Method for identifying engineering drawing detail table and generating BOM table

Abstract

The invention discloses a method for identifying an engineering drawing detail table and generating a BOM table, and belongs to the crossing field of automation technology and image processing, and the method comprises the steps: obtaining a scanning or electronic image of an engineering drawing; through preset datum line positioning, recursively detecting a nested rectangular region conforming to an area difference threshold value, and determining a title bar, a detail list region coordinate and a table image; identifying lines and cross points, analyzing the line and column boundaries of the table, and constructing a topological structure; performing character recognition by adopting multi-engine OCR integration; the characters and the cells are associated, a header is recognized through semantics, and analysis data subjected to integrity verification are generated; and automatically generating a structured BOM table in a preset standard format based on the data, and outputting an editable file. According to the method, full-process automation is achieved, manual intervention is not needed, and manual input cost and personal errors are greatly reduced.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of automation technology and image processing, and more specifically, relates to a method for recognizing engineering drawing detail sheets and generating a Bill of Materials (BOM). Background Technology

[0002] In the entire manufacturing supply chain, the detailed lists accompanying engineering drawings are the core data source for material requirements planning, production scheduling, procurement management, and cost accounting. Their accuracy directly determines the smoothness and economy of subsequent production processes. The Bill of Materials (BOM), as a structured representation of the detailed list data, is a crucial data link connecting design, production, and warehousing departments, and is the fundamental support for enterprises to achieve digital production and lean management. However, the conversion from engineering drawing detailed lists to structured BOMs has long been hampered by bottlenecks in both efficiency and accuracy, becoming a prominent pain point restricting the digital transformation of the manufacturing industry.

[0003] Traditional BOM (Bill of Materials) processing relies entirely on manual data entry, which has three major drawbacks: First, it is extremely inefficient; if multiple versions of drawings are involved in changes and iterations, the cost of repetitive labor increases exponentially. Second, it has a high error rate; manual data entry is prone to errors such as misreading numbers, omitting units, and confusing fields, which can directly lead to material shortages, production rework, or inventory backlogs. Third, it lacks data standardization; differences in operating habits among different data entry personnel can result in chaotic BOM formats, requiring additional manpower for data standardization and making it impossible to directly integrate with the enterprise's digital management system.

[0004] With the popularization of Industry 4.0 and intelligent manufacturing concepts, the manufacturing industry's requirements for automation and precision in material management continue to increase. Enterprises urgently need an intelligent bill of materials (BOM) recognition technology that can adapt to multiple drawing formats, be compatible with multiple fonts, and meet multiple industry standards, enabling one-click conversion from scanned drawings to standardized BOMs. Against this backdrop, developing an efficient, accurate, and universal method for recognizing engineering drawing BOMs and generating BOMs has become a key requirement for promoting the digital upgrade of material management in the manufacturing industry. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying engineering drawing detail sheets and generating BOM (Bill of Materials) sheets, which solves the shortcomings of traditional manual processing, such as low efficiency and high error rate, as well as the poor adaptability of existing general identification technologies, and improves the automation level of material management in manufacturing enterprises.

[0006] To achieve the above objectives, the present invention provides a method for identifying engineering drawing detail sheets and generating a Bill of Materials (BOM), the method comprising: S1, acquire scanned or electronic images of the target engineering drawings; S2, locates the nested rectangular area by a preset baseline, recursively detects the nested rectangular area, and continues to shrink inward when it meets the preset area difference threshold, and finally determines the coordinates of the title bar and detail table area and the table area image; S3 identifies lines and intersections in an image, and based on the positional relationship between lines and intersections, parses the row and column boundaries of the table and constructs the table topology. S4 uses multi-engine OCR integration for text recognition; S5: Based on the table topology, identify the association between text and corresponding cells; establish a two-dimensional grid mapping through the detected cell coordinates; identify the table header fields through semantic recognition technology; and generate parsed data that meets the BOM table field association requirements and has passed data integrity verification. S6. Based on the parsed data, a structured BOM table in a preset standard format is generated by the program software, and an editable structured BOM file is output.

[0007] Optionally, in S2, the preset baseline is the longest vertical line on the right side of the drawing; the area difference threshold is that the area difference between adjacent rectangular areas is ≤20%.

[0008] Optionally, S2 further includes: using OpenCV to perform grayscale processing and adaptive binarization processing on the table area image; verifying the accuracy of the detail table area through typical features, the typical features including: the width of the detail table is consistent with the title bar, and it contains preset table header key fields, the preset table header key fields including serial number, name, quantity, material and remarks.

[0009] Optionally, the specific steps of S3 are as follows: S3.1 uses morphological image processing to detect horizontal and vertical lines. First, it performs an opening operation through the adapted structuring element, performs connected component analysis on the detected lines, filters out short line segments and noisy lines that meet the preset length threshold, and finally applies a smoothing operation to reduce jagged edges, thus obtaining the boundary lines of the table rows and columns. S3.2, use bitwise operations to find the intersection of horizontal and vertical lines, apply image enhancement operations to enhance the visibility of the intersection, calculate image moments to determine the precise center position of the intersection, filter out false intersections that meet the preset area threshold, and obtain the core coordinate nodes of the table grid; S3.3 merges all lines and performs connected component analysis, identifies the connected region with the largest area as the main table, filters isolated line segments and optimizes the table edges to reduce jagged edges, and divides the row and column range of each cell according to the coordinate nodes of the intersection points and the boundary lines to construct a complete table topology structure.

[0010] Optionally, in S3.1, the adapted structural element is a 5×1 pixel rectangular structural element, and the preset length threshold is a line length < 50 pixels; in S3.2, the preset area threshold is an intersection area < 9 pixels.

[0011] Optionally, in S4, the multi-engine OCR integration includes a first OCR engine and a second OCR engine. The first OCR engine is used as the basic recognition engine, and the second OCR engine is used to enhance the processing capability of complex fonts. The engine is dynamically switched through a confidence threshold. When the recognition confidence of the first OCR engine is lower than the preset confidence threshold, the second OCR engine is automatically called for secondary recognition.

[0012] Optionally, the first OCR engine is TesseractOCR, and the second OCR engine is PaddleOCR; the complex fonts include handwritten fonts, scanned blurred fonts, and process annotation fonts unique to engineering drawings; the preset reliability threshold is 85%.

[0013] Optionally, in S5, the semantic recognition technology is NLP technology; the preset rules for data integrity verification include verification that the material code conforms to the GB / T standard format, verification that the quantity field is a positive integer or a decimal, and verification that the weight field has a numerical format.

[0014] Optionally, in S6, the preset standard format is GB / T20093-2006 Mechanical Industry Standard or SJ / T10668-2006 Electronic Industry Standard, and the preset standard format includes a material number field, a name field, a material field, a quantity field, and a weight field; the editable structured file format is Excel or XML format.

[0015] Optionally, in S6, BOM table generation also includes: calculating the total quantity and total weight fields; generating a tree-structured BOM table with a hierarchical structure to clarify the ownership relationship of each component.

[0016] The beneficial effects of this invention are as follows: It provides a method for recognizing engineering drawing detail tables and generating BOM tables. This method combines image processing, multi-engine OCR recognition, and semantic recognition to achieve full-process automation from engineering drawings to structured BOM tables. No manual intervention is required; it automates the entire process from engineering drawing image acquisition, region positioning, table recognition, text recognition, data parsing to BOM table generation, significantly reducing manual operation costs and human error. This method supports processing engineering drawings of different formats and resolutions, and is adaptable to detail table layouts in various industries such as machinery and electronics. It is optimized for complex backgrounds, diverse fonts (and table distortion issues) in engineering drawings, resulting in strong recognition robustness. Through the dual guarantee of multi-engine OCR integrated recognition and preset rule data verification, the text recognition accuracy and data compliance both reach over 98%, significantly reducing the error rate of traditional manual input and general recognition technologies.

[0017] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0018] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the invention.

[0019] Figure 1 A flowchart of the method for identifying and generating a Bill of Materials (BOM) table according to Embodiment 1 of the present invention is shown.

[0020] Figure 2 A flowchart of the method for identifying and generating a Bill of Materials (BOM) table according to Embodiment 2 of the present invention is shown.

[0021] Figure 3 An example of BOM table generation in Embodiment 2 of the present invention is shown. Detailed Implementation

[0022] Preferred embodiments of the invention will now be described in more detail. While preferred embodiments of the invention are described below, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0023] Example 1

[0024] See Figure 1 This disclosure provides a method for identifying and generating a Bill of Materials (BOM) from engineering drawings, the method comprising: S1, acquire scanned or electronic images of the target engineering drawings; S2, locates the nested rectangular area by a preset baseline, recursively detects the nested rectangular area, and continues to shrink inward when it meets the preset area difference threshold, and finally determines the coordinates of the title bar and detail table area and the table area image; In this step, the preset baseline is the longest vertical line on the right side of the drawing; the area difference threshold is ≤20% for the area difference between adjacent rectangular areas. Using the longest vertical line on the right side of the drawing as the preset baseline adapts to the conventional layout of the title block and detail table of the engineering drawing, improving the targeting of area positioning; by setting the area difference threshold, missed detections or false detections are avoided during recursive detection, ensuring the accuracy and consistency of the area positioning of the title block and detail table.

[0025] Optionally, this step also includes: performing grayscale and adaptive binarization processing on the table area image using OpenCV; verifying the accuracy of the detail table area through typical features, including: the detail table width is consistent with the title bar, and it contains preset key header fields, which include serial number, name, quantity, material, and remarks. Performing OpenCV grayscale and adaptive binarization processing on the table area image enhances the contrast between table lines and background, laying a high-quality image foundation for subsequent table recognition; through dual feature verification of consistent width and preset key header fields, misjudging other areas of the drawing as detail tables is avoided, further improving the accuracy of area positioning; the preset key header fields are clearly defined as "serial number," "name," "quantity," "material," and "remarks," covering the core fields of the engineering drawing detail table, ensuring the uniqueness of the identified object.

[0026] S3, identify lines and intersections in the image, and based on the positional relationship between lines and intersections, parse the table row and column boundaries to construct the table topology; the specific steps are as follows: S3.1 uses morphological image processing to detect horizontal and vertical lines. First, it performs an opening operation through the adapted structuring element, performs connected component analysis on the detected lines, filters out short line segments and noisy lines that meet the preset length threshold, and finally applies a smoothing operation to reduce jagged edges, thus obtaining the boundary lines of the table rows and columns. S3.2, use bitwise operations to find the intersection of horizontal and vertical lines, apply image enhancement operations to enhance the visibility of the intersection, calculate image moments to determine the precise center position of the intersection, filter out false intersections that meet the preset area threshold, and obtain the core coordinate nodes of the table grid; S3.3 merges all lines and performs connected component analysis, identifies the connected region with the largest area as the main table, filters isolated line segments and optimizes the table edges to reduce jagged edges, and divides the row and column range of each cell according to the coordinate nodes of the intersection points and the boundary lines to construct a complete table topology structure.

[0027] This step, through a step-by-step design of line detection, intersection analysis, and cell segmentation, ensures the integrity and accuracy of table frame recognition, solving the recognition challenges caused by noise in engineering drawing tables and blurred intersections. It first filters short line segments and false intersections, then constructs a topological structure to avoid interference from invalid data, ensuring the accuracy of cell segmentation and providing a guarantee for precise association between text and cells. By optimizing table edge processing, it improves the stability of the table topological structure, providing a reliable spatial positioning foundation for subsequent text recognition and data parsing.

[0028] Optionally, in S3.1, the adapted structural element is a 5×1 pixel rectangular structural element, with a preset length threshold of line length < 50 pixels; in S3.2, the preset area threshold is an intersection area < 9 pixels. This selectively filters out fine red noise and false intersections, further improving the accuracy of line detection and intersection recognition, and ensuring the reliability of the table topology construction.

[0029] S4 uses multi-engine OCR integration for text recognition; In this step, the multi-engine OCR integration includes a first OCR engine and a second OCR engine. The first OCR engine serves as the basic recognition engine, while the second OCR engine enhances the processing capabilities for complex fonts. Engines are dynamically switched using a confidence threshold. When the recognition confidence of the first OCR engine falls below a preset confidence threshold, the second OCR engine is automatically invoked for secondary recognition. This dual-engine architecture of a basic engine and an enhanced engine balances the recognition efficiency of regular fonts with the recognition accuracy of complex fonts, solving the recognition challenges of handwritten and blurred fonts in engineering drawings. Dynamically switching engines using a confidence threshold avoids the waste of efficiency caused by blindly using high-precision engines, achieving a balance between efficiency and accuracy.

[0030] In this embodiment, the first OCR engine is TesseractOCR, and the second OCR engine is PaddleOCR; complex fonts include handwritten fonts, blurred scanned fonts, and process annotation fonts unique to engineering drawings; the preset reliability threshold is 85%. This embodiment avoids the blind switching of engines by defining the specific types of complex fonts; through verification with a large number of samples, the 85% preset reliability threshold is clearly defined, which can accurately judge the recognition reliability and ensure the necessity and effectiveness of secondary recognition.

[0031] S5: Based on the table topology, identify the association between text and corresponding cells; establish a two-dimensional grid mapping through the detected cell coordinates; identify the table header fields through semantic recognition technology; and generate parsed data that meets the BOM table field association requirements and has passed data integrity verification. In this step, semantic recognition technology is NLP technology. Pre-set rules for data integrity verification include verifying that the material code conforms to the GB / T standard format, that the quantity field is a positive integer or decimal, and that the weight field has a numerical format. Using NLP technology to identify header fields ensures high matching accuracy, guaranteeing a precise association between header fields and cell data and avoiding data mapping confusion. Pre-set rules for verifying the material code according to the GB / T standard and the quantity / weight format filter out incorrectly identified or abnormally formatted data, ensuring the compliance and accuracy of the parsed data and providing double assurance for the quality of the BOM (Bill of Materials).

[0032] S6, based on parsed data, generates a structured BOM table in a preset standard format through software, and outputs an editable structured BOM file.

[0033] In this step, the default standard format is GB / T20093-2006 (Mechanical Industry Standard) or SJ / T10668-2006 (Electronic Industry Standard). The default standard format includes material number, name, material, quantity, and weight fields. The editable structured file format is Excel or XML. The BOM (Bill of Materials) conforms to GB / T20093-2006 (Mechanical) or SJ / T10668-2006 (Electronic) industry standards, ensuring the BOM's universality and standardization, meeting the material management needs of enterprises in different industries. It supports editable Excel or XML output formats, adapting to different digital management scenarios, facilitating secondary data editing and system integration, and improving usability.

[0034] In this embodiment, BOM generation also includes: calculating and summarizing the total quantity and total weight fields; generating a hierarchical tree-structured BOM to clarify the ownership relationship of each component. Automatically calculating the total quantity and total weight provides direct data support for parts procurement planning and cost control, reducing subsequent data aggregation workload; generating a tree-structured hierarchical BOM clarifies the ownership relationship of each component, facilitating material hierarchy management and production assembly process advancement, and solving the problem that traditional flat BOMs struggle to reflect assembly logic.

[0035] Specifically, this method automates the entire process from engineering drawing detail sheets to BOM (Bill of Materials) tables without manual intervention, significantly reducing manual data entry costs and human error, and solving the problems of low efficiency and high error rate in traditional manual processing. Through a closed-loop design of regional positioning, table topology construction, multi-engine recognition, data parsing, and standard BOM generation, it adapts to complex backgrounds, diverse fonts, and table deformation scenarios in engineering drawings, making it more adaptable than general recognition technologies. It generates structured BOM tables that conform to industry standards, which can be directly connected to enterprise ERP systems, improving the level of digitalization in material management and providing accurate data support for procurement planning and cost control.

[0036] Example 2

[0037] like Figure 2 As shown, this embodiment provides a method for identifying engineering drawing detail sheets and generating a Bill of Materials (BOM), including the following steps: Step 1, Engineering Drawing Image Acquisition: Obtain scanned or electronic images of the target engineering drawings as the raw data carrier for subsequent processing.

[0038] In the specific implementation process, it is necessary to ensure that the acquired images are free from fatal defects such as severe blurring, incompleteness, reflection, or occlusion, so as to guarantee the basic effectiveness of the subsequent recognition process.

[0039] Step 2, Image Preprocessing: Positioning is achieved using a preset baseline, and nested rectangular regions are recursively detected (continuing to shrink inwards when the area difference meets a preset threshold). The accuracy of the region is verified by combining typical features of the detail table, ultimately determining the coordinates of the title block and detail table regions, as well as the image of the table region. Typical features include: the detail table width is consistent with the title block, and it contains preset key header fields. This dual feature approach avoids misidentifying other areas of the drawing as detail tables, ensuring accurate positioning. OpenCV is used for grayscale conversion and adaptive binarization.

[0040] In the specific implementation process, the longest vertical line on the right side of the drawing is used as the baseline. This choice is compatible with the common layout habits of title blocks and detail tables in engineering drawings, resulting in higher positioning accuracy. A preset threshold is set so that if the area difference between adjacent rectangular areas is ≤20%, further shrinking is implemented. This threshold has been verified through extensive experiments and can accurately distinguish title blocks, detail tables, and other drawing areas, avoiding missed or false detections. Preset key fields for the table header include "Serial Number," "Name," "Quantity," "Material," and "Remarks," covering the core required fields of the engineering drawing detail table. Combined with the geometric feature that "the width of the detail table is consistent with the title block," a dual verification of geometric dimensions and content attributes is formed, further ensuring the accuracy of area positioning. OpenCV calls the cv2.COLOR_BGR2GRAY function to perform grayscale processing, converting the color image to a grayscale image; and calls the cv2.ADAPTIVE_THRESH_GAUSSIAN_C function to perform adaptive binarization processing, enhancing the contrast between the table area and the background, laying the foundation for subsequent table frame recognition.

[0041] Step 3, Table Framework Identification: Through three sub-steps—line detection, intersection analysis, and table extraction—a complete and accurate table topology is constructed step by step. Line detection employs morphological image processing, using adapted structural elements for opening operations to preprocess short line segments and noisy lines that meet preset length thresholds. Intersection analysis obtains intersections through bitwise operations, enhances visibility through image enhancement, and filters out pseudo-intersections that meet preset area thresholds. Table extraction selects the main table by merging lines and performing connected component analysis, filtering isolated line segments and optimizing edges. Each sub-step progresses progressively to ensure the integrity and accuracy of the table framework.

[0042] In practice, line detection uses a 5×1 pixel rectangular structural element. This size is designed for the common width of table lines in engineering drawings, effectively preserving valid lines and filtering out minor red noise. The preset length threshold is set to filter short line segments and noisy lines with a length of less than 50 pixels to avoid interference from non-table lines. After line detection, a smoothing operation is performed through median filtering to reduce line jagged edges and optimize line integrity.

[0043] In intersection point analysis, a dilation operation using 3×3 pixel rectangular structural elements is employed to enhance the visibility of intersection points in the image, avoiding missed detections due to thin or blurry lines. A preset area threshold is set to filter out false intersection points with an area less than 9 pixels. This value corresponds to the minimum effective pixel area of ​​the intersection point, accurately eliminating false intersection points caused by noise. Bitwise operations are used to obtain candidate regions for the intersection of horizontal and vertical lines. The image moments of each candidate region are calculated to determine the precise center coordinates of the intersection points, providing core coordinate data for constructing the table topology.

[0044] In the table extraction process, all optimized horizontal and vertical lines are merged to generate a merged line image, ensuring line continuity. Connectivity component analysis is performed on the merged image, and the connected region with the largest area is selected as the main table and the main body of the detail table. Based on the identified intersection coordinates, isolated line segments not associated with the intersections are filtered out, and then the table edges are optimized through an edge smoothing algorithm. Finally, a complete table topology structure is output, providing a stable foundation for the subsequent association of text and cells.

[0045] Step 4, Text Recognition: A multi-engine OCR integration solution is adopted, with the first OCR engine as the basic recognition engine and the second OCR engine to enhance the processing capability of complex characters. The engine is dynamically switched through the confidence threshold. When the recognition confidence of the first OCR engine is lower than the preset confidence threshold, the second OCR engine is automatically called for secondary recognition, forming a closed loop of basic recognition, threshold judgment and secondary supplementary recognition, which takes into account both recognition efficiency and accuracy.

[0046] In practice, the first OCR engine selected is TesseractOCR, which is highly efficient and fast in recognizing standard fonts and suitable for most common scenarios. The second OCR engine is PaddleOCR, which performs better in complex font recognition scenarios such as handwritten fonts, blurred fonts, and process annotation fonts unique to engineering drawings. The two engines working together can cover all font types in engineering drawings. Complex fonts include handwritten fonts, blurred scanned fonts, and process annotation fonts unique to engineering drawings. The applicable scenarios for the second OCR engine are clearly defined to avoid blind switching between engines. The preset confidence threshold is set at 85%. This threshold has been tested with a large number of samples to accurately determine the recognition reliability of TesseractOCR. When the confidence level is ≥85%, its recognition result is directly used to ensure efficiency; when the confidence level is <85%, PaddleOCR is switched to supplement recognition to ensure accuracy. In the recognition process, based on the table topology structure output by S3, the image region of each cell is segmented in row and column order; the cell images are input into TesseractOCR in sequence, and the recognition results and confidence scores are output; based on the comparison between the confidence score and the 85% threshold, it is decided whether to call PaddleOCR for secondary recognition; the result with the higher confidence score between the two recognitions is taken as the final text data, and invalid characters such as spaces and line breaks are removed to obtain standardized cell text information, providing clean input data for subsequent data parsing.

[0047] Step 5, Data Parsing and Mapping: Following the standardized text data in Step 4, the text is associated with corresponding cells according to the table topology. A two-dimensional grid mapping is established using cell coordinates, and the header fields are identified using semantic recognition technology. Data integrity verification is performed based on preset rules, and finally, parsed data that meets the BOM table field association requirements is generated, completing the transformation from text data to structured data.

[0048] In practice, NLP (Natural Language Processing) technology is used to identify header fields such as "Serial Number," "Material," "Quantity," "Weight," and "Remarks" through a keyword matching model, achieving a matching accuracy of over 98% to ensure the precision of header recognition. Based on the row and column coordinates of the cells, a 29-row × 8-column two-dimensional array grid structure (corresponding to the number of rows and columns in the detail table in this example) is created. The standardized text data obtained in step 4 is filled into the grid according to the corresponding coordinates, forming the original table data matrix and clarifying the one-to-one correspondence between text and cells. Verification is performed based on preset rules, including verification that the material code conforms to the GB / T standard format, that the quantity field is a positive integer or decimal, and that the weight field is a numerical format. Verification eliminates data with identification errors or abnormal formats, ensuring the compliance and accuracy of the parsed data.

[0049] Step 6, BOM generation: Based on the parsed data obtained in Step 5, the program software automatically maps the data to a preset standard format template to generate a structured BOM. The output is an editable structured BOM file, which also includes extended summary calculation and hierarchical structure generation functions to meet the actual needs of manufacturing enterprises' material management.

[0050] In practical implementation, the preset standard format adopts the GB / T20093-2006 mechanical industry standard or the SJ / T10668-2006 electronic industry standard, covering core fields such as material number, name, material, quantity, unit weight, total weight, and remarks, adapting to the material management specifications of multiple industries such as machinery and electronics. The structured file format supports Excel or XML output. The Excel file contains three worksheets: "Basic Information Table," "Material Details Table," and "Summary Table." The XML file organizes data according to preset tag specifications, facilitating integration with enterprise ERP systems and meeting the digital management needs of different enterprises. Based on the "Quantity" and "Unit Weight" fields in the parsed data, the total weight of each material is automatically calculated, and the total quantity and weight of all materials are summarized, providing direct data support for parts procurement planning and cost control. Based on the assembly relationships of materials, a tree-structured hierarchical BOM table is generated, clearly defining the ownership relationships of each component, facilitating hierarchical material management and the advancement of the production assembly process.

[0051] The method described in this embodiment is used to identify and generate a Bill of Materials (BOM) for an engineering drawing. The result is as follows: Figure 3 As shown, the entire process took 45 seconds, which is 53 times more efficient than manual data entry. The BOM data was compared with the standard data after manual review, and there were no omissions or errors. The tree-like hierarchical structure is clear and meets the actual needs of parts procurement planning, cost control and production assembly, thus verifying the practicality and superiority of the technical solution of this invention.

[0052] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method for identifying and generating a Bill of Materials (BOM) from engineering drawings, characterized in that, The method includes: S1, acquire scanned or electronic images of the target engineering drawings; S2, locates the nested rectangular area by a preset baseline, recursively detects the nested rectangular area, and continues to shrink inward when it meets the preset area difference threshold, and finally determines the coordinates of the title bar and detail table area and the table area image; S3 identifies lines and intersections in an image, and based on the positional relationship between lines and intersections, parses the row and column boundaries of the table and constructs the table topology. S4 uses multi-engine OCR integration for text recognition; S5: Based on the table topology, identify the association between text and corresponding cells; establish a two-dimensional grid mapping through the detected cell coordinates; identify the table header fields through semantic recognition technology; and generate parsed data that meets the BOM table field association requirements and has passed data integrity verification. S6. Based on the parsed data, a structured BOM table in a preset standard format is generated by the program software, and an editable structured BOM file is output.

2. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 1, characterized in that, In S2, The preset baseline is the longest vertical line on the right side of the drawing; The area difference threshold is defined as an area difference of ≤20% between adjacent rectangular regions.

3. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 1, characterized in that, S2 also includes: The table region image was processed using OpenCV for grayscale conversion and adaptive binarization. The accuracy of the details table area is verified by typical features, which include: the width of the details table is consistent with the title bar, and it contains preset key header fields, including serial number, name, quantity, material and remarks.

4. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 1, characterized in that, The specific steps for S3 are as follows: S3.1 uses morphological image processing to detect horizontal and vertical lines. First, it performs an opening operation through the adapted structuring element, performs connected component analysis on the detected lines, filters out short line segments and noisy lines that meet the preset length threshold, and finally applies a smoothing operation to reduce jagged edges, thus obtaining the boundary lines of the table rows and columns. S3.2, use bitwise operations to find the intersection of horizontal and vertical lines, apply image enhancement operations to enhance the visibility of the intersection, calculate image moments to determine the precise center position of the intersection, filter out false intersections that meet the preset area threshold, and obtain the core coordinate nodes of the table grid; S3.3 merges all lines and performs connected component analysis, identifies the connected region with the largest area as the main table, filters isolated line segments and optimizes the table edges to reduce jagged edges, and divides the row and column range of each cell according to the coordinate nodes of the intersection points and the boundary lines to construct a complete table topology structure.

5. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 4, characterized in that, In S3.1, the adapted structural element is a 5×1 pixel rectangular structural element, and the preset length threshold is a line length < 50 pixels; In S3.2, the preset area threshold is that the area of ​​the intersection point is less than 9 pixels.

6. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 1, characterized in that, In S4, The multi-engine OCR integration includes a first OCR engine and a second OCR engine. The first OCR engine serves as the basic recognition engine, while the second OCR engine enhances the ability to process complex fonts. The engine is dynamically switched through a confidence threshold. When the recognition confidence of the first OCR engine is lower than the preset confidence threshold, the second OCR engine is automatically called for secondary recognition.

7. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 6, characterized in that, The first OCR engine is TesseractOCR, and the second OCR engine is PaddleOCR; The complex fonts include handwritten fonts, blurred scan fonts, and process annotation fonts unique to engineering drawings; The preset confidence threshold is 85%.

8. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 1, characterized in that, In S5, The semantic recognition technology is NLP technology; The preset rules for data integrity verification include verification that the material code conforms to the GB / T standard format, verification that the quantity field is a positive integer or a decimal, and verification that the weight field has a numerical format.

9. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 1, characterized in that, In S6, The preset standard format is GB / T20093-2006 Mechanical Industry Standard or SJ / T10668-2006 Electronic Industry Standard, and the preset standard format includes material number field, name field, material field, quantity field and weight field; The editable structured file format is Excel or XML.

10. The method for identifying and generating a Bill of Materials (BOM) from engineering drawings according to claim 1, characterized in that, In S6, BOM generation also includes: calculating and summarizing the total quantity and total weight fields; generating a hierarchical tree-structured BOM table to clarify the ownership relationship of each component.

Images