An intelligent laboratory raw record automatic identification method and system

Abstract

This invention discloses an intelligent laboratory raw record automatic identification method and system. The method includes: preprocessing the raw record dataset to obtain a preprocessed raw record dataset; performing dynamic knowledge fusion based on the preprocessed raw record dataset to obtain a structured raw record dataset; performing three-level collaborative verification on the structured raw record dataset to obtain verification results, and selecting a subset of data that passes verification based on the verification results; converting the data subset according to the field mapping rules of the LIMS system data to obtain a converted dataset; inputting the converted dataset into the LIMS system for automatic storage and generating a standardized experimental report for the experiment to be identified. This invention can achieve intelligent identification and deep fusion of multimodal raw records, and perform dynamic and in-depth experimental logic verification, improving the automation, intelligence, and compliance level of laboratory data management.

Description

Technical Field

[0001] This invention relates to the field of intelligent laboratory analysis technology, and in particular to an intelligent laboratory raw record automatic identification method and system. Background Technology

[0002] In the daily operation of modern laboratories, original records are the legal carriers and core evidence of experimental processes, observation results, and data output. With the advancement of laboratory digitization, the data sources of original records are becoming increasingly heterogeneous, mainly presenting three modalities: paper handwritten or printed records, semi-structured or unstructured electronic documents (such as PDF reports, Word records, and Excel data tables), and time-series / waveform data streams directly output by instruments.

[0003] However, accurately, efficiently, and traceably transforming such diverse and heterogeneous raw records into structured data usable by a laboratory information management system faces severe technical challenges. Existing technical solutions have limitations at multiple levels: (I) Low accuracy in multimodal record recognition and poor adaptability to professional contexts; Traditional optical character recognition technologies based on template matching or general models typically have an accuracy rate of less than 65% when dealing with complex laboratory scenarios, such as handwritten chemical formulas, professional symbols, and spectral chart annotations; Furthermore, they are weak in handling complex tables commonly found in experimental reports (such as nested tables and merged cells), with a row and column omission rate exceeding 40% and a merged cell missegmentation rate as high as 35%; In addition, existing technologies treat paper records, electronic documents, and instrument waveform data in isolation, lacking an automatic alignment and association mechanism for cross-modal data, and data integration highly relies on manual transcription and proofreading, which is inefficient and prone to errors. (II) Shallow data verification mechanisms and serious omissions of experimental logic contradictions; Existing data verification mostly stays at the level of basic format verification (such as data type and numerical range), completely failing to address the inherent business logic and scientific rationality of experimental data. This leads to a large number of implicit logical contradictions becoming data quality black holes. (III) There is a "double gap" in the integration with the LIMS system (Laboratory Information Management System), which damages the integrity and credibility of the data. First, there is a semantic gap, where there is a lack of intelligent mapping between the free text and industry terms in the original records and the standardized fields preset by the LIMS system, resulting in information loss or misreading. Second, there is a traceability gap, where the complete "data lineage" chain from the data collection source (such as a scanner or an operator) to the final storage is broken. Any subsequent data modification or questioning cannot be traced back to the specific location and operation of the original record, making it difficult to meet the requirements of data integrity and audit traceability.

[0004] In summary, existing technologies generally suffer from systemic defects when dealing with multi-source heterogeneous raw records in laboratories, including low recognition accuracy, lack of logical verification, low system integration efficiency, and insufficient reliability. Therefore, there is an urgent need for an innovative technical solution capable of intelligent recognition and deep fusion of multimodal raw records, performing dynamic and in-depth experimental logical verification, and ultimately achieving seamless, reliable, and closed-loop integration with LIMS systems to comprehensively improve the automation, intelligence, and compliance levels of laboratory data management. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent laboratory raw record automatic identification method and system. This method can achieve intelligent identification and deep fusion of multimodal raw records, perform dynamic and in-depth experimental logic verification, and ultimately achieve seamless, reliable, and closed-loop integration with the LIMS system, so as to comprehensively improve the automation, intelligence and compliance level of laboratory data management.

[0006] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a method for automatic identification of original records in an intelligent laboratory, comprising: Obtain the original record dataset of the experiment to be identified, and preprocess the original record dataset to obtain the preprocessed original record dataset; Dynamic knowledge fusion is performed on the preprocessed raw record dataset to obtain a structured raw record dataset; A three-level collaborative validation was performed on the structured raw record dataset to obtain the validation results, and a subset of data that passed the validation was selected based on the validation results; The data subset is converted according to the field mapping rules of the LIMS system data to obtain the converted dataset. The transformed dataset is input into the LIMS system for automatic storage and a standardized experimental report for the experiment to be identified is generated.

[0007] Optionally, the original record dataset includes: paper record data, electronic document data, and waveform data; the electronic document data includes electronic documents in PDF, Word, and Excel formats.

[0008] Optionally, the preprocessing includes: Content recognition is performed on the paper record data to obtain scanned data; The PDF parsing library is used to extract text and image information from PDF format electronic documents to obtain PDF data; The content of the Word format electronic document is read using the Office API interface to obtain Word data; Use a table parsing tool to extract cell data from an Excel-formatted electronic document; For the waveform data, digital signal processing algorithms are used for feature extraction and digitization to obtain waveform feature data; The scanned data, PDF data, Word data, cell data, and waveform feature data are standardized to obtain the preprocessed raw record dataset.

[0009] Optionally, the dynamic knowledge fusion based on the preprocessed original record dataset to obtain a structured original record dataset includes: Document structure analysis is performed on the preprocessed raw record dataset to identify title, body text, table, and chart elements in the preprocessed raw record dataset, and a document structure tree is constructed based on the spatial relationship between the elements; Based on the document structure tree, the text region in the preprocessed original record dataset is located, and combined with the preset domain knowledge graph, character feature extraction and semantic matching are performed on the text content of the text region to obtain knowledge-enhanced text content. Based on the document structure tree, locate the table and chart areas in the preprocessed original record dataset, extract the data from the table and chart areas, and reorganize them into structured format data; By associating and integrating knowledge-enhanced text content with structured format data, a structured raw record dataset is obtained.

[0010] Optionally, the domain knowledge graph is scalable and integrates one or more of the following: a terminology dictionary, experimental operating procedures, equipment parameter database, and chemical substance attribute database.

[0011] Optionally, the step of performing three-level collaborative verification on the structured raw record dataset to obtain verification results, and then filtering the data subset that passes verification based on the verification results, includes: The structured raw record dataset is subjected to rule-level verification, knowledge-level verification, and learning-level verification in sequence. Data that passes all three levels of verification is marked as verified and included in the data subset; data that fails any level of verification is marked as suspicious data and fed back to the human-computer collaboration interface for updating the domain rule base.

[0012] Optionally, the rule-layer verification includes: verifying whether there are logical contradictions in the data in the structured original record dataset according to preset format rules and numerical range rules; if there are, the rule-layer verification is passed; otherwise, the rule-layer verification is not passed.

[0013] Optionally, the knowledge layer verification includes: verifying the semantic rationality of the data in the structured original record dataset in the context of the experiment to be identified based on the dynamically loaded domain rule base; if it is reasonable, the knowledge layer verification is passed; otherwise, the knowledge layer verification is not passed.

[0014] Optionally, the learning layer verification includes: using a pre-trained anomaly detection model to identify anomalies in the data in the structured raw record dataset; if no anomaly is identified, the learning layer verification is passed; otherwise, the learning layer verification is not passed.

[0015] Secondly, the present invention provides an intelligent laboratory raw record automatic identification system, comprising: The preprocessing module is used to obtain the original record dataset of the experiment to be identified, and to preprocess the original record dataset to obtain the preprocessed original record dataset. The knowledge fusion module is used to perform dynamic knowledge fusion based on the preprocessed raw record dataset to obtain a structured raw record dataset. The collaborative verification module is used to perform three-level collaborative verification on the structured raw record dataset, obtain the verification results, and select the data subset that passes the verification based on the verification results; The conversion module is used to convert the data subset according to the field mapping rules of the LIMS system data to obtain the converted dataset; The generation module is used to input the transformed dataset into the LIMS system for automatic data entry and generate standardized experimental reports for the experiments to be identified.

[0016] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: This invention provides an intelligent laboratory raw record automatic identification method and system. The method first acquires raw record datasets from paper scans, electronic documents, and instrument waveform data, and preprocesses them. Then, through dynamic knowledge fusion technology, the preprocessed data undergoes document structure analysis, knowledge graph-enhanced professional text recognition, and specialized extraction of tables and charts, transforming it into a structured raw record dataset rich in semantic relationships. This achieves high-precision identification and structured extraction of multimodal laboratory records, improving the accuracy of handling handwritten professional symbols and complex tables. Next, a three-level collaborative verification consisting of a rule layer, a knowledge layer, and a learning layer is performed on the structured dataset, automatically selecting a high-quality data subset that passes verification based on the verification results. Then, this data subset is standardized and converted according to the field mapping rules of a Laboratory Information Management System (LIMS) to form a directly interfaceable system dataset. Finally, the converted dataset is automatically entered into the LIMS system through an interface, and standardized electronic laboratory reports are generated simultaneously, thus completing an end-to-end automated processing loop from multi-source heterogeneous raw records to a reliable, traceable, and directly usable systemic data asset. Attached Figure Description

[0017] Figure 1 The diagram shown is a flowchart illustrating an automatic identification method for original records in an intelligent laboratory according to one embodiment of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0019] Example 1

[0020] like Figure 1 As shown in the figure, this invention provides a method and system for automatic identification of original records in an intelligent laboratory, comprising the following steps: Step S1: Obtain the original record dataset of the experiment to be identified, and preprocess the original record dataset to obtain the preprocessed original record dataset; Step S2: Perform dynamic knowledge fusion based on the preprocessed original record dataset to obtain a structured original record dataset; Step S3: Perform three-level collaborative validation on the structured raw record dataset to obtain validation results, and select the data subset that passes the validation based on the validation results; Step S4: Convert the data subset according to the field mapping rules of the LIMS system data to obtain the converted dataset; Step S5: Input the transformed dataset into the LIMS system for automatic data entry and generate a standardized experimental report for the experiment to be identified.

[0021] In this embodiment, step S1: Obtain the original record dataset of the experiment to be identified, and preprocess the original record dataset to obtain the preprocessed original record dataset, specifically including: S11: Data Acquisition The original record dataset includes: paper record data, electronic document data, and waveform data; the electronic document data includes electronic documents in PDF, Word, and Excel formats.

[0022] Specifically, paper-based data is recorded using a professional-grade document scanner with an optical resolution configured to be no less than 600 dpi, ensuring high-fidelity capture of details such as handwriting, minute symbols, and seals. The scanned images are typically in 24-bit true color or 8-bit grayscale TIFF format to preserve the original information to the greatest extent possible. Waveform data is acquired using oscilloscopes and spectrum analyzers. To ensure complete capture of high-speed transient signals, the sampling rate is configured to be no less than 1 GHz, with corresponding analog bandwidth and vertical resolution, ensuring distortion-free recording of the original waveform data.

[0023] Before entering the preprocessing stage, the data in the original record dataset also needs to be time-aligned with an accuracy of ±0.1 seconds. This allows for precise correlation between the peak waveform captured by the instrument at a certain moment and the "abnormal event" moment marked by the experimenter on the paper record, or with the description entered in the electronic log, laying the foundation for constructing a complete experimental context.

[0024] S12: Preprocessing: S121: Paper records: Content recognition is performed on the paper record data to obtain scanned data; Specifically, content recognition can employ general-purpose OCR engines based on deep learning, specialized models TRBA or PARSeq optimized for handwriting, and pre-trained models based on the IAM Handwriting Database. These are all known to those skilled in the art and will not be explained in detail here. S122: Electronic Documents: The PDF parsing library is used to extract text and image information from PDF format electronic documents to obtain PDF data; Specifically, it calls a PDF parsing library to perform deep analysis of PDF files. It not only extracts plain text but also accurately locates the coordinates and font information of the text; at the same time, it extracts embedded vector graphics (such as chemical structural formulas) and raster images (such as gel electrophoresis photographs) from the document and saves them as independent image elements.

[0025] The content of the Word format electronic document is read using the Office API interface to obtain Word data; Use a table parsing tool to extract cell data from an Excel-formatted electronic document; Specifically, using the Apache POI or OpenPyXL library, all worksheets in the workbook are read. This not only retrieves the final value of each cell but also parses cell formulas, comments, and cell formatting (such as number format and units). For charts, their data source references and chart type are extracted for subsequent reconstruction.

[0026] For the waveform data, digital signal processing algorithms are used for feature extraction and digitization to obtain waveform feature data; Specifically, depending on the instrument type, appropriate digital filters (such as Savitzky-Golay smoothing filters for spectral data or low-pass filters for time-series data) are applied for filtering, and key feature points in the waveform, such as peaks, inflection points, and plateau periods, are automatically detected and marked, and these features are parameterized.

[0027] S123: Standardize the scanned data, PDF data, Word data, cell data, and waveform feature data to obtain the preprocessed raw record dataset.

[0028] Specifically, standardization processes include, but are not limited to: unified encoding of text content, unified representation of numerical values, unified processing of coordinates, and unified processing of timestamps.

[0029] In this embodiment, experimental data from various sources such as paper records, electronic documents, and instrument waveforms are automatically aligned and correlated, effectively solving the problem of information fragmentation caused by inconsistent data formats and scattered sources in traditional methods. The entire processing flow requires no manual intervention, significantly improving the integrity and efficiency of data integration.

[0030] In this embodiment, step S2 involves dynamic knowledge fusion based on the preprocessed original record dataset to obtain a structured original record dataset, including: S21: Perform document structure analysis on the preprocessed original record dataset, identify the title, body text, table, and chart elements in the preprocessed original record dataset, and construct a document structure tree based on the spatial relationship between the elements; Specifically, a spatial-semantic association parsing engine is introduced to identify and classify the various elements in the document, and to precisely mark the position and scope of each identified element, thereby analyzing the spatial relationships and logical hierarchy between elements. For example, it is determined that paragraph paragrah5 is visually adjacent to table table2, and semantically likely describes that table, thus establishing a descriptive relationship between paragrah5 and table2. Ultimately, all elements and their interrelationships are organized into a document structure tree. S22: Based on the document structure tree, locate the text region in the preprocessed original record dataset, and combine it with the preset domain knowledge graph to extract character features and perform semantic matching on the text content of the text region to obtain knowledge-enhanced text content. Specifically, the character sequences are initially identified based on text detection and recognition models such as PaddleOCR, and then interact with the domain knowledge graph in real time to obtain knowledge-enhanced text content.

[0031] For example, when the initial identification result is "NaCl buffer": Search the knowledge graph using "NaCl" and "buffer solution" as query terms; The knowledge graph returns the following related information: There is a standardized term "NaCl" (sodium chloride), which is a common solute in "buffer solution"; "buffer solution" has the property "pH value".

[0032] Based on character shape similarity ("1" vs. "l") and semantic context ("buffer"), semantic matching is performed to correct "NaC1" and bind it to an entity in the knowledge graph; The final output is "[Chemical: NaCl] buffer, pH value is [Value: 8.5]".

[0033] Through the above process, the original text is transformed into knowledge-enhanced text content. Each key term is linked to a unique concept ID in the knowledge graph, along with its semantic type (such as chemical substance, instrument, parameter).

[0034] S23: Based on the document structure tree, locate the table and chart areas in the preprocessed original record dataset, extract the data from the table and chart areas, and reorganize them into structured format data; Specifically, firstly, a detection model based on Table Transformer is used to accurately locate the outer border of the table and the boundaries of all internal cells; then, the alignment relationship and spacing between cells are analyzed to construct a row and column logical structure model of the table and correctly handle merged cells; next, for the content in each cell, a dynamic knowledge binding recognition system (which shares the same core as text recognition, namely character feature extraction and semantic matching) is called to recognize and semantically bind the content, ensuring that the professional terms in the table are also corrected and enhanced; finally, the recognized content is filled according to the constructed row and column model to generate a two-dimensional data structure. S24: Link and integrate the knowledge-enhanced text content and structured format data to obtain a structured raw record dataset.

[0035] Furthermore, the domain knowledge graph is scalable and integrates one or more of the following: a terminology dictionary, experimental operating procedures, equipment parameter database, and chemical substance attribute database.

[0036] In this embodiment, a built-in dynamic knowledge fusion mechanism can automatically identify and correct laboratory-specific professional terms, chemical formulas, and physical symbols, transforming them into standardized expressions. This process, combined with contextual semantic analysis and a domain knowledge base, effectively overcomes the recognition bottleneck of traditional recognition methods when dealing with handwritten text, nested tables, and non-standard symbols, significantly improving the recognition accuracy of key information.

[0037] In this embodiment, step S3 performs three-level collaborative verification on the structured raw record dataset to obtain verification results, and filters the data subset that passes verification based on the verification results, including: The structured raw record dataset is subjected to rule-level verification, knowledge-level verification, and learning-level verification in sequence. Data that passes all three levels of verification is marked as verified and included in the data subset; data that fails any level of verification is marked as suspicious data and fed back to the human-computer collaboration interface for updating the domain rule base.

[0038] Specifically, after suspicious data is fed back to the human-computer collaboration interface, the experimenter makes a final decision on the human-computer collaboration interface: confirming an error, confirming a correctness, or making corrections. Subsequently, based on the decision result, the new rules or corrected rules are confirmed to update the domain rule base, or used as training samples to be fed back to the dynamic training engine for incremental optimization of the anomaly detection model and other models.

[0039] The rule-level verification includes: verifying whether there are logical contradictions in the data in the structured original record dataset according to preset format rules and numerical range rules. If there are, the rule-level verification passes; otherwise, it fails.

[0040] The knowledge layer verification includes: verifying the semantic rationality of the data in the structured original record dataset in the context of the experiment to be identified, based on the dynamically loaded domain rule base. If the data is reasonable, the knowledge layer verification is passed; otherwise, it fails.

[0041] The learning layer verification includes: using a pre-trained anomaly detection model to identify anomalies in the structured raw record dataset; if no anomalies are identified, the learning layer verification is passed; otherwise, it fails the learning layer verification.

[0042] This embodiment introduces a three-level collaborative verification system, which combines rule-based, knowledge-based, and learning-based verification. This system not only verifies the format and scope of the data, but also performs in-depth review of the experimental logic based on domain rules. Furthermore, it can continuously optimize the identification strategy by learning historical error patterns. This mechanism significantly reduces the false negative rate of deep logic errors and enhances the credibility and scientific rigor of the experimental data.

[0043] In this embodiment, steps S4-S5 are as follows: the data subset is converted according to the field mapping rules of the LIMS system data to obtain the converted dataset; the converted dataset is input into the LIMS system for automatic storage and a standardized experimental report of the experiment to be identified is generated. Specifically, steps S4-S5 together complete the final delivery of data from the intelligent processing system to the production business system.

[0044] First, the validated subset of data is processed by a configurable standardization engine converter, which performs precise format conversion and semantic mapping based on a predefined field mapping rule base. This converter automatically performs operations such as direct copying, unit conversion, date formatting, and encoding conversion, transforming the knowledge-enhanced structured data into a standardized dataset that fully conforms to the target LIMS system's data model. This is typically encapsulated in a specific JSON or XML exchange format, thereby completely eliminating semantic gaps between systems.

[0045] Subsequently, by calling the RESTful API provided by LIMS, the transformed dataset is automatically and atomically written into the relevant database tables of LIMS, including those for samples, tasks, results, and reagent consumption. Key receipt information, such as the work order number returned by LIMS, is captured in real time, establishing a two-way traceability link with the original records in this system to ensure synchronized data updates. After successful data entry, the report generation service is automatically triggered. Based on authoritative data in LIMS, it drives a pre-set standardized report template, dynamically filling in basic experimental information, conditions, results, and automatically calculated conclusions, ultimately generating a PDF experimental report that meets quality system requirements. This report can be automatically archived to LIMS, distributed via email, or published online, while retaining complete version and audit logs.

[0046] Thus, the entire process achieves end-to-end, fully automated closed-loop processing from multi-source heterogeneous original records to trusted business data assets and deliverable standardized reports. This reduces data entry time from hours to seconds and report generation time from hours of manual work to minutes. While greatly improving efficiency, it ensures data consistency, accuracy, compliance, and traceability throughout the entire chain; it also greatly reduces the workload of manual configuration and verification, and improves the automation level of system integration. Example 2

[0047] This embodiment provides an intelligent laboratory original record automatic identification system, including: The preprocessing module is used to obtain the original record dataset of the experiment to be identified, and to preprocess the original record dataset to obtain the preprocessed original record dataset. The knowledge fusion module is used to perform dynamic knowledge fusion based on the preprocessed raw record dataset to obtain a structured raw record dataset. The collaborative verification module is used to perform three-level collaborative verification on the structured raw record dataset, obtain the verification results, and select the data subset that passes the verification based on the verification results; The conversion module is used to convert the data subset according to the field mapping rules of the LIMS system data to obtain the converted dataset; The generation module is used to input the transformed dataset into the LIMS system for automatic data entry and generate standardized experimental reports for the experiments to be identified.

[0048] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0049] Example 3

[0050] This embodiment provides an automatic identification method and system for intelligent laboratory raw records based on Embodiments 1 and 2. Taking chemical titration experiments and EMC radiation interference tests as examples, the analysis is as follows: (I) Chemical Titration Experiment Background: In chemistry labs, raw records of titration experiments often include handwritten chemical formulas, such as the molecular formula of glucose, C6H2O. 12 O6. The formula for calculating the amount of substance conservation in titration analysis (V1C1 = V2C2) and nested tables are difficult for traditional OCR to accurately recognize handwritten symbols and cannot verify whether the reaction logic is reasonable. V1 represents the volume of standard titrant consumed; C1 represents the known concentration of standard titrant; V2 represents the volume of the sample being titrated; and C2 represents the concentration to be measured in the sample being titrated.

[0051] Implementation

[0052] Preprocessing: Scan paper titration records (≥600dpi), simultaneously acquire CSV waveform data output by the electronic titrator, and align them using spatiotemporal tags; Knowledge Integration: Identifying "C6H" 12 "O6" is bound to "glucose", "V1C1" is identified and associated with the titration reaction formula; Collaborative verification: Check whether V1C1 = V2C2 ± 2% is satisfied; if the deviation exceeds the limit, it is marked as abnormal. LIMS Integration: Automatically maps the identified fields such as "titration volume", "concentration", and "reactants" to the corresponding fields in the LIMS system, generating a certificate with digital signature and data lineage; Results: Recognition accuracy increased to 98%, compared to only 65% ​​for traditional OCR; the rate of missed logical errors decreased to 0.7%, compared to >60% for traditional solutions; and the rate of manual intervention decreased from 40% to 5%.

[0053] (ii) EMC radiated interference test Background: Electromagnetic compatibility test reports typically include spectrum diagrams and tabular data, such as frequency points and field strength values, as well as technical symbols, such as dBμV / m. Traditional methods require manual extraction of spectrum peaks and mapping to LIMS, which is inefficient and prone to errors. Implementation

[0054] Preprocessing: Simultaneously acquire PDF test reports and CSV waveform data output by the spectrum analyzer, and construct a cross-media index; Knowledge fusion: Identify sections such as "Test Configuration" and "Result Data" in the report, locate the spectrum graph and table areas; parse "Peak@1GHz" in the table as "radiated interference peak value" and align it with the peak value in the spectrum graph; Collaborative verification: Verify whether the electric field strength conforms to the E∝1 / r attenuation model. If it does not conform, an anomaly will be indicated. In the formula, E is the electric field strength; ∝ is proportional to the sign; and r is the distance. LIMS Integration: Automatically match “radiation field strength @763MHz” to the “RF peak” field of LIMS through semantic mapping and generate a visual lineage certificate; Results: The accuracy of spectrum annotation and resolution reached 97%, the detection rate of field strength attenuation anomalies increased to 99.3%, and the data processing time was reduced from 30 minutes / item to seconds.

[0055] In summary, based on Embodiments 1 and 2, an intelligent laboratory raw record automatic identification method and system are provided. Through multimodal data identification, three-level logical verification and automated system integration, the entire process from raw records to trusted data assets is made intelligent, improving data processing efficiency by two orders of magnitude while increasing identification accuracy.

[0056] Example 4

[0057] This embodiment provides a computer-readable storage medium storing a computer program that, when executed, implements the automatic identification method for original records in an intelligent laboratory as described in claim 1.

[0058] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for automatic identification of smart laboratory raw records, characterized in that, include: Obtain the original record dataset of the experiment to be identified, and preprocess the original record dataset to obtain the preprocessed original record dataset; Dynamic knowledge fusion is performed on the preprocessed raw record dataset to obtain a structured raw record dataset; A three-level collaborative validation was performed on the structured raw record dataset to obtain the validation results, and a subset of data that passed the validation was selected based on the validation results; The data subset is converted according to the field mapping rules of the LIMS system data to obtain the converted dataset. The transformed dataset is input into the LIMS system for automatic storage and a standardized experimental report for the experiment to be identified is generated.

2. The intelligent laboratory raw record automatic identification method according to claim 1, characterized in that, The original record dataset includes: paper record data, electronic document data, and waveform data; the electronic document data includes electronic documents in PDF, Word, and Excel formats.

3. The intelligent laboratory raw record automatic identification method according to claim 2, characterized in that, The preprocessing includes: Content recognition is performed on the paper record data to obtain scanned data; The PDF parsing library is used to extract text and image information from PDF format electronic documents to obtain PDF data; The content of the Word format electronic document is read using the Office API interface to obtain Word data; Use a table parsing tool to extract cell data from an Excel-formatted electronic document; For the waveform data, digital signal processing algorithms are used for feature extraction and digitization to obtain waveform feature data; The scanned data, PDF data, Word data, cell data, and waveform feature data are standardized to obtain the preprocessed raw record dataset.

4. The intelligent laboratory raw record automatic identification method according to claim 3, characterized in that, The dynamic knowledge fusion based on the preprocessed raw record dataset yields a structured raw record dataset, including: Document structure analysis is performed on the preprocessed raw record dataset to identify title, body text, table, and chart elements in the preprocessed raw record dataset, and a document structure tree is constructed based on the spatial relationship between the elements; Based on the document structure tree, the text region in the preprocessed original record dataset is located, and combined with the preset domain knowledge graph, character feature extraction and semantic matching are performed on the text content of the text region to obtain knowledge-enhanced text content. Based on the document structure tree, locate the table and chart areas in the preprocessed original record dataset, extract the data from the table and chart areas, and reorganize them into structured format data; By associating and integrating knowledge-enhanced text content with structured format data, a structured raw record dataset is obtained.

5. The intelligent laboratory raw record automatic identification method according to claim 4, characterized in that, The domain knowledge graph is scalable and integrates one or more of the following: a terminology dictionary, experimental operating procedures, equipment parameter database, and chemical substance property database.

6. The intelligent laboratory raw record automatic identification method according to claim 5, characterized in that, The three-level collaborative validation of the structured raw record dataset yields validation results, and a subset of data that passes validation is selected based on these results, including: The structured raw record dataset is subjected to rule-level verification, knowledge-level verification, and learning-level verification in sequence. Data that passes all three levels of verification is marked as verified and included in the data subset; data that fails any level of verification is marked as suspicious data and fed back to the human-computer collaboration interface for updating the domain rule base.

7. The intelligent laboratory raw record automatic identification method according to claim 6, characterized in that, The rule-layer verification includes: verifying whether there are logical contradictions in the data in the structured original record dataset according to preset format rules and numerical range rules. If there are contradictions, the rule-layer verification is passed; otherwise, the rule-layer verification is not passed.

8. The intelligent laboratory raw record automatic identification method according to claim 7, characterized in that, The knowledge layer verification includes: verifying the semantic rationality of the data in the structured original record dataset in the context of the experiment to be identified, based on the dynamically loaded domain rule base; if it is reasonable, the knowledge layer verification is passed; otherwise, the knowledge layer verification is not passed.

9. The intelligent laboratory raw record automatic identification method of claim 8, wherein, The learning layer verification includes: using a pre-trained anomaly detection model to identify anomalies in the data in the structured raw record dataset; if no anomalies are identified, the learning layer verification is passed; otherwise, the learning layer verification is failed.

10. An intelligent laboratory raw record automatic identification system, characterized in that, include: The preprocessing module is used to obtain the original record dataset of the experiment to be identified, and to preprocess the original record dataset to obtain the preprocessed original record dataset. The knowledge fusion module is used to perform dynamic knowledge fusion based on the preprocessed raw record dataset to obtain a structured raw record dataset. The collaborative verification module is used to perform three-level collaborative verification on the structured raw record dataset, obtain the verification results, and select the data subset that passes the verification based on the verification results; The conversion module is used to convert the data subset according to the field mapping rules of the LIMS system data to obtain the converted dataset; The generation module is used to input the transformed dataset into the LIMS system for automatic data entry and generate standardized experimental reports for the experiments to be identified.

Images