A method for automatic reading of a detection result field based on spectral map region positioning

Abstract

The application discloses a kind of based on spectral map area positioning's detection result field automatic reading method, it is related to image processing and detection data automatic reading technical field.The method obtains the result type candidate field in original detection record image, spectral map area positioning result and page association information, determines spectral map process section based on spectral map area positioning result and page association information, and determines the sample result section corresponding to target sample number based on page association information;Again, according to the paragraph attribution relationship of result type candidate field relative to spectral map process section and sample result section, field is shunted.Result type candidate field belonging to sample result section and not excluded by spectral map process section is read as target sample detection result field, and result type candidate field belonging to spectral map process section or paragraph attribution cannot be confirmed is not directly read as target sample detection result field, to reduce the risk of spectral map association process field miswriting target sample detection result.

Description

Technical Field

[0001] This invention relates to the field of image processing and automatic data reading technology, and in particular to a method for automatically reading detection result fields based on spectral region localization. Background Technology

[0002] Image processing and automatic data reading technologies are commonly used for the automatic entry of instrument detection data. In this process, the system typically needs to identify detection factors, detection results, units, spectra, and related numerical information in the original detection record images, and convert fields related to the target sample's detection results into structured reading results. The data in the original detection records can then serve as a source for subsequent detection data management or result verification.

[0003] Existing methods for automatically reading raw test records typically first perform text recognition, table recognition, or template region matching on the raw test record page, and then write the recognized fields into the structured reading results based on field names, header positions, or the proximity of fields and values. For raw test record pages with relatively stable field layouts and concentrated result areas, this type of method can complete the automatic reading of test result fields to a certain extent.

[0004] The existing technology has the following shortcomings: On the one hand, in the original detection record containing the spectral region, the spectrum itself is part of the original detection record, and the area surrounding the spectral region usually forms spectral association process content corresponding to the spectrum. Fields in this type of content may appear as result-type fields such as concentration, response value, content, and result, while the detection result aggregation range corresponding to the target sample number may also use similar field names, numerical forms, or table formats. In this case, identifying a field as a result-type candidate field only indicates that it has the candidate appearance for reading detection results, not that it belongs to the target sample's detection results. The consequence is that fields in the spectral association process content are easily mistakenly written into the structured reading results of the target sample, and this problem cannot be solved simply by improving the accuracy of text recognition or table line recognition.

[0005] On the other hand, existing methods typically use page proximity, header relationships, or field value correspondences as the basis for result reading. However, in raw spectral detection records, adjacent page ranges may serve as spectral process descriptions and target sample result collection, respectively. If reading is still based solely on field names, header positions, or proximity numerical relationships, it is difficult to distinguish between process fields and target sample detection result fields, which both present the appearance of result-type fields. The essence of this misreading risk is not the inability to identify fields, but rather the lack of further judgment regarding the attribution of field readings after the fields have already formed candidate result-type fields.

[0006] Based on this, the present invention provides an automatic reading method for detection result fields based on spectral region localization, so as to determine the reading direction of the field according to the paragraph belonging relationship of the field relative to the spectral process segment and the sample result segment after the result candidate field is formed. Summary of the Invention

[0007] To achieve the above objectives, the present invention provides the following technical solution: An automatic reading method for detection result fields based on spectral region localization includes: S101, Obtain the result-type candidate fields, spectral region localization results, and page association information from the original detection record image; S102, Based on the spectral region location results and page association information, determine the range of pages continuously associated with the spectral region, and use this range of pages as the spectral process segment; S103, Based on the page association information, determine the detection result collection range corresponding to the target sample number, and use the detection result collection range as the sample result segment; S104. Based on the segment affiliation relationship of the result-type candidate fields with respect to the spectrum process segment and the sample result segment, the result-type candidate fields are divided into fields. Among them, the result-type candidate fields that belong to the sample result segment and are not excluded by the spectrum process segment are read as target sample detection result fields, while the result-type candidate fields that belong to the spectrum process segment or whose segment affiliation cannot be confirmed are not read directly as target sample detection result fields.

[0008] In a preferred embodiment, the determination of the spectrum process segment includes: taking the spectrum region corresponding to the spectrum region location result as the starting range, and expanding to the adjacent page range based on page association information; when there is a continuous relationship in page structure, reading order, or chart combination between the adjacent page range and the spectrum region, the adjacent page range is incorporated into the spectrum process segment.

[0009] In a preferred embodiment, when determining the spectrum process segment, the end boundary is determined based on the paragraph switching content that can indicate the end of the current spectrum process segment; the paragraph switching content includes a new target sample number, a result summary title, the next spectrum region, or the next calibration process title; when paragraph switching content appears, the expansion of the current spectrum process segment is stopped, and the page range that has been incorporated before the stop position is taken as the spectrum process segment corresponding to the current spectrum region.

[0010] In a preferred embodiment, when there are multiple spectral regions in the original detection record image, a corresponding spectral process segment is established with each spectral region as the starting range, and the boundary of each spectral process segment is determined according to the title switching, reading order switching or page separation relationship between adjacent spectral regions; when the result-type candidate field is located between adjacent spectral process segments and its affiliation cannot be confirmed according to the page association information, the result-type candidate field is not directly read as the target sample detection result field.

[0011] In a preferred embodiment, when performing field splitting on the result-type candidate fields, the spectrum process segment forms a reading exclusion constraint on the result-type candidate fields; when a result-type candidate field belongs to the spectrum process segment, and it cannot be confirmed that it belongs to the sample result segment based on the detection result aggregation range corresponding to the target sample number, the result-type candidate field is treated as a process field candidate and is not written into the target sample detection result field in the structured reading result.

[0012] In a preferred embodiment, page association information is formed through a page structure diagram; the page structure diagram uses fields, tables, titles, spectral regions or pagination regions in the original detection record image as page objects, and records the positional relationship, reading order relationship and structural belonging relationship between page objects; the paragraph belonging of the result-type candidate field is determined based on the association path between the page object where the result-type candidate field is located and the spectral region, target sample number or detection result collection range.

[0013] In a preferred embodiment, obtaining result-type candidate fields includes: performing text recognition, layout recognition, or table structure recognition on the original detection record image; when the recognized text field or table field forms a field value correspondence with the numerical value, unit, detection factor, or result table header that can represent the candidate use of the detection result, the text field or table field is determined as a result-type candidate field; page titles, legends, coordinate axis labels, headers, or footers that do not form a field value correspondence are not considered as result-type candidate fields.

[0014] In a preferred embodiment, determining the sample result segment includes: identifying the page object corresponding to the target sample number in the page structure diagram; searching for the detection result aggregation range defined by the target sample number based on page association information; the detection result aggregation range is a page range that can be traced back to the target sample number and includes the content corresponding to the detection factors and result values; and incorporating the detection result aggregation range into the sample result segment.

[0015] In a preferred embodiment, when the sample result segment continues across pages, it is determined whether the result-type candidate fields on the next page inherit the sample result segment affiliation from the previous page, based on the continuation table title, repeated table header, target sample number in the header, or the continuity of the table structure between the previous and next pages. When a new target sample number, quality control segment title, spectrum process segment starting range, or result table header structure changes on the next page, the continuation of the sample result segment across pages on the previous page is stopped.

[0016] In a preferred embodiment, the result-type candidate field is divided into fields, including: when a result-type candidate field belongs to both the spectrum process segment and the sample result segment, or when the segment assignment results obtained by page position, reading order, or structural assignment are inconsistent, the result-type candidate field is marked as a field to be reviewed and is not written into the target sample detection result field in the structured reading result; when a result-type candidate field belongs to neither the spectrum process segment nor the sample result segment, the result-type candidate field is marked as a non-target aggregation field; the structured reading result records the detection factor, field value, unit, page position, and field division result corresponding to the target sample detection result field being read.

[0017] The effects and advantages of the automatic reading method for detection result fields based on spectral region localization in this invention are as follows: Before writing result-type candidate fields into the structured reading results, this invention first uses spectral region positioning results and page association information to determine the spectral process segment, and then uses the detection result aggregation range corresponding to the target sample number to determine the sample result segment. This changes the field reading process from directly writing based on field name, header position, or proximity to determining whether to read it as a target sample detection result field based on segment attribution. Therefore, even if the spectral association process field and the target sample detection result field have similar appearances, the risk of mistakenly writing the spectral association process field into the target sample detection result is reduced. For fields with conflicting or unconfirmed segment attributions, they are not directly written into the target sample detection result field, thereby improving the consistency and traceability of the structured reading results. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram illustrating the field separation between the spectral process segment and the sample result segment in Example 1; Figure 3 This is a schematic diagram of the boundary processing of the spectrum process segment under the continuous arrangement of multiple correction curves in Example 2; Figure 4 This is a schematic diagram illustrating the generation of page association information and determination of sample result segments in the page structure diagram of Example 3. Detailed Implementation

[0019] The technical solution of the present invention will be described below with reference to the accompanying drawings and specific embodiments. The embodiments are used to illustrate the method implementation process of the present invention and do not limit the scope of protection of the present invention.

[0020] In this invention, the record field used to distinguish target samples in the original detection record is called the target sample number; the target sample number can be the sample number, sample name, sample code, or other record fields that can trace the detection results back to the same target sample.

[0021] This invention provides an automatic reading method for detection result fields based on spectral region localization. It is applicable to scenarios involving automatic entry of instrument detection data, particularly when the original detection record page contains both spectral-related process data and target sample detection result data, and these two data are similar in field name, numerical format, or table structure. The original detection record may include spectra, calibration process content, fitting instructions, standard sample calculation data, quality control data, detailed sample detection results, and other original data content used for detection data management or result verification. The spectral region is part of the original detection record, not merely an additional image area. Conventional processing typically first identifies text, tables, or template areas, and then writes the identified fields into the structured reading result K0 based on field names, header positions, or the proximity of fields to values. However, in original detection records containing spectral regions, fitting instructions, standard sample calculation tables, or process verification content are often present around the spectral region. These contents may also appear as result-type fields such as "concentration," "response value," "content," or "result." Identifying a field as a result-type candidate field does not necessarily mean that the field should be read as a target sample detection result field.

[0022] This invention does not involve adjustments to the detection instrument, detection method, original detection record generation rules, or spectrum calculation process. Instead, after forming result-type candidate fields in the original detection record image, it further utilizes page association information to determine their paragraph affiliation relative to the spectrum process segment and sample result segment, and completes field splitting accordingly.

[0023] The original detection record image can be a single page or a sequence of images formed by multiple pages of original detection record images. When processing the original detection record image, the field reading system Q0 can obtain page objects such as text blocks, numerical blocks, unit blocks, table blocks, title nodes, pagination areas, and spectral areas through text recognition, layout recognition, table structure recognition, and image region recognition.

[0024] The spectral region represents the graphical record area in the original detection record page corresponding to the detection process, and typically includes at least the graphical area where the main spectral element is located. The spectral title, legend, axis labels, or adjacent spectral descriptions can be determined as related content of the spectral region based on layout recognition results, page structure relationships, or original detection record layout rules. The spectral region location result records the page position of the spectral region in the original detection record image and the corresponding page object. This location result can come from image region recognition, layout recognition, or preset original detection record layout markers. When subsequently determining the spectral process segment, it is not required that the spectral title, legend, or axis labels be incorporated into the spectral region itself; as long as the aforementioned page objects can establish a correspondence with the main spectral element through page association information, they can serve as auxiliary criteria for determining the continuous range of the spectral process segment.

[0025] In one specific processing method, Q0 can first filter graphic regions in the original detection record image based on the layout recognition results, and then combine them with curve trajectories, coordinate axis annotations, legends, spectrum titles, or adjacent explanatory text to form candidate spectrum regions. For graphic regions that cannot establish a page association with curve trajectories, coordinate axis annotations, legends, spectrum titles, or adjacent explanatory text, Q0 will not consider them as candidate spectrum regions. If it is not stable to confirm whether the same graphic region belongs to a spectrum region, Q0 can mark the corresponding spectrum region location result as pending confirmation; fields adjacent to the spectrum region to be confirmed and having the appearance of result-type fields will not be directly written into the target sample detection result field during subsequent field splitting, but will be further judged based on the sample result segment attribution, and if the attribution cannot be confirmed, they will enter the field processing for pending review.

[0026] Result-type candidate fields refer to field objects that have the potential to be read after text recognition and table structure recognition. Q0 can combine the row-to-row, column-to-column, adjacency, or same-table-block relationships between text blocks, table fields, numerical blocks, unit blocks, detection factors, and result table headers to determine whether the page content forms a field value correspondence. Text objects or table fields that form a field value correspondence enter the scope of result-type candidate fields; text objects that only appear as page titles, legends, axis labels, original detection record headers, or original detection record footers, and do not form a field value correspondence with numerical blocks, unit blocks, detection factors, or result table headers, do not enter the scope of result-type candidate fields.

[0027] Page association information is used to record the relationships between page objects in the original detection record image. It includes at least the page position relationships used to determine the paragraph affiliation of result-type candidate fields, and is used in conjunction with at least one of reading order relationships and structural affiliation relationships. Page position relationships reflect the relative positions or adjacency of page objects; reading order relationships reflect the order in which page objects are read within the original detection record page; structural affiliation relationships reflect the affiliation relationships between fields, result rows, header areas, table blocks, title nodes, spectral areas, or pagination areas. Title nodes are used to represent page objects that can influence paragraph affiliation, such as the title of the target sample number, the result summary title, the correction process title, or the quality control section title.

[0028] When generating page association information, Q0 can establish a page structure diagram based on the page recognition results of the original detection record image. Nodes in the page structure diagram can include text block nodes, numerical block nodes, unit block nodes, table block nodes, title nodes, spectrogram region nodes, pagination region nodes, table header nodes, result row nodes, and field value nodes. Q0 establishes connecting edges based on the inclusion relationships, vertical adjacency relationships, horizontal adjacency relationships, row relationships, column relationships, table header constraint relationships, title limitation relationships, and continuation relationships between preceding and following pages of page objects. The reading order can be determined according to page number order, column order, vertical position, horizontal position, and the row and column order within a table block. When the original detection record page has a multi-column layout or spans multiple pages, Q0 can first determine the column or continuation relationship, and then determine the reading order of page objects within the corresponding column or continuation range. Therefore, page association information is not a single geometric distance relationship, but is formed by a combination of page position relationships, reading order relationships, and structural affiliation relationships.

[0029] In this invention, the page range continuously associated with the spectral region refers to the range of page content that can be determined based on page association information to still serve the current spectral region. This continuous association can be confirmed through page structure continuity, reading order continuity, or chart combination attribution. When the page content is close to the spectral region, but a title change, target sample number change, result summary title change, or next spectral region change has occurred, it is no longer considered as the page range continuously associated with the current spectral region.

[0030] The scope of test results corresponding to a target sample number refers to the page range that can be traced back to the same target sample number through page association information and contains the corresponding content of test factors and result values. The scope of test results may include result tables, result rows, result columns, or cross-page areas defined by the target sample number; fields that cannot be traced back to the current target sample number are not within the scope of test results corresponding to that target sample number.

[0031] Based on the above design, this invention constructs an automatic reading process for the detection result field, consisting of steps S101 to S104. (Refer to...) Figure 1 , Figure 1 This is a schematic diagram of the method flow of the present invention.

[0032] Step S101: Obtain the result-type candidate fields, spectral region location results, and page association information from the original detection record image. This step does not directly determine whether the result-type candidate fields belong to the target sample detection results based on field names, numerical forms, or table proximity relationships. Instead, it simultaneously obtains the location results of the spectral region on the page and the page association information used to determine the field paragraph affiliation, so as to further determine whether the result-type candidate fields actually belong to the spectral association process content or the target sample detection result content.

[0033] Step S102: Based on the spectral region localization result and page association information, determine the range of pages continuously associated with the spectral region, and use this range as the spectral process segment. The spectral region localization result is used to determine the page position or starting range of the spectral region in the original detection record image, and the page association information is used to determine whether the content of the pages surrounding the spectral region is still continuously associated with that spectral region. Q0 does not use the closest geometric distance as the sole criterion, but combines the continuity of page structure, the continuity of reading order, or the association of chart combinations to determine the extended range of the spectral process segment.

[0034] When determining the spectral process segment, structural continuity on the page does not simply equate to closest geometric distance. Q0 continues reading adjacent page objects from the spectral region along the page reading order, and determines whether subsequent page content still serves the current spectral region by considering title nodes, table block affiliations, and chart combination relationships. Fitting instructions, correlation coefficient instructions, standard sample calculation tables, or process verification instructions maintain continuity with the spectral region in terms of reading order and structural affiliation, and no new target sample number, result summary title, quality control section title, next spectral region, or next calibration process title appears in between; these can be incorporated into the spectral process segment as the spectral process content corresponding to the current spectral region. If adjacent page objects are close to the spectral region, but their title affiliation, table block affiliation, or field set has been transferred to the sample result segment, another calibration process segment, or other original detection record segment, they will not be incorporated into the current spectral process segment. If page position relationships, reading order relationships, and structural affiliations do not consistently point to the same spectral process segment, relevant result-type candidate fields are marked as fields to be reviewed.

[0035] Step S103: Based on the page association information, determine the detection result aggregation range corresponding to the target sample number, and use this detection result aggregation range as the sample result segment. The target sample number is used to limit the aggregation range of its subsequent result tables, result rows, or result fields; the detection result aggregation range is not simply equivalent to the entire result table, but refers to the page range that can be traced back to the same target sample number through the page association information and includes the content corresponding to the detection factors and result values.

[0036] A result-type candidate field belonging to the sample result segment means that the field's bounding box position, its location in the page structure diagram node, or its reading order position can be determined to belong to the detection result aggregation range corresponding to the target sample number. If the attribution results obtained from the bounding box position, reading order relationship, and structural attribution relationship are inconsistent, or if the same result-type candidate field can be traced back to multiple target sample numbers and a unique target sample number cannot be identified, it will not be directly read as a target sample detection result field.

[0037] Step S104: Based on the segment affiliation of the result-type candidate fields relative to the spectral process segment and the sample result segment, the result-type candidate fields are divided into fields. The core of field division is that only result-type candidate fields belonging to the sample result segment and not excluded by the spectral process segment are read as target sample detection result fields; result-type candidate fields belonging to the spectral process segment or whose segment affiliation cannot be confirmed are not directly read as target sample detection result fields.

[0038] Q0 processes each result-type candidate field in the same judgment order: first, it determines whether the field can be confirmed to belong to the spectral process segment; then, it determines whether the field can be confirmed to belong to the sample result segment through the detection result aggregation range corresponding to the target sample number. A result-type candidate field belonging to a segment means that the field's bounding box position, its location on the page structure diagram node, or its reading order position can be determined to belong to the page range corresponding to that segment. Specifically, Q0 can form a positional attribution result based on the overlap between the bounding box and the segment range; a structural attribution result based on the connection of the field's node to the spectral region or the page object corresponding to the target sample number via containment edges, row / column attribution edges, reading order edges, or title subordinate edges; and a reading order attribution result based on the field's reading order position on the original detection record page. When at least two of the above attribution results consistently point to the same paragraph, the candidate field for the result type is confirmed to belong to that paragraph. When the above attribution results point to the spectral process segment and the sample result segment respectively, or when the same candidate field for the result type can be traced back to multiple target sample numbers, multiple spectral process segments, or inconsistent title nodes, the unique paragraph attribution is not directly confirmed, but the candidate field for the result type is marked as a field to be reviewed. The above attribution confirmation method can set an overlap ratio threshold, a path length threshold, or a relationship consistency threshold according to the actual original detection record layout. These thresholds are used to adapt to different page layouts and do not constitute a processing category independent of field diversion. For example, for original detection record pages with a single-column vertical layout and relatively clear paragraph boundaries, the overlap ratio threshold can be set to 60%. For pages where attribution is traced back through the page structure diagram or where the table continues across pages, the path length threshold can be set to 4 connecting edges. For cases where the attribution of adjacent paragraph boundaries cannot be confirmed based on a single relationship, the aforementioned relationship consistency threshold, where at least two attribution results are consistent, is used.

[0039] When a result-type candidate field can only be confirmed to belong to the spectral process segment but not to the sample result segment, Q0 treats it as a process field candidate and does not write it into the target sample detection result field in the structured reading result K0. When a result-type candidate field can only be confirmed to belong to the sample result segment, or although it is adjacent to the spectral process segment but cannot be confirmed to belong to the spectral process segment, but can be confirmed to belong to the sample result segment through the detection result aggregation range corresponding to the target sample number, Q0 reads it as a target sample detection result field. When a result-type candidate field satisfies both spectral process segment and sample result segment affiliation, or when the segment affiliation results obtained from the recognition box position, reading order relationship, and structural affiliation relationship are inconsistent, Q0 does not write the field into the target sample detection result field but marks it as a field to be reviewed. When a result-type candidate field cannot be confirmed to belong to either the spectral process segment or the sample result segment, Q0 marks it as a non-target aggregation field.

[0040] If a field in the detection conditions, method information, internal standard information, or reference information does not belong to the spectral process section or sample result section, it shall be recorded as a non-target collection field and no separate sub-category shall be set.

[0041] The target sample detection result field in the structured read result K0 only records fields that have been confirmed to belong to the sample result segment and have not been excluded by the spectral process segment. For process field candidates, fields to be reviewed, and non-target aggregation fields, K0 can record their field values, units, page positions, adjacent paragraphs, and split results in the field splitting record for manual confirmation or rule review.

[0042] The implementation process of the method of the present invention will be described below with reference to specific embodiments. The type of original detection record, instrument platform, and page layout in the following embodiments can be set according to the actual original detection record automatic entry system. In the following embodiments, an image of an original detection record containing calibration curves, fitting descriptions, standard sample calculation data, and sample detection result details is used as an example.

[0043] In Example 1, refer to Figure 2 , Figure 2 This is a schematic diagram illustrating the field separation between the spectral process segment and the sample result segment in Example 1. The method of the present invention will be described using an example of a page where the calibration curve and the page area displayed as "Detailed Results" are arranged continuously.

[0044] The original detection record image P1, read sequentially along the page, includes the detection condition area, the calibration curve graph, the fitting instructions and standard sample calculation table below the calibration curve, and the subsequent page area displayed as "Detailed Results". In the "Detailed Results" page area, the detection records are correlated according to the target sample number, detection factor, and result value. After Q0 performs text recognition, layout recognition, and table structure recognition on P1, it forms page objects and their associated page information, and determines multiple result-type candidate fields based on the field value correspondence. Detection conditions, method descriptions, record fields, numerical fields, legends, or text objects in the header and footer of the original detection record, if they do not form a field value correspondence, are not included in the result-type candidate field range.

[0045] In P1, the concentration and error fields in the calibration process table and the concentration field in the result table displayed as "detailed results" all have the appearance of result-type fields, but their page affiliations are different. Q0 does not directly write the above fields into the structured read result K0 based on the field name or numerical form, but continues to determine their paragraph affiliation based on the page association information.

[0046] In this embodiment, the calibration curve is used as a representation of the spectral region. Based on the spectral region location result, Q0 uses the calibration curve G1 as the starting range of the spectral process segment A1. As the page continues to be read in the order it was read, the fitting instructions and standard sample calculation table below G1 still explain the establishment and verification process of the calibration curve. Therefore, Q0 defines the aforementioned region as the page range continuously associated with the calibration curve G1 and uses this page range as the spectral process segment A1. Thus, the spectral process segment A1 is not solely composed of the calibration curve itself, but rather covers the fitting instructions and standard sample calculations continuously associated with the calibration curve.

[0047] Within the spectral process segment A1, although the result-type candidate field F1 in the calibration process table has the appearance of "concentration" or a similar result field, its value is used to represent the standard sample calculation or calibration verification process and is not limited by the target sample number. Q0 treats F1 as a process field candidate and does not write it into the target sample detection result field.

[0048] When the page reading order reaches the result summary title or the page position that can indicate the start of the sample result segment, Q0 determines that the subsequent page area has broken away from the process continuity of the current calibration curve, stops the expansion of the spectrum process segment A1, and takes the page area that has been incorporated before the stop position as A1.

[0049] Subsequently, Q0 determines the sample result segment B1 based on the page association information. The page area displayed as "Detailed Results" is not necessarily entirely equivalent to the sample result segment. Q0 identifies the page object corresponding to the target sample number within this page area and searches for the detection result aggregation range defined by that target sample number. If a result-type candidate field can be traced back to the target sample number through a table structure and forms corresponding content with the corresponding detection factor and result value, it is then assigned to the sample result segment B1. In this embodiment, the result-type candidate field F2 is in the same result record as the target sample number and its corresponding detection factor can be determined through a table structure. Q0 confirms that F2 belongs to the sample result segment B1. Since F2 is not excluded by the spectral process segment A1, Q0 reads F2 as a target sample detection result field and writes it into the structured read result K0.

[0050] In this embodiment, the overlap ratio threshold for a single-column vertical layout page can be set to 60%. The overlap ratio is determined by the proportion of the overlap area between the recognition box of a result-type candidate field and the corresponding paragraph range to the area of ​​the recognition box. When the overlap ratio between the recognition box of a result-type candidate field and the corresponding page range of the spectrum process segment A1 or the sample result segment B1 is not less than 60%, a corresponding positional attribution result is formed. For example, if the main body of the recognition box of F1 is located within the spectrum process segment A1, and its overlap ratio with the corresponding page range of A1 is not less than 60%, a positional attribution result of F1 pointing to A1 is formed; if the main body of the recognition box of F2 is located within the sample result segment B1, and its overlap ratio with the corresponding page range of B1 is not less than 60%, a positional attribution result of F2 pointing to B1 is formed. The positional attribution result, along with the structure attribution result and the reading order attribution result, is still used together to confirm the paragraph attribution of the field.

[0051] If the result-type candidate field F3 is located at the boundary between the spectrum process segment A1 and the sample result segment B1, and the page position relationship shows that it is close to the result table displayed as "detailed results", but the reading order relationship or structural belonging relationship still cannot rule out that it belongs to the previous correction process segment, then Q0 will not read F3 directly as the target sample detection result field, but will mark F3 as a field to be reviewed, and record the page position, adjacent paragraphs and belonging conflict status of F3 in the field diversion record in K0.

[0052] For detection condition areas, method description areas, or other numerical fields not governed by spectral process segment A1 and sample result segment B1, Q0 does not include them in the target sample detection results simply because they have numerical or unit forms. These fields are marked as non-target aggregation fields and recorded in the field splitting record. Q0 records the target sample number, detection factor, field value, unit, page position, and field splitting result in K0. Specifically, the splitting result for F1 is a process field candidate, the splitting result for F2 is the target sample detection result field, and the splitting result for F3 is a field awaiting review.

[0053] Through this embodiment, Q0 distinguishes the process fields within the spectral process segment, the target sample detection result fields within the sample result segment, and the fields to be verified where the segment affiliation cannot be confirmed in the same original detection record image, thus avoiding the miswriting of result fields in the correction, fitting, or standard sample calculation process into the target sample detection result.

[0054] In Example 2, refer to Figure 3 , Figure 3 This is a schematic diagram of the boundary processing of the spectral process segment under the continuous arrangement of multiple calibration curves in Example 2. Taking the page in the original detection record image P2 where the calibration curves corresponding to multiple detection factors appear continuously as an example, the boundary determination of the spectral process segment and the verification processing of the boundary fields are explained.

[0055] In this embodiment, the formation method of result-type candidate fields and page association information can follow that of Embodiment 1. Multiple calibration curve units are arranged continuously along the page reading order on P2. Each calibration curve unit includes a calibration curve graph, a fitting description corresponding to the calibration curve, and a corresponding standard sample calculation table. Because the end of the standard sample calculation table of the previous calibration curve unit is close to the title or calibration curve graph of the next calibration curve unit, some result-type candidate fields may exhibit inconsistencies in page position relationships, reading order relationships, and structural attribution relationships. If read directly based on the adjacent positions of the fields, it is easy to incorrectly classify fields at the boundaries into adjacent spectral process segments, or further misprocess them as target sample detection result fields.

[0056] Q0 first uses the first calibration curve G21 as the starting range of the first spectral process segment A21. Continuing to read the page in sequence, the fitting instructions and standard calculation table below G21 still illustrate the establishment and verification process of the first calibration curve. Therefore, Q0 defines this area as the page range continuously associated with G21 and incorporates it into A21. For the result-type candidate field F21, which is clearly located within the standard calculation table, Q0 confirms that F21 belongs to A21. Although F21 has the appearance of result fields such as concentration, error, or intensity, its page assignment points to calibration process data. Therefore, Q0 treats F21 as a process field candidate and does not read it as a target sample detection result field.

[0057] When the page reading order reaches the second calibration curve G22, Q0 identifies G22 as the next spectral region and stops the expansion of the first spectral process segment A21, taking the page region already incorporated before G22 as A21. Subsequently, Q0 uses G22 as the starting range of the second spectral process segment A22 and establishes the second spectral process segment A22. For subsequent consecutive calibration curve units, Q0 also uses the spectral region corresponding to each calibration curve as the starting range, and determines whether there is a title switch, reading order switch, or page separation between adjacent spectral regions based on the adjacent calibration curve titles, the starting position of the calibration curve, the ending position of the standard calculation table, and the page reading order switching position, and determines the boundary of each spectral process segment accordingly, instead of merging multiple consecutive calibration curve units into one spectral process segment.

[0058] When multiple calibration curve units are arranged consecutively, Q0 does not only define boundaries based on page distance. The same title level can be confirmed by the hierarchical relationship of the calibration curve title or process title in the page structure diagram; the same chart combination can be confirmed by the continuity of the calibration curve graph, fitting description, and standard calculation table in the reading order; the affiliation of adjacent table blocks can be determined by the table title, header structure, table block position, and adjacent explanatory text; the continuity of process description content can be confirmed by the absence of paragraph transitions between the fitting description, correlation coefficient description, standard calculation table, and process verification description. If there is a clear blank dividing area between two adjacent spectrogram regions, and the title level, table block affiliation, or reading order boundary before and after the blank dividing area has changed, Q0 uses this blank dividing area as an auxiliary boundary between the two spectrogram process segments; if the above relationships still cannot uniformly point to the same spectrogram process segment, the boundary field remains pending review and stream division results.

[0059] In P2, if the result-type candidate field F22 is located between the end of A21 and the beginning of G22 (for example, if the identification box of F22 is close to the last row of the first standard sample calculation table and is immediately below the second calibration curve G22), causing the page position relationship, reading order relationship, and title subordination relationship to not consistently point to A21 or A22, then Q0 will not directly confirm the paragraph assignment of F22, nor will it directly read F22 as a target sample detection result field. Q0 will mark F22 as a field to be reviewed and record the page position of F22, adjacent spectral process segments, assignment conflict status, and the review result of the subordination in the field subordination record in K0.

[0060] In this embodiment, when multiple correction curve units are arranged consecutively on a page, the relationship consistency threshold is determined by the method of judging that at least two of the three types of attribution results are consistent. For F21, its position attribution result, structure attribution result, and reading order attribution result all point to the first spectrum process segment A21, so it is confirmed that F21 belongs to A21. For F22 located near the boundary between A21 and A22, if the position attribution result points to A21, while the reading order attribution result points to A22, and the structure attribution result cannot be confirmed due to title switching or table block attribution, then the relationship consistency threshold of at least two types of attribution results being consistent has not been reached. Q0 does not confirm the unique paragraph attribution of F22 and keeps F22 as a field to be reviewed.

[0061] If F22 is subsequently confirmed to belong to either the first standard sample calculation table or the second calibration curve unit, it will be treated as a candidate process field according to the confirmed spectral process segment assignment. If its segment assignment still cannot be confirmed, it will remain in the pending verification state and will not be written into the target sample detection result field. Thus, Q0 can establish spectral process segments and control the field destination at the boundaries of adjacent segments in a page with multiple calibration curves arranged consecutively.

[0062] In Example 3, refer to Figure 4 , Figure 4 This is a schematic diagram illustrating the formation of page association information and the determination of sample result segments in the page structure diagram of Example 3. Taking the result summary page and the cross-page continuation result table in the original detection record image P3 as examples, the formation of page association information, the determination of sample result segments, the processing of non-target collection fields, and the processing of cross-page continuation of sample result segments are explained. This example mainly describes the collection process on one side of the sample result segment; the method for determining the spectral process segment can follow that of Examples 1 and 2.

[0063] After performing text recognition, layout recognition, and table structure recognition on the original detection record image P3, Q0 generates a page structure diagram D3. Nodes in page structure diagram D3 can include text blocks, table blocks, spectral regions, pagination regions, page objects corresponding to target sample numbers, header regions, result rows, field values, and detection factors; connecting edges are used to record page position relationships, reading order relationships, row and column affiliation relationships, and title affiliation relationships. When a result-type candidate field can be traced back to the same target sample number along the result row, header region, and table block, and the tracing path does not pass through a new target sample number, quality control section title, or the starting range of the spectral process section, Q0 confirms that the field belongs to the detection result collection range corresponding to the current target sample number; if the tracing path is interrupted, or the same field can be traced back to multiple target sample numbers without confirming a unique collection object, Q0 does not directly confirm its sample result section affiliation.

[0064] In the results summary page P31 of P3, the target sample number appears at the top of the page, and the results table T31 appears in the middle or bottom of the page. The results rows in T31 are corresponding according to the detection factors and results values. Q0 establishes the structural belonging relationship between the page object corresponding to the target sample number and the results table T31 in the page structure diagram D3, and further establishes the row and column belonging relationship between the header area, results rows and field values. For the result-type candidate field F31, if F31 is located in the result row corresponding to the detection factor, and can be traced back to the result row, header area, results table T31 and the page object corresponding to the target sample number in sequence, then Q0 confirms that F31 belongs to the sample result segment B31; since F31 is not excluded by the spectral process segment, Q0 reads F31 as the target sample detection result field and writes it into the structured read result K0. For fields related to method information, acquisition time, calibration method, dilution information, instrument conditions, internal standard, reference, or quality control, even if they contain numerical, unit, or time information, as long as their page structure does not correspond to the target sample test results, Q0 will still mark the corresponding result-type candidate field F32 or F33 as a non-target aggregation field and record it in the field splitting record.

[0065] In the page structure diagram D3 of this embodiment, the path length threshold can be set to 4 connecting edges. The path length is determined by the number of connecting edges that the node containing the result-type candidate field passes through sequentially to the page object corresponding to the target sample number or the page object within the scope of the detection result collection. For F31, its field value node is sequentially connected to the page object corresponding to the target sample number through the result row, the header area, and the result table T31. The associated path does not exceed 4 connecting edges, and the path does not pass through a new target sample number, quality control section title, or the starting range of the spectrum process section. Therefore, F31 is assigned to the sample result segment B31. For F34, which continues across pages, its result row is connected to the page object corresponding to the target sample number in the previous page through the continuous relationship of the result rows of the previous and next pages, repeated headers, or pagination area nodes. When the associated path does not exceed 4 connecting edges, and there is no paragraph switching content in the path, F34 is assigned to the sample result segment of the previous page. If the path has more than 4 connecting edges, or if the path passes through a new target sample number, quality control section title, spectrum process section starting range, or result table header structure switching position, then the current sample result section will not be determined based on the path.

[0066] In another continuous page processing step, P3 also includes adjacent result pages P32 and P33. P32 contains the header of the result table and some result records. P33 continues with the same or corresponding header structure as P32, and lists the target sample number, detection factor, and detection results in the order of page reading. Q0 determines whether the result-type candidate fields in P33 inherit the sample result segment affiliation from P32 based on the pagination area nodes, the order of page reading, the header correspondence, and the continuity of result rows in the page structure diagram D3. The result-type candidate fields on the next page only inherit the sample result segment affiliation from the previous page if they can be traced back to the same target sample through the target sample number, the target sample number in the header, a repeated header, or a continuous result row structure. If the next page contains a new target sample number, quality control segment title, spectrum process segment starting range, calibration curve title, or the result header structure has changed, Q0 stops the cross-page continuation of the sample result segment from the previous page and re-determines the result-type candidate fields after the stop position according to the new segment affiliation in the current page. For the result-type candidate field F34 that meets the inheritance conditions, if F34 belongs to the detection result collection range and is not excluded by the spectral process segment, then Q0 reads F34 as the target sample detection result field and records its page position and cross-page inheritance status in K0. For the result-type candidate field F35 after inheritance stops, if F35 cannot be traced back to the current target sample number and does not belong to the confirmed sample result segment, then Q0 treats F35 as a non-target collection field; if F35 is within the detection result collection range limited by the new target sample number, then it is judged separately according to the new sample result segment, and is not written into the structured reading result K0 of the previous target sample. Q0 records in K0 the target sample number, detection factor, field value, unit, page position, field splitting result, and necessary cross-page inheritance status corresponding to the read field.

[0067] In this embodiment, Q0 utilizes page structure diagram D3 to organize the page association information between the target sample number, header area, result row, field values, and pagination area, so that the reading of result-type candidate fields no longer depends solely on numerical form, unit form, or table proximity. For fields that can be traced back to the target sample number and belong to the detection result aggregation range, Q0 reads them as the target sample detection result; for method information, internal standard information, reference information, and fields after cross-page inheritance stops, Q0 does not mistakenly write them into the target sample detection result.

[0068] As a comparative example, without the method of this invention, if the field reading system only reads based on field names, result headers, or the proximity of fields and values, fields such as "concentration," "response value," "error," and "correlation coefficient" in the calibration curve, fitting description section, or standard sample calculation table are easily written into the target sample detection result because they have the appearance of result-type fields. With the method of this invention, Q0 first determines the spectral process segment based on the calibration curve and its continuously associated fitting description and standard sample calculation table, and then determines the sample result segment based on the detection result aggregation range corresponding to the target sample number. Fields belonging to the spectral process segment are processed as process field candidates, and fields with boundary conflicts are marked as fields to be reviewed. Only fields confirmed to belong to the sample result segment and not excluded by the spectral process segment are written into the structured reading result K0.

[0069] The field reading system Q0 in the above embodiments can be implemented by software, hardware, firmware, or a combination thereof. When implemented by software, the program instructions that implement the above processing can be stored in a computer-readable storage medium and read and executed by a processor to complete processes such as acquiring the original detection record image, determining the result-type candidate field, reading the spectral region positioning result, forming page association information, determining the spectral process segment, determining the detection result aggregation range, determining the sample result segment, field splitting, and generating the structured reading result K0.

[0070] Those skilled in the art can make routine adjustments to page object recognition, page structure diagram construction, spectral process segment boundary determination, test result aggregation range determination, cross-page continuation judgment, field splitting records, and output methods for fields to be reviewed, based on the deployment methods of the actual original test record automatic entry system, test instrument recording format, page recognition module, table parsing module, and structured reading result output system. As long as it still revolves around the technical concept of using spectral region positioning results and page association information to determine the spectral process segment and the sample result segment corresponding to the target sample number, and determining whether to read the result-type candidate field as the target sample test result field based on the segment affiliation relationship of the result-type candidate field with respect to the spectral process segment and the sample result segment, it should all fall within the protection scope of this invention.

Claims

1. A method for automatically reading detection result fields based on spectral region localization, characterized in that, include: S101, Obtain the result-type candidate fields, spectral region localization results, and page association information from the original detection record image; S102, Based on the spectral region location results and page association information, determine the range of pages continuously associated with the spectral region, and use this range of pages as the spectral process segment; S103, Based on the page association information, determine the detection result collection range corresponding to the target sample number, and use the detection result collection range as the sample result segment; S104. Based on the segment affiliation relationship of the result-type candidate fields with respect to the spectrum process segment and the sample result segment, the result-type candidate fields are divided into fields. Among them, the result-type candidate fields that belong to the sample result segment and are not excluded by the spectrum process segment are read as target sample detection result fields, while the result-type candidate fields that belong to the spectrum process segment or whose segment affiliation cannot be confirmed are not read directly as target sample detection result fields.

2. The method for automatically reading detection result fields based on spectral region localization according to claim 1, characterized in that, The process of determining the spectrum includes: taking the spectrum region corresponding to the spectrum region location result as the starting range, and expanding to the range of adjacent pages based on page association information; when there is a continuous relationship of page structure, reading order, or chart combination between the adjacent page range and the spectrum region, the adjacent page range is incorporated into the spectrum process.

3. The method for automatically reading detection result fields based on spectral region localization according to claim 2, characterized in that, When determining the spectrum process segment, the end boundary is determined based on the paragraph switching content that can indicate the end of the current spectrum process segment; the paragraph switching content includes the new target sample number, the result summary title, the next spectrum area or the next calibration process title; when paragraph switching content appears, the expansion of the current spectrum process segment is stopped, and the page range that has been incorporated before the stop position is taken as the spectrum process segment corresponding to the current spectrum area.

4. The method for automatically reading detection result fields based on spectral region localization according to claim 2, characterized in that, When there are multiple spectral regions in the original detection record image, a corresponding spectral process segment is established with each spectral region as the starting range. The boundary of each spectral process segment is determined according to the title switching, reading order switching or page separation relationship between adjacent spectral regions. When the result-type candidate field is located between adjacent spectral process segments and its ownership cannot be confirmed according to the page association information, the result-type candidate field is not directly read as the target sample detection result field.

5. The method for automatically reading detection result fields based on spectral region localization according to claim 1, characterized in that, When performing field splitting on result-type candidate fields, the spectrum process segment forms a read exclusion constraint on the result-type candidate fields; when a result-type candidate field belongs to the spectrum process segment, and its belonging to the sample result segment cannot be confirmed based on the detection result aggregation range corresponding to the target sample number, the result-type candidate field is treated as a process field candidate and is not written into the target sample detection result field in the structured read results.

6. The method for automatically reading detection result fields based on spectral region localization according to claim 1, characterized in that, Page association information is formed through a page structure diagram; the page structure diagram uses fields, tables, titles, spectral regions or pagination regions in the original detection record image as page objects, and records the positional relationship, reading order relationship and structural belonging relationship between page objects; The paragraph affiliation of the result-type candidate field is determined based on the association path between the page object where the result-type candidate field is located and the spectral region, target sample number, or detection result aggregation range.

7. The method for automatically reading detection result fields based on spectral region localization according to claim 1, characterized in that, Obtaining result-type candidate fields includes: performing text recognition, layout recognition, or table structure recognition on the original detection record image; when the recognized text field or table field forms a field value correspondence with the numerical value, unit, detection factor, or result table header that can represent the candidate use of the detection result, the text field or table field is determined as a result-type candidate field; page titles, legends, coordinate axis labels, headers, or footers that do not form a field value correspondence are not considered as result-type candidate fields.

8. The method for automatically reading detection result fields based on spectral region localization according to claim 6, characterized in that, Determining the sample result segment includes: identifying the page object corresponding to the target sample number in the page structure diagram; finding the detection result aggregation range limited by the target sample number based on page association information; the detection result aggregation range is a page range that can be traced back to the target sample number and includes the content corresponding to the detection factors and result values; and incorporating the detection result aggregation range into the sample result segment.

9. The method for automatically reading detection result fields based on spectral region localization according to claim 8, characterized in that, When the sample result segment continues across pages, the system determines whether the result-type candidate fields on the next page inherit the sample result segment affiliation from the previous page, based on the continuation table title, repeated table header, target sample number in the header, or the continuity of the table structure between the previous and next pages. When a new target sample number, quality control segment title, spectrum process segment starting range, or result table header structure changes on the next page, the continuation of the sample result segment across pages on the previous page is stopped.

10. The method for automatically reading detection result fields based on spectral region localization according to claim 1, characterized in that, Field splitting is performed on result-type candidate fields, including: when a result-type candidate field belongs to both the spectrum process segment and the sample result segment, or when the segment assignment results obtained by page position, reading order, or structural assignment are inconsistent, the result-type candidate field is marked as a field to be reviewed and is not written into the target sample detection result field in the structured reading result; when a result-type candidate field belongs to neither the spectrum process segment nor the sample result segment, the result-type candidate field is marked as a non-target settling field; the structured reading result records the detection factor, field value, unit, page position, and field splitting result corresponding to the read target sample detection result field.

Images