Training method of recognition model, recognition method of information table in drawing
By training a recognition model and combining it with an item information database, and using convolutional neural networks and loss function adjustments, the problem of low confidence in the recognition of drawing information tables was solved, achieving accurate recognition and error correction at the field level, thus improving recognition accuracy and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA NUCLEAR IND FIFTH CONSTR CO LTD
- Filing Date
- 2023-07-28
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies fail to effectively combine engineering expertise for detailed breakdown when identifying information tables in drawings, resulting in low confidence levels in the identification results and a lack of field-level review and error correction capabilities.
By training a recognition model and combining it with an item information database, the model can identify information tables in drawings. The model parameters are adjusted to improve recognition accuracy. Convolutional neural networks and other structures are used, and cross-entropy loss functions are employed to adjust the model, thereby achieving accurate recognition of item names, attributes, and categories.
It improved the confidence level of information tables in drawings, enabled field-level review and error correction, reduced manual intervention, and improved recognition accuracy and work efficiency.
Smart Images
Figure CN116824614B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image recognition, and particularly to a training method for a recognition model for recognizing information tables in drawings, a method for recognizing information tables in drawings, a device for recognizing information tables in drawings, and a computer-readable storage medium. Background Technology
[0002] Drawings typically include both image and text content, requiring recognition technology to identify and digitize the content. Currently, some related technologies use OCR (Optical Character Recognition) to extract text from drawings and export recognition results. However, the exported results are often large blocks of text, such as a description of an item; the review and error correction functions remain limited to reviewing and correcting errors between recognized content. Current drawing recognition technologies lack fine-grained breakdown based on engineering expertise, merely applying OCR technology in a rudimentary way, resulting in insufficient confidence in the exported material lists; and they haven't accumulated a database of fields or item information identified from drawings, only comparing confidence levels between recognized content, failing to establish field-level review and error correction capabilities. Summary of the Invention
[0003] To overcome the problems existing in the related art, exemplary embodiments of this disclosure provide a training method for a recognition model for recognizing information tables in drawings, a method for recognizing information tables in drawings, a device for recognizing information tables in drawings, and a computer-readable storage medium.
[0004] An exemplary embodiment of the first aspect of this disclosure provides a training method for a recognition model used to identify information tables in a drawing. The training method includes: identifying information tables in the drawing using the recognition model to obtain name recognition information, at least one attribute recognition information, and category recognition information corresponding to each item in the information table, wherein the information table contains the name of each item in the drawing and descriptive text containing all attributes of each item; calling an item information database, wherein the item information database contains the name of each item among a plurality of items, at least one attribute of each item, and the category corresponding to the at least one attribute of each item; adjusting the parameters of the recognition model based on the comparison results of the name recognition information, attribute recognition information, and category recognition information with the names, attributes, and categories of items in the item information database; and generating a trained recognition model.
[0005] In some embodiments, adjusting the parameters of the recognition model based on the comparison results between the name recognition information, attribute recognition information, and category recognition information and the name, attribute, and category of the item in the item information database includes: obtaining the corresponding name, category, and one or more attributes in the item information database based on the name recognition information and category recognition information; and adjusting the parameters of the recognition model if the attribute recognition information does not exist in the one or more attributes.
[0006] In some embodiments, the parameters of the recognition model are adjusted based on one or more of the following loss functions: cross-entropy loss function, mean squared error loss function, and connection-time classification loss function.
[0007] In some embodiments, the item information database is constructed in the following manner: based on the names of items in the information tables of multiple drawings, the names of the same item are associated to obtain the name of each item; based on the descriptive text of the items in the information tables of multiple drawings, the attributes of each item and the category corresponding to the attributes are obtained; based on the attributes of each item and the category corresponding to each item attribute, a dictionary category containing all item categories and the attribute of each item corresponding to the dictionary category are obtained; based on the name of each item, the dictionary category containing all item categories, and the attribute of each item corresponding to the dictionary category, the item information database is formed.
[0008] In some embodiments, obtaining the attributes of each item and the category corresponding to the attributes based on the descriptive text of items in the information tables of multiple drawings includes: splitting the descriptive text to obtain the attributes of each item; and obtaining the category corresponding to the attributes based on the attributes of each item.
[0009] In some embodiments, obtaining a dictionary category containing all item categories and the attribute of each item corresponding to the dictionary category based on the attributes of each item and the category corresponding to each item attribute includes: retrieving the category of each attribute of the first item to form a dictionary category corresponding to each attribute; retrieving the category of each attribute of the next item, and if it is the same as the category in the dictionary category, associating it with the same category; if it is different from the category in the dictionary category, adding a new category in the dictionary category and updating the dictionary category, repeating the process until the category of each attribute of all items is retrieved; obtaining a dictionary category containing all item categories and the attribute of each item corresponding to the dictionary category.
[0010] Secondly, according to some other exemplary embodiments, this disclosure also provides a method for identifying information tables in a drawing, the method comprising: acquiring a drawing with information tables; identifying the information tables in the drawing using an identification model generated by the training method described in the first aspect, and obtaining name identification information, at least one attribute identification information, and category identification information corresponding to each item in the information tables.
[0011] In some embodiments, the identification method further includes at least one of the following steps: comparing the name identification information, attribute identification information, and category identification information with the name, attribute, and category of items in the item information database; if the item information database does not contain any attribute identification information of any item, then issuing a reminder; if the item information database does not contain any attribute identification information of any item, and the attribute identification information is correctly identified, then adding the attribute identification information of any item and the category corresponding to the attribute identification information to the item information database.
[0012] Thirdly, according to some other exemplary embodiments, this disclosure also provides an identification device for information tables in drawings, characterized in that the identification device includes: an acquisition unit for acquiring drawings with information tables; and an identification unit for identifying information tables in drawings using an identification model generated by the training method described in the first aspect, and obtaining name identification information, at least one attribute identification information, and category identification information corresponding to each item in the information table.
[0013] Fourthly, according to some other exemplary embodiments, this disclosure also provides a computer-readable storage medium, characterized in that it stores instructions that, when executed by a processor, perform a training method for a recognition model for recognizing information tables in a drawing as described in the first aspect, or perform a recognition method for information tables in a drawing as described in the second aspect.
[0014] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: by combining knowledge of specific fields with that of items in drawings for deep integration, the long-segment attribute descriptions of items are structured and split, and each attribute and category of each attribute of the item are identified, thereby improving the confidence in the identification of information tables in drawings and enabling field-level review and error correction.
[0015] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description
[0016] The invention can be better understood by describing exemplary embodiments of the invention in conjunction with the accompanying drawings, in which:
[0017] Figure 1 This is a flowchart illustrating a training method for a recognition model used to identify information tables in drawings.
[0018] Figure 2 This is a flowchart illustrating another training method for a recognition model used to identify information tables in drawings.
[0019] Figure 3 This is a flowchart illustrating another training method for a recognition model used to identify information tables in drawings.
[0020] Figure 4 This is a flowchart illustrating another training method for a recognition model used to identify information tables in drawings.
[0021] Figure 5 This is a flowchart illustrating another training method for a recognition model used to identify information tables in drawings.
[0022] Figure 6 This is a schematic diagram of an information table shown in an embodiment of this disclosure.
[0023] Figure 7 This is a flowchart illustrating a method for identifying information tables in drawings according to an embodiment of this disclosure.
[0024] Figure 8 This is a schematic diagram of a module for identifying an information table in a drawing, as shown in an embodiment of this disclosure. Detailed Implementation
[0025] The following describes specific embodiments of the present invention. It should be noted that, in order to provide a concise description, this specification cannot exhaustively describe all features of the actual embodiments. It should be understood that, in the actual implementation of any embodiment, just as in any engineering or design project, various specific decisions are often made to achieve the developer's specific goals and to meet system-related or business-related constraints, and this can change from one embodiment to another. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this invention, some design, manufacturing, or production modifications based on the technical content disclosed herein are merely conventional technical means and should not be construed as insufficient content of this disclosure.
[0026] Unless otherwise defined, the technical or scientific terms used in the claims and description shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in the patent application description and claims of this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including" and similar terms mean that the element or object preceding "comprising" or "including" encompasses the element or object listed following "comprising" or "including" and its equivalents, and do not exclude other elements or objects. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, nor are they limited to direct or indirect connections.
[0027] To address the aforementioned technical issues, this disclosure provides a training method for a recognition model used to identify information tables in drawings, a method for identifying information tables in drawings, a device for identifying information tables in drawings, and a computer-readable storage medium. By deeply integrating knowledge of specific domains and items in drawings, the long-segment attribute descriptions of items are structured and broken down, identifying each attribute of the item and the category of each attribute, thereby improving the confidence level of identifying information tables in drawings and enabling field-level review and error correction.
[0028] like Figure 1 As shown, Figure 1 A flowchart illustrating a training method for a recognition model used to identify information tables in drawings is shown. The training method may include: Step S110, identifying information tables in the drawing using the recognition model to obtain name recognition information, at least one attribute recognition information, and category recognition information corresponding to each item in the information table, wherein the information table contains the name of each item in the drawing and descriptive text containing all attributes of each item; Step S120, calling an item information database, wherein the item information database contains the name of each item among multiple items, at least one attribute of each item, and the category corresponding to at least one attribute of each item; Step S130, adjusting the parameters of the recognition model based on the comparison results of the name recognition information, attribute recognition information, and category recognition information with the names, attributes, and categories of items in the item information database; Step S140, generating the trained recognition model.
[0029] This disclosure provides a training method for a recognition model. In step S110, a drawing can be recognized using a pre-set initial recognition model. This involves inputting the drawing into the recognition model, which then calculates and outputs the name recognition information of each item, at least one attribute recognition information corresponding to the item, and the category recognition information corresponding to each attribute. The name recognition information is the name of the item output by the recognition model, the attribute recognition information is the attribute corresponding to the item output by the recognition model, and the category recognition information is the category corresponding to each attribute output by the recognition model.
[0030] The recognition model can include a Convolutional Neural Network (CNN) model, which may include one or more convolutional layers, fully connected (FC) layers, softmax output layers, and other structures. The recognition model is trained by comparing the recognition results with corresponding fields in an object information database, thereby continuously adjusting parameters. Finally, once the recognition results converge to a threshold, training is complete, and the trained recognition model is generated.
[0031] This disclosure primarily concerns the identification of drawings, which, in addition to their image components, also contain textual components. The image components of a drawing may include the structure and connections of various objects, while the textual components may contain object information, such as the object's name and attributes. In a drawing, an object typically has a name and may have multiple attributes. Attributes can be specific details such as the object's model, material, height, length, width, density, and thickness. Each attribute can be represented by a few numbers, letters, symbols, or Chinese characters, and multiple attributes are recorded in the drawing as a longer descriptive text.
[0032] In this embodiment of the disclosure, the drawing can be a piping diagram or other types of drawings. Taking a piping diagram as an example, the image portion can be the connection structure of the pipe, and the items can include pipes, valves, flanges, tees, etc. The text portion can record the name and attributes of each item, such as the pipe's attributes including model, material, length, diameter, wall thickness, etc.
[0033] Table 1 provides an example of all or part of the text content in a drawing.
[0034] Table 1:
[0035] name Description text tube SCH40S, A312 TP304 / 304L, SMLS, PE, ASMBB36.19M
[0036] Among them, SCH40S is a pipe attribute, representing the wall thickness series; A312TP304 / 304L is another pipe attribute, representing the material; SMLS, PE, and ASMBB36.19M are also different pipe attributes, representing the pipe type, connection form, and standard, respectively.
[0037] The drawing may include an information table, which may contain the names of one or more items and descriptive text containing all attributes for each item. In this embodiment, the information table may record item information in tabular form. For example, the information table may include two columns, as shown in Table 1 above. The first column records the names of each item in the drawing, and the second column records all attributes corresponding to each item in the first column. All attributes are recorded in the form of a descriptive text. The information table may also include more columns, such as a serial number column, a quantity column, etc. Since each item corresponds to different attributes, the categories of attributes are different, and the number of attributes is also different, and the area of the drawing is limited, the image part needs to be displayed in a larger area, while the text part needs to save the area occupied by the drawing. Therefore, the attributes of an item need to be listed as a descriptive text in a table of the information table, with the attribute descriptive text of each item forming a column, thereby saving space. The recognition model trained by the training method of this disclosure can identify each attribute field of each item and make a confidence judgment based on each field, thereby improving the accuracy of recognition.
[0038] The recognition model disclosed herein can identify each attribute field separately, and each item can identify one or more attribute identification information corresponding to one or more attributes of the item. In other words, the recognition model disclosed herein can identify each attribute in the descriptive text about the attributes of the item in the drawing, rather than just identifying a whole paragraph of descriptive text.
[0039] In step S120, the item information database can be invoked, wherein the item information database may contain the name of each item among multiple items, at least one attribute of each item, and the category corresponding to at least one attribute of each item.
[0040] In this embodiment of the disclosure, the recognition model may be designed for recognizing drawings in a specific field, and the item information database may include all or a relatively comprehensive set of item information in that field. In some embodiments, the recognition model may also be designed for recognizing drawings in a broader field or multiple fields, in which case the item information database also needs to include all or a relatively comprehensive set of item information related to the corresponding field during the training process.
[0041] The item information database can contain the names of multiple items within a domain. Each item name is associated with one or more attributes, and each attribute corresponds to an attribute category. Every attribute of an item corresponds to an attribute category, and an attribute category may correspond to multiple attributes for a single item. For example, if the item is a pipe, the attribute category could include wall thickness, and the attribute could include multiple fields for different wall thicknesses. An attribute category can correspond to only one item or multiple items. For instance, the attribute category could include material; pipes, tees, and valves can all be associated with specific attributes corresponding to their materials.
[0042] In step S130, the parameters of the recognition model can be adjusted based on the comparison results between the name recognition information, attribute recognition information, and category recognition information and the names, attributes, and categories of items in the item information database. In step S140, training can be completed and the trained recognition model can be generated after the recognition confidence of the recognition model exceeds a threshold, such as 99%.
[0043] In this embodiment, the recognition model can recognize the text in the drawing to obtain name recognition information about the item's name and attribute recognition information about the item's attributes. It should be noted that in the drawing, multiple attributes of an item are recorded in the form of a long descriptive text, such as the multiple attributes of a pipe shown in Table 1 above. The recognition model of this disclosure can identify multiple attribute recognition information of the item, not just a single long text. Furthermore, the item information database also records each of the item's multiple attributes separately and associates them with the item's name. This allows for comparison between the attribute recognition information obtained by the recognition model and the attributes in the item information database. Based on the comparison results, the parameters of the recognition model are adjusted, resulting in higher recognition accuracy.
[0044] In some embodiments, such as Figure 2As shown, step S130, adjusting the parameters of the recognition model based on the comparison results of the name recognition information, attribute recognition information, and category recognition information with the name, attribute, and category of the item in the item information database, may include: step S131, obtaining the corresponding name, category, and one or more attributes in the item information database based on the name recognition information and category recognition information; step S132, adjusting the parameters of the recognition model if the attribute recognition information does not exist in one or more attributes. In this embodiment of the disclosure, the output result of the recognition model may include the recognized item name (i.e., name recognition information), one or more attributes (i.e., attribute recognition information), and the category corresponding to each attribute (i.e., category recognition information). Based on the name and category output by the recognition model, one or more attributes associated with the item and the category can be confirmed in the item information database. Then, the one or more attributes are compared with the recognized attribute recognition information to determine whether one or more attributes contain attribute recognition information, thereby adjusting the parameters of the recognition model. By comparing whether each attribute is correct, rather than examining whether the entire descriptive text is correctly recognized, the accuracy of the recognition model is improved, and the recognition results are more convenient for subsequent processing and statistics.
[0045] In some embodiments, the parameters of the recognition model can be adjusted based on one or more of the following loss functions: cross-entropy loss, mean-squared loss, and connectionist temporal classification loss. The loss can be confirmed based on the comparison results, and the parameters of the recognition model can be adjusted accordingly. The loss function can be one or more of cross-entropy loss, mean-squared loss (MSE Loss), and connectionist temporal classification loss (CTC Loss). The recognition model training method provided in this disclosure can improve the recognition accuracy of the recognition model for each attribute and category.
[0046] In some embodiments, such as Figure 3 As shown, the item information database is constructed in the following manner: Step S310, based on the names of items in the information tables of multiple drawings, the names of the same item are associated to obtain the associated name of each item; Step S320, based on the descriptive text of items in the information tables of multiple drawings, the attributes of each item and the category corresponding to the attributes are obtained; Step S330, based on the attributes of each item and the category corresponding to each item attribute, a dictionary category containing all item categories and the attributes of each item's corresponding dictionary category are obtained; Step S340, based on the name of each item, the dictionary category containing all item categories, and the attributes of each item's corresponding dictionary category, the item information database is formed.
[0047] In this embodiment of the disclosure, an item information database is established, which includes the name of each item, a dictionary category containing all item categories, and the attributes of each item corresponding to the dictionary category. This enables the attribute fields identified by the recognition model to be compared and verified separately, thereby improving the confidence of the recognition results of the recognition model.
[0048] In the item information database, each item's name corresponds to one or more attributes, and each attribute corresponds to a dictionary category. An attribute category for an item may have multiple attributes; for example, if the item is a pipe, the attribute category could include wall thickness, and the attribute could include multiple fields for different wall thicknesses. An attribute category can correspond to only one item or multiple items; for example, the attribute category could include material, and pipes, tees, and valves could all be associated with specific material attributes.
[0049] When constructing an item information database, information tables from multiple drawings can be obtained. These tables include the names of multiple items and associated descriptive text containing at least one attribute. The name and corresponding descriptive text of each item are identified from the information tables. The attributes of each item are obtained from the descriptive text in the information tables, and the category of each attribute is determined based on these attributes. The name of any item, its attributes, and the category of the attribute are all interconnected. Finally, the categories of all items are compiled into a dictionary of categories, and each attribute of all items is associated with a dictionary of categories. Using this method, item information can be easily obtained from the information tables of drawings, enabling the acquisition of large amounts of item information in the required field, forming a comprehensive dictionary of categories, and providing a sufficient data foundation for training the recognition model.
[0050] In step S310, the names of items in the information tables of multiple drawings can be associated to obtain the name of each item. In this embodiment of the disclosure, item information can be obtained through drawings, which may come from different projects and different units. For item attributes, such as the material of pipes, there are unified writing standards, but for item names, different projects and different units may have different names for the same item. For example, as shown in Table 2:
[0051] Table 2:
[0052]
[0053] In this embodiment, the names of the same item are associated. For example, different names for pipes, such as PIPE, steel pipe, seamless steel pipe, and Pipe, are associated, so that different names for the same item can be associated with the item's attributes. This allows for the acquisition of more data from different drawings and avoids errors caused by inconsistent item names. It also facilitates the acquisition of large amounts of item information in the required field, forming a comprehensive dictionary category and providing a sufficient data foundation for training the recognition model.
[0054] In some embodiments, such as Figure 4 As shown, step S320, based on the descriptive text of items in the information tables of multiple drawings, obtains the attributes of each item and the category corresponding to the attributes, which may include: step S321, splitting the descriptive text to obtain the attributes of each item; step S322, obtaining the category corresponding to the attributes based on the attributes of each item.
[0055] In this embodiment, information tables from multiple drawings are obtained, and the descriptive text corresponding to each item is identified. The descriptive text regarding item attributes in the information tables is then broken down, for example, based on spaces, punctuation marks, etc., to obtain at least one attribute corresponding to each item. Based on at least one attribute of each item, the category of each attribute is confirmed. This can be done manually based on attributes. Using this method, the attributes of items can be obtained from the information tables of the drawings, and their categories can be confirmed. This allows for the convenient acquisition of large amounts of item information in the required field, forming a comprehensive dictionary of types, providing a sufficient data foundation for training the recognition model.
[0056] In some embodiments, such as Figure 5As shown, step S330, based on the attributes of each item and the category corresponding to each item's attribute, obtains a dictionary category containing all item categories and the attribute of each item's corresponding dictionary category. This may include: step S331, retrieving the category of each attribute of the first item to form a dictionary category corresponding to each attribute; step S332, retrieving the category of each attribute of the next item; if it is the same as the category in the dictionary category, it is associated with the same category; if it is different from the category in the dictionary category, a new category is added to the dictionary category and the dictionary category is updated, repeating this process until the category of each attribute of all items is retrieved; step S333, obtaining a dictionary category containing all item categories and the attribute of each item's corresponding dictionary category. In this embodiment, step S331, from the information of all items obtained through the information table of the drawing, can retrieve the information of the first item to form a basic information database, including the name of the first item, one or more attributes of the first item, and the category of the attribute, with the category of each attribute forming an initial dictionary category. In step S332, the information for the next item can be retrieved, and the category of each attribute associated with the next item is compared with the current dictionary category. If each attribute category is included in the current dictionary category, then each attribute category of the item is associated with the corresponding dictionary category. If one or more attribute categories of the item do not exist in the current dictionary category, then the corresponding category can be added to the dictionary category, updating the dictionary category, thereby continuously expanding and improving the dictionary category. Step S332 is repeated to retrieve and compare the attribute categories of all retrieved items, continuously updating the dictionary category so that the dictionary category includes the attribute categories of all items. In step S333, the dictionary categories associated with the categories of all items and the attributes of the dictionary categories associated with each item can be obtained.
[0057] Since the attribute descriptions of items may differ in the information tables of each drawing, for example, some drawings may have more lenient restrictions on certain items, with fewer attribute descriptions and fewer categories covered in the information tables, while other drawings may have more stringent restrictions on these items, with more attribute descriptions and multiple categories covered in the information tables. In such cases, the method of constructing the item information database disclosed herein can build comprehensive item information, thereby laying the foundation for training a high-accuracy recognition model.
[0058] The training method for a recognition model used to identify information tables in drawings provided in this disclosure can, through deep integration with the knowledge of items in the drawings, structurally process and split the long attribute descriptions of items, confirm the category of each attribute, thereby improving the confidence in the recognition of item information in the information tables of the drawings, and can form field-level review and error correction.
[0059] The following section uses the training process of a recognition model for pipe drawings as an example to illustrate a training method for a recognition model.
[0060] Piping drawings can include general information such as drawing number, version number, and page number; they can also include image information, such as isometric views of the engineering piping, including isometric views of various components such as pipes, valves, and tees; and they can include information tables, which can be such as... Figure 6 As shown, it includes the serial number, the name of the item, and a complete descriptive text containing all the attributes of each item.
[0061] First, an item information database can be built. This can be achieved by obtaining the name, at least one attribute, and the category of each attribute from piping drawings from different units and projects. Specific methods may include: obtaining one or more aliases for any item; associating one or more aliases for any item with the name of any item; obtaining the descriptive text corresponding to each item; breaking down the descriptive text to obtain the attributes corresponding to each item; and confirming the category of each attribute based on the attributes of each item. A large number of piping drawings can be obtained from different units and projects, maximizing the acquisition of item information. The names and aliases of items from different units and projects should be unified and associated, as shown in Table 2 above. The system allows for the organization of different item attribute information based on various arrangement principles. Long descriptive texts can be broken down according to drawing rules, such as using commas or spaces. The same item from different projects within the same unit can be grouped together to form multiple item information databases for the same item across projects within the same unit. Identical items from each project are grouped together, corresponding to their various attributes and categories. Further aggregation creates an attribute information database for each item within each unit, with the same item within each unit being combined and statistically analyzed, corresponding to its various attributes and categories. This method allows for the convenient acquisition of a large amount of item information from numerous drawings.
[0062] After acquiring item information from numerous pipeline drawings across different units and projects, the method for constructing the item information database includes retrieving the category of each attribute of the first item to form a dictionary category; retrieving the category of each attribute of the next item and comparing it with the dictionary category; if it matches the category in the dictionary category, it is associated with the same category; if it does not match the category in the dictionary category, a new category is added to the dictionary category and the dictionary category is updated, repeating this process until the category of each attribute of all items is obtained; based on the name of each item, at least one attribute, and the updated dictionary category, the item information database is formed. This method enables the construction of a comprehensive and accurate item information system, containing all attributes of each item, with each attribute having an independent field and corresponding associated category.
[0063] After constructing the item information database, the drawings are input into the recognition model. The recognition model identifies the drawings to obtain the name recognition information, at least one attribute recognition information, and category recognition information for each item. The item information database is then accessed, containing the names of multiple items, at least one corresponding attribute for each item, and the category for each attribute. Based on the comparison results between the name recognition information, attribute recognition information, and category recognition information and the name, attribute, and category in the item information database, the parameters of the recognition model are adjusted. After the drawings are input into the recognition model, the model can output the corresponding recognition results, which may include the item's name, attribute, and category. For example, the recognition model might output a set of associated results such as "pipe, PE, connection method," corresponding to the item's name, attribute, and category, respectively. Based on the name recognition information and category recognition information of the output results—i.e., pipe and connection method—the corresponding associated attributes are confirmed in the item information database. In this example, multiple attributes may be confirmed: PE, BE, and NPT (representing: flat end, bevel, and thread, respectively). Next, it is determined whether the identified attribute information "PE" is included in multiple confirmed attributes in the item information database. If so, the identification is considered correct; otherwise, it is considered incorrect. The loss can be calculated based on the loss function, and the model parameters can be adjusted. The model is trained using multiple drawings, and the model parameters are adjusted until the final identification model's recognition structure converges and the confidence level is higher than a threshold, such as 95%. At this point, the identification model is considered to have completed training. Test sets and validation sets can also be set for testing.
[0064] The training method of this disclosure performs structured processing and splitting of long attribute descriptions of objects, sorts out the attributes of objects, organizes the corresponding categories into dictionary categories, deeply integrates them with the knowledge of objects in drawings, and then trains the recognition model to confirm each attribute and category, thereby improving the confidence of recognizing object information in drawings and enabling field-level review and error correction.
[0065] Based on the same inventive concept, this disclosure also provides a method for identifying information tables in drawings, such as... Figure 7 As shown, the identification method may include: step S710, obtaining a drawing with an information table; step S720, identifying the information table in the drawing using an identification model generated by the training method of any of the foregoing embodiments, and obtaining the name identification information, at least one attribute identification information, and the category identification information corresponding to each item in the information table.
[0066] The identification method provided in this disclosure can identify the contents of information tables in drawings, including the name of items, one or more attributes corresponding to each item, and the category corresponding to each attribute. Because it can identify each attribute, rather than just a text containing multiple attributes, it can group the same items within the same project to form the material quantity for that item, facilitating statistics, reducing manual merging work, and providing a fast and convenient solution for project material quantity management, bidding, and other tasks, improving work efficiency and reducing labor costs. It can provide a more accurate material list for projects.
[0067] In some embodiments, the identification method may further include: comparing the name identification information, attribute identification information, and category identification information with the name, attribute, and category of an item in an item information database; if the item information database does not contain any attribute identification information for any item, a warning is issued. In this embodiment, the identification result of the identification model can be compared again with the item information database. Based on the identified name identification information and category identification information, if the identified attribute identification information is confirmed to be included in multiple attributes associated with the name and category in the item information database, the identification is considered accurate, and no warning is needed; alternatively, the accurate identification result can be communicated. On the other hand, if the identified attribute identification information is not confirmed to be included in multiple attributes associated with the name and category in the item information database, i.e., the comparison result is questionable, a warning is issued to avoid subsequent statistical errors caused by identification errors. For example, when the confidence level reaches 99%, only individual characters with lower confidence levels are highlighted with different colors or wavy lines, allowing for manual intervention and correction, forming the first layer of review and correction; when the confidence level is below 99%, characters with lower confidence levels are highlighted, allowing for manual intervention and correction, forming the second layer of manual review and correction.
[0068] In some embodiments, the identification method may include: if the item information database does not contain any attribute identification information for any item, and any attribute identification information is correctly identified, then add any attribute identification information for any item and the category corresponding to any attribute identification information to the item information database. In this embodiment, if, based on the identified name identification information and category identification information, multiple attributes associated with the name and category in the item information database are found to be missing the identified attribute identification information, i.e., the comparison result is questionable, further confirmation is needed to determine if the identification result is incorrect. If the identification result is confirmed to be accurate, it may be due to the absence of categories and attributes in the item information database. In this embodiment, the item information database can be updated based on the identification result, mapping relationships can be added, and the item information database can be continuously improved. Through this embodiment, a good item data foundation can be provided for subsequent project management systems or BIM collaboration platforms, forming accurate and stable item data output. By accumulating the identification mapping of dictionary categories, the dictionary content such as material grades, pressure ratings, and standard requirements can be continuously expanded through autonomous learning, providing a good field application foundation for subsequent data applications such as robotic process automation (RPA) in pipeline engineering. Furthermore, by sorting out the names, attributes, and categories of items from different units, mapping relationships are continuously accumulated, and the human nervous system is simulated to gradually master the rules of pipeline design, providing a rule-based foundation for subsequent pipeline design.
[0069] Based on the same inventive concept, such as Figure 8 As shown, this embodiment of the present disclosure also provides an information table identification device 800 in a drawing, including: an acquisition unit 810, used to acquire a drawing with an information table; and an identification unit 820, used to identify the information table in the drawing through an identification model generated by the training method of any of the foregoing embodiments, and to obtain name identification information, at least one attribute identification information, and category identification information corresponding to each item in the information table.
[0070] Regarding the identification device 800 for the information table in the drawings of the above embodiments, the specific manner in which each unit performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated here.
[0071] The training method for the recognition model used to identify information tables in drawings, as described above, and the method for recognizing information tables in drawings, can be implemented in hardware, software, firmware, or any combination thereof. If implemented as a computer program product in software, the functions can be stored or transmitted as one or more instructions or codes on or through a computer-readable medium. Computer-readable media includes both computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one location to another. Storage media can be any available medium accessible to a computer. By way of example and not limitation, such computer-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then those coaxial cables, fiber optic cables, twisted pairs, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of media. As used herein, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs, where disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.
[0072] Although the present invention has been described with reference to specific embodiments, those skilled in the art should recognize that the above embodiments are merely illustrative of the invention, and various equivalent changes or substitutions can be made without departing from the spirit of the invention. Therefore, any changes or modifications to the above embodiments within the essential spirit of the invention will fall within the scope of the claims of this application.
Claims
1. A training method for a recognition model used to identify information tables in drawings, the training method comprising: The recognition model identifies the information table in the drawing to obtain the name recognition information, at least one attribute recognition information, and the category recognition information corresponding to each item in the information table. The information table contains the name of each item in the drawing and descriptive text containing all attributes of each item. The item information database is invoked, wherein the item information database contains the name of each item among a plurality of items, at least one attribute of each item, and the category corresponding to at least one attribute of each item; Based on the comparison results between the name recognition information, attribute recognition information, and category recognition information and the name, attribute, and category of the items in the item information database, the parameters of the recognition model are adjusted. Generate the trained recognition model; The item information database is constructed in the following manner: based on the names of items in the information tables of multiple drawings, the names of the same item are associated to obtain the name of each item; based on the descriptive text of the items in the information tables of multiple drawings, the attributes of each item and the category corresponding to the attributes are obtained; based on the attributes of each item and the category corresponding to each item attribute, a dictionary category containing all item categories and the attribute of each item corresponding to the dictionary category are obtained; based on the name of each item, the dictionary category containing all item categories, and the attribute of each item corresponding to the dictionary category, the item information database is formed.
2. The training method according to claim 1, characterized in that, The step of adjusting the parameters of the recognition model based on the comparison results between the name recognition information, attribute recognition information, and category recognition information and the names, attributes, and categories of items in the item information database includes: Based on the name recognition information and category recognition information, obtain the corresponding name, category, and one or more attributes in the item information database; If the attribute identification information does not exist in one or more attributes, adjust the parameters of the identification model.
3. The training method according to claim 2, characterized in that, The parameters of the recognition model are adjusted based on one or more of the following loss functions: cross-entropy loss function, mean squared error loss function, and connection-time classification loss function.
4. The training method according to claim 1, characterized in that, The description text of items in the information table based on multiple drawings is used to obtain the attributes of each item and the corresponding category of each attribute, including: The descriptive text is broken down to obtain the attributes of each item; Based on the attributes of each item, the category corresponding to the attribute is obtained.
5. The training method according to claim 1, characterized in that, The process of obtaining a dictionary category containing all item categories and the attribute of each item corresponding to each item category based on the attributes of each item and the category corresponding to each item category includes: Retrieve the category of each attribute of the first item to form a dictionary category corresponding to each attribute; Retrieve the category of each attribute of the next item. If it is the same as the category in the dictionary category, associate it with the same category. If it is not the same as the category in the dictionary category, add a new category in the dictionary category and update the dictionary category. Repeat this process until the category of each attribute of all items is retrieved. Obtain a dictionary category containing all items and the attributes of each item corresponding to that dictionary category.
6. A method for identifying information tables in a drawing, the method comprising: Obtain drawings with information tables; The recognition model generated by the training method as described in any one of claims 1-5 identifies the information table in the drawing and obtains the name recognition information, at least one attribute recognition information, and the category recognition information corresponding to each item in the information table.
7. The identification method according to claim 6, characterized in that, The identification method further includes at least one of the following steps: The name identification information, attribute identification information, and category identification information are compared with the name, attribute, and category of the items in the item information database. If the item information database does not contain any attribute identification information of any item, a reminder is issued. or If the item information database does not contain any attribute identification information for any item, and the attribute identification information is correctly identified, then the attribute identification information for any item and the category corresponding to the attribute identification information are added to the item information database.
8. A device for identifying information tables in drawings, characterized in that, The identification device includes: The acquisition unit is used to acquire drawings with information tables; and The identification unit is configured to identify the information table in the drawing using the identification model generated by the training method as described in any one of claims 1-5, and obtain the name identification information, at least one attribute identification information, and the category identification information corresponding to each item in the information table.
9. A computer-readable storage medium, characterized in that, The system stores instructions that, when executed by a processor, perform the training method as described in any one of claims 1-5 or the recognition method as described in claim 6 or 7.