Method, system and medium for extracting structured data tables from power grid engineering drawings

Abstract

The present invention relates to a method, system, and medium for extracting structured data tables from power grid engineering drawings. The method comprises: obtaining a PDF image file of a power grid engineering drawing and gray-scaling it; inputting the gray-scaled PDF image file into a pre-built and trained intelligent image recognition model, obtaining images of each cell area within a table area in the PDF image file, and recording the corresponding row and column numbers; performing text recognition on each cell area image using a text OCR detection algorithm to obtain a rough text recognition result; matching the rough text recognition result with a pre-built power grid engineering terminology library to obtain a precise text recognition result, and constructing a structured data table based on the corresponding row and column numbers. Compared with the prior art, the present invention improves the accuracy of the table area recognition results and reduces the training requirements for the neural network recognition model.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid engineering drawing recognition, and in particular to a method, system and medium for extracting structured data tables from power grid engineering drawings. Background Art

[0002] After the design of a power grid project is completed, the results need to be reviewed. Currently, this review process relies on manual review by expert reviewers, relying heavily on their personal experience. Review results are not reusable, resulting in low review efficiency and a high workload. Power grid project design results include drawings, documents, and tables. Drawings contain elements such as graphics, text, and tables, and are rich in content. However, the reviewers have limited time to review them, placing high demands on the review process. Against this backdrop, the need for intelligent power grid project reviews has emerged.

[0003] Currently, there is an intelligent review platform based on pre-trained large models. It uses natural language large models to perform text recognition and semantic understanding on design documents, extract key fields, and then compare them with the review specifications to draw review conclusions.

[0004] Existing technical solutions mainly focus on intelligently identifying and extracting key fields from document-based design results. There are few intelligent recognition solutions for drawing content, which makes it impossible to support the intelligent review of drawings. The invention with publication number CN117275022A discloses a method and device for complex table recognition and structured data based on PDF files, including: performing image preprocessing on the PDF file to obtain a first image; performing table line detection on the first image to obtain the positional relationship of the lines and the coordinates of each point; obtaining a simplified table structure and the connection relationship between each point and adjacent points based on preset merging rules and the positional relationship of the lines; obtaining multiple cell images based on the coordinates of each point and the connection relationship; obtaining text recognition results and data information based on the multiple cell images; organizing and integrating the text recognition results and data information based on the simplified table structure to obtain structured data.

[0005] The above scheme obtains a simplified table structure through table line detection, but the scanned PDF image file may have missing and misplaced lines. Reconstructing the table based solely on the line detection results can easily lead to recognition errors. After identifying the table structure, text recognition is performed through cell images, but the fonts in the scanned PDF image file may have some missing lines, which can easily lead to text recognition errors. Therefore, the above existing scheme still has the problem of low accuracy in the structured data table extraction results of PDF image files. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method, system and medium for extracting structured data tables from power grid engineering drawings, thereby improving the extraction accuracy of structured data tables.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A method for extracting structured data tables from power grid engineering drawings comprises the following steps:

[0009] Obtain the PDF image file of the power grid engineering drawing and convert it into grayscale;

[0010] Input the grayscaled PDF image file into a pre-built and trained intelligent image recognition model, obtain the images of each cell area in the table area of ​​the PDF image file, and record the corresponding row and column numbers;

[0011] Use the text OCR detection algorithm to perform text recognition on each cell area image to obtain the rough text recognition results;

[0012] The text rough recognition result is matched with a pre-built power grid engineering terminology library to obtain the text fine recognition result, and a structured data table is constructed according to the corresponding row and column numbers.

[0013] Furthermore, the intelligent image recognition model includes a table neural network recognition model and a cell neural network recognition model. The table neural network recognition model is used to identify the table area from the PDF image file and crop the table area image; the cell neural network recognition model is used to identify the cell area from the table area image, crop each cell area image and record the corresponding row and column numbers.

[0014] Furthermore, the processing of the intelligent image recognition model includes:

[0015] S201: Inputting the PDF image file to be tested into the table neural network recognition model to obtain the table area recognition result;

[0016] S202: acquiring an edge outer local area and an edge inner local area based on the table area recognition result;

[0017] S203: Determine whether the average grayscale value of non-pure white pixels in the local area outside the edge is less than 128. If so, execute step S204; otherwise, execute step S207;

[0018] S204: Determine whether the average grayscale value of non-pure white pixels in the local area inside the edge is greater than 128. If so, execute step S205; otherwise, execute step S206;

[0019] S205: After clipping the table area recognition result based on the local area inside the edge, re-input it into the table neural network recognition model to obtain an updated table area recognition result, and execute step S207;

[0020] S206: After expanding the table area recognition result based on the local area outside the edge, an updated table area recognition result is obtained, and step S207 is executed;

[0021] S207: Outputting the final table area recognition result;

[0022] S208: Input the final table area recognition result into the cell neural network recognition model to identify the cell area.

[0023] Furthermore, the process of respectively obtaining the local area outside the edge and the local area inside the edge based on the table area recognition result is specifically as follows:

[0024] Based on the four sides of the table area recognition result, square areas are respectively intercepted to the outside of the table area recognition result with a preset expansion width to obtain four edge outer local areas; based on the four sides of the table area recognition result, square areas are respectively intercepted to the inside of the table area recognition result with a preset expansion width to obtain four edge inner local areas.

[0025] Furthermore, step S203 specifically includes: determining whether the average grayscale value of non-pure white pixels in each local area outside the edge is less than 128, and if at least one local area outside the edge satisfies the condition, executing step S204; otherwise, executing step S207;

[0026] Step S204 specifically includes: determining whether the average grayscale value of non-pure white pixels in each local area inside the edge is greater than 128. If at least one local area inside the edge satisfies the condition, step S205 is executed; otherwise, step S206 is executed.

[0027] Step S205 specifically includes: clipping the table area recognition result based on the local area inside the edge that meets the conditions of step S204, re-inputting it into the table neural network recognition model to obtain an updated table area recognition result, and executing step S207;

[0028] Step S206: After expanding the table area recognition result based on the local area outside the edge that meets the conditions of step S203, an updated table area recognition result is obtained, and step S207 is executed.

[0029] Furthermore, the training process of the intelligent image recognition model includes:

[0030] Obtain a PDF image set with structured data tables for training, mark the table area of ​​each image in the PDF image set with a box, and use it as the table label corresponding to each PDF image, thereby obtaining a table training set; crop each PDF image according to the table label to obtain a table area image, and mark the cell area of ​​each table area image with a box, and use it as the cell label corresponding to each table area image, thereby obtaining a cell training set;

[0031] Input the PDF images in the table training set into the table neural network recognition model, obtain the table area recognition results, and perform model training based on the corresponding table labels;

[0032] The table area pictures in the cell training set are input into the cell neural network recognition model to identify the cell area, and the model training is performed based on the corresponding cell labels.

[0033] Furthermore, the process of constructing the table training set and the cell training set further includes:

[0034] The images in the obtained table training set and cell training set are rotated, flipped, and noise is added to obtain augmented data.

[0035] Furthermore, the method regenerates a table according to the acquired row and column numbers, and fills the generated table with the corresponding text recognition results to obtain a structured data table.

[0036] The present invention also provides a system for extracting structured data tables from power grid engineering drawings, comprising a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of the above method.

[0037] The present invention also provides a computer-readable storage medium, on which a computer program is stored. The computer program is used by a processor to execute the steps of the above method.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] (1) The present invention first identifies the cell area and the corresponding row and column numbers in the PDF image file through an intelligent image recognition model based on a neural network, and then performs text recognition in the cell area in two steps. First, rough text recognition is achieved through a text OCR detection algorithm. Then, combined with the standardization requirements for economic and technical table terms in power grid engineering drawings, a power grid engineering terminology library is collected and established. By matching the rough text recognition results with the power grid engineering terminology library, a precise text recognition result that meets the power grid engineering drawing standards is obtained, thereby extracting a more accurate structured data table, reducing the impact of missing or unclear pixels in the PDF image file, and improving the accuracy of the extraction results.

[0040] (2) The present invention uses two neural networks to perform table area recognition and cell area recognition in sequence. On the one hand, the accuracy of the recognition results can be improved by training them separately. On the other hand, in order to address the possible error problem in table area recognition, a method of combining pixel points to determine whether it is a table area or a labeling area can be used to improve the accuracy of table area recognition.

[0041] (3) The present invention is aimed at the PDF image file of the power grid engineering drawing, in which the economic and technical indicator table is located in the outer frame, and there is an explanatory annotation at the bottom of the economic and technical indicator table, and the pixel grayscale value of the explanatory annotation is generally low. The table area recognition result based on the table neural network recognition model is used to obtain the local area outside the edge and the local area inside the edge respectively, and judge whether there are lines or fonts of non-light-colored explanatory annotations in the local area outside the edge, and whether there are dark-colored lines or fonts in the local area inside the edge, so as to further judge whether the table area recognition result is the outer frame or the inside of the table, thereby realizing expansion or cropping, further improving the accuracy of the table area recognition result, and reducing the training requirements for the table neural network recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic flow chart of a method for extracting structured data tables from power grid engineering drawings provided in an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of a process for determining whether a table area or a marked area is used in combination with pixel points according to an embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram of a PDF image file of a power grid engineering drawing provided in an embodiment of the present invention. It is only a schematic diagram of the position of the table in the power grid engineering drawing. The unclear parts do not affect the main scheme to be protected by the present invention.

[0045] Figure 4 for Figure 3A schematic diagram of a local area, wherein the red frame area is the outer frame of the economic and technical indicator table, the blue frame area is the explanation and annotation part, the purple frame part is the economic and technical indicator table, and the unclear text in the remaining part does not affect the main scheme to be protected by the present invention;

[0046] Figure 5 This is a result of constructing a structured data table provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0048] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0049] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0050] Example 1

[0051] like Figure 1 As shown, this embodiment provides a method for extracting structured data tables from power grid engineering drawings, comprising the following steps:

[0052] S1: Obtain the PDF image file of the power grid engineering drawing and convert it into grayscale;

[0053] S2: Input the grayscaled PDF image file into the pre-built and trained intelligent image recognition model, obtain the images of each cell area in the table area of ​​the PDF image file, and record the corresponding row and column numbers;

[0054] S3: Use the text OCR detection algorithm to perform text recognition on each cell area image to obtain the rough text recognition result;

[0055] S4: Match the text coarse recognition results with the pre-built power grid engineering terminology library to obtain the text fine recognition results, and construct a structured data table based on the corresponding row and column numbers.

[0056] This solution first identifies the cell area and corresponding row and column numbers in the PDF image file through an intelligent image recognition model based on a neural network, and then performs text recognition in the cell area in two steps. First, rough text recognition is achieved through the text OCR detection algorithm. Then, combined with the standardization requirements for economic and technical table terms in power grid engineering drawings, a power grid engineering terminology library is collected and established. By matching the rough text recognition results with the power grid engineering terminology library, precise text recognition results that meet the standards of power grid engineering drawings are obtained, thereby extracting more accurate structured data tables, reducing the impact of missing or unclear pixels in PDF image files, and improving the accuracy of the extraction results.

[0057] Specifically, the intelligent image recognition model includes a table neural network recognition model and a cell neural network recognition model. The table neural network recognition model is used to identify the table area from the PDF image file and crop the table area image; the cell neural network recognition model is used to identify the cell area from the table area image, crop each cell area image and record the corresponding row and column numbers.

[0058] By using two neural networks to identify the table area and the cell area respectively, on the one hand, the accuracy of the recognition results can be improved through separate training. On the other hand, in order to address the possible errors in the table area recognition, the following method of combining pixel points to determine whether the table area is the labeling area can be used to improve the accuracy of the table area recognition.

[0059] like Figure 3 and Figure 4 As shown, for the processing object of this solution: the PDF image file of the power grid engineering drawing, the economic and technical indicators table area has the following characteristics: the economic and technical indicators table is located in the outer frame, and the outer frame is as follows: Figure 4 In the red box area, there are annotations at the bottom of the economic and technical indicators table, such as Figure 4 The blue box area in the figure is generally low in grayscale value, that is, the color is lighter, and the grayscale value of most pixels is not less than 128, that is, the K value is less than 50%.

[0060] The PDF image file to be recognized may have line breakpoints and offsets, which may lead to errors in the table neural network recognition model's recognition. There is also the problem that the table neural network recognition model's recognition result is the outer frame of the economic and technical indicators table, or the recognition result is a part of the economic and technical indicators table.

[0061] In this regard, Figure 2 As shown in the figure, the processing process of the intelligent image recognition model proposed in this solution specifically includes:

[0062] S201: Inputting the PDF image file to be tested into the table neural network recognition model to obtain the table area recognition result;

[0063] S202: Obtaining an edge outer local area and an edge inner local area based on the table area recognition result;

[0064] S203: Determine whether the average grayscale value of non-pure white pixels in the local area outside the edge is less than 128. If so, execute step S204; otherwise, execute step S207;

[0065] If the average grayscale value of non-pure white pixels in the local area outside the edge is less than 128, then there are non-light-colored explanatory annotation lines or fonts in the local area outside the edge, and the current table area recognition result may be the outer frame of the economic and technical indicator table or a part of the economic and technical indicator table. Therefore, further judgment of the local area inside the edge is required;

[0066] If the average grayscale value of non-pure white pixels in the local area outside the edge is greater than 128, then there are explanatory annotations in the local area outside the edge. At this time, the recognition result of the current table area is likely to be correct.

[0067] S204: Determine whether the average grayscale value of non-pure white pixels in the local area inside the edge is greater than 128. If so, execute step S205; otherwise, execute step S206;

[0068] If the average grayscale value of the non-pure white pixels in the local area inside the edge is greater than 128, it indicates that there are light-colored lines or fonts, that is, the inner side of the current table area recognition result is an annotation. It can be determined that the current table area recognition result is the outer frame of the economic and technical indicator table. Then, the table is cropped in step S205 and the recognition of the table neural network recognition model is performed again.

[0069] If the average grayscale value of the non-pure white pixels in the local area inside the edge is less than 128, it means that there are dark lines or fonts, that is, there are dark lines or fonts on both the inside and outside of the current table area recognition result. It can be judged that the current table area recognition result is part of the economic and technical indicator table. After the table is expanded in step S206, an updated table area recognition result is obtained.

[0070] The reason why it only needs to be expanded once is that the error between the recognition result of the tabular neural network recognition model and the true value generally does not exceed 10%, and a single expansion can meet the accuracy requirements.

[0071] S205: After the table area recognition result is cropped based on the local area inside the edge, it is re-input into the table neural network recognition model to obtain an updated table area recognition result, and step S207 is executed;

[0072] S206: After expanding the table area recognition result based on the local area outside the edge, an updated table area recognition result is obtained, and step S207 is executed;

[0073] S207: Outputting the final table area recognition result;

[0074] S208: Input the final table area recognition result into the cell neural network recognition model to identify the cell area.

[0075] Specifically, in step S202, the process of respectively obtaining the local area outside the edge and the local area inside the edge based on the table area recognition result is as follows:

[0076] Based on the four sides of the table area recognition result, square areas are cut outward of the table area recognition result with a preset expansion width to obtain four local areas outside the edge; based on the four sides of the table area recognition result, square areas are cut outward of the table area recognition result with a preset expansion width to obtain four local areas inside the edge.

[0077] The expansion width defined above can be adjusted according to actual conditions and tested to achieve the required accuracy.

[0078] In this regard, step S203 specifically includes: determining whether the average grayscale value of non-pure white pixels in each local area outside the edge is less than 128. If at least one local area outside the edge satisfies the condition, it indicates that the table area recognition result is a portion of the outer frame or the economic and technical indicator table, and step S204 is executed; otherwise, step S207 is executed.

[0079] Step S204 specifically includes: determining whether the average grayscale value of non-pure white pixels in each local area inside the edge is greater than 128. If at least one local area inside the edge satisfies the condition, it indicates that the table area recognition result is part of the economic and technical indicator table, and step S205 is executed; otherwise, step S206 is executed.

[0080] Step S205 specifically includes: clipping the table area recognition result based on the local area inside the edge that meets the conditions of step S204, re-inputting it into the table neural network recognition model to obtain an updated table area recognition result, and executing step S207;

[0081] Step S206: After expanding the table area recognition result based on the local area outside the edge that meets the conditions of step S203, an updated table area recognition result is obtained, and step S207 is executed.

[0082] Preferably, the training process of the intelligent image recognition model includes:

[0083] Obtain a PDF image set with structured data tables for training, mark the table area of ​​each image in the PDF image set with a box, and use it as the table label corresponding to each PDF image, thereby obtaining a table training set; crop each PDF image according to the table label to obtain a table area image, and mark the cell area of ​​each table area image with a box, and use it as the cell label corresponding to each table area image, thereby obtaining a cell training set;

[0084] Input the PDF images in the table training set into the table neural network recognition model, obtain the table area recognition results, and perform model training based on the corresponding table labels;

[0085] The table area images in the cell training set are input into the cell neural network recognition model to identify the cell area and perform model training based on the corresponding cell labels.

[0086] This is equivalent to training the table neural network recognition model and the cell neural network recognition model separately.

[0087] Preferably, the process of constructing the table training set and the cell training set further includes:

[0088] The images in the obtained table training set and cell training set are rotated, flipped and noise-added to obtain expanded data, thereby improving the recognition accuracy of the table neural network recognition model and the cell neural network recognition model.

[0089] Preferably, the method regenerates the table according to the obtained row and column numbers, and fills the corresponding text recognition results in the generated table to obtain a structured data table, such as Figure 5 shown.

[0090] The process of regenerating a table according to row and column numbers mainly realizes the tabular display of text recognition results. It is necessary to define the coordinate positions of the table lines and each table cell according to the row and column numbers.

[0091] Example 2

[0092] This embodiment provides a system for extracting structured data tables from power grid engineering drawings, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the method for extracting structured data tables from power grid engineering drawings as described in Example 1.

[0093] This embodiment further provides a computer-readable storage medium having a computer program stored thereon. The computer program is executed by a processor to perform the steps of the method for extracting structured data tables from power grid engineering drawings in embodiment 1.

[0094] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. A computer-readable storage medium can be a machine-readable signal medium or a machine-readable storage medium. A computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0095] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. Obtain a PDF image file of a power grid engineering drawing and convert it to grayscale; in the PDF image file of the power grid engineering drawing, a table area is provided with an outer frame outside and an explanatory annotation is provided at the bottom, and the pixel value of the font in the table area is larger than the pixel value of the font in the explanatory annotation; Input the grayscaled PDF image file into a pre-built and trained intelligent image recognition model, obtain the images of each cell area in the table area of ​​the PDF image file, and record the corresponding row and column numbers; Use the text OCR detection algorithm to perform text recognition on each cell area image to obtain the rough text recognition results; Matching the rough text recognition results with a pre-built power grid engineering terminology library to obtain precise text recognition results, and constructing a structured data table based on the corresponding row and column numbers; The intelligent image recognition model includes a table neural network recognition model and a cell neural network recognition model. The table neural network recognition model is used to identify the table area from the PDF image file and crop the table area image; the cell neural network recognition model is used to identify the cell area from the table area image, crop each cell area image and record the corresponding row and column numbers; The processing process of the intelligent image recognition model includes: S201: Inputting the PDF image file to be tested into the table neural network recognition model to obtain the table area recognition result; S202: acquiring an edge outer local area and an edge inner local area based on the table area recognition result; S203: Determine whether the average grayscale value of non-pure white pixels in the local area outside the edge is less than 128. If so, execute step S204; otherwise, execute step S207; S204: Determine whether the average grayscale value of non-pure white pixels in the local area inside the edge is greater than 128. If so, execute step S205; otherwise, execute step S206; S205: After clipping the table area recognition result based on the local area inside the edge, re-input it into the table neural network recognition model to obtain an updated table area recognition result, and execute step S207; S206: After expanding the table area recognition result based on the local area outside the edge, an updated table area recognition result is obtained, and step S207 is executed; S207: Outputting the final table area recognition result; S208: Inputting the final table area recognition result into the cell neural network recognition model to identify the cell area; The specific process of obtaining the edge outer local area and the edge inner local area based on the table area recognition result is as follows: Based on the four sides of the table area recognition result, square areas are respectively cut outward of the table area recognition result with a preset expansion width to obtain four edge outer local areas; based on the four sides of the table area recognition result, square areas are respectively cut outward of the table area recognition result with a preset expansion width to obtain four edge inner local areas; Step S203 specifically includes: determining whether the average grayscale value of non-pure white pixels in each local area outside the edge is less than 128, and if at least one local area outside the edge satisfies the condition, executing step S204; otherwise, executing step S207; Step S204 specifically includes: determining whether the average grayscale value of non-pure white pixels in each local area inside the edge is greater than 128. If at least one local area inside the edge satisfies the condition, step S205 is executed; otherwise, step S206 is executed. Step S205 specifically includes: clipping the table area recognition result based on the local area inside the edge that meets the conditions of step S204, re-inputting it into the table neural network recognition model to obtain an updated table area recognition result, and executing step S207; Step S206: After expanding the table area recognition result based on the local area outside the edge that meets the conditions of step S203, an updated table area recognition result is obtained, and step S207 is executed.

2. The method for extracting structured data tables from power grid engineering drawings according to claim 1, characterized in that: The training process of the intelligent image recognition model includes: Obtain a PDF image set with structured data tables for training, mark the table area of ​​each image in the PDF image set with a box, and use it as the table label corresponding to each PDF image, thereby obtaining a table training set; crop each PDF image according to the table label to obtain a table area image, and mark the cell area of ​​each table area image with a box, and use it as the cell label corresponding to each table area image, thereby obtaining a cell training set; Input the PDF images in the table training set into the table neural network recognition model, obtain the table area recognition results, and perform model training based on the corresponding table labels; The table area pictures in the cell training set are input into the cell neural network recognition model to identify the cell area, and the model training is performed based on the corresponding cell labels.

3. The method for extracting structured data tables from power grid engineering drawings according to claim 2, characterized in that: The process of constructing the table training set and the cell training set also includes: The images in the obtained table training set and cell training set are rotated, flipped, and noise is added to obtain augmented data.

4. The method for extracting structured data tables from power grid engineering drawings according to claim 1, characterized in that: The method regenerates a table according to the acquired row and column numbers, and fills the generated table with the corresponding text recognition results to obtain a structured data table.

5. A system for extracting structured data tables from power grid engineering drawings, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of any one of the methods according to claims 1 to 4.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which is used by a processor to execute the steps of the method according to any one of claims 1 to 4.

Images