System and method for ai-based data collection and labeling pipeline processing for maritime reports

Abstract

Provided are a system and a method for AI-based data collection and labeling pipeline processing for maritime reports. The present invention relates to a system and a method for AI-based data collection and labeling pipeline processing for maritime reports and, more particularly, to a system and a method for AI-based data collection and labeling pipeline processing for maritime reports, which enable processing various data formats, simplifying data collection, and quickly providing highly reliable reports on the basis of the latest data.

Description

AI-based data collection and labeling pipeline processing system and method for maritime reporting

[0001] The present invention relates to an AI-based data collection and labeling pipeline processing system and method for maritime reporting, and more particularly, to an AI-based data collection and labeling pipeline processing system and method for maritime reporting, which can process various data formats, simplify data collection, and quickly provide highly reliable reports based on up-to-date data.

[0002] The maritime industry generates vast and diverse data in its daily operations. This data can be generated across critical business areas such as vessel operations, safety management, environmental monitoring, and regulatory compliance, including emails, electronic logbooks, various forms of reports, sensor data, and satellite communication logs. Traditional data processing methods rely on manual processes. These collection and data extraction methods are time-consuming, prone to human error, and inefficient for supporting real-time decision-making. Furthermore, given the global nature of the maritime industry and its 24 / 7 operations, these manual processes can become even more complex due to time zone differences and communication delays.

[0003] Existing data management solutions often require fundamental changes to a company's entire reporting structure. This can generate significant resistance among maritime workers and be a major factor in delaying the adoption and adoption of new systems. Therefore, the maritime industry requires a flexible system that can seamlessly integrate with existing workflows while automating data processing. Furthermore, a system that can effectively handle diverse maritime data formats, streamline data collection processes, and ensure accurate data labeling and classification is essential.

[0004] In order to solve the problems of the prior art as described above, an AI-based data collection and labeling pipeline processing system and method for maritime reporting according to an embodiment of the present invention automatically collects and extracts maritime navigation data from various formats of files, accurately labels them through interaction with workers, and provides an AI-based data collection and labeling pipeline processing system and method for maritime reporting that can quickly provide users with highly reliable reports based on the latest data.

[0005] According to one aspect of the present invention for solving the above-mentioned problem, the present invention comprises: a data collection unit that periodically monitors the email accounts of registered maritime industry workers and extracts and collects maritime operation document files including email bodies and attachments from a corresponding webmail server; a preprocessing unit that extracts and analyzes text and visual data from the maritime operation document files collected by the data collection unit using an OCR (Optical Character Recognition) tool and an LLM (Large Language Model) and reprocesses the data based on context to generate text data; a labeling unit that applies maritime operation information labels to the text data generated by the preprocessing unit using an active learning labeling tool and NLP (Natural Language Processing) to generate label data; a data storage unit that stores the original data collected by the data collection unit, the preprocessed data generated by the preprocessing unit, and the label data generated by the labeling unit; a version management unit that tracks and manages changes to a machine learning model of labeling pipeline processing and data stored in the data storage unit; A model training and verification unit that trains and verifies the machine learning model using label data stored in the data storage unit using the AI ​​platform; and

[0006] An AI-based data collection and labeling pipeline processing system for maritime reports is provided, including an application section that provides an interface and API for generating maritime reports based on the latest version of label data using the latest version of a machine learning model and providing the same to a user terminal.

[0007] In one embodiment, the data storage unit can store the label data classified into structured data and semi-structured data.

[0008] According to another aspect of the present invention, a method for processing an AI-based data collection and labeling pipeline for a maritime report is provided, including: a step of periodically monitoring an email account of a registered maritime industry worker to extract and collect maritime operation document files including email bodies and attachments from a corresponding webmail server; a preprocessing step of extracting and analyzing data and generating reprocessed text data considering the context between the data; a labeling step of applying a maritime operation information label to the data to generate label data; a data storage step of storing original data, preprocessed data, and label data; a version management step of tracking changes to a machine learning model and data; a step of training and verifying a machine learning model; and an application step of generating and displaying data and reports on a user terminal.

[0009] In one embodiment, the preprocessing unit extracts the text and the visual data from the maritime navigation document file, reprocesses the extracted text and visual data into text data considering the context of the extracted text and visual data and the data context according to the arrangement of the table, and integrates and standardizes the reprocessed text data.

[0010] In one embodiment, the labeling unit may generate a list of labeling tasks corresponding to data, determine the order of labeling tasks included in the generated list, and perform labeling tasks according to the determined order, while providing a predicted label by the machine learning model to a user terminal and receiving correction or additional feedback on the predicted label from the user terminal to generate the label data.

[0011] An AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention automatically collects and extracts maritime operation data from various formats of files, and accurately labels them through interaction with workers, thereby shortening the data collection time for training a machine learning model, and enabling training with high-quality training data, thereby quickly providing a highly reliable report based on the latest data to a user terminal, thereby supporting accurate decision-making regarding maritime industry operations in a timely manner.

[0012] In addition, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention can shorten data collection time and improve the efficiency of subsequent preprocessing processes by collecting maritime operation document files from various sources and converting them into a consistent format.

[0013] In addition, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention automatically extracts text and visual data from files of various formats, reprocesses them to reflect context, and standardizes and preprocesses them, thereby accurately extracting meaningful data while maintaining the context of the data even from manually written data and documents with complex structures.

[0014] In addition, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention can accurately label the latest data based on new labels and feedback from workers and repeatedly update the machine learning model, thereby gradually improving the performance of the machine learning model while reflecting the latest information into the machine learning model and generating high-quality data for training the machine learning model.

[0015] In addition, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention systematically classifies and stores data in a database suitable for the characteristics of the data format, thereby improving the stability and scalability of data storage and simultaneously improving the data access speed of the machine learning model, thereby improving the efficiency of data management and utilization.

[0016] In addition, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention provides a change history management function for data and machine learning models and a restoration function to a previous version, thereby securing systematicity and visibility in the experimental process for performance optimization of machine learning models, and objectively comparing the performance of multiple versions of machine learning models and selecting an appropriate model as needed, thereby enabling systematic and efficient management of machine learning models.

[0017] In addition, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention provides an interface for generating and displaying maritime reports, thereby allowing user terminals to access data regardless of time and place through the web and external systems, thereby enabling accurate and timely compliance and decision-making, thereby improving the stability of maritime industry operations and enhancing work efficiency.

[0018] In addition, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention determines whether the target accuracy of the machine learning model before training is achieved, and compares the performance of the previous version of the machine learning model to determine whether the accuracy of the current version of the machine learning model is improved after training, thereby continuously improving the machine learning model and ensuring improved accuracy, and continuously optimizing the system to meet the special requirements of the maritime industry.

[0019] FIG. 1 is a configuration diagram of an AI-based data collection and labeling pipeline processing service system for maritime reporting according to one embodiment of the present invention.

[0020] FIG. 2 is a block diagram of an AI-based data collection and labeling pipeline processing system for maritime reporting according to one embodiment of the present invention.

[0021] FIG. 3 is a block diagram of a preprocessing unit of an AI-based data collection and labeling pipeline processing system for maritime reporting according to one embodiment of the present invention.

[0022] FIG. 4 is a block diagram of a labeling unit of an AI-based data collection and labeling pipeline processing system for maritime reporting according to one embodiment of the present invention.

[0023] FIG. 5 is a block diagram of a storage unit of an AI-based data collection and labeling pipeline system for maritime reporting according to one embodiment of the present invention.

[0024] FIG. 6 is a flowchart of an AI-based data collection and labeling pipeline processing method for a maritime report according to one embodiment of the present invention.

[0025] FIG. 7 is a flowchart of a preprocessing procedure of an AI-based data collection and labeling pipeline processing method for a maritime report according to one embodiment of the present invention.

[0026] FIG. 8 is a flowchart of a labeling procedure of an AI-based data collection and labeling pipeline processing method for a maritime report according to one embodiment of the present invention.

[0027] Hereinafter, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily practice the present invention. The present invention may be implemented in various different forms and is not limited to the embodiments described herein. In the drawings, parts irrelevant to the description have been omitted for clarity of description, and the same reference numerals designate identical or similar components throughout the specification.

[0028] Hereinafter, with reference to the drawings, an AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention will be described in more detail.

[0029] FIG. 1 is a schematic diagram of an AI-based data collection and labeling pipeline processing system for maritime reporting according to one embodiment of the present invention.

[0030] Referring to FIG. 1, an AI-based data collection and labeling pipeline processing service system (1) for maritime reports may include an AI-based data collection and labeling pipeline processing system (100) for maritime reports, a webmail server (10), and a user terminal (20).

[0031] The AI-based data collection and labeling pipeline processing service system (1) for maritime reports is a system for providing maritime reports generated based on maritime operation data collected in real time. The AI-based data collection and labeling pipeline processing system (100) for maritime reports learns maritime operation data collected in real time from a webmail server (10) based on AI to generate maritime reports and display the same on a user terminal (20).

[0032] An AI-based data collection and labeling pipeline processing system (100) for maritime reporting periodically monitors a webmail server (10) linked to the email account of a registered maritime industry worker to collect files in various formats, extract and analyze maritime operation data, and provide a comprehensive report necessary for operational decision-making in the maritime industry.

[0033] The webmail server (10) may include a storage space that functions as a mailbox corresponding to each user's email account. The webmail server (10) may store emails sent from the sender's terminal. Furthermore, the webmail server (10) may transmit the stored emails to the recipient's webmail server (10) via the Internet using a sending protocol.

[0034] The user terminal (20) can remotely access the AI-based data collection and labeling pipeline processing system (100) for maritime reporting via a communications network. At this time, the user terminal (20) can display a comprehensive maritime operation report. For example, the user terminal (20) may be a wireless communication electronic device such as a smartphone, laptop, or smart pad, or a wired communication electronic device such as a desktop.

[0035] FIG. 2 is a block diagram of an AI-based data collection and labeling pipeline processing system for maritime reporting according to one embodiment of the present invention.

[0036] Referring to FIG. 2, an AI-based data collection and labeling pipeline processing system (100) for a maritime report according to one embodiment of the present invention may include a data collection unit (110), a preprocessing unit (120), a labeling unit (130), a data storage unit (140), a version management unit (150), a model training and verification unit (160), and an application unit (170).

[0037] The AI-based data collection and labeling pipeline processing system (100) for maritime reporting is a system for generating and providing comprehensive reports necessary for decision-making in maritime industry operations based on collected maritime data. The system automatically collects and extracts maritime operation data from various formats of files, and accurately labels them through interaction with workers, thereby shortening the data collection time for training a machine learning model, and enabling training with high-quality training data. Therefore, a highly reliable report based on the latest data can be quickly provided to a user terminal (20), thereby supporting accurate decision-making in a timely manner regarding maritime industry operations.

[0038] The data collection unit (110) can periodically monitor the email accounts of marine industry workers registered in the email account management server from the webmail server (10) and extract and collect marine operation document files including email bodies and attachments. At this time, the data collection unit (110) can collect marine operation document files from the webmail server (10) linked to the email account. For example, the data collection unit (110) can collect and process marine operation document files in various formats, including emails, Excel sheets, PDFs, and image files.

[0039] More specifically, the data collection unit (110) can automatically extract emails and attachments periodically monitored using an email parser from the webmail server (10). Furthermore, the data collection unit (110) can directly collect files from a customer system using an API. Furthermore, the data collection unit (110) can process specific data formats of collected files using custom scripts that include special processing logic suited to each data format. This allows the data collection unit (110) to collect new data formats by adding new scripts whenever new data formats are added, thereby ensuring system flexibility.

[0040] In this way, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention can shorten data collection time and improve the efficiency of the subsequent preprocessing process by collecting maritime operation document files from various sources and converting them into a consistent format. The preprocessing unit (120) can extract and analyze text and visual data from the maritime operation document files collected by the data collection unit (110) using an Optical Character Recognition (OCR) tool and a Large Language Model (LLM). At this time, the preprocessing unit (120) can also process structured information and tabular data, and can extract the context of the image using an LLM with vision capabilities.

[0041] Additionally, the preprocessing unit (120) can reprocess the extracted text and visual data based on context to generate text data. At this time, the preprocessing unit (120) can reprocess the extracted text and visual data into text that reflects the context between the data using a Large Language Model (LLM). Furthermore, the preprocessing unit (120) can standardize the reprocessed text data using a custom script and convert it into a consistent format.

[0042] In this way, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention automatically extracts text and visual data from various formats of files, reprocesses them to reflect context, and standardizes and preprocesses them, thereby accurately extracting meaningful data while maintaining the context of the data even from manually written data and documents with complex structures.

[0043] The labeling unit (130) can create label data by applying a maritime navigation information label to text data created in the preprocessing unit (120) using an active learning labeling tool and NLP (Natural Language Processing).

[0044] The labeling unit (130) can perform labeling by interacting with the worker using an active learning-type labeling tool. At this time, the labeling unit (130) can generate a labeling task list using the active learning-type labeling tool and provide the labeling task to the worker. Specifically, the labeling unit (130) can iteratively improve the machine learning model by preferentially labeling uncertain data using the active learning-type labeling tool. At this time, the labeling unit (130) can gradually adjust the weights of the existing model using only newly labeled data at each labeling repetition cycle during the active learning process.

[0045] Additionally, the labeling unit (130) can receive label data provided as feedback from the operator. For example, the labeling unit (130) can use Prodigy as an active learning labeling tool.

[0046] Additionally, the labeling unit (130) can generate and display predicted label data. For example, the labeling unit (130) can display predicted labels of data to a worker based on the context of preprocessed data using an NLP transformer model including BERT and GPT.

[0047] At this time, the labeling unit (130) can label the timestamp, vessel information, operational data, location, and abnormal conditions for the log entry. For example, the labeling unit (130) can label the vessel name, type, and identifier for the vessel data. As another example, the labeling unit (130) can label speed, fuel consumption, and cargo information for the operational data. As yet another example, the labeling unit (130) can label the port, waypoint, and coordinates for the location data. In addition, the labeling unit (130) can store the label and the labeled data in the data storage unit (140).

[0048] In this way, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention performs accurate labeling of the latest data based on new labels and feedback from workers and repeatedly updates the machine learning model, thereby gradually improving the performance of the machine learning model while reflecting the latest information into the machine learning model and generating high-quality data for training the machine learning model.

[0049] The data storage unit (140) can store original data collected by the data collection unit (110), preprocessed data generated by the preprocessing unit (120), and label data generated by the labeling unit (130). At this time, the data storage unit (140) can store the original maritime operation document file collected by the data collection unit (110). For example, the data storage unit (140) can store and manage emails, Excel sheets, PDFs, and image files as object storage.

[0050] Additionally, the data storage unit (140) can store text and visual data that has been preprocessed by reprocessing based on the context between data in the preprocessing unit (120). For example, the data storage unit (140) can store and manage data preprocessed as object storage in an unstructured format.

[0051] Additionally, the data storage unit (140) can store label data generated by the labeling unit (130). For example, the data storage unit (140) can store and manage label data structured according to a predefined schema in a relational database. As another example, the data storage unit (140) can store and manage semi-structured label data in a non-relational database.

[0052] In this way, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention systematically classifies and stores data in a database suitable for the characteristics of the data format, thereby improving the stability and scalability of data storage and simultaneously improving the data access speed of the machine learning model, thereby improving the efficiency of data management and utilization.

[0053] The version control unit (150) can track and manage changes to the machine learning model and data stored in the data storage unit (140) for labeling pipeline processing. For example, the version control unit (150) can use DVC (Data Version Control), which can be integrated with GIT, a code version control tool, to perform version control of the data set and machine learning model, and can also perform code version control at the same time.

[0054] As another example, the version control unit (150) can use Kubeflow, a platform for machine learning model orchestration, to define, train, and deploy a pipeline for machine learning models for maritime reporting. In this case, the version control unit (150) can define and automate data learning, evaluation, and deployment as code, enabling rapid reproducibility of previous machine learning models.

[0055] In this way, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention provides a change history management function for data and machine learning models and a restoration function to a previous version, thereby securing systematicity and visibility in the experimental process for performance optimization of machine learning models, and objectively comparing the performance of multiple versions of machine learning models and selecting an appropriate model as needed, thereby enabling systematic and efficient management of machine learning models.

[0056] The model training and validation unit (160) can use the AI ​​platform to train and validate a machine learning model using labeled data stored in the data storage unit (140). For example, the model training and validation unit (160) can use Kubeflow Pipelines to train a machine learning model using labeled data. Thus, the model training and validation unit (160) can train a machine learning model using labeled data, thereby improving the accuracy and robustness of the model.

[0057] The application unit (170) may provide an interface and API for generating a maritime report based on the latest version of label data using the latest version of a machine learning model and providing the report to the user terminal (20). For example, the application unit (170) may provide a web interface for interaction with a dashboard screen area that displays a maritime report including visualized data on the user terminal (20). As another example, the application unit (170) may include a RESTful API to enable access to maritime data and reports processed in an external system. Accordingly, the application unit (170) may display an accurate and timely maritime report on the user terminal (20) connected to a web browser or linked to an external system via the Internet.

[0058] In this way, the AI-based data collection and labeling pipeline processing system and method for maritime reports provide an interface for generating and displaying maritime reports, thereby allowing a user terminal (20) to access data regardless of time and place through the web and external systems, thereby enabling accurate and timely compliance and decision-making, thereby improving the stability of maritime industry operations and enhancing work efficiency.

[0059] FIG. 3 is a block diagram of a preprocessing unit of an AI-based data collection and labeling pipeline processing system for maritime reporting according to one embodiment of the present invention.

[0060] Referring to FIG. 3, the preprocessing unit (120) may include a data extraction unit (122), a data reprocessing unit (124), and a data standardization unit (126).

[0061] The data extraction unit (122) can extract text and visual data from maritime operation document files. At this time, the data extraction unit (122) can convert unstructured data into machine-readable text. Specifically, the data extraction unit (122) can extract text data from maritime operation document files in image format containing a mixture of machine-written and hand-written text. For example, the data extraction unit (122) can extract text data from scanned image files and PDFs using an OCR (Optical Character Recognition) tool. As another example, the data extraction unit (122) can extract data using AWS Textract and Google Cloud Vision APIs.

[0062] In addition, the data extraction unit (122) can analyze and process structured information. At this time, the data extraction unit (122) can extract data in a visually structured table format. For example, the data extraction unit (122) can extract a table format of an Excel file, and can extract contextual data by understanding the context according to the table arrangement from an Excel file in which multiple tables exist. For example, the data extraction unit (122) can extract contextual data from an Excel file including at least two tables and sheets. At this time, the data extraction unit (122) can extract contextual data that reflects the causal relationship or precedence relationship between a sheet displaying fuel consumption data by navigation equipment, a table recording daily and time-period sea conditions within each sheet, a table recording daily fuel consumption, a table recording fuel consumption per voyage, and a table recording CO2 emissions.

[0063] Additionally, the data extraction unit (122) can understand and extract data expressed as merged cells in a table. For example, the preprocessing unit (120) can extract the context of an image using an LLM with vision capabilities.

[0064] The data reprocessing (124) can reprocess the text and visual data extracted from the data extraction unit (122) into text data that takes into account the data context according to the context and table arrangement. At this time, the data reprocessing unit (124) can receive the text and visual data and context data extracted from the data extraction unit (122) as input data. In addition, the data reprocessing unit (124) can comprehensively analyze the structure of the document, the arrangement of the table, and the relationship between data based on the context data. In addition, the data reprocessing unit (124) can identify and integrate the contextual correlation between the individual data elements extracted based on the context data and reprocess them. For example, the data reprocessing unit (124) can integrate and reprocess the data into text data that includes the context between the data using an LLM (Large Language Model). As a result, the data reprocessing unit (124) can reprocess the data extracted from the data extraction unit (122) by converting it into a more structured and meaningful form. For example, data reprocessing (124) can automatically insert expressions including correlations between maritime environmental conditions, navigation equipment, and overall ship performance based on contextual data, thereby reprocessing the data into text data that includes the context it contains.

[0065] The data standardization unit (126) can integrate and standardize reprocessed text data. At this time, the data standardization unit (126) can organize and standardize the extracted text using a custom script. The data standardization unit (126) can convert the expression format of the extracted data into a consistent standard format. For example, the data standardization unit (126) can unify and convert date formats, numeric notations, currency units, etc.

[0066] The data standardization unit (126) can unify and convert various expressions representing the same concept into a single standard term. For example, the data standardization unit (126) can unify and convert the expressions "wave height," "wave height," and "wave height" into "wave height." Thus, the data standardization unit (126) can prevent the dispersion of semantically identical data and eliminate data duplication. Accordingly, the data standardization unit (126) can maintain data consistency.

[0067] In this way, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention automatically extracts text and visual data from various formats of files, reprocesses them to reflect context, and standardizes and preprocesses them, thereby accurately extracting meaningful data while maintaining the context of the data even from manually written data and documents with complex structures.

[0068] FIG. 4 is a block diagram of a labeling unit of an AI-based data collection and labeling pipeline processing system for maritime reporting according to one embodiment of the present invention.

[0069] Referring to FIG. 4, the labeling unit (130) may include a labeling task generation unit (132), a labeling task ordering unit (134), a prediction label display unit (136), and a label data generation unit (138).

[0070] The labeling task generation unit (132) can generate a labeling task list corresponding to the data in order to provide labeling tasks to the labeling worker. That is, the labeling task generation unit (132) can generate a labeling task list to be processed on the user terminal (20) based on the preprocessed data. At this time, the labeling task generation unit (132) can generate an optimized task list by considering the characteristics of the data and labeling requirements.

[0071] The labeling task ordering unit (134) can determine the order of the labeling tasks included in the generated labeling task list. At this time, the labeling task ordering unit (134) can apply the active learning principle to arrange the generated labeling tasks in an order of importance for improving the performance of the learning model. In other words, the labeling task ordering unit (134) can determine the order of tasks so that data with high model uncertainty or data that can significantly contribute to learning are labeled first.

[0072] In addition, the labeling task sequence arrangement unit (134) can display the labeling tasks in a determined order on the user terminal (20). That is, the labeling task sequence arrangement unit (134) can provide a screen area for displaying data to be labeled to the user terminal (20).

[0073] The prediction label display unit (136) can generate a predicted label for data. For example, the prediction label display unit (136) can use NLP (Natural Language Processing) to understand the context of the data and generate an appropriate predicted label. In addition, the prediction label display unit (136) can provide the user terminal (20) with a screen area that displays a predicted label for the data. Accordingly, the prediction label display unit (136) can support the worker's labeling process.

[0074] The label data generation unit (138) can provide the user terminal (20) with an interface that can modify or add a predicted label displayed from the predicted label display unit (136). In addition, the label data generation unit (138) can store the modified or added label in the data storage unit (140).

[0075] FIG. 5 is a block diagram of a storage unit of an AI-based data collection and labeling pipeline system for maritime reporting according to one embodiment of the present invention.

[0076] Referring to FIG. 5, the storage unit (140) may include structured data (142), semi-structured data (144), and original and preprocessed data (146).

[0077] Structured data (142) may be one or more relational data output from the labeling unit (130) corresponding to a single piece of data. In this case, the structured data (142) may include a series of metadata, labels, and labeled text generated from the labeling unit (130). For example, the structured data (142) may be organized and displayed in a table format to facilitate searching.

[0078] Semi-structured data (144) may be labeled data that cannot be structured. Furthermore, semi-structured data (144) may be in various data formats. Semi-structured data (144) may be stored and managed in a non-relational database. In this case, semi-structured data (144) may not be displayed in a complete table format, but may be expressed in XML and JSON file formats. For example, semi-structured data (144) may be a maritime navigation log file labeled with a timestamp.

[0079] The original and preprocessed data (146) may include unstructured data including emails, Excel sheets, PDFs, and image files collected from the data collection unit (110) and data extracted from the preprocessing unit (120).

[0080] FIG. 6 is a flowchart of an AI-based data collection and labeling pipeline processing method for a maritime report according to one embodiment of the present invention.

[0081] Referring to FIG. 6, the AI-based data collection and labeling pipeline processing method (200) for maritime reporting includes a data collection step (S210), a preprocessing step (S220), a labeling step (S230), a data storage step (S240), a version management step (S250), a step of determining whether the target accuracy of the machine learning model before training is achieved (S255), a model training and verification step (S260), a step of determining whether the accuracy of the machine learning model after training is improved (S265), and a report generation and display step (S270).

[0082] More specifically, as illustrated in FIG. 6, the AI-based data collection and labeling pipeline processing system (100) for maritime reports periodically monitors email accounts of maritime industry workers pre-registered in the email account management server from the webmail server (10) to extract and collect maritime operation document files including email bodies and attachments (step S210). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can automatically extract emails and attachments that are periodically monitored from the webmail server (10) using an email parser. In addition, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can directly retrieve files from a customer system using an API. Furthermore, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can process specific data formats of collected files using custom scripts that include special processing logic suited to each data format.

[0083] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting extracts, analyzes, and preprocesses data to generate text data that is reprocessed by considering the context between the data (step S220). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can generate text data by reprocessing the data based on the context between the data using an OCR tool and LLM.

[0084] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reports applies maritime navigation information labels to the data to generate label data (step S230). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can generate label data by applying maritime navigation information labels to the text data generated in step S220 using an active learning labeling tool and NLP. At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can incrementally update the existing machine learning model by adjusting the weights using only the newly labeled data.

[0085] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reports stores original data, preprocessed data, and label data (step S240). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can store and manage emails, Excel sheets, PDFs, and image files as object storage. In addition, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can store and manage preprocessed data in an unstructured format as object storage. In addition, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can store and manage structured label data according to a predefined schema in a relational database, and can store and manage semi-structured label data in a non-relational database.

[0086] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting tracks changes in machine learning models and data to manage versions (step S250). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting uses DVC, which can be integrated with GIT, to manage the version of the dataset and machine learning model, and can also perform code version management simultaneously. In addition, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can define, train, and deploy a pipeline of machine learning models for maritime reporting using Kubeflow, a platform for machine learning model orchestration.

[0087] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting determines whether the target accuracy of the pre-training machine learning model has been achieved (step S255). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can measure the performance of the machine learning model using DVC and Kubeflow. More specifically, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can determine whether further adjustments to the machine learning model are required for new data.

[0088] If it is determined in step S255 that the target accuracy has not been achieved, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting may return to step S230 and repeatedly perform steps S230 to S250 until the target accuracy is achieved.

[0089] If the target accuracy is determined to have been achieved in Step S255, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting trains and verifies a machine learning model (Step S260). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can train the machine learning model using all accumulated labeled data using Kubeflow Pipelines.

[0090] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting determines whether the accuracy of the current version of the machine learning model has improved after training by comparing the performance of the machine learning model with that of the previous version (step S265). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can measure the performance of the machine learning model using DVC and Kubeflow and compare it with the performance of the previous version of the machine learning model. More specifically, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can determine that the accuracy of the current version of the machine learning model is lower than that of the other versions by comparing the performance of the machine learning model with that of the other versions.

[0091] If it is determined in step S265 that the accuracy of the current version of the machine learning model has not improved compared to the performance of other versions of the machine learning model, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting may return to step S250 and repeatedly perform steps S250 to S260 until the accuracy of the machine learning model is improved.

[0092] If it is determined in step S265 that the accuracy of the current version of the machine learning model has been improved compared to the performance of other versions of the machine learning model, the AI-based data collection and labeling pipeline processing system (100) for maritime reports generates and displays data and reports on the user terminal (20) (step S270). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reports may provide an interface and API for generating a maritime report based on the latest version of label data using the latest version of the machine learning model and providing the generated maritime report to the user terminal (20). For example, the AI-based data collection and labeling pipeline processing system (100) for maritime reports may provide the user terminal (20) with a web interface for interaction with a dashboard screen area that displays a maritime report including visualized data. Thereafter, the AI-based data collection and labeling pipeline processing system (100) for maritime reports completes a series of processes.

[0093] In this way, the AI-based data collection and labeling pipeline processing system and method for maritime reporting according to one embodiment of the present invention can continuously improve the machine learning model and ensure accuracy improvement by determining whether the target accuracy of the machine learning model before training is achieved and comparing the performance of the previous version of the machine learning model to determine whether the accuracy of the current version of the machine learning model is improved after training, and can continuously optimize the system to meet the special requirements of the maritime industry.

[0094] FIG. 7 is a flowchart of a preprocessing procedure of an AI-based data collection and labeling pipeline processing method for a maritime report according to one embodiment of the present invention.

[0095] Referring to FIG. 7, the preprocessing procedure (S220) includes a data extraction step (S222), a data reprocessing step (S224), and a data standardization step (S226).

[0096] More specifically, as illustrated in FIG. 7, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting extracts text and visual data from maritime operational document files (step S222). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can extract text data from scanned image files and PDFs using an OCR tool.

[0097] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting reprocesses the extracted text and visual data into text data that takes into account the context of the extracted text and visual data and the data context according to the arrangement of the table (step S224). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can extract the context of the image from the extracted text and visual data using an LLM equipped with vision capabilities, integrate the extracted text and visual data into text data that includes the context between the data, and reprocess it.

[0098] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reports integrates and standardizes the reprocessed text data (step S226). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can convert the expression format of the extracted data into a consistent standard format using a custom script. In addition, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can store the preprocessed data in the data storage unit (140). Thereafter, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can terminate a series of processes and return to step S230.

[0099] FIG. 8 is a flowchart of a labeling procedure of an AI-based data collection and labeling pipeline processing method for a maritime report according to one embodiment of the present invention.

[0100] Referring to FIG. 8, the labeling procedure (S230) includes a step of generating a labeling task (S231), a step of arranging the order of the generated labeling tasks (S232), a step of displaying a predicted label (S233), and a step of generating label data (S234).

[0101] More specifically, as illustrated in FIG. 8, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting generates a labeling task list corresponding to the data (step S231). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can generate a labeling task list to be processed on the user terminal (20) based on the preprocessed data.

[0102] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting determines the order of the labeling tasks included in the generated list (step S232). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reporting can apply active learning principles to arrange the generated labeling tasks in an order of importance for improving the performance of the learning model.

[0103] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reports performs labeling tasks according to a determined order, and provides predicted labels by the machine learning model to the user terminal (20) (step S233). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can display the labeling tasks in the determined order on the user terminal (20). In addition, the AI-based data collection and labeling pipeline processing system (100) for maritime reports can understand the context of the data using NLP, generate appropriate predicted labels, and provide a screen area for displaying the predicted labels on the user terminal (20).

[0104] Next, the AI-based data collection and labeling pipeline processing system (100) for maritime reports receives feedback on corrections or additions to the predicted labels from the user terminal (20) to generate label data (step S234). At this time, the AI-based data collection and labeling pipeline processing system (100) for maritime reports may provide the user terminal (20) with an interface for corrections or additions to the predicted labels. In addition, the AI-based data collection and labeling pipeline processing system (100) for maritime reports may store the corrections or additions to the labels in the data storage unit (140). Thereafter, the AI-based data collection and labeling pipeline processing system (100) for maritime reports may terminate a series of processes and return to step S240.

[0105] The above methods can be implemented by an AI-based data collection and labeling pipeline processing system (100) for maritime reporting as illustrated in FIG. 1, and in particular, can be implemented by software programs that perform these steps, in which case these programs can be stored in a computer-readable recording medium or transmitted by a computer data signal coupled with a carrier wave in a transmission medium or a communication network.

[0106] At this time, the computer-readable recording medium includes all kinds of recording devices in which data readable by a computer system is stored, and may be, for example, ROM, RAM, CD-ROM, DVD-ROM, DVD-RAM, magnetic tape, floppy disk, hard disk, optical data storage device, etc.

[0107] Although one embodiment of the present invention has been described above, the spirit of the present invention is not limited to the embodiments presented in this specification, and those skilled in the art who understand the spirit of the present invention will be able to easily propose other embodiments by adding, changing, deleting, or adding components within the scope of the same spirit, but this will also be considered to fall within the spirit of the present invention.

[0108] It can be used in ship operations, where route optimization and fuel efficiency improvement are required through real-time integrated analysis of various data, and in carbon emission compliance, which is required to comply with the IMO (International Maritime Organization) carbon emission regulations.

Claims

1. A data collection unit that periodically monitors the email accounts of registered marine industry workers and extracts and collects marine operation document files, including email bodies and attachments, from the relevant webmail server; A preprocessing unit that extracts and analyzes text and visual data from maritime operation document files collected by the data collection unit using an OCR (Optical Character Recognition) tool and an LLM (Large Language Model) and reprocesses the data based on context to create text data; A labeling unit that creates label data by applying maritime navigation information labels to text data created in the above preprocessing unit using an active learning labeling tool and NLP (Natural Language Processing); A data storage unit that stores original data collected from the data collection unit, preprocessed data generated from the preprocessing unit, and label data generated from the labeling unit; A version control unit that tracks and manages changes to the machine learning model of labeling pipeline processing and data stored in the data storage unit; A model training and verification unit that trains and verifies the machine learning model using label data stored in the data storage unit using the AI ​​platform; and An application section that provides an interface and API for generating a marine report based on the latest version of label data using the latest version of a machine learning model and providing it to the user terminal; AI-based data collection and labeling pipeline processing system for maritime reports including .

2. In paragraph 1, The above data storage unit is an AI-based data collection and labeling pipeline processing system for maritime reports that stores the label data classified into structured data and semi-structured data.

3. A step of periodically monitoring the email accounts of registered marine industry workers and extracting and collecting marine operation document files including email bodies and attachments from the relevant webmail server; A preprocessing step that extracts data, analyzes it, and creates text data that is reprocessed by considering the context between the data; A step of creating label data by applying a marine navigation information label to the data; Step of saving original data, preprocessed data, and label data; Steps to manage versions by tracking changes in machine learning models and data; Steps for training and validating a machine learning model; and A step of generating and displaying data and reports on a user terminal; Method for processing AI-based data collection and labeling pipeline for maritime reports including .

4. In paragraph 1, The above preprocessing unit, Extract the text and visual data from the above marine operation document file, The extracted text and visual data are reprocessed into text data that takes into account the context of the data according to the context and table arrangement. An AI-based data collection and labeling pipeline processing system for maritime reporting that integrates and standardizes the above-mentioned reprocessed text data.

5. In paragraph 1, The above labeling part, Create a list of labeling tasks corresponding to the data, Determine the order of the labeling tasks included in the above generated list, The labeling task is performed according to the above-determined order, and the predicted label by the machine learning model is provided to the user terminal. An AI-based data collection and labeling pipeline processing system for generating label data by receiving correction or additional feedback on the predicted label from the user terminal.

Images