An intelligent tracking system for intermodal transportation of molten iron based on OCR recognition

Abstract

The application discloses an iron-water combined transportation intelligent tracking system based on OCR identification applied to the technical field of intelligent logistics informatization, builds a data supermarket of the system itself and a whole-link data blood relationship graph of 'daily demand-waybill-goods tracking', realizes full-process visualized tracking and abnormal rapid positioning of goods state, and effectively changes the state of previous information islands and difficult tracking; in addition, by additionally arranging portable intelligent input terminals, a double-track data acquisition and fusion system of 'terminal input+OCR identification' is formed, data recording on a form can be checked at the same time, data quality problems caused by careless writing, format errors and logical contradictions are greatly reduced from the most front end of data generation, data accuracy is further improved, subsequent error correction cost is greatly reduced, and the rapid establishment of a data blood relationship graph is facilitated.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics information technology, and in particular to an intelligent tracking system for rail-water intermodal transport based on OCR recognition. Background Technology

[0002] Currently, China's inland river port rail-water intermodal transport mainly adopts the traditional transshipment model. Under this model, container transport involves several stages: unloading from the ship to a truck, transporting from the truck to the port yard, loading from the port yard onto another truck, transporting from the truck to the railway freight yard, and finally loading from the railway freight yard onto a train. Examples include the integrated rail-water intermodal transport information platform disclosed in Chinese patent CN114444944B and the rail-water intermodal transport message exchange system disclosed in Chinese patent CN203522782U. However, due to the numerous intermediate stages and poor coordination, traditional rail-water intermodal transport information systems suffer from the following problems: Core technologies rely on foreign countries: core components such as databases and middleware are mostly foreign products, posing supply chain security risks. Low data collection efficiency: Paper forms such as railway timetables, waybills, and weighbridge slips rely on manual input or traditional OCR recognition, which has low recognition accuracy for forms with illegible handwriting and unclear images. Data flow is not transparent: the flow of goods between railways and ports is difficult to track in real time, and an end-to-end visualized link cannot be formed; Data quality control lags behind: Data errors are often only discovered late in the business process, resulting in high correction costs. While OCR technology is widely used in existing technologies, and is frequently employed in rail-water intermodal transport systems to replace manual data entry for batch recognition and uploading of paper forms, it is prone to errors when recognizing manually entered forms. Issues such as form quality (creases, stains) and insufficient clarity of handwriting lead to significant discrepancies between the generated data and the actual situation, making it difficult for cargo owners to clearly understand the cargo's transit information. Summary of the Invention

[0003] The core of this invention lies in enhancing the accuracy of OCR recognition and establishing a data lineage graph based on the recognized data to achieve full-chain visual tracking of goods, thereby solving the problem of unclear goods flow in existing technologies. Simultaneously, core components such as the database and middleware are replaced with domestically produced products to reduce supply chain security risks.

[0004] To solve the above problems, the present invention adopts the following technical solution.

[0005] A smart tracking system for rail-water intermodal transport based on OCR recognition includes: The domestic infrastructure layer includes domestically produced servers, domestically produced operating systems, and network security equipment; The data support layer includes domestically produced databases and middleware; The application presentation layer provides a web front-end interface for displaying recognition results, lineage charts, and real-time cargo tracking status. The business service layer includes: The intelligent acquisition and preprocessing module is used to perform quality assessment, enhancement and repair, and layout analysis on the input form images; A multimodal OCR recognition engine is used to call multiple OCR engines for collaborative recognition and to integrate business rules and semantic information to optimize the results; The data quality verification and closed-loop learning module is used to verify the recognition results in real time and iteratively optimize the recognition model through human feedback data. A data lineage analysis engine is used to build and visualize the end-to-end relationships between daily demand data, waybill data, and cargo tracking data; The data supermarket module provides unified data query, cleaning, and standardization services.

[0006] Furthermore, the intelligent data acquisition and preprocessing module includes a scenario-based processing unit, which is designed for railway timetables, weighbridge slips, and ship data. Figure 3 The following preprocessing strategies are applied to different types of business forms: A1. Perform table layout analysis, cell association, and page continuity processing on railway timetables; A2. Perform template matching, handwritten number area enhancement, and strong weight logic verification on the weighbridge slip; A3. Separate the ship diagram from the graphic and the table, and associate the container number verification algorithm with the three-dimensional stowage coordinates.

[0007] Furthermore, the multimodal OCR recognition engine includes: The multi-engine parallel calling unit is used to simultaneously call at least two OCR engines to recognize the same image region; The weighted voting fusion unit performs weighted voting on the recognition results based on the historical accuracy and current confidence of each OCR engine, and outputs the optimal text. The business rules and semantic error correction unit uses a pre-built business dictionary, format rules, and natural language processing model to correct the recognized text.

[0008] Furthermore, the data quality verification and closed-loop learning module includes: The dual verification unit scores the confidence level of the identified fields and performs cross-validation with pre-defined business rules. The manual review push unit pushes fields with low confidence or those that fail the verification to the manual review interface; The model iteration and optimization unit uses the manually verified correct results and the corresponding original images as high-quality training samples to perform incremental training on the preprocessing model and the OCR recognition model on a regular basis.

[0009] Optionally, the intelligent tracking system for rail-water intermodal transport also includes an intelligent input terminal. The intelligent input terminal includes a terminal body, with alignment micro-protrusions fixedly connected to the four corners of the upper end of the terminal body. Magnetic grooves are provided at both the left and right ends of the terminal body, and paper strips are matched and adsorbed on the magnetic grooves. The paper form has the same area as the upper surface of the terminal body, and the four corners of the paper form are beveled, with each of the four beveled corners corresponding to and matching the four alignment micro-protrusions.

[0010] Furthermore, a pressure sensing layer is provided inside the screen of the terminal body. The effective area of ​​the screen on the terminal body is completely consistent with that of the paper form. The terminal body stores an electronic form that is one-to-one with the paper form, and only the effective area of ​​the electronic form is displayed on the screen. The paper pressing strip includes a magnetic strip that matches the magnetic groove, a reinforcing rib fixedly connected to the top of the upper end of the magnetic strip, and a diagonal rod fixedly connected to the end of the reinforcing rib. The depth of the magnetic groove is no more than 2mm.

[0011] Furthermore, the terminal body also integrates: The real-time handwriting recognition and verification module is used to convert electronic handwriting into structured text data and perform real-time compliance verification and reminders. The camera module is used to take images of the paper form after it has been filled out. The data synchronization module associates the validated structured text data with the captured image data of the paper form and synchronizes it to the business service layer. The synchronized data is directly stored in the database as high-confidence structured text data, and the high-confidence structured text data and its corresponding form images constitute labeled training pairs for training the multimodal OCR recognition engine.

[0012] Furthermore, the business service layer also includes a data fusion and decision-making module, which includes: The data association unit uses a unique identifier to associate the data from the smart input terminal of the same form with the OCR recognition result; The difference calculation unit calculates the degree of difference between the data from the intelligent input terminal and the OCR recognition results in key fields; The decision-making execution unit executes one of the following preset strategies: B11. If smart input terminal data exists and the difference is less than the threshold, then the smart input terminal data shall be used as the final data. B12. If there is data from the intelligent input terminal and the difference is greater than or equal to the threshold, the manual review process will be triggered, and the review result will be the final data. B2. If no data is available from the intelligent input terminal, the OCR recognition result shall be used as the final data.

[0013] A smart tracking method for rail-water intermodal transport based on multimodal OCR and data lineage analysis includes the following steps: S1, Dual-track data acquisition: S11. Collect on-site operation data through an intelligent input terminal. The on-site operation data includes high-confidence structured text data and its corresponding form images. S12. Collect image data of historical paper forms and third-party paper documents by scanning; S2, Data Processing: S21. Perform preprocessing and enhancement processing on the acquired image data; S22. Recognize and optimize paper form images using a multimodal OCR recognition engine; S3, Data Fusion Decision-Making: S31. Compare and correlate the data from the smart input terminal of the same form with the OCR recognition results; S32. Determine the final valid data based on preset strategies, including: prioritizing smart terminal data and triggering manual review when the difference is too large; S4. Quality Verification and Optimization: S41. Verify the confidence level and business rules of the final data; S42. Use the results of manual review as training samples to iteratively optimize the OCR recognition model; S5. Establishing a lineage relationship: Based on the final valid data, extract key identifiers from daily demand data, waybill data, and cargo tracking data, establish the relationship between the three types of data, and form a data lineage relationship map; S6. Data Application and Presentation: S61. Store the final valid data and kinship map in a domestic database; S62. Provide data query, lineage visualization and real-time cargo tracking services through a web front-end interface; S63. Use kinship maps to trace the source and analyze the impact of data anomalies.

[0014] Furthermore, in step S11, the step of collecting data and images through the intelligent input terminal specifically includes: C1. Align the four corners of the paper form with the four alignment micro-protrusions on the terminal body to achieve alignment of the effective area of ​​the paper form with the effective area of ​​the electronic form displayed on the terminal body screen. Then fix the paper form to the surface of the terminal body by pressing the paper strip. C2. Staff members write normally on paper forms, and simultaneously write their handwriting on electronic forms through a pressure-sensing layer. Handwriting recognition and compliance verification are performed in real time. If the verification fails, the user is immediately notified to correct the form content. C3. After completing the form, take a picture of the form and link the validated structured text data with the form image.

[0015] Compared with the prior art, the advantages of this invention are: (1) This solution replaces the original database with a domestic database, modifies the relevant system components and interface data in a synchronous manner to achieve full compatibility, replaces the relevant middleware of the system and ensures that it supports the domestic database language, and finally completes the migration of all historical data, ensuring the continuous normal operation of the system, while effectively improving supply chain security and data security.

[0016] (2) Through multimodal OCR collaboration, the OCR recognition accuracy of complex forms has been significantly improved.

[0017] (3) Build the system’s own data supermarket to enable users to query independently according to their needs. Improve data quality and consistency through unified cleaning, transformation and standardization. Establish a data security management system with the help of access control, data encryption and other means to ensure the integrity, security and confidentiality of data during use and transmission.

[0018] (4) Through the data lineage analysis engine, a complete data relationship map of "daily demand → waybill → cargo tracking" is automatically constructed. Business personnel can view the status of any cargo from planning and shipment to its current location in real time and intuitively, effectively changing the previous situation of information silos and difficulty in traceability. In addition, it can clearly identify the source and destination of data, realize the rapid location of the source of data anomalies, and improve the level of data quality control.

[0019] (5) By adding an intelligent data entry terminal, data can be verified at the same time as data is recorded on the form. This greatly reduces data quality problems caused by illegible handwriting, incorrect format, and logical contradictions from the very beginning of data generation, further improving data accuracy and significantly reducing subsequent error correction costs.

[0020] (6) The system is compatible with both smart data entry terminals and OCR data recognition modes. It can ensure the source quality of key operational data through terminals and efficiently process historical archives and third-party documents through OCR. The data from the two modes can be compared, verified and merged to form a double insurance for data quality, ensuring the reliability and continuity of the system in different business scenarios.

[0021] (7) The high-confidence structured text data generated by the intelligent input terminal can be used as the "standard answer" to train the OCR model, and the recognition results of OCR and the feedback of manual review continuously optimize the terminal's verification rules. Attached Figure Description

[0022] Figure 1 This is a main system block diagram of the present invention; Figure 2 This is a flowchart illustrating the usage of the present invention; Figure 3 This is a flowchart illustrating the usage of the present invention when an intelligent input terminal is added; Figure 4 This is a perspective view of the intelligent input terminal of the present invention in use; Figure 5 This is a schematic diagram illustrating how the data from the paper form after recognition is associated with the data from the intelligent data entry terminal through a unique identifier. Figure 6 This is an exploded view of the intelligent input terminal of the present invention.

[0023] The numbers in the diagram are explained as follows: 1 Terminal body, 2 Alignment micro-protrusion, 101 Magnetic groove, 3 Paper pressing strip, 31 Magnetic sticker, 32 Reinforcing rib, 33 Diagonal bar. Detailed Implementation

[0024] The technical solutions will now be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention.

[0025] First implementation method: like Figure 1 A smart tracking system for rail-water intermodal transport based on OCR recognition includes: The domestic infrastructure layer includes domestic servers (configured with at least 16-core CPU, 16GB memory, and 500GB storage, supporting virtualization deployment), domestic operating systems (domestic Linux systems such as Kylin and Euler, developed in Java), and network security equipment. Among them, network security equipment mainly involves deploying domestic firewalls, intrusion prevention systems, and other security devices, and establishing a role-based access control mechanism and log auditing function. The data support layer includes domestic databases and middleware. The domestic database used is DM (DaMeng), which is widely used in key sectors such as finance and government, boasting a transaction processing capacity (TPS) of up to millions. Historical data can be completely migrated using professional migration tools. After migration, comprehensive data verification is performed to ensure accuracy. The middleware is replaced with Kingdee Tianyan, a domestic middleware. Simultaneously, system components and interfaces are adapted for compatibility, ensuring seamless integration of database-related views, stored procedures, triggers, etc., with the domestic environment. Web interface parameters and return values ​​remain consistent between the old and new systems. This system replaces the database and middleware with domestic alternatives and simultaneously modifies related system components and interfaces to fully support domestic databases, ensuring normal system operation and achieving complete historical data migration. The application presentation layer provides a web front-end interface for displaying recognition results, lineage charts, and real-time cargo tracking status. The business service layer includes an intelligent data acquisition and preprocessing module, a multimodal OCR recognition engine, a data quality verification and closed-loop learning module, and a data lineage analysis engine.

[0026] The intelligent acquisition and preprocessing module is used to perform quality assessment, enhancement and repair, and layout analysis on the input form images; The multimodal OCR recognition engine is used to call multiple OCR engines for collaborative recognition and to integrate business rules and semantic information to optimize the results. The multimodal OCR recognition engine includes: a multi-engine parallel calling unit, a weighted voting fusion unit, and a business rule and semantic error correction unit. The multi-engine parallel invocation unit is used to simultaneously invoke at least two OCR engines to recognize the same image region; the weighted voting fusion unit performs weighted voting on the recognition results based on the historical accuracy and current confidence of each OCR engine to output the optimal text; and the business rules and semantic error correction unit uses a pre-set business dictionary, format rules and natural language processing model to correct the recognized text.

[0027] The data quality verification and closed-loop learning module is used to verify the recognition results in real time and to iteratively optimize the recognition model through human feedback data. The data quality verification and closed-loop learning module includes a dual verification unit, a human review and push unit, and a model iterative optimization unit. The dual verification unit scores the confidence level of the recognition fields and cross-validates them with the preset business rules; the manual review and push unit pushes fields with low confidence or those that fail verification to the manual review interface; the model iteration and optimization unit uses the correct results confirmed by the manual review and the corresponding original images as high-quality training samples, and regularly performs incremental training on the preprocessing model and the OCR recognition model.

[0028] A data lineage analysis engine is used to build and visualize the end-to-end relationships between daily demand data, waybill data, and cargo tracking data; The data supermarket module provides unified data query, cleaning, transformation, and standardization services, ensuring data consistency and integrity and reducing usage errors. It also constructs a data security management system through access control, data encryption, and other technologies to ensure the security and confidentiality of data use and transmission. Furthermore, users can independently query and filter the data they need within the data supermarket. The intelligent data acquisition and preprocessing module includes a scenario-based processing unit, which is designed for railway timetables, weighbridge slips, and ship data. Figure 3 The following preprocessing strategies are applied to different types of business forms: A1. Perform table layout analysis, cell association, and page continuity processing on railway timetables; A2. Perform template matching, handwritten number area enhancement, and strong weight logic verification on the weighbridge slip; A3. Separate the ship diagram from the graphic and the table, and associate the container number verification algorithm with the three-dimensional stowage coordinates.

[0029] like Figure 2 A smart tracking method for rail-water intermodal transport based on OCR recognition includes the following steps: Step 1: Paper form collection: Convert paper forms (railway timetables, waybills, weighbridge slips, ship diagrams, etc.) into digital images using scanners, document scanners, and other equipment; The second step, image preprocessing and enhancement, involves denoising, binarizing, tilt correction, perspective transformation, and other processing of the image, and employing differentiated preprocessing strategies for different types of forms. The third step, multimodal OCR recognition, involves simultaneously calling multiple OCR engines (such as Baidu OCR engine, Alibaba Cloud OCR engine, Tencent Cloud OCR engine, etc.) to recognize the same image region; then, a weighted vote is performed based on the historical accuracy and current confidence of each OCR engine to output the optimal recognition result.

[0030] Step 4: Business Rules and Semantic Correction: The recognition results are corrected using business dictionaries (station names, cargo categories, etc.) and format rules (train number codes, date formats, etc.), and natural language processing technology is used to semantically optimize the text.

[0031] Step 5: Data quality verification: Confidence scores are given to the identified fields, and fields below the threshold are marked as suspicious data; and rule verification is performed on the executed business, such as weight balance verification and time logic verification. Similarly, data that does not meet the verification rules is marked as suspicious data. Step 6: Manual review and model optimization: Push suspicious data to the manual review interface, then add the correct results confirmed by the review to the training set, and optimize the OCR recognition model regularly; Step 7: Establishing a lineage: Extract key identifiers (request acceptance number, waybill number, and cargo ID) from the final confirmed data, then establish the association between the three types of data to form a data lineage graph, and finally associate the data lineage graph with the business data and store it in the domestic database.

[0032] Step 8, Data Application and Display: The web interface provides functions such as data query, lineage visualization, and real-time cargo tracking. Real-time cargo tracking allows for both forward and reverse tracking. Specifically: Positive tracking: Query the real-time location of associated waybills and goods based on the request acceptance number; Reverse tracking: Based on the abnormal status of goods, reverse the location of relevant waybills and demand data; When tracking goods through a web interface, data lineage maps can be used to quickly locate the source of data anomalies and help identify redundant data to optimize storage costs.

[0033] In summary, this solution firstly replaces the original database with a domestic database, simultaneously modifies relevant system components and interface data to achieve full compatibility, replaces relevant system middleware and ensures its support for domestic database languages, and finally completes the migration of all historical data, ensuring the continuous normal operation of the system, while effectively improving supply chain security and data security. Secondly, by replacing manual data extraction with OCR recognition technology, the efficiency and accuracy of railway timetable data processing are greatly improved, human error is reduced, and the accuracy of OCR recognition of complex forms is significantly improved through multimodal OCR collaboration. Furthermore, the system has built its own data supermarket, enabling users to query data independently according to their needs. Through unified cleaning, transformation, and standardization, the system improves data quality and consistency. It also establishes a data security management system using access control, data encryption, and other means to ensure the integrity, security, and confidentiality of data during use and transmission. Furthermore, the data lineage analysis engine automatically constructs a complete data relationship graph from "daily demand → waybill → cargo tracking". Business personnel can view the entire process status of any cargo from planning and shipment to its current location in real time and intuitively, effectively changing the previous situation of information silos and difficulty in traceability. It can also clearly identify the source and destination of data, enabling rapid location of the source of data anomalies and improving the level of data quality control.

[0034] Second implementation method: This embodiment adds an intelligent input terminal to the first embodiment, while the rest remains the same as the first embodiment.

[0035] like Figure 4 and Figure 6 The intelligent tracking system for rail-water intermodal transport also includes an intelligent data entry terminal. The intelligent data entry terminal includes a terminal body 1, with alignment micro-protrusions 2 fixedly connected to the four corners of the upper end of the terminal body 1. Magnetic grooves 101 are provided at both ends of the terminal body 1, and paper pressure strips 3 are matched and adsorbed on the magnetic grooves 101. The paper form has the same area as the upper surface of the terminal body 1. The four corners of the paper form are beveled, and the four bevels are respectively matched with the four alignment micro-protrusions 2. Through the matching of the alignment micro-protrusions 2 and the bevels, the effective area of ​​the paper form and the effective area on the screen can be aligned. Subsequently, through the magnetic attraction between the paper pressure strips 3 and the magnetic grooves 101, the position of the paper form can be limited, thereby enabling the effective area on the screen of the terminal body 1 and the effective area of ​​the paper form to be synchronously aligned, improving the accuracy of subsequent data entry and making it less likely for data misalignment to occur on the electronic form.

[0036] The terminal body 1 has a pressure sensing layer inside its screen. The effective area of ​​the screen on the terminal body 1 is completely consistent with that of the paper form. The terminal body 1 stores an electronic form that is one-to-one with the paper form, and only the effective area of ​​the electronic form is displayed on the screen. This allows the staff to record relevant content on the paper form, and under the action of the pressure sensing layer, the corresponding position of the electronic form in the terminal body 1 can be synchronized in real time. This makes it less likely for the content entered to be displayed in the electronic form, thus effectively ensuring the consistency between the paper form data and the electronic form data. The paper pressing strip 3 includes a magnetic strip 31 that matches the magnetic groove 101, a reinforcing rib 32 fixedly connected to the top of the magnetic strip 31, and a diagonal bar 33 fixedly connected to the end of the reinforcing rib 32. The reinforcing rib 32 can effectively improve the strength of the magnetic strip 31, making it less likely to break due to being too long and thin. The diagonal bar 33 makes it easy to remove the entire paper pressing strip 3, facilitating the handling of paper forms.

[0037] The depth of the magnetic groove 101 is no more than 2mm. After the paper form is aligned with the effective area inside the terminal body 1 by the four alignment micro-protrusions 2, when the paper pressing strip 3 magnetically attracts the paper form and the inside of the magnetic groove 101, its small depth is unlikely to damage the paper.

[0038] The terminal body 1 also integrates a real-time handwriting recognition and verification module, a shooting module, and a data synchronization module: The real-time handwriting recognition and verification module converts electronic handwriting into structured text data and performs real-time compliance verification and alerts. Compliance verification rules include weight balance verification and time logic verification. When manually entered data by staff does not meet the verification rules, the real-time handwriting recognition and verification module can directly provide feedback to terminal 1. Terminal 1 can provide relevant alerts, such as voice prompts and local highlighting, enabling staff to promptly manually verify and correct the data. Compared to existing technologies, this approach, where anomalies are detected after OCR recognition followed by manual review, allows for correction at the data source, effectively ensuring the accuracy of the data source and reducing the workload of subsequent data recognition and processing, thus improving the efficiency and accuracy of establishing data lineage. The image capture module takes a picture of the paper form after it is filled out. The data synchronization module associates the verified structured text data with the captured paper form image data and synchronizes it to the business service layer. Among them, the synchronized data is directly entered into the database as high-confidence structured text data, and the high-confidence structured text data and its corresponding form images constitute labeled training pairs for training the multimodal OCR recognition engine.

[0039] The business service layer also includes a data fusion and decision-making module, which includes a data association unit, a difference calculation unit, and a decision execution unit.

[0040] The data association unit associates smart input terminal data (high-confidence structured text data) with OCR recognition results from the same form using a unique identifier; the difference calculation unit calculates the difference between smart input terminal data and OCR recognition results on key fields; and the decision execution unit executes one of the following preset strategies: B11. If smart input terminal data exists and the difference is less than the threshold, then the smart input terminal data shall be used as the final data. B12. If there is data from the intelligent input terminal and the difference is greater than or equal to the threshold, the manual review process will be triggered, and the review result will be the final data. B2. If no data is available from the intelligent input terminal, the OCR recognition result shall be used as the final data.

[0041] like Figure 5 The unique identifier is generated as follows: each time a staff member fills out a paper form, a QR code or barcode with a digital code is generated on the one-to-one electronic form on the smart input terminal. After the staff member finishes filling out the form and removes it from the smart input terminal, the digital code is copied to a designated location. This same digital code can be used as a "unique identifier" to assist in associating high-confidence structured text data with OCR-recognized form image data.

[0042] like Figure 3 A smart tracking method for rail-water intermodal transport based on multimodal OCR and data lineage analysis includes the following steps: S1, Dual-track data acquisition: S11. Collect on-site operation data through intelligent input terminals. The on-site operation data includes high-confidence structured text data and its corresponding form images. S12. Collect image data of historical paper forms and third-party paper documents by scanning; S2, Data Processing: S21. Perform preprocessing and enhancement processing on the acquired image data; S22. Recognize and optimize paper form images using a multimodal OCR recognition engine; S3, Data Fusion Decision-Making: S31. Compare and correlate the data from the smart input terminal of the same form with the OCR recognition results; S32. Determine the final valid data based on preset strategies, including: prioritizing smart terminal data and triggering manual review when the difference is too large; S4. Quality Verification and Optimization: S41. Verify the confidence level and business rules of the final data; S42. Use the results of manual review as training samples to iteratively optimize the OCR recognition model; S5. Establishing a lineage relationship: Based on the final valid data, extract key identifiers from daily demand data, waybill data, and cargo tracking data, establish the relationship between the three types of data, and form a data lineage relationship map; S6. Data Application and Presentation: S61. Store the final valid data and kinship map in a domestic database; S62. Provide data query, lineage visualization and real-time cargo tracking services through a web front-end interface; S63. Use kinship maps to trace the source and analyze the impact of data anomalies.

[0043] In step S11, the steps of acquiring data and images through the intelligent input terminal specifically include: C1. Align the four corners of the paper form with the four alignment micro-protrusions 2 on the terminal body 1 to achieve alignment of the effective area of ​​the paper form with the effective area of ​​the electronic form displayed on the screen of the terminal body 1. Then fix the paper form on the surface of the terminal body 1 by pressing the paper strip 3. C2. Staff members write normally on paper forms, and simultaneously write their handwriting on electronic forms through a pressure-sensing layer to form electronic handwriting. Handwriting recognition and compliance verification are performed in real time. If the verification fails, the user is immediately notified to correct the form content. C3. After completing the form, take a picture of the form and link the validated structured text data with the form image (also linked by a unique identifier).

[0044] In summary, this implementation method, by adding an intelligent data entry terminal, enables data verification while recording data on the form. This significantly reduces data quality issues caused by illegible handwriting, formatting errors, and logical inconsistencies from the very beginning of data generation, further improving data accuracy and greatly reducing subsequent error correction costs. Furthermore, this implementation method is compatible with both smart data entry terminals and OCR-recognized data modes. It can ensure the source quality of critical operational data through terminals, while efficiently processing historical archives and third-party documents through OCR. The data from the two modes can be compared, verified, and merged, forming a double guarantee of data quality and ensuring the reliability and continuity of the system in different business scenarios. Furthermore, the high-confidence structured text data generated by the intelligent input terminal can be used as a "standard answer" to train the OCR model, and the OCR recognition results and human review feedback continuously optimize the terminal's verification rules. The above description is merely a preferred embodiment of the present invention; it encompasses all the protection scope of the present invention. Any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the protection scope of the present invention.

Claims

1. A smart tracking system for rail-water intermodal transport based on OCR recognition, characterized in that, include: The domestic infrastructure layer includes domestically produced servers, domestically produced operating systems, and network security equipment; The data support layer includes domestically produced databases and middleware; The application presentation layer provides a web front-end interface for displaying recognition results, lineage charts, and real-time cargo tracking status. The business service layer includes: The intelligent acquisition and preprocessing module is used to perform quality assessment, enhancement and repair, and layout analysis on the input form images; A multimodal OCR recognition engine is used to call multiple OCR engines for collaborative recognition and to integrate business rules and semantic information to optimize the results; The data quality verification and closed-loop learning module is used to verify the recognition results in real time and iteratively optimize the recognition model through human feedback data. A data lineage analysis engine is used to build and visualize the end-to-end relationships between daily demand data, waybill data, and cargo tracking data; The data supermarket module provides unified data query, cleaning, and standardization services.

2. The intelligent tracking system for rail-water intermodal transport based on OCR recognition according to claim 1, characterized in that, The intelligent data acquisition and preprocessing module includes a scenario-based processing unit, which performs the following preprocessing strategies for three types of business forms: railway timetables, weighbridge slips, and ship maps: A1. Perform table layout analysis, cell association, and page continuity processing on railway timetables; A2. Perform template matching, handwritten number area enhancement, and strong weight logic verification on the weighbridge slip; A3. Separate the ship diagram from the graphic and the table, and associate the container number verification algorithm with the three-dimensional stowage coordinates.

3. The intelligent tracking system for rail-water intermodal transport based on OCR recognition according to claim 1, characterized in that, The multimodal OCR recognition engine includes: The multi-engine parallel calling unit is used to simultaneously call at least two OCR engines to recognize the same image region; The weighted voting fusion unit performs weighted voting on the recognition results based on the historical accuracy and current confidence of each OCR engine, and outputs the optimal text. The business rules and semantic error correction unit uses a pre-built business dictionary, format rules, and natural language processing model to correct the recognized text.

4. The intelligent tracking system for rail-water intermodal transport based on OCR recognition according to claim 1, characterized in that, The data quality verification and closed-loop learning module includes: The dual verification unit scores the confidence level of the identified fields and performs cross-validation with pre-defined business rules. The manual review push unit pushes fields with low confidence or those that fail the verification to the manual review interface; The model iteration and optimization unit uses the manually verified correct results and the corresponding original images as high-quality training samples to perform incremental training on the preprocessing model and the OCR recognition model on a regular basis.

5. The intelligent tracking system for rail-water intermodal transport based on OCR recognition according to claim 1, characterized in that, It also includes an intelligent input terminal, which includes a terminal body (1). Each of the four corners of the upper end of the terminal body (1) is fixedly connected with a positioning micro-protrusion (2). Both the left and right ends of the terminal body (1) are provided with magnetic grooves (101). A paper strip (3) is matched and adsorbed on the magnetic groove (101). The paper form has the same area as the upper surface of the terminal body (1). The four corners of the paper form are all cut off, and the four cut corners are respectively matched with the four positioning micro-protrusions (2).

6. The intelligent tracking system for rail-water intermodal transport based on OCR recognition according to claim 5, characterized in that, The terminal body (1) has a pressure sensing layer inside the screen. The effective area of ​​the screen on the terminal body (1) is completely consistent with that of the paper form. The terminal body (1) stores an electronic form that is one-to-one with the paper form. Only the effective area of ​​the electronic form is displayed on the screen. The paper pressing strip (3) includes a magnetic strip (31) that matches the magnetic groove (101), a reinforcing rib (32) fixedly connected to the top of the upper end of the magnetic strip (31), and a diagonal rod (33) fixedly connected to the end of the reinforcing rib (32). The depth of the magnetic groove (101) is no more than 2 mm.

7. The intelligent tracking system for rail-water intermodal transport based on OCR recognition according to claim 6, characterized in that, The terminal body (1) also integrates: The real-time handwriting recognition and verification module is used to convert electronic handwriting into structured text data and perform real-time compliance verification and reminders. The camera module is used to take images of the paper form after it has been filled out. The data synchronization module associates the validated structured text data with the captured image data of the paper form and synchronizes it to the business service layer. The synchronized data is directly stored in the database as high-confidence structured text data, and the high-confidence structured text data and its corresponding form images constitute labeled training pairs for training the multimodal OCR recognition engine.

8. The intelligent tracking system for rail-water intermodal transport based on OCR recognition according to claim 7, characterized in that, The business service layer also includes a data fusion and decision-making module, which includes: The data association unit uses a unique identifier to associate the data from the smart input terminal of the same form with the OCR recognition result; The difference calculation unit calculates the degree of difference between the data from the intelligent input terminal and the OCR recognition results in key fields; The decision-making execution unit executes one of the following preset strategies: B11. If smart input terminal data exists and the difference is less than the threshold, then the smart input terminal data shall be used as the final data. B12. If there is data from the intelligent input terminal and the difference is greater than or equal to the threshold, the manual review process will be triggered, and the review result will be the final data. B2. If no data is available from the intelligent input terminal, the OCR recognition result shall be used as the final data.

9. The tracking method of the intelligent tracking system for rail-water intermodal transport based on OCR recognition according to claim 8, characterized in that, Includes the following steps: S1, Dual-track data acquisition: S11. Collect on-site operation data through an intelligent input terminal. The on-site operation data includes high-confidence structured text data and its corresponding form images. S12. Collect image data of historical paper forms and third-party paper documents by scanning; S2, Data Processing: S21. Perform preprocessing and enhancement processing on the acquired image data; S22. Recognize and optimize paper form images using a multimodal OCR recognition engine; S3, Data Fusion Decision-Making: S31. Compare and correlate the data from the smart input terminal of the same form with the OCR recognition results; S32. Determine the final valid data based on preset strategies, including: prioritizing smart terminal data and triggering manual review when the difference is too large; S4. Quality Verification and Optimization: S41. Verify the confidence level and business rules of the final data; S42. Use the results of manual review as training samples to iteratively optimize the OCR recognition model; S5. Establishing a lineage relationship: Based on the final valid data, extract key identifiers from daily demand data, waybill data, and cargo tracking data, establish the relationship between the three types of data, and form a data lineage relationship map; S6. Data Application and Presentation: S61. Store the final valid data and kinship map in a domestic database; S62. Provide data query, lineage visualization and real-time cargo tracking services through a web front-end interface; S63. Use kinship maps to trace the source and analyze the impact of data anomalies.

10. The intelligent tracking method for rail-water intermodal transport based on OCR recognition according to claim 9, characterized in that, In step S11, the step of collecting data and images through the intelligent input terminal specifically includes: C1. Align the four corners of the paper form with the four alignment micro-protrusions (2) on the terminal body (1) to achieve alignment of the effective area of ​​the paper form with the effective area of ​​the electronic form displayed on the screen of the terminal body (1). Then fix the paper form on the surface of the terminal body (1) by pressing the paper strip (3). C2. Staff write normally on paper forms, and the handwriting is simultaneously transferred to the electronic form through a pressure sensor layer to form electronic handwriting. Handwriting recognition and compliance verification are performed in real time. If the verification fails, the user is immediately notified to correct the form content. C3. After the form is filled out, an image of the form is taken, and the verified structured text data is linked and synchronized with the form image.

Images