An intelligent analysis method for primary wiring diagram of substation

Abstract

The invention discloses an intelligent parsing method for a primary wiring diagram of a substation, comprising: 1) realizing position detection and rotation angle recognition of graphic elements in the primary wiring diagram based on a target detection algorithm and a template matching method; 2) using Hough line detection to recognize the main connecting lines in the wiring diagram, and detecting the end points of the graphic elements according to the positions of the graphic elements; 3) using background colors to fill the pixel areas where the graphic elements are located, obtaining a primary wiring diagram containing only connecting lines, and extracting connecting line pixels by using threshold segmentation; 4) selecting the end points of any graphic elements as the starting growth points, obtaining the connected areas with the starting growth points, and using or copying to first detect and fit the connected areas; the invention has the advantages that: structured information such as components, topological relationships, keywords, etc. can be intelligently extracted from an unstructured primary wiring diagram, and a wiring table can be automatically generated, so that the existing primary wiring diagram can be effectively digitally managed, and the work efficiency of substation operation and maintenance management personnel can be greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital image recognition, and in particular to an intelligent analysis method for a primary wiring diagram of a transformer substation. Background Art

[0002] The primary wiring diagram of the substation is the core basis for the dispatching of the power system. The digital management of the primary wiring diagram of the substation will greatly facilitate the dispatching of the power system. At present, the primary wiring diagram of the substation is mainly read and entered manually. With the rapid construction of the power grid, the workload of manually entering the wiring diagram is huge and it is difficult to meet the needs of actual operation and maintenance. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent parsing method for a substation primary wiring diagram, which can intelligently identify the topological structure of the substation primary wiring diagram through image data and form structured data that can be understood by a computer, thereby realizing intelligent recognition and automatic entry of the substation primary wiring diagram.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] A method for intelligently analyzing a primary wiring diagram of a substation, comprising the following steps:

[0006] Step 1: Detect the position of the primary wiring diagram element and identify the rotation angle based on the target detection algorithm and template matching method;

[0007] Step 2: Use Hough line detection to identify the main connecting lines in the wiring diagram, and detect the endpoints of the primitives according to the primitive positions;

[0008] Step 3, fill the pixel area where the primitive is located with the background color to obtain a primary wiring diagram containing only connecting lines, and extract the connecting line pixels using threshold segmentation;

[0009] Step 4: Select any endpoint of a primitive as the starting growth point, obtain the connected area with the starting growth point, use or copy to first detect and fit the connected area, obtain the straight connecting lines, endpoints and devices connected to the endpoints in the area, and add the devices in the connected domain to the connected device set;

[0010] Step 5: Select the endpoints on the connected device that have not been detected for the connectivity domain, and repeat step 4 to expand the connectivity domain until all endpoints are connected;

[0011] Step 6: For devices that are not added to the connected domain, repeat steps 4 and 5 to form a new connected domain until all devices on the wiring diagram are connected;

[0012] Step 7: Use optical character recognition to detect the text logo in the primary wiring diagram, and match the text with the graphic element according to the closest matching relationship;

[0013] Step 8: The recognition results are stored according to the categories of graphics, text, and connecting lines. The text needs to be associated with the corresponding components, and the connecting lines need to be associated with the corresponding components and ports.

[0014] In the above-mentioned intelligent analysis method for a substation primary wiring diagram, the implementation of step 1 can be specifically divided into the following steps:

[0015] Step 1.1, mark all the graphics elements in the primary wiring diagram of the substation, and cut the wiring diagram into several image blocks for model training and target detection testing. After the test, the graphics element position needs to be mapped to the original wiring diagram. The target detection result needs to get the type and position of the graphics element; the graphics elements include 11 categories: circuit breaker, disconnector, grounding switch, two-winding transformer, three-winding transformer, capacitor, reactor, busbar, line, reactive compensation device, generator;

[0016] Step 1.2: Create a standard graphic element template library, which mainly includes templates of various graphic elements with different rotation angles. For example, a single symmetry axis graphic element needs to include four types of templates: 0°, 90°, 180°, and 270°; a double symmetry axis graphic element needs to include two types of templates: 0° and 180°.

[0017] Step 1.3: Perform template matching on the graphics element obtained by target detection and the standard graphics element template library to obtain the rotation angle of the graphics element, and at the same time, obtain the distribution order of the graphics element terminals based on electrical expertise.

[0018] In the above-mentioned intelligent analysis method for a substation primary wiring diagram, the implementation of step 2 can be specifically divided into the following steps:

[0019] Step 2.1, pre-process the original wiring diagram using Gaussian filtering and threshold segmentation to reduce the interference of irrelevant pixels;

[0020] Step 2.2, use Hough line detection to fit and merge the straight lines in the wiring diagram to obtain the connecting lines between the graphics elements;

[0021] Step 2.3, perform an AND operation on the rectangular frame and connecting lines containing the graphic element. The intersection point of the rectangle and the connecting line can be used as the terminal position of the corresponding graphic element. At the same time, the graphic element terminals are numbered according to the terminal distribution order in step 1.3.

[0022] In the above-mentioned intelligent analysis method for a substation primary wiring diagram, the implementation of step 3 can be specifically divided into the following steps:

[0023] Step 3.1: In order to effectively extract the specific position of the connecting line, the pixels corresponding to the primitives in the image need to be filled with the background color, where the upper left vertex of the primitive rectangle is (x min ,y min ), the coordinates of the lower right vertex are (x max ,y max ), so that P i,j {i∈(x min ,x max ),j∈(y min ,y max )}=255, eliminate the influence of primitive pixels on connection line extraction;

[0024] Step 3.2: Perform Gaussian smoothing and threshold segmentation on the wiring diagram with the graphics elements removed to extract the pixels of the connecting lines.

[0025] In the above-mentioned intelligent analysis method for a substation primary wiring diagram, the implementation of step 4 can be specifically divided into the following steps:

[0026] Step 4.1, randomly select a primitive, take one of the terminals on the primitive as the starting growth point, use the expansion algorithm to expand the starting point, and solve the connected domain of the starting point according to the 8-connected domain rule. The connected domain is the connecting line that can pass from the starting point;

[0027] Step 4.2, perform Hough line detection on the connected domain to obtain the line segments corresponding to each connecting line;

[0028] Step 4.3, for the detected line segments, calculate the intersection points between every two line segments, and divide the intersection points into "L"-shaped intersection points, "T"-shaped intersection points, and "X"-shaped intersection points. For the "X"-shaped intersection points, cut the two line segments into two segments at the intersection points to obtain new connecting lines.

[0029] Step 4.4, divide the endpoints of all line segments into connection line endpoints and component endpoints, wherein the connection line endpoints are obtained in step 4.3, and the component endpoints can be obtained according to the position of the primitive terminal in step 2.3, and the primitives connected to the component endpoints are obtained at the same time to form a topological relationship between the connection line and the primitive; if for a connection line, both of its endpoints are not connection line endpoints or component endpoints, the connection line is considered to be a busbar;

[0030] Step 4.5: The primitives connected to the component endpoints can be added to the existing connected domain, and the new connected domain is used as the identified area, and all primitive endpoints in the connected domain that are not connected to the components or connecting lines are used as candidate starting growth points.

[0031] In the above-mentioned intelligent analysis method for a substation primary wiring diagram, the implementation of step 5 can be specifically divided into the following steps:

[0032] Step 5.1: Select any point from the candidate starting growth points, and calculate a new connected domain from this point until all component endpoints in the connected domain are connected to the connecting lines or primitive endpoints.

[0033] In the above-mentioned intelligent analysis method for a substation primary wiring diagram, the implementation of step 6 can be specifically divided into the following steps:

[0034] Step 6.1: After executing step 5, if there are still primitives that are not included in the connected domain, continue to select primitives from these primitives and execute steps 4 and 5.

[0035] In the above-mentioned intelligent analysis method for a substation primary wiring diagram, the implementation of step 7 can be specifically divided into the following steps:

[0036] Step 7.1, extracting the character string and its position in the wiring diagram using optical character recognition technology;

[0037] Step 7.2: For each character string, calculate the closest graphic element to it, and if the distance between the outer boundary of the graphic element and the outer boundary of the character string does not exceed a certain threshold, then the character string is used as the keyword of the graphic element, otherwise the character string is discarded.

[0038] Compared with the prior art, the intelligent parsing method for the substation primary wiring diagram described in the present invention can intelligently extract structured information such as components, topological relationships, keywords, etc. from the unstructured primary wiring diagram, automatically generate a wiring table, and effectively digitally manage the existing primary wiring diagram, thereby greatly improving the work efficiency of substation operation and maintenance management personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is an overall flow chart of a primary wiring diagram intelligent identification method in one embodiment of the present invention;

[0040] Figure 2 This is a primitive erasing effect diagram in one embodiment of the present invention;

[0041] Figure 3 This is a connection line extraction effect diagram in one embodiment of the present invention;

[0042] Figure 4 is a visualization diagram of recognition results in one embodiment of the present invention; DETAILED DESCRIPTION

[0043] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0044] This embodiment combines existing image processing, target detection and optical character recognition algorithms to intelligently identify the primary wiring diagram of the substation in the power system, and realizes intelligent analysis of the primary wiring diagram from image to structured data, which is of great significance for improving the level of intelligent operation and maintenance of the substation.

[0045] This embodiment is implemented by the following technical solution, a method for intelligently analyzing a primary wiring diagram of a substation, comprising the following steps:

[0046] S1: Based on the target detection algorithm and template matching method, the position detection and rotation angle recognition of the wiring diagram elements are realized.

[0047] S2: Use Hough line detection to identify the main connecting lines in the wiring diagram, and detect the endpoints of the primitives according to the positions of the primitives;

[0048] S3: Fill the pixel area where the primitive is located with the background color to obtain a primary wiring diagram containing only connecting lines, and extract the connecting line pixels using threshold segmentation;

[0049] S4: Select an endpoint of any primitive as the starting growth point, obtain the connected area with the starting growth point, use Hough line detection to fit the connected area, obtain the straight line connection lines, endpoints and devices connected to the endpoints in the area, and add the devices in the connected area to the connected device set;

[0050] S5: Select endpoints on the connected device that have not been detected for the connectivity domain in turn, and repeat step 4 to expand the connectivity domain until all endpoints are connected;

[0051] S6: For devices that are not added to the connected domain, repeat steps 4 and 5 to form a new connected domain until all devices on the wiring diagram are connected.

[0052] S7: Use optical character recognition to detect text labels in the primary wiring diagram, and match the text with the graphic element based on the closest matching relationship.

[0053] S8: The recognition results are classified and stored according to the graphics, texts, and connecting lines. The texts need to be associated with the corresponding components, and the connecting lines need to be associated with the corresponding components and ports.

[0054] In specific implementation, the intelligent analysis method of a substation primary wiring diagram proposed by the present invention can be implemented by the following steps:

[0055] 1. Realize the position detection and rotation angle recognition of the wiring diagram element based on the target detection algorithm and template matching method;

[0056] 1) This embodiment develops a primary wiring diagram primitive recognition algorithm based on the YOLO v4 model. The primary wiring diagram primitives are divided into 11 categories, including circuit breakers, isolating switches, grounding switches, two-winding transformers, three-winding transformers, capacitors, reactors, busbars, lines, reactive power compensation devices, and generators. This embodiment labels the above primitive types according to the VOC standard, and trains on the labeled data set to obtain a primitive recognition model.

[0057] 2) YOLO v4 needs to resample the input image to 608*608 before primitive recognition, and the actual primary wiring diagram is much larger than this size. If it is directly scaled, it will make primitive recognition difficult. Therefore, a wiring diagram cropping strategy is proposed in this embodiment, specifically: the original image of the primary wiring diagram is cropped into several images of 608*608 size, and the first image in the upper left corner is the (1,1)th cropped image, then the pixel range of the (m,n)th image is [m*550:m*550+608,n*550:n*550+608];

[0058] 3) Based on the primitive recognition results, it is necessary to further confirm the ports of each primitive, so the present embodiment uses the template matching method to detect the rotation angle and port of the primitive; specifically: for each type of primitive, it is necessary to prepare primitive templates including different forms and different rotation angles (such as single-axis symmetry primitives need to include four types of 0°, 90°, 180°, and 270°, and double-axis symmetry primitives need to include two types of 0° and 180°), and each template needs to manually set the two attributes of rotation angle and connection port; for the recognition result obtained by the primitive recognition algorithm, the template matching method is used to match it with the pre-made template set, and the template with the highest matching pair with the recognition result is found, and the attributes of the template are passed to the recognition result to obtain the rotation angle and connection port number of the recognized primitive;

[0059] Second, use Hough line detection to identify the main connecting lines in the wiring diagram, and detect the endpoints of the primitives according to the primitive positions;

[0060] 1) In this embodiment, Gaussian filtering and threshold segmentation algorithms are first used to perform image preprocessing on the primary wiring diagram to reduce the interference of noise on subsequent recognition. Specifically, in the implementation process, Gaussian filtering function and threshold segmentation function in the OpenCV algorithm library are used for processing;

[0061] 2) Based on the image obtained in the above steps, the Hough straight line fitting algorithm in the OpenCV algorithm library is used to fit the connecting lines in the image, and the fitted straight lines are divided into horizontal connecting lines and vertical connecting lines;

[0062] 3) For the pixel blocks of the primitives detected in step 1, calculate the straight lines that intersect with them, and use the intersections of the straight lines and the primitive boundaries as the connection points of the primitives. Specifically, assuming that the position of the current primitive can be expressed as (x1, y1, x2, y2), for each straight line fitted in the previous steps, calculate the intersection of the straight line and the primitive pixel area detected in step 1, as the connection endpoints of the corresponding primitive; specifically, if calculating the intersection of a horizontal straight line and a primitive, assuming that the two endpoints of the straight line are A(x1, y1, x2, y2) and A(x2, y2). A ,y A ) and B(x B ,y B ), we can see that y A =y B For the endpoint A of a line, if the following conditions are met, it can be considered to be connected to the primitive, and the endpoint A is recorded as the connection point:

[0063]

[0064] Wherein T is the discontinuity threshold, in this embodiment T=5, and the unit is pixel. The calculation method of endpoint B is the same as above.

[0065] If we calculate the intersection of a vertical line and a primitive, assuming that the two endpoints of the line are A(x A ,y A ) and B(x B ,y B ), we can see that x A =x B , the basis for judging whether the endpoint is connected to the primitive is:

[0066]

[0067] For each primitive, record all connection endpoint information.

[0068] 3. Fill the pixel area where the primitive is located with the background color to obtain a primary wiring diagram containing only connecting lines, and extract the connecting lines using threshold segmentation;

[0069] 1) In step 1, the primitives in the wiring diagram have been identified, and many primitives have structures similar to connecting lines. Therefore, first, the primitive part is covered with the background. Specifically, assuming that the upper left vertex of the primitive rectangle is (x min ,y min ), the coordinates of the lower right vertex are (x max ,y max ), so that P i,j {i∈(x min ,x max ),j∈(y min ,y max)}=255, thereby eliminating the influence of primitive pixels on the extraction of connecting lines; the effect after primitive erasure is shown in the figure Figure 2 As shown;

[0070] 2) The wiring diagram after removing the primitives mainly includes the connecting lines and text parts. This embodiment further uses OpenCV to perform Gaussian smoothing and threshold segmentation to reduce the interference of stray points and noise; the effect diagram after the connecting line extraction is as follows Figure 3 As shown;

[0071] 4. Select any endpoint of a primitive as the starting growth point, obtain the connected area with the starting growth point, use Hough line detection to fit the connected area, obtain the straight line connection lines, endpoints and devices connected to the endpoints in the area, and add the devices in the connected domain to the connected device set;

[0072] 1) Select an element at random, and use one of the terminals on the element as the starting growth point. Use the expansion algorithm to expand the starting point, and solve the connected domain of the starting point according to the 8-connected domain rule. The connected domain is the connecting line that can pass from the starting point;

[0073] 2) Perform Hough line detection on the connected domain to obtain the line segments corresponding to each connecting line;

[0074] 3) For the detected line segments, the intersection points between every two line segments are calculated, and the intersection points are divided into "L"-shaped intersection points, "T"-shaped intersection points, and "X"-shaped intersection points. For the "X"-shaped intersection points, the two line segments are cut into two segments at the intersection points to obtain new connecting lines;

[0075] 4) The endpoints of all line segments are divided into connection line endpoints and component endpoints, where the connection line endpoints are obtained by 3), and the component endpoints can be obtained according to the terminal positions of the primitives in step 2, and the primitives connected to the component endpoints are obtained at the same time to form a topological relationship between the connection line and the primitives; if for a connection line, both of its endpoints are not connection line endpoints or component endpoints, the connection line is considered to be a busbar;

[0076] 5) The primitives connected to the component endpoints can be added to the existing connected domain, and the new connected domain is used as the identified area, and all primitive endpoints in the connected domain that are not connected to the components or connecting lines are used as candidate starting growth points;

[0077] 5. Select any point from the candidate starting growth points, and calculate a new connected domain from this point until all component endpoints in the connected domain are connected to the connecting lines or primitive endpoints;

[0078] 6. For devices that are not added to the connected domain, repeat steps 4 and 5 to form a new connected domain until all devices on the wiring diagram are connected;

[0079] 7. Use optical character recognition to detect text labels in a wiring diagram and match the text with the graphic element based on the nearest matching relationship;

[0080] 1) The optical character recognition detection in this embodiment is implemented based on PaddlePaddle OCR. Through this module, the content and position of each text block in the wiring diagram can be identified;

[0081] 2) Considering that the text in the wiring diagram may appear in multiple lines or vertical text, and PaddlePaddle OCR can only detect single-line text, this embodiment merges multiple lines of text. Specifically, if the vertical distance between two text blocks does not exceed the average height of the two text blocks, the two text blocks are considered to be merged into one, and the text contents are spliced ​​from top to bottom. The text position is the smallest rectangular box containing the two text blocks;

[0082] 3) For each character string, calculate the closest primitive to it, and if the distance between the outer boundary of the primitive and the outer boundary of the character string does not exceed a certain threshold, then the character string is used as the keyword of the primitive, otherwise the character string is discarded; in this embodiment, the distance threshold is set according to the size of the primitive, and the average value of the long sides of all circuit breaker primitives is taken as the distance judgment threshold;

[0083] 8. The recognition results are stored according to the categories of graphics, texts, and connecting lines. The texts need to be associated with the corresponding components, and the connecting lines need to be associated with the corresponding components and ports.

[0084] The recognition results of this embodiment are visualized as shown in the attached figure. Figure 4 shown.

[0085] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0086] Although the specific embodiments of the present invention are described above in conjunction with the accompanying drawings, it should be understood by those skilled in the art that these are only examples, and that various variations or modifications may be made to these embodiments without departing from the principles and essence of the present invention. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for intelligent analysis of a primary wiring diagram of a substation, characterized in that: The following steps are involved: Step 1: Detect the position of the primary wiring diagram element and identify the rotation angle based on the target detection algorithm and template matching method; Step 2: Use Hough line detection to identify the connection lines in the wiring diagram, and detect the endpoints of the primitives according to the primitive positions; Step 3, fill the pixel area where the primitive is located with the background color to obtain a primary wiring diagram containing only connecting lines, and extract the connecting line pixels using threshold segmentation; Step 4: Select any endpoint of a primitive as the starting growth point, obtain the connected area with the starting growth point, first detect and then fit the connected area, obtain the straight connecting lines, endpoints and devices connected to the endpoints in the area, and add the devices in the connected area to the connected device set; The implementation consists of the following steps: Step 4.1, randomly select a primitive, take one of the terminals on the primitive as the starting growth point, use the expansion algorithm to expand the starting point, and solve the connected domain of the starting point according to the 8-connected domain rule. The connected domain is the connecting line that can pass from the starting point; Step 4.2, perform Hough line detection on the connected domain to obtain the line segments corresponding to each connecting line; Step 4.3, for the detected line segments, calculate the intersection points between every two line segments, and divide the intersection points into "L"-shaped intersection points, "T"-shaped intersection points, and "X"-shaped intersection points. For the "X"-shaped intersection points, cut the two line segments into two segments at the intersection points to obtain new connecting lines; Step 4.4, divide the endpoints of all line segments into connection line endpoints and component endpoints, wherein the connection line endpoints are obtained in step 4.3, and the component endpoints are obtained according to the position of the primitive terminal in step 2.3, and the primitives connected to the component endpoints are obtained at the same time, so as to form a topological relationship between the connection line and the primitive; if for a connection line, both of its endpoints are not connection line endpoints or component endpoints, then the connection line is considered to be a busbar; Step 4.5: The primitives connected to the component endpoints are added to the existing connected domain, and the new connected domain is used as the identified area. All primitive endpoints in the connected domain that are not connected to the components or connecting lines are used as candidate starting growth points. Step 5: Select the endpoints on the connected device that have not been detected for the connectivity domain, and repeat step 4 to expand the connectivity domain until all endpoints are connected; Step 6: For devices that are not added to the connected domain, repeat steps 4 and 5 to form a new connected domain until all devices on the wiring diagram are connected; Step 7: Use optical character recognition to detect text labels in the primary wiring diagram, and match the text with the graphic element according to the closest matching relationship; Step 8: The recognition results are stored according to the categories of graphics, text, and connecting lines. The text needs to be associated with the corresponding components, and the connecting lines need to be associated with the corresponding components and ports.

2. The intelligent analysis method of a substation primary wiring diagram according to claim 1, characterized in that: The implementation of step 1 includes the following steps: Step 1.1, mark all the graphics elements in the primary wiring diagram of the substation, and cut the wiring diagram into several image blocks for model training and target detection testing. After the test, the graphics element position needs to be mapped to the original wiring diagram. The target detection result needs to get the type and position of the graphics element; the graphics elements include circuit breakers, disconnect switches, grounding switches, two-winding transformers, three-winding transformers, capacitors, reactors, busbars, lines, reactive compensation devices and generators, a total of 11 categories; Step 1.2: Create a standard graphic element template library, including templates of various graphic elements at different rotation angles; Step 1.3: Perform template matching on the graphics element obtained by target detection and the standard graphics element template library to obtain the rotation angle of the graphics element, and at the same time, obtain the distribution order of the graphics element terminals based on electrical expertise.

3. A method for intelligent analysis of a primary wiring diagram of a substation as claimed in claim 2, characterized in that: The implementation of step 2 includes the following steps: Step 2.1, pre-process the original wiring diagram using Gaussian filtering and threshold segmentation to reduce the interference of irrelevant pixels; Step 2.2, use Hough line detection to fit and merge the straight lines in the wiring diagram to obtain the connecting lines between the graphics elements; Step 2.3, perform an AND operation on the rectangular frame and connecting lines containing the graphic element. The intersection point of the rectangle and the connecting line can be used as the terminal position of the corresponding graphic element. At the same time, the graphic element terminals are numbered according to the terminal distribution order in step 1.

3.

4. The intelligent analysis method of a substation primary wiring diagram according to claim 1, characterized in that: The implementation of step 3 includes the following steps: Step 3.1: In order to effectively extract the specific position of the connecting line, the pixels corresponding to the primitives in the image need to be filled with the background color, where the upper left vertex of the primitive rectangle is (x min ,y min ) ， The coordinates of the lower right vertex are ( x max ,y max)， Make P i,j {i∈(x min ,x max ),j∈(y min ,y max )} ＝ 255 ， Eliminate the influence of primitive pixels on connection line extraction; Step 3.2: Perform Gaussian smoothing and threshold segmentation on the wiring diagram with the graphics elements removed to extract the pixels of the connecting lines.

5. The intelligent analysis method of a substation primary wiring diagram according to claim 4, characterized in that: The implementation of step 5 includes the following steps: Step 5.1: Select any point from the candidate starting growth points, and calculate a new connected domain from this point until all component endpoints in the connected domain are connected to the connecting lines or primitive endpoints.

6. A method for intelligent analysis of a primary wiring diagram of a substation as claimed in claim 5, characterized in that: The implementation of step 6 includes the following steps: Step 6.1: After executing step 5, if there are still primitives that are not included in the connected domain, continue to select primitives from these primitives and execute steps 4 and 5.

7. A method for intelligent analysis of a primary wiring diagram of a substation as claimed in claim 6, characterized in that: The implementation of step 7 includes the following steps: Step 7.1, extracting the character string and its position in the wiring diagram using optical character recognition technology; Step 7.2: For each character string, calculate the closest graphic element to it, and if the distance between the outer boundary of the graphic element and the outer boundary of the character string does not exceed a certain threshold, then the character string is used as the keyword of the graphic element, otherwise the character string is discarded.

Images