Invoice automatic auditing and intelligent settlement system for automobile road rescue

Abstract

The invention discloses an invoice automatic auditing and intelligent settlement system for automobile road rescue, and relates to the technical field of intelligent settlement systems, a correction module integrates a machine learning model to analyze historical bills, automatically corrects unreasonable settlement amount and generates a correction suggestion report, a prompt module builds an intelligent priority matrix, and the intelligent priority matrix is sent to a server. The reminding frequency is dynamically adjusted according to parameters such as the historical settlement scale and the service life of a supplier, the settlement module recognizes an abnormal invoicing terminal by combining an invoicing equipment fingerprint recognition technology, a payment process is automatically triggered after verification is passed, a supplier credit portrait system is established, and a dynamic settlement scheme is supported. The settlement system establishes an intelligent priority matrix to realize important supplier high-frequency reminding and low-risk project low-frequency intervention, so that the whole process timeliness is guaranteed, the defects of a traditional rule engine in coping with complex scenes are effectively overcome, and the response capability and the autonomous decision-making capability of the system to abnormal behaviors are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent settlement systems, and particularly relates to an invoice automatic auditing and intelligent settlement system for automobile road rescue. Background Art

[0002] With the continuous growth of the automobile ownership, the road traffic environment has become increasingly complex, and the probability of vehicles breaking down or having accidents during driving has also increased accordingly. The demand for road rescue services has increased year by year. To improve customer satisfaction and service response efficiency, more and more insurance companies, automobile manufacturers, and third-party rescue service agencies provide 7×24-hour road rescue services, including various service contents such as towing, battery jump-starting, tire replacement, and emergency fuel delivery.

[0003] While road rescue services are becoming increasingly popular, the accompanying invoice management and cost settlement problems have become increasingly prominent. The settlement system combines means such as OCR recognition technology, business rule engines, big data comparison, and smart contracts to realize the automated and intelligent processing of the entire process of rescue invoices from uploading, auditing to settlement, improve operation efficiency, reduce management costs, and enhance system security.

[0004] The existing technologies have the following defects:

[0005] 1. Currently, the industry generally uses rule engines for expense auditing, which is difficult to handle variable and unstructured abnormal situations. For example, in a specific period, the rescue frequency in a remote area has increased abnormally. The traditional system cannot actively perceive or give an early warning, which easily leads to a large number of false settlements flowing into the financial system. It lacks the fusion analysis of multi-source data such as historical bills, geographical trajectories, and behavior patterns, and has weak abnormal recognition ability, bringing a high risk of financial losses.

[0006] 2. Most existing systems only communicate with suppliers about invoice verification matters through simple to-do lists, manual email reminders, etc. They do not establish a dynamic priority model and lack an intelligent reminder and emergency event handling mechanism. The importance of suppliers is disconnected from the communication strategy, which easily leads to high-priority services being affected by delayed reminders and affecting the overall process timeliness.

[0007] Based on this, the present invention proposes an invoice automatic auditing and intelligent settlement system for automobile road rescue, establishes an intelligent priority matrix to achieve "high-frequency reminders for important suppliers and low-frequency intervention for low-risk projects", guarantees the overall process timeliness, effectively makes up for the deficiencies of traditional rule engines in dealing with complex scenarios, and improves the system's response ability and autonomous decision-making ability to abnormal behaviors. Summary of the Invention

[0008] The purpose of the present invention is to provide an invoice automatic auditing and intelligent settlement system for automobile road rescue to solve the deficiencies in the background art.

[0009] To achieve the above object, the present invention provides the following technical solutions: An automatic invoice review and intelligent settlement system for automobile road rescue, including a correction module, a prompt module, a storage module, an anti-counterfeiting identification module, and a settlement module;

[0010] Correction module: Integrate a machine learning model to analyze historical data, warn of high-incidence abnormal settlement items, automatically correct unreasonable settlement amounts, and generate a correction suggestion report;

[0011] Prompt module: Establish an intelligent priority matrix to dynamically adjust the reminder frequency;

[0012] Storage module: Use blockchain technology to distribute and store data;

[0013] Anti-counterfeiting identification module: When receiving an appeal request, associate historical data. After the supplier uploads the invoice, automatically associate the settlement form and extract information into the database, and prevent false printed invoices through OCR + 3D watermark identification technology;

[0014] Settlement module: Combine the fingerprint recognition technology of the invoicing device to identify abnormal invoicing terminals, and automatically trigger the payment process after passing the verification.

[0015] Preferably, each time the supplier issues an invoice, the settlement module collects the unique identifier of the invoicing device through the fingerprint recognition technology of the invoicing device, and the device fingerprint is used to identify each invoicing device;

[0016] Compare the collected device fingerprint with the records in the historical device fingerprint library to determine whether the device is a trusted device. If an abnormal device fingerprint in the historical record is matched, it is marked as an abnormal terminal, and manual review or further verification is triggered. If an abnormal invoicing terminal is detected, the invoicing operation of this terminal will be automatically stopped, and the administrator will be notified for verification.

[0017] Preferably, after passing the device fingerprint verification, the settlement module will further verify the invoice content uploaded by the supplier, compare the supplier's past settlement records with the current invoice content, and determine whether the invoice conforms to the conventional settlement mode. If the invoice amount significantly deviates from the historical data or exceeds the preset range, a warning will be issued and manual review will be required;

[0018] If the device fingerprint verification is passed and the invoice information is legal, the payment process will be automatically triggered, and the payment process will determine whether to settle in real time according to factors such as the supplier's credit level, settlement amount, and invoice validity;

[0019] Automatically generate and update the supplier's credit profile based on the supplier's historical settlement behavior, invoice authenticity, and performance data.

[0020] Preferably, the supplier credit profile includes the following information:

[0021] Credit score: Comprehensive score based on the supplier's settlement history and invoice compliance indicators;

[0022] Settlement behavior: Including the timeliness of settlement time and invoice accuracy;

[0023] Performance: The timeliness and quality of the supplier's contract performance;

[0024] After each settlement, the credit profile is automatically updated according to the new settlement information.

[0025] Preferably, upon receiving a settlement appeal request initiated by the supplier, the anti-counterfeiting identification module retrieves the corresponding historical rescue work order pictures, GPS trajectory data, verified settlement statements and bill amendment reports from the alliance chain storage module through the appeal order number or associated bill ID, automatically creates an identification task order, and pushes it to the anti-counterfeiting identification queue;

[0026] The supplier uploads the invoice to be verified, and uses a multi-modal OCR engine to extract fields, including invoice number, invoice code, issue date, amount, tax rate, issuing unit, purchaser information, project details, and total amount. At the same time, the position coordinates of the fields are recorded.

[0027] The identified field information is automatically matched with the associated bill database. If a field is missing, the amount is inconsistent, or the issue date does not fall within the time range, the system marks it as initially abnormal.

[0028] Perform a light and shadow depth analysis on the security watermark area in the scanned image to check for the presence of legitimate angular reflection features and micro-printing information, and use a watermark signature matching model to determine whether the invoice is an original machine-printed invoice.

[0029] Use an image feature point matching algorithm to compare with the historical seal library of the issuing unit, and detect forgery situations including seal offset, seal overlap, and blurred font; further verify the authenticity in combination with the device fingerprint to which the seal belongs.

[0030] Preferably, the anti-counterfeiting identification module uses image segmentation technology to extract the seal area in the invoice, and uses the SIFT or SURF algorithm to extract the feature points of the seal area. The extracted feature points include: key point position, scale, direction, and descriptor information for subsequent feature matching.

[0031] Load the real seal image registered by the issuing unit from the historical seal library, compare the feature points of the invoice seal uploaded with the feature points of the historical seal image, and judge the matching degree through the feature point matching algorithm.

[0032] If the matching is qualified, compare the offset of the seal in the invoice image from the preset standard position. If the offset is greater than the offset threshold, mark it as abnormal.

[0033] Use the image overlap algorithm to determine whether the seal overlaps with other elements, and use the image blurriness measurement algorithm to evaluate whether the text in the seal is blurred.

[0034] Preferably, the storage module deploys a multi-node consortium blockchain network including: a platform operator node, a third-party audit node, and a core supplier node;

[0035] Use the identity authentication module to manage the access rights of each participant, and set a smart contract template for standardizing the storage rules of bill reconciliation, confirmation, modification, and settlement status;

[0036] Each bill reconciliation process will automatically generate a reconciliation transaction object. The reconciliation transaction object generates a data fingerprint through a hash digest. The data fingerprint is packaged together with the original data and submitted to the consortium blockchain network for storage;

[0037] Each node participates in the consensus to confirm the validity of the reconciliation transaction. If the consensus is passed, the data is packaged into a block to form an on-chain record.

[0038] Preferably, the prompt module obtains the basic data and historical cooperation information of each supplier from the database, including the total amount of cumulative settlement of the supplier, the cooperation years, the settlement exception rate, and the response timeliness, to form a portrait of each supplier;

[0039] Normalize all the obtained parameters, then calculate the comprehensive score, and divide all suppliers into multiple levels according to the comprehensive score, including level A, level B, level C, and level D, to form a priority matrix.

[0040] Preferably, the correction module automatically collects historical bill data and corresponding GPS rescue trajectory data. The historical bill includes rescue time, service type, settlement amount, supplier information, and service area. The GPS rescue trajectory data includes the positioning points before and after vehicle rescue, driving path, and rescue start and end time information;

[0041] The machine learning model embedded in the correction module trains the preprocessed data, extracts the normal rescue frequency, rescue type, and amount distribution patterns in each region and each period, identifies the feature boundaries of normal settlement behaviors, and constructs a benchmark behavior pattern library;

[0042] When new bill data enters the audit process, compare it with the benchmark behavior pattern to identify abnormal high-incidence settlement items.

[0043] Preferably, the correction module constructs a feature vector:

[0044] , where is the feature vector of the i-th rescue, is the duration of the i-th rescue, is the distance of the i-th rescue, is the amount of the i-th rescue;

[0045] After performing K-Means clustering on all feature vectors, all clustering clusters are obtained, and the total within-class distance squared sum is minimized: , where, is the total within-class distance squared sum, is the number of clustering categories, is the sample set included in the k-th cluster, is the centroid of the k-th cluster;

[0046] After obtaining the new bill, the new bill is classified into the corresponding clustering cluster according to the feature vector of the new bill, and the deviation value is obtained by subtracting the amount corresponding to the centroid of the clustering cluster from the new bill amount. If the deviation value is greater than the deviation threshold, it is determined that the new bill is abnormal;

[0047] The new bill amount is readjusted, and the expression is: , where, is the amount of the new bill after adjustment, is the amount corresponding to the centroid of the clustering cluster, is the standard deviation of the clustering cluster amount, is the adjustment factor, and .

[0048] In the above technical solution, the technical effects and advantages provided by the present invention:

[0049] The present invention analyzes historical bills and GPS rescue trajectory data through a correction module integrated with a machine learning model, warns of high-incidence abnormal settlement items, automatically corrects unreasonable settlement amounts and generates a correction suggestion report. The reminder module establishes an intelligent priority matrix, dynamically adjusts the reminder frequency according to parameters such as the historical settlement scale and cooperation years of the supplier, and automatically triggers a call robot outbound call in case of an emergency. The settlement module combines the fingerprint recognition technology of the invoicing device to identify abnormal invoicing terminals, and automatically triggers the payment process after passing the verification, and establishes a supplier credit portrait system to support a dynamic settlement plan. The settlement system establishes an intelligent priority matrix to achieve "high-frequency reminder for important suppliers and low-frequency intervention for low-risk projects", ensures the timeliness of the overall process, effectively makes up for the deficiencies of the traditional rule engine in dealing with complex scenarios, and improves the system's response ability and autonomous decision-making ability to abnormal behaviors. Brief Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings.

[0051] Figure 1 This is the system architecture diagram of the present invention.

[0052] Figure 2 This is the system timing diagram of the present invention.

[0053] Figure 3 This is the system framework diagram of the present invention. Detailed implementation manners

[0054] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] Embodiment 1: Please refer to Figures 1 - 3 As shown, a system for automatic invoice review and intelligent settlement for vehicle roadside assistance in this embodiment includes a correction module, a reminder module, a storage module, an anti-counterfeiting identification module, and a settlement module;

[0056] Correction module: Integrate a machine learning model to analyze historical bills and GPS rescue trajectory data, warn of settlement items with high abnormal incidence (such as a sharp increase in rescue frequency in remote areas), automatically correct unreasonable settlement amounts and generate a correction suggestion report, and send the corrected settlement amount and suggestion report to the storage module;

[0057] Reminder module: Establish an intelligent priority matrix, dynamically adjust the reminder frequency according to parameters such as the historical settlement scale and cooperation years of the supplier, automatically trigger an outbound call by a telephone robot in case of an emergency, and send the priority matrix to the storage module;

[0058] Storage module: Adopt the alliance chain technology to distribute and store the verification process and confirmation results to ensure the auditability of the operation process, and send the distributed storage results to the anti-counterfeiting identification module and the settlement module;

[0059] Anti-counterfeiting identification module: When receiving an appeal request, automatically associate original data such as historical rescue work order pictures and GPS trajectories. After the supplier uploads the invoice, automatically associate the settlement form and extract key information for storage, and prevent false printed invoices through OCR + 3D watermark identification technology. Send the invoice anti-counterfeiting identification result to the settlement module;

[0060] Settlement module: Identify abnormal invoicing terminals by combining invoicing device fingerprint recognition technology. After passing the verification, automatically trigger the payment process, establish a supplier credit profiling system, and support a dynamic settlement plan - high-quality suppliers can settle basic payments in real time, while suppliers that do not meet the standards have partial amounts settled with a delay.

[0061] This application integrates a machine learning model through a correction module to analyze historical bills and GPS rescue trajectory data, warn of abnormally high-incidence settlement items, automatically correct unreasonable settlement amounts and generate a correction suggestion report. The reminder module establishes an intelligent priority matrix, dynamically adjusts the reminder frequency according to parameters such as the supplier's historical settlement scale and cooperation years, automatically triggers a phone robot outbound call in case of an emergency. The settlement module combines invoicing device fingerprint recognition technology to identify abnormal invoicing terminals, and automatically triggers the payment process after passing the verification, establishing a supplier credit profiling system to support a dynamic settlement plan. The settlement system establishes an intelligent priority matrix to achieve "high-frequency reminders for important suppliers and low-frequency intervention for low-risk projects", ensuring the timeliness of the overall process, effectively making up for the deficiencies of traditional rule engines in dealing with complex scenarios, and enhancing the system's response ability and autonomous decision-making ability to abnormal behaviors.

[0062] The specific working process of the settlement system is as follows:

[0063] The settlement system integrates a machine learning model to analyze historical bills and GPS rescue trajectory data, warn of abnormally high-incidence settlement items (such as a sharp increase in rescue frequencies in remote areas), automatically correct unreasonable settlement amounts and generate a correction suggestion report, establish an intelligent priority matrix, dynamically adjust the reminder frequency according to parameters such as the supplier's historical settlement scale and cooperation years, automatically trigger a phone robot outbound call in case of an emergency, use the consortium chain technology to distribute and store the verification process and confirmation results to ensure that the operation process is auditable and traceable. When receiving an appeal request, automatically associate original data such as historical rescue work order pictures and GPS trajectories. After the supplier uploads the invoice, automatically associate the settlement form and extract key information for storage, prevent false printed invoices through OCR + 3D watermark recognition technology, combine invoicing device fingerprint recognition technology to identify abnormal invoicing terminals, automatically trigger the payment process after passing the verification, establish a supplier credit profiling system, and support a dynamic settlement plan - high-quality suppliers can settle basic payments in real time, while suppliers that do not meet the standards have partial amounts settled with a delay.

[0064] Embodiment 2: The correction module integrates a machine learning model to analyze historical bills and GPS rescue trajectory data, warn of abnormally high-incidence settlement items (such as a sharp increase in rescue frequencies in remote areas), automatically correct unreasonable settlement amounts and generate a correction suggestion report. The corrected settlement amount and suggestion report are sent to the storage module.

[0065] The correction module automatically collects historical billing data and corresponding GPS rescue trajectory data from the system. Historical billing includes, but is not limited to, rescue time, service type, settlement amount, supplier information, and service area. GPS rescue trajectory data includes information such as the vehicle's location before and after the rescue, the driving route, and the rescue start and end times. The collected data undergoes preprocessing operations such as format standardization, missing value completion, and outlier removal to establish a unified data input format.

[0066] The machine learning model embedded in the module (such as the cluster analysis model or the time series anomaly detection model) is trained on the preprocessed data to extract the normal rescue frequency, rescue type and amount distribution pattern in each region and time period, identify the characteristic boundaries of normal settlement behavior, and thus build a benchmark behavior pattern library.

[0067] Correction module to construct feature vector: , where is the feature vector of the i-th rescue, is the duration of the i-th rescue, is the distance of the i-th rescue, is the amount of the rescue in Article i.

[0068] After performing K-Means clustering on all feature vectors, all clusters are obtained to minimize the total intra-class distance sum of squares: , where is the total intra-class distance sum of squares, is the number of cluster categories, is the sample set contained in the k-th cluster, is the centroid of the kth cluster (i.e., the mean of the eigenvectors of the samples in this cluster).

[0069] After obtaining a new bill, the new bill is divided into the corresponding cluster according to its feature vector. The deviation value is obtained by subtracting the amount corresponding to the cluster centroid from the new bill amount. If the deviation value is greater than the deviation threshold, the new bill is determined to be abnormal and the new bill amount is readjusted. The expression is: , where The adjusted amount for the new bill. is the amount corresponding to the centroid of the cluster, is the standard deviation of the cluster amount, is the regulating factor, and .

[0070] When new bill data enters the audit process, the system calls the model to compare with the benchmark behavior pattern and identifies high-incidence abnormal settlement items. If the rescue frequency in a certain geographical area is much higher than the historical average within a specific time period, or the rescue cost significantly deviates from the historical range, it can be marked as an "abnormal settlement warning area", and the abnormal parameters (such as the surge ratio of frequency, type of abnormality, etc.) are recorded.

[0071] The reminder module establishes an intelligent priority matrix, dynamically adjusts the reminder frequency according to parameters such as the historical settlement scale of the supplier and the cooperation years, automatically triggers the outbound call of the phone robot in case of emergency, and sends the priority matrix to the storage module.

[0072] The reminder module obtains the basic data and historical cooperation information of each supplier from the database, including: total settlement amount scale: the total amount accumulated by the supplier for settlement, cooperation years: the number of years the supplier has cooperated with the platform, settlement abnormality rate: the proportion of settlement abnormalities (such as being corrected, delayed, rejected) occurring in the past year, response timeliness: the average response time for bill processing. These parameters are used as feature vectors to input into the model to form a portrait of each supplier.

[0073] Normalize all the obtained parameters, and then calculate the comprehensive score by combining the set weights: , where is the comprehensive score of the i-th supplier, , , , are weights (adjustable), is the total amount accumulated by the supplier for settlement, is the number of years of cooperation, is the settlement abnormality rate, is the response timeliness.

[0074] All suppliers are divided into multiple levels according to the comprehensive score, including level A, level B, level C, and level D, to form a priority matrix. The priority matrix can be as shown in Table 1:

[0075] Table 1

[0076] Level Score Range Reminder Strategy Grade A S≥80 Normal Rhythm Reminder (Once a Day) Grade B 60≤S<80 Increase Frequency (Twice a Day) Grade C S<60 Strong Reminder (Once Every 4 Hours) Grade D Abnormal Mark Emergency Notice + Outbound Call

[0077] The system writes information such as the current level, reminder frequency, and trigger conditions of each supplier into the "priority matrix structure table". The system runs regularly and executes corresponding reminder tasks according to the priority matrix: for level B and level C, the system sends reminder notices through App messages, text messages, or emails; for level D, if statuses such as "bill rejection confirmation" and "not responding for more than X hours" are detected, the phone robot is immediately called for voice outbound calls.

[0078] In this application, the weights (adjustable) are manually set by business experts or project teams according to experience for each parameter weight. An example of weight setting (adjustable) is shown in Table 2:

[0079] Table 2

[0080] Parameter Item Description Weight (Example) Total Settlement Amount Scale Reflect Service Activity 0.30 Cooperation Years Indicate Cooperation Stability 0.20 Response Timeliness Embody Service Timeliness 0.25 Settlement Abnormality Rate Indicate Bill Credibility 0.25

[0081] The storage module uses the consortium blockchain technology to distribute and store the verification process and confirmation results, ensuring the auditability of the operation process. The distributed storage results are sent to the anti-counterfeiting identification module and the settlement module.

[0082] The storage module deploys a multi-node consortium blockchain network (such as based on Hyperledger Fabric or FISCO BCOS), including the platform operator node, the third-party audit node, and the core supplier node. A unified identity authentication module (such as a CA certificate) is used to manage the access rights of each participant; a smart contract template is set to standardize the storage rules for bill verification, confirmation, modification, and settlement status.

[0083] For each bill verification process, the system will automatically generate a "verification transaction object", the content of which includes: bill number, verifier, verification time, verification content, abnormal marking type, summary of correction suggestions, operation IP address, and device ID (for later audit). The above information generates a unique "data fingerprint" through a hash digest. The data fingerprint and the original data are packaged together and submitted to the consortium blockchain network for storage.

[0084] Each node participates in consensus (such as PBFT or RAFT) to confirm the validity of the verification transaction; once the consensus is passed, the system packages the data into a block to form a record on the chain. The storage content of each verification transaction includes: original summary information, verification operation details, transaction processing status (such as "pending confirmation", "correction suggestions have been sent"), operation timestamp, and signature. Each block carries the hash value of the previous block, forming an immutable data chain.

[0085] When the anti-counterfeiting identification module receives an appeal request, it automatically associates with original data such as historical rescue work order pictures and GPS trajectories. After the supplier uploads the invoice, it automatically associates with the settlement form and extracts key information for storage. It uses OCR + 3D watermark recognition technology to prevent false printed tickets, and the invoice anti-counterfeiting identification result is sent to the settlement module.

[0086] The anti-counterfeiting identification module receives the settlement appeal request initiated by the supplier; through the appeal order number or associated bill ID, it obtains the corresponding: historical rescue work order pictures (such as vehicle rescue site photos, repair orders, etc.), GPS trajectory data (including the start and end positions and time of the rescue), the verified settlement form and bill correction report from the consortium blockchain storage module, automatically creates an identification task order, and pushes it to the anti-counterfeiting identification queue.

[0087] The supplier uploads invoices to be verified (supporting formats such as PDF, images, etc.); the anti-counterfeiting identification module automatically performs the following processing:

[0088] 1) OCR text recognition processing: Use a multi-modal OCR engine (supporting printed text + handwritten text) to extract the following key fields: invoice number, invoice code, invoice date, amount, tax rate, invoicing unit, purchaser information, project details, and total amount. At the same time, record the field position coordinates for subsequent layout verification.

[0089] 2) Structured information storage in the database: Automatically match the identified field information with the associated bill database. If a field is missing, the amount is inconsistent, or the invoice date does not fall within the time range, the system marks it as "initially abnormal".

[0090] The anti-counterfeiting identification module performs anti-counterfeiting identification, including:

[0091] 1) 3D watermark feature verification: Perform light and shadow depth analysis on the security watermark area in the scanned image to check for the presence of legal angular reflection features and micro-printing information, and use the watermark signature matching model to determine whether the invoice is an original machine-printed invoice.

[0092] 2) Invoice seal image comparison: Use an image feature point matching algorithm (such as SIFT / SURF) to compare with the historical seal library of the invoicing unit, and detect forgery situations such as seal offset, seal overlap, and blurred font; further verify the authenticity in combination with the device fingerprint to which the seal belongs.

[0093] The anti-counterfeiting identification module uses image segmentation technology (such as the Sobel operator) to extract the seal area in the invoice, and uses the SIFT (Scale-Invariant Feature Transform) or SURF (Speeded-Up Robust Features) algorithm to extract the feature points of the seal area. The extracted feature points include information such as the key point position, scale, direction, and descriptor, which are used for subsequent feature matching.

[0094] Load the real seal image registered by the invoicing unit from the historical seal library, compare the feature points of the uploaded invoice seal with the feature points of the historical seal image, and judge the matching degree through a feature point matching algorithm (such as brute-force matching or FLANN matcher). The comparison result will return the number of matching feature points and the matching degree. A lower matching degree indicates that the seal may be forged.

[0095] If the match is qualified, compare the offset of the seal in the invoice image from the preset standard position. If the offset is greater than the offset threshold, mark it as "abnormal". Use the image overlay algorithm to determine whether the seal overlaps with other elements (such as text, graphics, etc.). If there is an overlap, it indicates that there may be forgery behavior. Use an image blurriness measurement algorithm (such as Laplacian transform) to evaluate whether the text in the seal is blurred. If it is blurred, it may be a forged seal.

[0096] Suppose a company uploads an invoice that contains a seal and the seal may be forged. We will explain in detail how to use the above techniques for forgery detection step by step.

[0097] Input: Invoice image file.

[0098] The Sobel operator is used for edge detection. First, convert the invoice image to a grayscale image. Then, apply the Sobel operator to detect the edges in the image, especially the edges of the seal area. By calculating the gradients of the pixels in the image, the Sobel operator can highlight the contours of the seal area.

[0099] Seal area extraction: Based on the edge information detected by the Sobel operator, the area where the seal is located can be extracted through an image segmentation algorithm.

[0100] In the extracted seal area, use the SIFT (Scale-Invariant Feature Transform) or SURF (Speeded-Up Robust Features) algorithm to extract key feature points. The feature points include position, scale, orientation, descriptor, etc. These feature points are robust to seals at different angles, scales, and illuminations, ensuring that even if the seal changes, it can still be matched.

[0101] Suppose there is a real seal image of the company in the historical seal library. Load the historical image into the system and extract its feature points.

[0102] Use a brute-force matcher or a FLANN (Fast Library for Approximate Nearest Neighbors) matcher to compare the feature points of the seal in the invoice with the feature points of the historical seal image.

[0103] After detecting the seal area, the system will compare the position of the seal in the image with the preset standard position (such as the standard position in the upper right corner of the invoice). By calculating the deviation of the position of the seal in the image from the standard position, if the offset exceeds the set threshold (such as 5 mm), it is marked as "abnormal".

[0104] Use an image overlapping algorithm (such as image stitching technology) to determine whether the seal overlaps with other elements (such as text, graphics). If there is an overlap, there may be forgery because a normal seal does not overlap with other elements.

[0105] Use an image blurriness measurement algorithm such as Laplacian transform to check whether the text in the seal area is blurred. If the text is blurred, it is usually because the seal is forged using low-quality equipment.

[0106] Integrate the results of various detections (feature matching, offset, overlap, blurriness) to obtain the final verification result of the invoice seal. If there are abnormalities (such as large seal offset, overlap or blurriness), mark the invoice as "forged". Otherwise, the invoice is considered valid and legal.

[0107] Obtain the invoicing device number or device signature information shown on the invoice (such as the anti-counterfeiting invoicing code); combine with the "invoicing device fingerprint library" in the settlement module: verify whether it is a registered device; determine whether it frequently appears in abnormal invoices (such as a device issuing abnormally high-value invoices within a day); integrate the recognition and analysis results to form an anti-counterfeiting recognition report, including: invoice key information extraction form, anti-counterfeiting verification status (passed / suspicious / forged), details of suspicious items (such as: amount mismatch, watermark failure, seal abnormality, etc.), and send the recognition result to the settlement module automatically through the interface for subsequent payment judgment; if the recognition result is "suspicious" or "forged", the system automatically suspends the corresponding settlement process and issues a warning. If there is a risk with the device, immediately mark it as a "suspicious terminal" and push it to the credit portrait module.

[0108] The settlement module combines invoicing device fingerprint recognition technology to identify abnormal invoicing terminals. After passing the verification, it automatically triggers the payment process, establishes a supplier credit portrait system, and supports a dynamic settlement plan - high-quality suppliers can settle basic payments in real time, and suppliers that do not meet the standards implement partial amount deferred settlement.

[0109] Each time a supplier issues an invoice, the system collects the unique identifier of the invoicing device (such as device hardware fingerprint, MAC address, operating system information, etc.) through invoicing device fingerprint recognition technology. This device fingerprint is used to identify each invoicing device to ensure that the invoice is issued by a real and compliant device. Compare the collected device fingerprint with the records in the historical device fingerprint library to determine whether the device is a trusted device. If an abnormal device fingerprint in the historical record is matched (such as an unregistered device, illegal device, etc.), mark it as an "abnormal terminal" and trigger manual review or further verification. If an abnormal invoicing terminal is detected, the system will automatically stop the invoicing operation of this terminal and notify the administrator for verification to prevent the risk of forged invoices.

[0110] After passing the device fingerprint verification, the system will further verify the invoice content uploaded by the supplier to ensure that the invoice amount, tax number, invoice number, etc. are consistent with the preset standards, avoiding settlement anomalies caused by human errors or forgery. It will compare the supplier's past settlement records with the current invoice content to determine whether the invoice conforms to the regular settlement pattern. If the invoice amount significantly deviates from the historical data or exceeds the preset range, the system will issue a warning and require manual review.

[0111] If the device fingerprint verification is passed and the invoice information is legal, the system will automatically trigger the payment process. The payment process will determine whether to settle in real-time based on factors such as the supplier's credit rating, settlement amount, invoice validity, etc. It supports multiple payment methods (e.g., bank transfer, third-party payment platform, etc.) and automatically selects the most suitable payment method according to the credit ratings of different suppliers.

[0112] Based on data such as the supplier's historical settlement behavior, invoice authenticity, and performance, automatically generate and update the supplier's credit profile. The supplier credit profile includes the following information:

[0113] Credit score: Comprehensive scoring based on indicators such as the supplier's settlement history and invoice compliance.

[0114] Settlement behavior: Including the timeliness of settlement time, invoice accuracy, etc.

[0115] Performance: The timeliness and quality of the supplier's contract performance.

[0116] Credit profile update: After each settlement, the system will automatically update the credit profile according to the new settlement information. If the supplier has abnormal behavior (such as late invoice submission, frequent invoice forgery, etc.), the system will lower its credit score and may affect its subsequent settlement conditions.

[0117] High-quality suppliers (i.e., suppliers with high credit scores, compliant invoices, and good performance) can enjoy real-time settlement, that is, the settlement amount will be paid immediately after passing the verification, improving their capital turnover efficiency. Suppliers that do not meet the standards (i.e., suppliers with low credit scores, invoice problems, or poor performance) will have partial amount delayed settlement. Specifically, a partial amount of the settlement amount (such as the basic amount) may need to be delayed until the supplier improves its credit behavior. Dynamically adjust the settlement conditions according to the supplier's credit profile and historical performance. For example, high-quality suppliers will be automatically marked as "priority settlement", while suppliers that do not meet the standards need to further improve their credit ratings to enjoy real-time settlement treatment.

[0118] This application ensures the security, transparency, and compliance of the settlement process through methods such as fingerprint recognition of invoicing devices, invoice data verification, and construction of supplier credit portraits. The dynamic settlement plan can automatically adjust settlement conditions according to the credit situation of suppliers, provide real-time settlement services for high-quality suppliers, and implement delayed settlement for suppliers that do not meet the standards, avoiding financial risks and promoting the standardization of suppliers' behaviors.

[0119] In the description of this specification, the description with reference to terms such as "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0120] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and utilize the present invention well. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. An automatic invoice review and intelligent settlement system for automobile roadside assistance, characterized in that: It includes a correction module, a reminder module, a storage module, an anti-counterfeiting identification module, and a settlement module; Correction module: Integrate a machine learning model to analyze historical data, warn of abnormally high-incidence settlement items, automatically correct unreasonable settlement amounts, and generate a correction suggestion report; Reminder module: Establish an intelligent priority matrix to dynamically adjust the reminder frequency; Storage module: Use blockchain technology to store data distributively; Anti-counterfeiting identification module: When receiving an appeal request, associate historical data. After the supplier uploads an invoice, automatically associate the settlement form and extract information for storage. Use OCR + 3D watermark identification technology to prevent false printed invoices; Settlement module: Combine the fingerprint recognition technology of the invoicing device to identify abnormal invoicing terminals. After passing the verification, automatically trigger the payment process.

2. The automatic invoice review and intelligent settlement system for automobile road rescue according to claim 1, wherein: When the settlement module invoices each time for the supplier, it collects the unique identifier of the invoicing device through the fingerprint recognition technology of the invoicing device. The device fingerprint is used to identify each invoicing device; Compare the collected device fingerprint with the records in the historical device fingerprint database to determine whether the device is a trusted device. If an abnormal device fingerprint in the historical record is matched, it is marked as an abnormal terminal, and manual review or further verification is triggered. If an abnormal invoicing terminal is detected, the invoicing operation of this terminal will be automatically stopped, and the administrator will be notified for verification.

3. The automatic invoice review and intelligent settlement system for vehicle roadside assistance according to claim 2, wherein: After passing the device fingerprint verification, the settlement module will further verify the content of the invoice uploaded by the supplier. Compare the supplier's past settlement records with the content of the current invoice to determine whether the invoice conforms to the regular settlement mode. If the invoice amount significantly deviates from the historical data or exceeds the preset range, an alarm will be issued and manual review will be required; If the device fingerprint verification is passed and the invoice information is legal, the payment process will be automatically triggered. The payment process will determine whether to settle in real time based on factors such as the supplier's credit level, settlement amount, and invoice validity; Automatically generate and update the supplier's credit profile based on the supplier's historical settlement behavior, invoice authenticity, and performance data.

4. The automatic invoice review and intelligent settlement system for automobile roadside assistance according to claim 3, wherein: The supplier credit profile includes the following information: Credit score: Comprehensive score based on the supplier's settlement history and invoice compliance indicators; Settlement behavior: Includes the timeliness of the settlement time and the accuracy of the invoice; Performance: The timeliness and quality of the supplier's contract performance; After each settlement, the credit profile is automatically updated according to the new settlement information.

5. An automatic invoice review and intelligent settlement system for automobile road rescue according to claim 4, characterized in that: When the anti-counterfeiting identification module receives a settlement appeal request initiated by the supplier, it obtains the corresponding historical rescue work order pictures, GPS trajectory data, and the reconciled settlement form and bill correction report from the blockchain storage module through the appeal order number or associated bill ID, automatically creates an identification task order, and pushes it to the anti-counterfeiting identification queue; The supplier uploads the invoice to be verified, and uses a multi-modal OCR engine to extract fields, including invoice number, invoice code, invoicing date, amount, tax rate, invoicing unit, purchaser information, project details, and total amount. At the same time, the position coordinates of the fields are recorded; The identified field information is automatically matched with the associated bill database. If a field is missing, the amount is inconsistent, or the invoicing date does not fall within the time range, the system marks it as initially abnormal; Perform light and shadow depth analysis on the security watermark area in the scanned image, check for the existence of legal angular reflection features and micro-printed pattern information, and use the watermark signature matching model to determine whether the ticket is an original machine-printed ticket; Use the image feature point matching algorithm to compare with the historical seal library of the invoicing unit, and detect forgery situations including seal offset, seal overlap, and blurred font; further verify the authenticity in combination with the device fingerprint of the seal.

6. The automatic invoice review and intelligent settlement system for vehicle roadside assistance according to claim 5, characterized in that: The anti-counterfeiting recognition module uses image segmentation technology to extract the seal area in the invoice, and uses the SIFT or SURF algorithm to extract the feature points of the seal area. The extracted feature points include: key point position, scale, direction, and descriptor information for subsequent feature matching; Load the real seal image registered by the invoicing unit from the historical seal library, compare the feature points of the invoice seal uploaded with the feature points of the historical seal image, and judge the matching degree through the feature point matching algorithm; If the matching is qualified, compare the offset of the seal in the invoice image with the preset standard position. If the offset is greater than the offset threshold, mark it as abnormal; Use the image overlap algorithm to judge whether the seal overlaps with other elements, and use the image blurriness measurement algorithm to evaluate whether the text in the seal is blurred.

7. An automatic invoice review and intelligent settlement system for automobile road rescue according to claim 6, characterized in that: The storage module deploys a multi-node consortium chain network including: the platform operator node, the third-party audit node, and the core supplier node; Use the identity authentication module to manage the access rights of each participant, and set the smart contract template for standardizing the storage rules of bill checking, confirmation, modification, and settlement status; Each bill checking process will automatically generate a checking transaction object. The checking transaction object generates a data fingerprint through a hash digest. The data fingerprint and the original data are packaged together and submitted to the consortium chain network for storage; Each node participates in the consensus to confirm the validity of the checking transaction. If the consensus passes, the data is packaged into a block to form a chain record.

8. An automatic invoice review and intelligent settlement system for automobile roadside assistance according to claim 7, characterized in that: The prompt module obtains the basic data and historical cooperation information of each supplier from the database, including the total amount of cumulative settlement, cooperation years, settlement exception rate, and response timeliness of the supplier, and forms a portrait of each supplier; Normalize all the obtained parameters, then calculate the comprehensive score, and divide all suppliers into multiple levels according to the comprehensive score, including level A, level B, level C, and level D, to form a priority matrix.

9. The automatic invoice review and intelligent settlement system for automobile road rescue according to claim 8, characterized in that: The correction module automatically collects historical bill data and the corresponding GPS rescue trajectory data. The historical bill includes rescue time, service type, settlement amount, supplier information, and service area. The GPS rescue trajectory data includes the positioning points before and after vehicle rescue, driving path, and rescue start and end time information; The machine learning model embedded in the correction module trains the preprocessed data, extracts the normal rescue frequency, rescue type, and amount distribution patterns in each region and each time period, identifies the feature boundaries of normal settlement behaviors, and constructs a benchmark behavior pattern library; When the new bill data enters the audit process, compare it with the benchmark behavior pattern to identify the settlement items with high abnormal incidence.

10. The automatic invoice review and intelligent settlement system for vehicle roadside assistance according to claim 9, characterized in that: The correction module constructs a feature vector: , where is the feature vector of the i-th rescue,[[]]END]] is the duration of the i-th rescue,[[]]END]] is the distance of the i-th rescue,[[]]END]] is the amount of the i-th rescue; After performing K-Means clustering on all feature vectors, all clusters are obtained, and the total within-class distance squared sum is minimized: , where is the total within-class distance squared sum,[[]] is the number of clustering categories,[[]] is the sample set included in the k-th cluster,[[]] is the centroid of the k-th cluster; After obtaining a new bill, divide the new bill into the corresponding clustering cluster according to the feature vector of the new bill, and obtain the deviation value by subtracting the amount corresponding to the centroid of the clustering cluster from the new bill amount. If the deviation value is greater than the deviation threshold, determine that the new bill is abnormal; Readjust the new bill amount, with the expression: , where is the amount after the new bill adjustment, is the amount corresponding to the centroid of the cluster, is the standard deviation of the cluster amount, is the adjustment factor, and .

Images