresponding to a skilled and normal operation path.

[0043] 3. Average jump length : The system calculates the average distance of page jumps based on the jump path, which can be expressed as:

[0044] Where: : The logical distance between pages, which can be set to 1 for direct jumps; : Represents the frequency of this jump; this indicator can reflect whether the operation path is reasonable. For example, too long a jump may indicate an abnormal process.

[0045] Secondly, the system statistically calculates the operation frequency of the user within 24 hours of a day based on the operation timestamp information and constructs an operation rhythm vector:

[0046] Where: is the number of operations of the user in the th hour segment. This vector reflects the user's time behavior pattern. For example, high frequency concentrated in the working hours may represent compliant operations.

[0047] The system matches the current operation rhythm with the historical rhythm mean or the center of the user group to obtain a time consistency score:

[0048] Where: is the dot product of vectors, representing the direction similarity; : Represents the 2-norm (Euclidean length) of the vector; the higher the similarity (close to 1), the more consistent the behavior time pattern is with the historical / normal users, and the higher the credibility.

[0049] Finally, the system concatenates the above structural indicators and the time consistency score into a set of vectors:

[0050] Among them, is the path entropy, is the average jump length, is the average out-degree, is the average in-degree, is the time consistency score.

[0051] This behavioral feature vector can be used as the input of the intelligent judgment module in the provident fund system, which helps to achieve efficient preliminary qualification review, risk identification and anomaly warning. It is especially suitable for identifying potential violations such as unreasonable process jumps, abnormally dense time or frequent path switching. This method combines structural modeling and time modeling, improving the system's characterization ability and recognition robustness for user behavior.

[0052] In a feasible implementation, as Figure 4 shown, Figure 4 is a flowchart showing the specific extraction operation steps of an identity feature information provided by an embodiment of this application. The specific extraction operation steps of the above identity feature information include: S410. Standardize the user identity authentication fields based on the preprocessed identity authentication data, and extract the geographical attributes; S420. Extract the identity compliance features related to the home purchase qualification based on the above preprocessed identity authentication data. Among them, the above identity compliance features include whether it is the first-time applicant, marital status, and family co-loan suitability; S430. Extract the identity stability features related to the provident fund deposit based on the above preprocessed identity authentication data. Among them, the above identity stability features include the deposit type, deposit history, and abnormal information of the payment amount; S440. Perform logical consistency comparison and analysis based on the above geographical attributes, the above identity compliance features, and the above identity stability features, and generate the fused identity feature information.

[0053] Exemplarily, first, standardize the user identity authentication fields based on the preprocessed identity authentication data, unify the field naming and coding methods, and extract the geographical attributes , where represents the geographical vector, including province code, city classification, and regional economic level.

[0054] Secondly, extract the identity compliance features related to the home purchase qualification , including: whether it is the first-time home purchase applicant ; marital status code , representing single, married, and divorced respectively; family co-loan suitability score , obtained based on the spouse identity and income matching degree. Therefore, the identity compliance vector can be constructed.

[0055] Then, extract the identity stability features related to the provident fund deposit , including: deposit type , indicating individual unit's agency payment; Deposit history stability score , estimated based on the deposit continuity and interruption times in the past 24 months; Payment amount volatility Calculated using the coefficient of variation (CV):

[0056] Wherein: is the average monthly payment amount; is the standard deviation; is a small constant to avoid division by zero. Finally, it forms the identity stability vector .

[0057] Take as the three types of vectors and input them into the consistency mapping function for logical comparison and fusion coding, defined as follows:

[0058] Wherein: represents the fused identity feature vector; is the weight matrix for linear projection in different feature dimensions; is the bias term; is the activation function (such as ReLU or tanh).

[0059] Next, based on the above geographical attributes , identity compliance vector and identity stability vector , perform logical consistency comparison analysis to identify potential conflicts or supportive relationships between identity information in different dimensions, and generate the fused identity feature information

[0060] First, construct the joint interaction structure between the two types of features, and define the consistency tensor mapping:

[0061] Wherein, represents the third-order outer product operation to generate the tensor for capturing the structural coupling and semantic relationships between different identity dimensions.

[0062] To simplify the high-order tensor calculation and retain the core interaction information, introduce the attention compression mechanism to map the tensor to the consistency kernel score:

[0063] Wherein, is the trainable or rule-set attention weight, used to reflect the importance of the interaction combination of the three types of fields, satisfying:

[0064] Meanwhile, to improve the interpretability of the identity consistency score, a cross-domain cosine similarity metric is further introduced:

[0065] The three types of similarities are combined in a weighted average form and defined as an auxiliary consistency scoring item:

[0066] where , and the coefficients of each item can be set according to the business priority.

[0067] Finally, a fused identity feature information representation is constructed , which is defined based on the logical consistency score and the original fields concatenation as follows:

[0068] where represents vector concatenation, is the linear mapping weight matrix, is the bias term, is the activation function (such as ReLU or , and the resulting represents the finally fused identity feature information, which has the ability to discriminate structural coupling, behavioral deviation, and logical rationality.

[0069] This fused feature can be used as the core input for subsequent intelligent judgment modules (such as loan eligibility assessment, abnormal risk screening), which helps to improve the system's comprehensive recognition ability of identity authenticity, compliance, and risk.

[0070] In a feasible implementation manner, as shown in Figure 5 , Figure 5 is a schematic flowchart of the specific extraction operation steps of a material consistency feature information provided by an embodiment of this application. The specific extraction operation steps of the above material consistency feature information include: Perform field-level structured recognition on the preprocessed image or document material data to extract keyword field information related to the provident fund business; Based on the above keyword field information, user-filled information, and identity authentication data, perform one-to-one field comparison to calculate the field consistency score; Cross-check the same fields among multiple materials uploaded in the same business application, construct a field covariance comparison matrix, and output the cross-material consistency score; Use a lightweight image forgery detection algorithm to analyze the authenticity of the image material and obtain the image authenticity score; Fuse the above field consistency score, the above cross-material consistency score, and the above image authenticity score to construct the above material consistency feature information.

[0071] Exemplarily, the extraction process of the material consistency feature information includes multiple steps such as structured recognition of images and document materials, field consistency comparison, cross-material verification, and image authenticity assessment, so as to achieve a comprehensive analysis of the materials uploaded by users in terms of content, structure, and authenticity. This process specifically includes the following consecutive operations: First, the system performs field-level structured recognition on the preprocessed image or document material data. Based on OCR technology, layout template matching, and natural language parsing algorithms, key content areas in various materials are recognized, and a set of keyword field information for provident fund business review is extracted , where each field is represented as a key-value pair , including the field name and its corresponding value. The extracted fields include but are not limited to name, ID number, unit name, deposit amount, etc.

[0072] Next, the system performs a one-to-one correspondence comparison at the field level based on the above keyword field information, user-filled information, and identity authentication data to construct a field consistency score. For each field , its value is compared with the corresponding field value in the reference data . If they are exactly equal, the score is 1; if there are slight differences, the edit distance (Levenshtein Distance) is used to calculate their similarity, and the score calculation formula is:

[0073] where is the edit distance function, is the smoothing coefficient, which is used to control the error tolerance range. The scores of all fields are averaged to obtain the material field consistency score:

[0074] Subsequently, the system cross-verifies the same fields in multiple materials uploaded by the user for the same business application. For each type of keyword field, such as "name", "ID number", and "unit address", etc., comparisons are made among all materials to construct a collaborative field comparison matrix , where the th row and the th column represent the matching degree of material and on this field, which is defined as the output of the similarity function, such as using the normalized edit distance or the cosine similarity after BERT encoding. The elements on the main diagonal are averaged to obtain the cross-material field consistency score:

[0075] Among them, is the number of common fields in all materials, indicating the self-similarity score of the fields in each material with the same fields in other materials.

[0076] Furthermore, to prevent image security risks such as forgery and tampering, the system uses a lightweight image forgery detection algorithm to evaluate the authenticity of the uploaded image materials. This algorithm judges the image content through means such as color spectrum distribution analysis, edge continuity detection, and frequency domain compression feature recognition, and outputs the forgery probability , and then obtains the image authenticity score:

[0077] The closer the image authenticity score is to 1, the more credible the shooting or scanning result of the image is.

[0078] Finally, the system combines the field consistency score , the cross-material consistency score and the image authenticity score to construct the final material consistency feature information vector :

[0079] In a feasible implementation manner, as Figure 6 shown, Figure 6 is a schematic flow chart of specific extraction operation steps for obtaining the fusion processing result provided by the embodiment of the present application. The above-mentioned local intelligent judgment operation is performed at the edge node using the local fusion model based on the above-mentioned feature information to obtain the fusion processing result, including: S1401. Obtain service type information; S1402. Generate derived feature information according to the above-mentioned feature information and the above-mentioned service type information; S1403. Perform sparse coding operation on the above-mentioned derived feature information and the above-mentioned feature information to obtain edge node feature information; S1404. Perform local intelligent judgment operation at the edge node using the edge node feature information with the local fusion model to obtain the fusion processing result.

[0080] In a feasible implementation manner, the above-mentioned local fusion model includes an input layer, a feature transformation layer, a multi-task output structure, and an output layer. The above-mentioned feature transformation layer includes a dynamic feature cross-encoding sub-module, a local attention feature enhancement sub-module, and a multi-channel feature fusion sub-module. The above-mentioned multi-task output structure includes a first output channel, a second output channel, and a third output channel. The first output channel is used to perform the task of initial qualification review judgment, the second output channel is used to perform the task of loan risk scoring, and the third output channel is used to perform the task of detecting abnormal behaviors.

[0081] In a feasible implementation manner, the first output channel includes a shallow multi-layer perceptron model, the second output channel includes a lightweight regression network, and the third output channel includes a multi-label classifier.

[0082] Exemplarily, the system performs intelligent analysis on user feature information through the local fusion model at the edge node to achieve fast and task-based judgment of the provident fund service request, improving the response efficiency and local autonomy ability on the edge side. The processing flow includes four stages: service type recognition, feature derivation, encoding processing, and fusion model inference, and finally outputs multi-dimensional service processing results, specifically including the following consecutive steps: First, the system obtains the service type information of the current service request at the edge node, that is, identifies that the currently processed service belongs to specific service scenarios such as "provident fund loan application", "provident fund withdrawal", "account information change", or "deposit status verification". This service type information is passed in through service request parameters and is represented in the form of one-hot encoding or an embedding vector, serving as an auxiliary input for subsequent feature derivation.

[0083] Subsequently, the system generates derived feature information based on the extracted user feature information (including user behavior features, identity features, and material consistency features) and the current service type information. This process captures the semantic interaction relationships between features under different service types by constructing combined features and conditional logic features. For example: in the loan application scenario, the system crosses "co-borrower suitability score" and "marital status" to form "co-borrower structure risk features" for identifying potential credit compensation relationships; in withdrawal-related services, the system focuses on activating fields related to account status, deposit duration, and unit information changes to determine the user's withdrawal eligibility; in the abnormal detection scenario, the system combines "user access path complexity" and "image authenticity score of uploaded materials" to construct abnormal recognition features to assist in identifying forged materials and bypassing process behaviors.

[0084] After the derived feature vector obtained through the above process is concatenated with the original features, the system performs sparse coding on it to form a unified input format for the edge nodes. Sparse coding includes: performing one-hot encoding on the categorical fields followed by linear projection, normalizing or standardizing the numerical fields, and screening and compressing the sparse feature dimensions, etc. Finally, an edge node feature vector is constructed:

[0085] Among them, represents the original three types of feature information, represents the derived feature vector, represents the concatenation operation.

[0086] Next, the system inputs this feature vector into a local fusion model (Local Fusion Model) deployed on the edge side for inference processing. The structure of this fusion model includes the following components: 1. Input layer: Receives the edge feature vector that has been sparsely encoded; 2. Feature transformation layer, including the following three sub-modules: Dynamic feature cross-coding sub-module: Automatically learns the non-linear interaction relationships between different feature combinations to enhance the feature expression ability; Local attention feature enhancement sub-module: Introduces a local attention mechanism to adjust the weights of the keyword fields concerned by different tasks, and realizes task-related feature selection; Multi-channel feature fusion sub-module: Processes the behavior class, identity class, and material class feature channels in parallel and performs unified fusion mapping to form a shared feature representation; 3. Multi-task output structure: Inputs the shared feature representation into three independent task channels respectively: The first output channel: Used to perform the preliminary judgment of loan eligibility, adopts a shallow multi-layer perceptron (MLP) model, and outputs a classification result of whether it meets the access conditions; The second output channel: Used to perform the loan risk scoring task, adopts a lightweight regression network structure, and scores the user risk with a continuous value (such as between 0 and 1); The third output channel: Used to perform the task of detecting abnormal behaviors, adopts a multi-label classifier structure, and outputs multiple abnormal labels at the same time, such as "forged application", "abnormal operation", and "frequent attempts", etc.

[0087] 4. Output layer: Integrates the results of the three channels and outputs the final fusion processing result , representing the preliminary eligibility review result, risk score value, and abnormal behavior annotation respectively.

[0088] Through the inference process of the local fusion model, the edge node can complete the intelligent response to the service request locally without uploading all the data to the central server, effectively improving the response efficiency, data security and system scalability. This structure is particularly suitable for the public fund business handling scenarios that are sensitive to processing latency and have high requirements for data privacy.

[0089] In a feasible implementation manner, the specific steps for the central server to perform business collaboration decision-making include: Perform field parsing and rule matching on the above fusion processing results to obtain qualification judgment results, risk score values, and abnormal behavior labels; Execute decision path shunting according to the above qualification judgment results, the above risk score values, and the above abnormal behavior labels to obtain a decision structure, where the above shunting includes any one of automatic transfer, manual review trigger, risk warning registration, and user correction notice; Generate a feedback information packet according to the above decision result and push it to the user terminal.

[0090] Exemplarily, after receiving the fusion processing result from the edge node, the central server executes the business collaboration decision-making process to achieve unified response, collaborative distribution, and personalized feedback for the provident fund service request. This process involves links such as field parsing, rule matching, path shunting, and feedback generation, ensuring that the system realizes automated, intelligent, and auditable business processing capabilities while meeting compliance requirements.

[0091] First, the central server performs field parsing and rule matching operations on the fusion processing result uploaded by the edge node. Specifically, the fusion processing result usually contains three core fields: the qualification judgment result is a binary or multi-class result indicating whether the user meets the access conditions for the provident fund business; the risk score value is a floating-point number between 0 and 1, used to quantify the potential credit or behavior risk of the user; the abnormal behavior label is a multi-label set that identifies multiple possible abnormal types in the current business.

[0092] After field parsing, the server performs matching according to the preset business rule table. For example: if and , it is regarded as a low-risk qualified user; if and , or , the manual review process is triggered; if , it is judged whether compliance can be restored through material correction.

[0093] Next, the server performs decision routing based on the above results to generate a clear business decision structure. The main routing strategies include: 1. Auto-Pass: When the user's qualifications are judged to be passed, the risk score is low and there is no abnormal label, the next business link (such as loan approval) is automatically completed; 2. Manual-Review Trigger: When there is a high risk score or abnormal behavior, the system pushes the business request to the manual background, and manual reviewers intervene to handle it; 3. Risk Alert Registration: If the risk score exceeds the warning threshold (such as 0.85) or the abnormal label involves high-risk behavior (such as forged materials, identity conflict), an internal risk control event registration will be generated and linked to the risk control model for further tracking; 4. User Correction Notice: If the qualification is not passed but there are fields that can be corrected (such as missing materials or incomplete information), the system will automatically identify the materials that need to be supplemented and generate a reminder for the user to submit them.

[0094] Finally, based on the above decision results, the server constructs a feedback information package and pushes it to the user terminal. The content includes: processing results (such as "passed", "pending review", "supplementary materials required"); risk description (such as "the score is 0.82, there are frequent operational abnormalities"); correction instructions and operation instructions (such as "please upload a complete income certificate, click here to upload"); review process status and estimated processing time.

[0095] The entire process is coordinated by the central server, and logs are recorded for auditing and model optimization, realizing the closed-loop management of the entire process of the provident fund business approval chain. This collaborative processing mechanism significantly improves the system's ability to handle high-concurrency, multi-heterogeneous business requests, while meeting the dual requirements of regulatory compliance and user experience.

[0096] Second, Figure 7 As shown, Figure 7 The present invention provides a structural diagram of a data fusion processing system based on edge computing provided in an embodiment of the present invention. The present invention proposes a data fusion processing system based on edge computing, including: A receiving unit 21 is used to access multiple data sources at the edge node, wherein the multiple data sources include user service processing data, identity authentication data, and image or document material data; A preprocessing unit 22, configured to perform a preprocessing operation on the data of the plurality of data sources to obtain preprocessed data, wherein the preprocessing operation includes data format unification, outlier processing, time series alignment and structured conversion; A feature extraction unit 23 is configured to perform feature extraction operations on the preprocessed data above to obtain feature information related to provident fund services, where the feature information includes user behavior feature information, identity feature information, and material consistency feature information; An acquisition unit 24 is configured to perform local intelligent judgment operations at an edge node based on the above feature information using a local fusion model to obtain a fusion processing result, where the local intelligent judgment operations include initial qualification review extraction, loan risk assessment, and abnormal behavior detection; A decision-making unit 25 is configured to send the above fusion processing result and its associated tags to a central server for business collaborative decision-making.

[0097] In a feasible implementation manner, the specific extraction operation steps of the above user behavior feature information include: Construct a user behavior sequence based on the preprocessed user service handling data, and use positional embedding coding combined with a sliding window mechanism to extract user access path pattern information; Based on the preprocessed user service handling data above, use a time-aware behavior clustering algorithm to extract the user's operation habit information, where the operation habit information includes operation frequency and interaction time distribution; Determine the above user behavior feature information according to the above user access path pattern information and the above operation habit information.

[0098] In a feasible implementation manner, the determining the above user behavior feature information according to the above user access path pattern information and the above operation habit information includes: Model the above user access path pattern information as a weighted directed graph, where the nodes of the weighted directed graph represent operation pages or function entrances, the edges of the weighted directed graph represent page jumps, and the edge weights of the weighted directed graph represent jump frequencies; Calculate structural indicators based on the above weighted directed graph, where the structural indicators include node in-degree and out-degree distributions, path entropy, and average jump length; Construct an operation rhythm vector based on the above operation habit information, and obtain a time consistency score through cosine similarity calculation with historical rhythm patterns; Construct the above user behavior feature information with the above structural indicators and the above time consistency score.

[0099] In a feasible implementation manner, the specific extraction operation steps of the above identity feature information include: Standardize user identity authentication fields based on the preprocessed identity authentication data, and extract regional attributes; Extract identity compliance features related to home purchase eligibility from the preprocessed identity authentication data above. Among them, the above identity compliance features include whether it is the first-time applicant, marital status, and family co-loan suitability; Extract identity stability features related to provident fund contributions from the preprocessed identity authentication data above. Among them, the above identity stability features include contribution type, contribution history, and abnormal information on contribution amount; Perform a logical consistency comparison and analysis based on the above geographical attributes, the above identity compliance features, and the above identity stability features to generate fused identity feature information.

[0100] In a feasible implementation manner, the specific extraction operation steps of the above material consistency feature information include: Perform field-level structured recognition on the preprocessed image or document material data, and extract keyword field information related to provident fund business; Based on the above keyword field information, user-filled information, and identity authentication data, perform one-to-one field comparison, and calculate the field consistency score; Cross-verify the same fields among multiple materials uploaded in the same business application, construct a field co-comparison matrix, and output the cross-material consistency score; Use a lightweight image forgery detection algorithm to perform authenticity analysis on the image material, and obtain the image authenticity score; Fuse the above field consistency score, the above cross-material consistency score, and the above image authenticity score to construct the above material consistency feature information.

[0101] In a feasible implementation manner, the above local intelligent judgment operation is performed on the edge node using the local fusion model based on the above feature information to obtain the fusion processing result, including: Obtain business type information; Generate derived feature information based on the above feature information and the above business type information; Perform sparse coding operation on the above derived feature information and the above feature information to obtain edge node feature information; Use the edge node feature information to perform a local intelligent judgment operation on the edge node using the local fusion model to obtain the fusion processing result.

[0102] In a feasible implementation manner, the above local fusion model includes an input layer, a feature transformation layer, a multi-task output structure, and an output layer. The above feature transformation layer includes a dynamic feature cross-encoding sub-module, a local attention feature enhancement sub-module, and a multi-channel feature fusion sub-module. The above multi-task output structure includes a first output channel, a second output channel, and a third output channel. The first output channel is used to perform the initial qualification judgment task. The second output channel is used to perform the loan risk scoring task. The third output channel is used to perform the abnormal behavior detection task.

[0103] In a feasible implementation manner, the first output channel includes a shallow multi-layer perceptron model. The second output channel includes a lightweight regression network. The third output channel includes a multi-label classifier.

[0104] In a feasible implementation manner, the specific steps for the central server to perform business collaborative decision-making include: Perform field parsing and rule matching on the above fusion processing results to obtain the qualification judgment result, risk score value, and abnormal behavior label; Perform decision path diversion according to the above qualification judgment result, the above risk score value, and the above abnormal behavior label to obtain a decision structure, where the above diversion includes any one of automatic transfer, manual review trigger, risk warning registration, and user correction notice; Generate a feedback information packet according to the above decision result and push it to the user terminal.

[0105] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A data fusion processing method based on edge computing, characterized in that, Including: Accessing multiple data sources at the edge node, where the multiple data sources include user service handling data, identity authentication data, and image or document material data; Performing preprocessing operations on the data of the multiple data sources to obtain preprocessed data, where the preprocessing operations include data format unification, outlier processing, time series alignment, and structured transformation; Performing feature extraction operations on the preprocessed data to obtain feature information related to the provident fund business, where the feature information includes user behavior feature information, identity feature information, and material consistency feature information; Performing local intelligent judgment operations at the edge node based on the feature information using a local fusion model to obtain a fusion processing result, where the local intelligent judgment operations include initial qualification review extraction, loan risk assessment, and abnormal behavior detection; Sending the fusion processing result and its associated tags to the central server for business collaborative decision-making.

2. The data fusion processing method based on edge computing according to claim 1, wherein, The specific extraction operation steps of the user behavior feature information include: Constructing a user behavior sequence based on the preprocessed user service handling data, and extracting user access path pattern information using position embedding coding combined with a sliding window mechanism; Extracting user operation habit information based on the preprocessed user service handling data using a time-aware behavior clustering algorithm, where the operation habit information includes operation frequency and interaction time distribution; Determining the user behavior feature information based on the user access path pattern information and the operation habit information.

3. The data fusion processing method based on edge computing according to claim 2, characterized in that, The determining the user behavior feature information based on the user access path pattern information and the operation habit information includes: Modeling the user access path pattern information as a weighted directed graph, where the nodes of the weighted directed graph represent operation pages or function entrances, the edges of the weighted directed graph represent page jumps, and the edge weights of the weighted directed graph represent jump frequencies; Calculating structural metrics based on the weighted directed graph, where the structural metrics include node in-degree and out-degree distributions, path entropy, and average jump length; Constructing an operation rhythm vector based on the operation habit information, and obtaining a time consistency score through cosine similarity calculation with historical rhythm patterns; Constructing the user behavior feature information from the structural metrics and the time consistency score.

4. The data fusion processing method based on edge computing according to claim 1, characterized in that The specific extraction operation steps of the identity feature information include: Standardizing user identity authentication fields based on the preprocessed identity authentication data, and extracting geographical attributes; Extracting identity compliance features related to housing purchase eligibility based on the preprocessed identity authentication data, where the identity compliance features include whether it is the first home applicant, marital status, and family co-loan suitability; Extracting identity stability features related to provident fund contributions based on the preprocessed identity authentication data, where the identity stability features include contribution type, contribution history, and abnormal information on contribution amounts; Performing logical consistency comparison and analysis based on the geographical attributes, the identity compliance features, and the identity stability features to generate fused identity feature information.

5. The data fusion processing method based on edge computing according to claim 1, wherein The specific extraction operation steps of the material consistency feature information include: Perform field-level structured recognition on the preprocessed image or document material data to extract key field information related to the provident fund business; Based on the above key field information, user-filled information, and identity authentication data, perform one-to-one field comparison to calculate the field consistency score; Cross-check the same fields among multiple materials uploaded in the same business application, construct a field covariance comparison matrix, and output the cross-material consistency score; Use a lightweight image forgery detection algorithm to analyze the authenticity of the image material and obtain the image authenticity score; Fuse the field consistency score, the cross-material consistency score, and the image authenticity score to construct the material consistency feature information.

6. The data fusion processing method based on edge computing according to claim 1, wherein Based on the above feature information, use a local fusion model to perform local intelligent judgment operations at the edge node to obtain the fusion processing result, including: Obtain the business type information; Generate derived feature information based on the feature information and the business type information; Perform sparse coding operations on the derived feature information and the feature information to obtain edge node feature information; Use the edge node feature information and a local fusion model to perform local intelligent judgment operations at the edge node to obtain the fusion processing result.

7. The data fusion processing method based on edge computing according to claim 6, wherein The local fusion model includes an input layer, a feature transformation layer, a multi-task output structure, and an output layer. The feature transformation layer includes a dynamic feature cross-coding sub-module, a local attention feature enhancement sub-module, and a multi-channel feature fusion sub-module. The multi-task output structure includes a first output channel, a second output channel, and a third output channel. The first output channel is used to perform the initial qualification judgment task, the second output channel is used to perform the loan risk scoring task, and the third output channel is used to perform the abnormal behavior detection task.

8. The data fusion processing method based on edge computing according to claim 7, characterized in that The first output channel includes a shallow multi-layer perceptron model, the second output channel includes a lightweight regression network, and the third output channel includes a multi-label classifier.

9. The data fusion processing method based on edge computing according to claim 8, wherein The specific steps for the central server to perform business collaborative decision-making include: Perform field parsing and rule matching on the fusion processing result to obtain the qualification judgment result, risk score value, and abnormal behavior label; Perform decision path diversion based on the qualification judgment result, the risk score value, and the abnormal behavior label to obtain the decision structure, where the diversion includes any one of automatic transfer, manual review trigger, risk warning registration, and user correction notice; Generate a feedback information packet based on the decision result and push it to the user terminal.

10. A data fusion processing system based on edge computing, characterized in that, Including: A receiving unit for accessing multiple data sources at the edge node, where the multiple data sources include user business handling data, identity authentication data, and image or document material data; A preprocessing unit for performing preprocessing operations on the data of the multiple data sources to obtain preprocessed data, where the preprocessing operations include data format unification, outlier processing, time series alignment, and structured transformation; A feature extraction unit for performing feature extraction operations on the preprocessed data to obtain feature information related to the provident fund business, where the feature information includes user behavior feature information, identity feature information, and material consistency feature information; An acquisition unit for performing local intelligent judgment operations at the edge node based on the feature information using a local fusion model to obtain a fusion processing result, where the local intelligent judgment operations include initial qualification review, loan risk assessment, and abnormal behavior detection; A decision-making unit for sending the fusion processing result and its associated label to the central server for business collaborative decision-making.

Images