Accounting information input system and input method thereof

Abstract

The invention discloses an accounting information entry system and an entry method thereof, which significantly improve the efficiency and accuracy of data processing, realize full-process automatic processing from data acquisition to storage by a multi-module cooperative working mechanism, greatly improve the overall processing speed, and improve the efficiency and accuracy of data processing. The intelligent recognition module adopts a deep learning technology and a context understanding capability, greatly improves the accuracy of character recognition, especially for complex documents such as handwritten documents and irregular typesetting, the recognition accuracy has qualitative leap compared with a traditional OCR technology, a multi-dimensional data verification mechanism is adopted, and the recognition efficiency is improved. Comprising mathematical verification, logic verification, historical comparison and rule verification, an all-around data quality guarantee system is formed, the data error rate is reduced to an extremely low level, and an adaptive learning mechanism can continuously optimize an identification and classification model, so that the system is kept and the accuracy is continuously improved in a long-term use process; the capacity of the continuous improvement is far better than that of a traditional static system.

Description

Technical Field

[0001] This invention relates to the field of financial technology, specifically to an accounting information entry system and its entry method. Background Technology

[0002] With the rapid development of financial services and the deepening of digital transformation, the accurate and efficient entry of accounting information has become a key requirement for financial institutions. Currently, accounting information entry mainly employs two methods: traditional manual entry and emerging intelligent entry systems. Traditional manual entry relies on operators manually inputting data, while intelligent entry systems utilize technologies such as Optical Character Recognition (OCR) and Natural Language Processing (NLP) to automatically extract and process document information. Some advanced systems also incorporate machine learning algorithms to improve the accuracy of data classification and verification.

[0003] However, existing accounting information entry methods still have many shortcomings. Manual entry is inefficient, prone to errors, and difficult to cope with large-scale data processing needs. Although intelligent systems have improved efficiency to some extent, they still face the following problems: (1) The accuracy of intelligent recognition is unstable, especially for complex or non-standard format documents; (2) The flexibility of data classification and verification is insufficient, making it difficult to adapt to diverse business scenarios; (3) There is a lack of effective exception handling mechanisms, often requiring a lot of manual intervention; (4) The system has poor scalability, making it difficult to cope with rapidly changing business needs and data volume growth; (5) Data security and privacy protection measures are not perfect, posing potential risks.

[0004] Given the limitations of existing technology, there is an urgent need to develop a new type of accounting information entry system.

[0005] The system should possess higher intelligent recognition accuracy, greater data processing flexibility, a more robust anomaly handling mechanism, better scalability, and stricter data security and privacy protection measures. These improvements will not only significantly enhance the efficiency and accuracy of accounting information entry but also provide financial institutions with a secure, reliable, and scalable data processing solution, effectively supporting the digital transformation and refined management of financial operations. Summary of the Invention

[0006] The purpose of this invention is to provide an accounting information entry system and method to solve the problems existing in the current accounting information entry system mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an accounting information entry system and its entry method, comprising: The system includes a data acquisition module, an intelligent recognition module, a data classification module, a data verification module, a manual review module, and a data storage module. The data acquisition module and the intelligent recognition module are connected through a data preprocessing interface; The intelligent recognition module is connected to the data classification module through a feature extraction channel; The data classification module is connected to the data verification module via a classification result transmission bus; The data verification module is connected to the manual review module through a two-way verification feedback channel; Both the manual review module and the data verification module are connected to the data storage module via a secure data transmission protocol. The data storage module adopts a distributed storage architecture, which includes data sharding and replication mechanisms, and is used to store processed accounting data, processing logs of each module, and intermediate results.

[0008] Preferably, the intelligent recognition module includes: an image preprocessing unit, a deep learning recognition unit, a context understanding unit, and a model update unit; in, The image preprocessing unit enhances the input image data and transmits the processing result to the deep learning recognition unit. The deep learning recognition unit processes the enhanced image based on an attention mechanism and outputs preliminary recognition results to the context understanding unit. The context understanding unit combines the historical classification information provided by the data classification module to perform semantic analysis and supplementation on the preliminary recognition results, forming the final recognition results. The model update unit continuously evaluates the recognition accuracy, and when the accuracy decreases, it triggers parameter adjustments in the deep learning recognition unit.

[0009] Preferably, the data classification module includes: a keyword matching unit, a pattern recognition unit, a machine learning classification unit, and an adaptive learning unit; in, The keyword matching unit and the pattern recognition unit process the output of the intelligent recognition module in parallel, and transmit the preliminary classification results to the machine learning classification unit. The machine learning classification unit integrates the results of keyword matching and pattern recognition, and combines them with the semantic information provided by the intelligent recognition module to generate a comprehensive classification result; The adaptive learning unit dynamically adjusts the parameters and weights of other units based on feedback from the manual review module, thereby optimizing the classification performance.

[0010] Preferably, the data verification module includes: a mathematical verification unit, a logical verification unit, a historical comparison unit, a rule verification unit, and an intelligent early warning unit; in, The mathematical verification unit, logical verification unit, historical comparison unit and rule verification unit perform multi-dimensional verification of the data in parallel, and the verification results of each unit are summarized into the intelligent early warning unit. The intelligent early warning unit calculates the comprehensive credibility score of the data based on multi-dimensional verification results. When the credibility score is lower than the preset threshold, the intelligent early warning unit generates a detailed anomaly report and transmits the data and report to the manual review module through a two-way verification feedback channel. The rule verification unit maintains a real-time connection with the data storage module to ensure that the verification rules are updated immediately.

[0011] Preferably, an accounting information entry system further includes: a data encryption module, an access control module, an audit log module, and a performance monitoring module; in, The data encryption module provides end-to-end encryption services during data transmission and storage; The access control module interacts with all other modules to verify permissions for each operation request; The audit log module continuously records the operation logs of each module in the system, and ensures the integrity and immutability of the logs through a chain structure; The performance monitoring module collects the performance indicators of each component of the system in real time, predicts performance bottlenecks through data analysis, and triggers the resource scheduling mechanism.

[0012] Preferably, an accounting information entry system includes the following steps in its entry method: S1: Receive multi-source heterogeneous accounting data through the data acquisition module; S2: Perform type determination and preprocessing on the received data; S3: Utilize the intelligent recognition module to perform optical character recognition and semantic understanding; S4: Use the data classification module to perform multi-dimensional adaptive classification; S5: Perform multi-level data verification through the data verification module; S6: Perform data splitting processing based on the verification results; S7: Use the manual review module to process abnormal data; S8: The processed data is persistently stored through the data storage module.

[0013] Preferably, step S2 specifically includes: S21: When the data is determined to be real-time data, it will be immediately transmitted to the intelligent recognition module; S22: When the data is determined to be historical data, the data is stored in a temporary buffer and subsequent processing is triggered according to the preset scheduling strategy. The preset scheduling strategy takes into account the current system load status and data processing priority.

[0014] Preferably, step S3 specifically includes: S31: Adaptive preprocessing of image data, including noise reduction, tilt correction and contrast enhancement; S32: Select an appropriate deep learning model for character recognition based on image features; S33: Utilize contextual information for semantic understanding and entity recognition; S34: Automatically adjust model parameters based on recognition accuracy: S341: When the accuracy is higher than the preset threshold, maintain the current model parameters; S342: When the accuracy is lower than the preset threshold, the model parameter fine-tuning process is triggered.

[0015] Preferably, step S5 specifically includes: S51: Parallel execution of mathematical verification, logical verification, historical comparison, and rule verification; S52: Calculate data credibility using a weighted scoring mechanism based on multi-dimensional verification results; S53: Categorize based on credibility score: S531: When the confidence level is higher than the high confidence threshold, the data is directly verified; S532: When the confidence level is between the high and low confidence thresholds, it is marked as data to be reviewed; S533: When the confidence level is below the low confidence threshold, the data along with a detailed anomaly report will be sent to the manual review module.

[0016] Preferably, a method for entering accounting information also includes system performance monitoring and security management steps: S9: Continuously monitor the performance indicators of each module in the system; S10: Use predictive analytics algorithms to identify potential performance bottlenecks; S11: Automatically adjust system resource allocation based on prediction results: S111: When prediction performance is normal, maintain the current resource configuration; S112: When a performance bottleneck is predicted, a dynamic resource expansion mechanism is triggered; S12: Regularly perform security vulnerability scans and risk assessments. S121: Take appropriate defensive measures against detected security threats; S122: Generate a comprehensive security audit report; S13: Implement appropriate privacy protection measures based on the data sensitivity level.

[0017] Compared with the prior art, the beneficial effects of the present invention are: First, this invention significantly improves the efficiency and accuracy of data processing. The multi-module collaborative working mechanism realizes fully automated processing from data collection to storage, greatly improving the overall processing speed. The intelligent recognition module adopts deep learning technology and contextual understanding capabilities, which greatly improves the accuracy of text recognition. In particular, for complex documents such as handwritten text and irregular layouts, the recognition accuracy has made a qualitative leap compared with traditional OCR technology. The multi-dimensional data verification mechanism, including mathematical verification, logical verification, historical comparison and rule verification, forms a comprehensive data quality assurance system, reducing the data error rate to an extremely low level. The adaptive learning mechanism can continuously optimize the recognition and classification model, enabling the system to maintain and continuously improve its accuracy during long-term use. This ability to continuously improve far exceeds that of traditional static systems. Secondly, this invention enhances the system's flexibility and intelligence. The modular design gives the system excellent scalability, allowing for the rapid addition or modification of functional modules according to the specific needs of different financial institutions, significantly shortening the system's deployment and upgrade cycle. The adaptive learning unit can dynamically adjust the classification strategy based on actual usage, flexibly adapting to different business scenarios. The intelligent early warning unit, based on multi-dimensional verification results, can intelligently identify potential data anomalies, greatly improving the early detection rate of problematic data. The processing of abnormal data combines automated processes with manual review, improving efficiency while ensuring accuracy and greatly reducing the need for manual intervention. This high degree of intelligence and automation not only significantly improves operational efficiency but also greatly reduces the risk of human error, enabling the system to cope with more complex and ever-changing business environments. Third, this invention significantly improves data security and system reliability. End-to-end data encryption ensures data security during transmission and storage. The attribute-based fine-grained access control mechanism effectively prevents unauthorized access and enables precise field-level permission management, greatly enhancing data security. The tamper-proof audit log system uses blockchain technology to ensure the traceability of all operations, providing a solid guarantee for system security. The distributed storage architecture and data replication mechanism significantly improve the system's fault tolerance and data reliability. Even in the event of partial hardware failure, the system can still maintain extremely high availability. Continuous performance monitoring and automated resource scheduling ensure stable system operation. Through predictive analysis, the system can identify most potential performance bottlenecks in advance and automatically adjust resources. These security and reliability measures not only meet stringent regulatory requirements but also provide comprehensive protection for the data assets of financial institutions. Finally, this invention optimizes overall operating costs and promotes business innovation. The highly automated processing flow significantly reduces manual operation costs and substantially lowers the required human resource investment. The intelligent resource scheduling mechanism optimizes hardware resource utilization and significantly saves on hardware investment. The high accuracy and robust anomaly handling mechanism greatly reduce the workload of subsequent data correction and significantly reduce financial losses caused by data errors. In addition, the system's efficiency and accuracy provide financial institutions with a reliable data foundation, strongly supporting the design and risk management of more complex financial products. The powerful data processing capabilities enable financial institutions to respond more quickly to market changes and launch new business models, thereby maintaining a leading position in the highly competitive financial market. This enhanced innovation capability has a profound impact on the long-term development of financial institutions. Attached Figure Description

[0018] Figure 1 This is an overall architecture diagram of an accounting information entry system according to the present invention; Figure 2 This is a flowchart of the intelligent recognition module of the present invention; Figure 3 This is a flowchart of the data verification module of the present invention; Figure 4 This is the main data processing flowchart of the present invention; Figure 5 This is a flowchart illustrating the system performance monitoring and optimization process of the present invention. Figure 6 This is a flowchart of an accounting information entry method according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In the description of the invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front end," "rear end," "both ends," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0021] In the description of the invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] Please see Figure 1-6 The present invention provides a technical solution: an accounting information entry system, comprising: The system includes a data acquisition module, an intelligent recognition module, a data classification module, a data verification module, a manual review module, and a data storage module. The data acquisition module and the intelligent recognition module are connected through a data preprocessing interface; The intelligent recognition module is connected to the data classification module through a feature extraction channel; The data classification module is connected to the data verification module via the classification result transmission bus; The data verification module is connected to the manual review module through a two-way verification feedback channel; Both the manual review module and the data verification module are connected to the data storage module via a secure data transmission protocol. The data storage module adopts a distributed storage architecture, which includes data sharding and replication mechanisms, and is used to store processed accounting data, processing logs of each module, and intermediate results.

[0023] Specifically, the data acquisition module receives raw accounting data from multiple data sources, achieving unified collection of heterogeneous data from multiple sources and providing a foundation for subsequent processing. The intelligent recognition module performs optical character recognition and natural language processing on the received data, significantly improving the accuracy of text recognition, especially for complex documents. The data classification module performs multi-dimensional classification based on the recognition results and metadata, improving the accuracy of data organization and facilitating subsequent processing and analysis. The data verification module checks the compliance and integrity of the classified data, ensuring data quality and reducing the error rate. The manual review module handles abnormal data requiring human intervention, resolving complex anomalies and improving the overall accuracy of the system. The data storage module stores the finally processed data into a distributed database system, ensuring secure storage and efficient access to the data.

[0024] Specifically, by using the data acquisition module, intelligent recognition module, data classification module, data verification module, manual review module, and data storage module in combination, the entire process from data acquisition to storage is automated, which significantly improves efficiency. The collaborative work of multiple modules greatly enhances the accuracy of data processing, and the use of a distributed storage architecture enhances the reliability and scalability of the system.

[0025] The intelligent recognition module includes: an image preprocessing unit, a deep learning recognition unit, a context understanding unit, and a model update unit; in, The image preprocessing unit enhances the input image data and transmits the processing results to the deep learning recognition unit; The deep learning recognition unit processes the enhanced image based on an attention mechanism and outputs preliminary recognition results to the context understanding unit. The context understanding unit combines the historical classification information provided by the data classification module to perform semantic analysis and supplementation on the preliminary recognition results, forming the final recognition results. The model update unit continuously evaluates the recognition accuracy, and when the accuracy decreases, it triggers parameter adjustments in the deep learning recognition unit.

[0026] Specifically, the image preprocessing unit performs denoising, correction, and enhancement on the received image data, which can improve the accuracy of subsequent recognition; the deep learning recognition unit uses a multi-model ensemble method to recognize text and tables to achieve high-accuracy text recognition, especially for complex documents; the context understanding unit combines historical classification information fed back by the data classification module to perform semantic understanding, which can improve the semantic accuracy of the recognition results; and the model update unit continuously monitors the recognition accuracy and triggers the model fine-tuning process to maintain the long-term high performance of the system.

[0027] Specifically, the multi-unit collaboration of the intelligent recognition module significantly improves the accuracy of text recognition, the adaptive learning mechanism ensures continuous optimization of system performance, and the contextual understanding capability enhances the system's ability to process complex documents.

[0028] The data classification module includes: a keyword matching unit, a pattern recognition unit, a machine learning classification unit, and an adaptive learning unit; in, The keyword matching unit and the pattern recognition unit process the output of the intelligent recognition module in parallel, and transmit the preliminary classification results to the machine learning classification unit; The machine learning classification unit integrates the results of keyword matching and pattern recognition, and combines them with the semantic information provided by the intelligent recognition module to generate a comprehensive classification result; The adaptive learning unit dynamically adjusts the parameters and weights of other units based on feedback from the manual review module, thereby optimizing the classification performance.

[0029] Specifically, the keyword matching unit performs preliminary classification based on a pre-defined multi-level keyword dictionary, quickly completing basic classification and improving overall efficiency; the pattern recognition unit identifies common patterns and structures in the data, capturing the structured features of the data and enhancing classification accuracy; the machine learning classification unit comprehensively considers the preliminary classification results and the semantic information provided by the intelligent recognition module to perform refined classification, achieving high-precision data classification; and the adaptive learning unit analyzes feedback from the manual review module, dynamically adjusting the weights and parameters of each classification unit, continuously optimizing the classification strategy and improving the long-term performance of the system.

[0030] Specifically, the multi-dimensional classification strategy of the data classification module significantly improves the accuracy and flexibility of classification, the adaptive learning mechanism enables the system to continuously optimize and adapt to different business scenarios, and the combination of machine learning and rules balances efficiency and accuracy.

[0031] The data verification module includes: a mathematical verification unit, a logical verification unit, a historical comparison unit, a rule verification unit, and an intelligent early warning unit; in, The mathematical verification unit, logical verification unit, historical comparison unit and rule verification unit perform multi-dimensional verification of the data in parallel, and the verification results of each unit are summarized into the intelligent early warning unit. The intelligent early warning unit calculates the comprehensive credibility score of the data based on multi-dimensional verification results. When the credibility score is lower than the preset threshold, the intelligent early warning unit generates a detailed anomaly report and transmits the data and report to the manual review module through a two-way verification feedback channel. The rule verification unit maintains a real-time connection with the data storage module to ensure that the verification rules are updated immediately.

[0032] Specifically, the mathematical verification unit checks the accuracy of numerical calculations, such as summation and balancing, and transmits the verification results to the intelligent early warning unit to ensure the accuracy of the numerical data. The logical verification unit checks the logical relationships between data, such as debit and credit balance and account relationships, and transmits the verification results to the intelligent early warning unit, ensuring the logical consistency of the data. The historical comparison unit compares and analyzes current data with historical data to identify abnormal fluctuations, and transmits the analysis results to the intelligent early warning unit, which can detect potential data anomalies. The rule verification unit performs compliance checks according to preset accounting standards and regulations, establishes a real-time connection with the data storage module, and transmits the verification results to the intelligent early warning unit, ensuring that the data meets the latest regulatory requirements. The intelligent early warning unit judges the credibility of the data based on multi-dimensional verification results. It receives results from all other verification units and connects with the manual review module. Through the intelligent early warning unit, data quality can be comprehensively evaluated, and anomalies can be detected and reported in a timely manner.

[0033] Specifically, the multi-dimensional verification mechanism of the data verification module significantly improves the accuracy and reliability of the data, the intelligent early warning function greatly reduces the need for manual intervention, and real-time rule updates ensure that the system always complies with the latest regulatory requirements.

[0034] An accounting information entry system further includes: a data encryption module, an access control module, an audit log module, and a performance monitoring module; in, The data encryption module provides end-to-end encryption services during data transmission and storage; The access control module interacts with all other modules to verify permissions for each operation request; The audit log module continuously records the operation logs of each module in the system, and ensures the integrity and immutability of the logs through a chain structure; The performance monitoring module collects performance metrics of each system component in real time, predicts performance bottlenecks through data analysis, and triggers resource scheduling mechanisms.

[0035] Specifically, the data encryption module connects to all modules involved in data transmission to ensure data security and prevent unauthorized access; the access control module implements fine-grained permission management and interacts with all other modules to prevent unauthorized access and protect sensitive information; the audit log module captures the operation logs of each module in real time, using an immutable storage mechanism, and connects to all other modules to ensure all operations are traceable and meet compliance requirements; the performance monitoring module continuously collects performance indicators of each system component, predicts potential performance bottlenecks, and ensures stable system operation, enabling proactive maintenance.

[0036] Specifically, by adding data encryption, access control, audit log, and performance monitoring modules, comprehensive security measures significantly improve the system's data protection capabilities. Fine-grained access control enhances the system's security and compliance, while performance monitoring and automatic optimization mechanisms ensure the system's long-term stable operation.

[0037] An accounting information entry system, the entry method of which includes the following steps: S1: Receives heterogeneous accounting data from multiple sources through the data acquisition module; S2: Perform type determination and preprocessing on the received data; S3: Utilizes the intelligent recognition module to perform optical character recognition and semantic understanding; S4: Use the data classification module for multi-dimensional adaptive classification; S5: Perform multi-level data validation through the data validation module; S6: Perform data splitting processing based on the verification results; S7: Use the manual review module to handle abnormal data; S8: Persistently store the processed data through the data storage module.

[0038] Specifically, steps S1-S8 realize the complete process from data reception to storage, achieving full-process automation and significantly improving efficiency. Multi-step verification ensures the accuracy and reliability of the data, and the manual review process handles complex anomalies, improving the system's adaptability.

[0039] Step S2 specifically includes: S21: When the data is determined to be real-time data, it will be immediately transmitted to the intelligent recognition module; S22: When the data is determined to be historical data, the data is stored in a temporary buffer and subsequent processing is triggered according to the preset scheduling strategy. The preset scheduling strategy takes into account the current system load status and data processing priority.

[0040] Specifically, by using different processing methods for real-time and historical data, the system's resource utilization has been optimized, processing efficiency has been improved, and flexible scheduling strategies have enhanced the system's adaptability.

[0041] Step S3 specifically includes: S31: Adaptive preprocessing of image data, including noise reduction, tilt correction and contrast enhancement; S32: Select an appropriate deep learning model for character recognition based on image features; S33: Utilize contextual information for semantic understanding and entity recognition; S34: Automatically adjust model parameters based on recognition accuracy: S341: When the accuracy is higher than the preset threshold, maintain the current model parameters; S342: When the accuracy is lower than the preset threshold, the model parameter fine-tuning process is triggered.

[0042] Specifically, through each step of intelligent recognition, adaptive image preprocessing improves recognition accuracy, dynamic selection of recognition models enhances system flexibility, and the automatic model update mechanism ensures long-term high performance.

[0043] Step S5 specifically includes: S51: Parallel execution of mathematical verification, logical verification, historical comparison, and rule verification; S52: Calculate data credibility using a weighted scoring mechanism based on multi-dimensional verification results; S53: Categorize based on credibility score: S531: When the confidence level is higher than the high confidence threshold, the data is directly verified; S532: When the confidence level is between the high and low confidence thresholds, it is marked as data to be reviewed; S533: When the confidence level is below the low confidence threshold, the data along with a detailed anomaly report will be sent to the manual review module.

[0044] Specifically, through multi-dimensional verification and credibility-based classification, multi-dimensional verification significantly improves the accuracy of data, credibility-based classification optimizes the efficiency of manual review, and automated anomaly detection reduces human error.

[0045] An accounting information entry method also includes system performance monitoring and security management steps: S9: Continuously monitor the performance indicators of each module in the system; S10: Use predictive analytics algorithms to identify potential performance bottlenecks; S11: Automatically adjust system resource allocation based on prediction results: S111: When prediction performance is normal, maintain the current resource configuration; S112: When a performance bottleneck is predicted, a dynamic resource expansion mechanism is triggered; S12: Regularly perform security vulnerability scans and risk assessments. S121: Take appropriate defensive measures against detected security threats; S122: Generate a comprehensive security audit report; S13: Implement appropriate privacy protection measures based on the data sensitivity level.

[0046] Specifically, through system performance monitoring, resource scheduling, and security management processes, predictive analytics improves system stability, dynamic resource scheduling optimizes system performance, and comprehensive security measures enhance data protection capabilities.

[0047] Example 1 Implementation Case of an Accounting Information Entry System and Entry Method of a Large Commercial Bank This case study details the practical application of this invention in a regional branch of a bank. This branch processes approximately 100,000 transactions daily, involving both personal and corporate business. The following is a breakdown of the system's operation on Saturday, April 15, 2023.

[0048] S1: Receives multi-source heterogeneous accounting data through the data acquisition module. Data collection status for the day: 1. Over-the-counter transaction system (core banking system): Number of transactions: 42,367 Data format: Proprietary binary format (.dat file) Data size: 2.1 GB Data collection time: 23:05-23:17 (batch export) 2. Online banking platforms: Number of transactions: 53,842 Data format: JSON Data size: 1.8 GB Data collection method: Real-time API interface, batch data collection every 5 minutes. 3. Interbank Settlement System: Number of transactions: 8,791 Data format: XML (following the CNAPS2 standard of the People's Bank of China) Data size: 650 MB Data collection time: 22:00-22:10 (after clearing by the People's Bank of China) Practical problems and solutions: At 23:09, the core system export was interrupted due to network fluctuations. The system automatically retried 3 times and succeeded.

[0049] The online banking data experienced a delay between 13:20 and 13:35. The system automatically adjusted to retrieve the data once per minute until the data stream returned to normal.

[0050] S2: Perform type determination and preprocessing on the received data. 1. Real-time data processing: Online banking transactions: 49,726 out of 53,842 transactions were processed in real time. Average processing latency: 1.2 seconds 2. Batch data processing: Over-the-counter transactions: All 42,367 transactions were processed in batches. Interbank settlement: All 8,791 transactions were processed in batches. 4,116 online banking transactions (mainly recurring transfers) were processed in batches. Batch processing start time: 23:30 Preprocessing steps: 1. Data cleaning: 326 duplicate records were removed. 1,205 date format errors have been corrected (e.g., "2023 / 4 / 15" should be corrected to "2023-04-15"). 78 missing customer ID records have been added (by querying associated accounts). 2. Data standardization: Convert all amounts to integers in units of cents. The unified encoding is UTF-8. 3. Data encryption: Use the bank's RSA public key to encrypt sensitive information (such as ID card number and mobile phone number). Practical problems and solutions: The system detected 50 cross-bank transactions with incorrect counterparty bank numbers and automatically queried the latest bank number table to correct them.

[0051] Two counter transactions had abnormal amounts (exceeding the marked amount) and were marked by the system as pending manual review.

[0052] S3: Utilizes an intelligent recognition module to perform optical character recognition and semantic understanding. This step mainly processes paper receipts from over-the-counter transactions, totaling 3,567 receipts.

[0053] 1. Image preprocessing: Image enhancement using the OpenCV library Average processing time: 0.3 seconds / image Tilt correction: 526 images were detected that required correction, with an average correction angle of 3.2°. 2. Character recognition: Text localization using a custom-trained YOLOv5 model Text recognition using a Transformer-based OCR model Average recognition time: 1.2 seconds / image Overall recognition accuracy: 99.7% 3. Semantic understanding: Named entity recognition using a BERT model pre-trained in the financial field. Identify key information: account number, amount, date, transaction type, etc. Average processing time: 0.5 seconds / image 4. Model Update: The 52 images whose recognition accuracy was below the threshold on that day were marked for the next model update. Incremental model training is performed every Sunday at 3:00 AM. Practical problems and solutions: Fifteen receipts were marked as "requiring manual review" because the stamps were blurry and difficult to identify.

[0054] The two receipts were in the new format, which the system successfully learned and recognized, and the new format was added to the template library.

[0055] S4: Use the data classification module for multi-dimensional adaptive classification Data classification process: 1. Keyword matching: Using a dictionary containing 12,735 financial terms It uses a double-array Trie tree structure, with a memory footprint of 138MB and an average matching time of 0.5 milliseconds per match. 2. Pattern Recognition: 7,892 payroll transactions were identified (characteristics: multiple small transfers, mostly to personal accounts). 3,156 recurring transfer transactions were identified (characteristics: fixed time, fixed amount, fixed counterparty). 3. Classification of Machine Learning: Fine-grained classification using the LightGBM model Feature dimensions: 187, including transaction amount, frequency, customer segmentation, historical behavior, etc. Classification accuracy: 99.85% 4. Adaptive learning: Based on feedback from manual review that day, 23 classification rules were updated. Online model updates: Incremental learning is performed after every 1,000 transactions. The classification results are shown in the table below:

[0056] Practical problems and solutions: A newly added wealth management product ("Smart Deposit") was initially incorrectly classified as a regular deposit. After manual correction, the system automatically added it to the special classification rules.

[0057] A new type of telecom fraud was detected that day, and the risk rule base was immediately updated and the transactions of that day were reclassified.

[0058] S5: Perform multi-level data validation via the data validation module. Data validation process: 1. Mathematical verification: Perform a loan balance check on all 100,000 transactions. Three transactions were found to have discrepancies in lending, with a total difference of 1,235.67 yuan. 2. Logical verification: 267 transactions were found to exceed the individual account single transaction limit (500,000 yuan). 18 transactions were found to have unusual timing (outside of business hours). 3. Historical Comparison: An ARIMA model was built using data from the past 90 days. The system detected that the total daily transfer volume increased by 23% compared to expectations, triggering an anomaly alert. 4. Rule Validation: Applying the latest anti-money laundering regulations, 126 large transactions (a single transaction exceeding 500,000 yuan or a cumulative total exceeding 2 million yuan in a single day) were flagged as requiring reporting. All 15 company transfers exceeding 10 million yuan were reviewed to confirm whether there were corresponding approval records. 5. Intelligent early warning: Using the Isolation Forest algorithm, 89 potentially abnormal transactions were detected. A risk score is calculated for each transaction, and the distribution is as follows:

[0059] Practical problems and solutions: Upon discovering a large, excessive transfer, the system automatically froze the transaction and notified bank executives. After verification that it was a multinational corporate acquisition payment, it was processed after obtaining proper authorization.

[0060] The system detected 50 small, high-frequency transfers from the same IP address, automatically flagged the relevant accounts as suspicious, and reported them to the anti-money laundering department.

[0061] S6: Perform data splitting based on the verification results. Based on the verification results, the system will distribute the transactions as follows: 1. Direct approval: 97,526 transactions (risk score < 0.3) These transactions require no human intervention and proceed directly to the subsequent processing steps. 2. Pending review: 2,385 cases (0.3 ≤ risk score < 0.7) These transactions were marked as requiring review, but processing was allowed to continue. The system generates review tasks and sorts them by priority. 3. Manual review: 89 cases (risk score ≥ 0.7) These transactions were immediately blocked and prioritized for manual review. The system automatically assigned these transactions to six dedicated reviewers. Post-splitting processing status: Transactions approved directly completed their follow-up processing within 10 minutes. Of the 2,103 transactions awaiting review, the review and processing were completed within 2 hours. All high-risk transactions underwent manual review within one hour. Practical problems and solutions: Around 23:45, the system detected a backlog in the queue of reviews and automatically notified two additional reviewers to join the processing.

[0062] S7: Handle abnormal data using the manual review module. Manual review status: 1. Auditor staffing: Routine audit: 6 full-time auditors Expedited support: 2 senior auditors (deputy manager level) Special case: 1 compliance officer, 1 risk control manager 2. Review interface functions: Dual-screen display: Transaction details on the left, and relevant historical transactions and customer information on the right. Intelligent suggestions: The system provides processing suggestions based on historical cases. Quick actions: One-click approval, rejection, reporting, or requesting more information. 3. Audit Results:

[0063] 4. Special Case Handling: One suspected money laundering transaction: A single excessive amount was broken up into 100 smaller transfers. The system automatically froze the relevant accounts and notified the anti-money laundering department in real time.

[0064] 5. Review Feedback: Each review decision is recorded and used for system learning. The review results that day led to the updating of 15 business rules, and the parameters of 3 machine learning models were also adjusted accordingly. Practical problems and solutions: Around 00:30, an auditor encountered a system lag, and IT support resolved the issue remotely within 5 minutes (the cause was a browser cache overflow).

[0065] S8: Persistently store the processed data through the data storage module. Stored procedures: 1. Data encryption: All sensitive fields are encrypted using the national cryptographic algorithm SM4. Encryption keys are managed through a Hardware Security Module (HSM). Average encryption time: 0.5 milliseconds / message 2. Data sharding: Sharding based on transaction date and account hash value Consistent hashing is used to ensure that data is evenly distributed across the 12 storage nodes. The size of a single fragment should be kept below 100GB. 3. Real-time backup: After the primary storage write is completed, an asynchronous backup is immediately performed at the local backup center. Average backup latency: 50 milliseconds Daily success rate: 99.9999% (Only one transaction was delayed due to network jitter and was automatically retried and succeeded within 5 seconds) 4. Data archiving: Data older than 3 months is automatically archived to the nearline storage system. Using compression algorithms, the average compression ratio reaches 1:5. Data archived that day prior to January 15, 2023, totaled 2.3TB. The archiving process takes place between 2:00 AM and 4:00 AM, and will not affect normal business operations. 5. Data Index: Use a B+ tree index structure to create indexes on key fields such as transaction date, account number, and amount. Index updates are performed incrementally, every 5 minutes. Approximately 500,000 new index entries were added that day. 6. Storage performance: Write speed: Average 8,000 rows / second, peak 12,000 rows / second Read speed: Average 50,000 records / second, peak 80,000 records / second Storage latency: 99% of write operations are completed within 10ms. Practical problems and solutions: At 1:37 AM, a storage node triggered a hard drive failure alarm. The system automatically migrated the data from that node to a hot standby node. The entire process took 12 minutes, during which there was no service interruption.

[0066] During the data archiving process, some data indexes from December 2022 were found to be abnormal. The system automatically initiated a repair program and completed the index reconstruction within 30 minutes.

[0067] Additional steps: System performance monitoring and security management S9: Continuously monitor the performance metrics of each module in the system. 1. Real-time monitoring metrics: CPU utilization: average 65%, peak 82% (occurring during batch processing from 23:30 to 23:45). Memory usage: Average 75%, Peak 88% Network throughput: Average 1.2 GB / s, peak 2.5 GB / s Disk I / O: Average 3,000 IOPS, Peak 5,500 IOPS 2. Module performance tracking: Data acquisition module: Average processing time 1.5ms / data entry Intelligent recognition module: Average processing time 1.8s / item (including OCR and semantic understanding) Data classification module: Average processing time 5ms / record Data validation module: Average processing time 8ms / data entry 3. Alarm Settings: An alarm is triggered when CPU usage exceeds 90% for 5 consecutive minutes. Memory usage > 95% triggers an alarm. An alarm is triggered if the average processing time of any module exceeds a preset threshold by 200%. Actual situation: At 23:42, a high CPU utilization alarm was triggered. The system automatically expanded by 2 compute nodes, and the load returned to normal within 5 minutes.

[0068] Three non-critical alarms were triggered throughout the day, all of which were handled automatically without manual intervention.

[0069] S10: Use predictive analytics algorithms to identify potential performance bottlenecks. 1. Load forecasting: Using an LSTM model, predict the system load for the next 24 hours based on historical data from the past 90 days. Prediction accuracy: 93% (root mean square error 7%) 2. Performance Trend Analysis: The linear regression model was used to analyze the performance trends of each module over time. The processing time of the data validation module has shown a slight upward trend over the past week (an average increase of 0.1ms per day). 3. Bottleneck identification: By comparing the processing capabilities and predicted load of each module, potential bottlenecks can be identified. The system predicts that the data verification module may become a bottleneck next Monday (early month business peak). S11: Automatically adjust system resource allocation based on prediction results 1. Resource elastic scaling: Based on load forecasting, the system automatically added two data processing nodes at 21:00. At 3:00 AM, a decrease in load was detected, and one node was automatically removed. 2. Load balancing optimization: Adjust the load balancing strategy in real time to allocate more requests to higher-performing nodes. The daily load imbalance decreased from 15% in the morning to 5% in the evening. 3. Database optimization: Based on access pattern analysis, the system automatically adjusted the database indexes at 22:00. After optimization, the response time for high-frequency queries was reduced by an average of 20%. S12: Regularly perform security vulnerability scans and risk assessments. 1. Routine security scans: A comprehensive security scan is performed every 4 hours. Using the latest vulnerability database (updated at 4:00 AM that day). Three low-risk vulnerabilities were discovered that day, all of which were in non-critical components. 2. Penetration Testing: A full penetration test is conducted on the last Saturday of each month. This test simulated 23 attack scenarios, and the system successfully defended against 21 of them. Two potential risk points have been identified, and remediation plans have been developed. 3. Risk Assessment: A security risk report is generated at the end of each day, including: abnormal access statistics, data breach risk assessment, and compliance check results. Daily risk score: 92 points (out of 100), an increase of 2 points from last week. S13: Implement appropriate privacy protection measures based on the data sensitivity level. 1. Data Classification: The processed data is categorized into four levels: Public, Internal, Confidential, and Secret. Of the data processed that day, 80% was classified as internal, 15% as confidential, and 5% as secret. 2. Access Control: Enable two-factor authentication for confidential data. A total of 23 unauthorized attempts to access sensitive data were blocked that day. 3. Data anonymization: Dynamic anonymization of data used for testing and development Homomorphic encryption is used to process data that needs to be analyzed but cannot be stored in plaintext. 4. Audit trail: Log all access to and operations on sensitive data. The audit log generated that day was 1.2GB in size. Practical problems and solutions: An employee was detected attempting to export a large amount of customer data. The system automatically blocked the attempt and reported it to the security team.

[0070] When processing a cross-border transaction, the system automatically applies the data protection regulations of the relevant country to ensure compliance.

[0071] Through the detailed implementation steps described above, we can see the specific application of this invention in a real banking environment. The system not only processes a large number of daily transactions efficiently and accurately, but also demonstrates strong adaptability and security. From data collection to storage, from anomaly detection to manual review, each step reflects the system's intelligent and refined management. Furthermore, continuous performance monitoring and predictive maintenance ensure the system's stable operation, while comprehensive security measures protect data integrity and privacy. This embodiment realistically demonstrates how this invention functions in a complex banking environment, providing financial institutions with a powerful, flexible, and secure accounting information processing solution.

[0072] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An accounting information entry system, characterized in that, include: The system includes a data acquisition module, an intelligent recognition module, a data classification module, a data verification module, a manual review module, and a data storage module. The data acquisition module and the intelligent recognition module are connected through a data preprocessing interface; The intelligent recognition module is connected to the data classification module through a feature extraction channel; The data classification module is connected to the data verification module via a classification result transmission bus; The data verification module is connected to the manual review module through a two-way verification feedback channel; Both the manual review module and the data verification module are connected to the data storage module via a secure data transmission protocol. The data storage module adopts a distributed storage architecture, which includes data sharding and replication mechanisms, and is used to store processed accounting data, processing logs of each module, and intermediate results.

2. The accounting information entry system according to claim 1, characterized in that, The intelligent recognition module includes: an image preprocessing unit, a deep learning recognition unit, a context understanding unit, and a model update unit; in, The image preprocessing unit enhances the input image data and transmits the processing result to the deep learning recognition unit. The deep learning recognition unit processes the enhanced image based on an attention mechanism and outputs preliminary recognition results to the context understanding unit. The context understanding unit combines the historical classification information provided by the data classification module to perform semantic analysis and supplementation on the preliminary recognition results, forming the final recognition results. The model update unit continuously evaluates the recognition accuracy, and when the accuracy decreases, it triggers parameter adjustments in the deep learning recognition unit.

3. The accounting information entry system according to claim 1, characterized in that, The data classification module includes: a keyword matching unit, a pattern recognition unit, a machine learning classification unit, and an adaptive learning unit; in, The keyword matching unit and the pattern recognition unit process the output of the intelligent recognition module in parallel, and transmit the preliminary classification results to the machine learning classification unit. The machine learning classification unit integrates the results of keyword matching and pattern recognition, and combines them with the semantic information provided by the intelligent recognition module to generate a comprehensive classification result; The adaptive learning unit dynamically adjusts the parameters and weights of other units based on feedback from the manual review module, thereby optimizing the classification performance.

4. The accounting information entry system according to claim 1, characterized in that, The data verification module includes: a mathematical verification unit, a logical verification unit, a historical comparison unit, a rule verification unit, and an intelligent early warning unit; in, The mathematical verification unit, logical verification unit, historical comparison unit and rule verification unit perform multi-dimensional verification of the data in parallel, and the verification results of each unit are summarized into the intelligent early warning unit. The intelligent early warning unit calculates the comprehensive credibility score of the data based on multi-dimensional verification results. When the credibility score is lower than the preset threshold, the intelligent early warning unit generates a detailed anomaly report and transmits the data and report to the manual review module through a two-way verification feedback channel. The rule verification unit maintains a real-time connection with the data storage module to ensure that the verification rules are updated immediately.

5. The accounting information entry system according to claim 1, characterized in that, Also includes: Data encryption module, access control module, audit log module, and performance monitoring module; in, The data encryption module provides end-to-end encryption services during data transmission and storage. The access control module interacts with all other modules to verify permissions for each operation request; The audit log module continuously records the operation logs of each module in the system, and ensures the integrity and immutability of the logs through a chain structure; The performance monitoring module collects the performance indicators of each component of the system in real time, predicts performance bottlenecks through data analysis, and triggers the resource scheduling mechanism.

6. The accounting information entry system according to claim 1, wherein the entry method includes the following steps: S1: Receive multi-source heterogeneous accounting data through the data acquisition module; S2: Perform type determination and preprocessing on the received data; S3: Utilize the intelligent recognition module to perform optical character recognition and semantic understanding; S4: Use the data classification module to perform multi-dimensional adaptive classification; S5: Perform multi-level data verification through the data verification module; S6: Perform data splitting based on the verification results; S7: Use the manual review module to process abnormal data; S8: The processed data is persistently stored through the data storage module.

7. The accounting information entry method according to claim 6, characterized in that, Step S2 specifically includes: S21: When the data is determined to be real-time data, it will be immediately transmitted to the intelligent recognition module; S22: When the data is determined to be historical data, the data is stored in a temporary buffer and subsequent processing is triggered according to the preset scheduling strategy. The preset scheduling strategy takes into account the current system load status and data processing priority.

8. The accounting information entry method according to claim 6, characterized in that, Step S3 specifically includes: S31: Adaptive preprocessing of image data, including noise reduction, tilt correction and contrast enhancement; S32: Select an appropriate deep learning model for character recognition based on image features; S33: Utilize contextual information for semantic understanding and entity recognition; S34: Automatically adjust model parameters based on recognition accuracy: S341: When the accuracy is higher than the preset threshold, maintain the current model parameters; S342: When the accuracy is lower than the preset threshold, the model parameter fine-tuning process is triggered.

9. The accounting information entry method according to claim 6, characterized in that, Step S5 specifically includes: S51: Parallel execution of mathematical verification, logical verification, historical comparison, and rule verification; S52: Calculate data credibility using a weighted scoring mechanism based on multi-dimensional verification results; S53: Categorize based on credibility score: S531: When the confidence level is higher than the high confidence threshold, the data is directly verified; S532: When the confidence level is between the high and low confidence thresholds, it is marked as data to be reviewed; S533: When the confidence level is below the low confidence threshold, the data along with a detailed anomaly report will be sent to the manual review module.

10. The accounting information entry method according to claim 6, characterized in that, It also includes system performance monitoring and security management steps: S9: Continuously monitor the performance indicators of each module in the system; S10: Use predictive analytics algorithms to identify potential performance bottlenecks; S11: Automatically adjust system resource allocation based on prediction results: S111: When prediction performance is normal, maintain the current resource configuration; S112: When a performance bottleneck is predicted, a dynamic resource expansion mechanism is triggered; S12: Regularly perform security vulnerability scans and risk assessments. S121: Take appropriate defensive measures against detected security threats; S122: Generate a comprehensive security audit report; S13: Implement appropriate privacy protection measures based on the data sensitivity level.

Images