2D structure engineering drawing reconstruction model system and method based on deep learning

Abstract

The invention discloses a 2D structural engineering drawing reconstruction model system and method based on deep learning, and belongs to the technical field of drawing recognition. A drawing processing module; a data storage module; a three-dimensional model generation module; wherein the drawing processing module comprises a drawing detection and identification unit used for carrying out object detection and identification on components and annotations in the 2D structural engineering drawing and outputting an object list of the detected components, including object types and pixel coordinates of objects; the annotation extracting and matching unit is used for extracting specific text contents from the recognized annotation areas, matching the annotations with corresponding components and constructing matching pairs formed by the components and the annotations; the coordinate system conversion unit is used for converting pixel coordinates of the component detected by the drawing detection and recognition unit into drawing coordinates; the technical problems of accuracy, economy, efficiency, applicability and the like of reconstructing the BIM model from a 2D structural engineering drawing in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of drawing recognition, and in particular to a system and method for reconstructing a model of a 2D structural engineering drawing based on deep learning. Background Art

[0002] Building Information Modeling (BIM) is a technology that digitally represents the physical and functional characteristics of buildings. It is an information base that contains detailed parameter information of building objects (such as geometric shapes, spatial relationships, attributes, etc.). BIM is mainly used in the construction industry to improve the quality of building design, construction, and operation and maintenance management. It can connect massive and discrete building data in series, realize the effective transmission of project information from design, construction to operation and maintenance, and promote data collaboration among various disciplines in the operation and maintenance stage.

[0003] Currently, for buildings that have been built for more than 20 years, most project information is saved and delivered in the form of 2D as-built drawings, lacking BIM management. To solve this problem, the industry has developed some methods for reconstructing 3D models from existing building as-built drawings. These methods mainly include reconstruction methods based on field data (such as point cloud data) and reconstruction methods based on CAD 2D drawings.

[0004] However, these prior arts have the following problems:

[0005] First, the reconstruction method based on field data (such as point cloud) cannot establish a complete and information-rich BIM model, and the equipment cost is high and the operation is difficult, which limits its application scope;

[0006] Second, the reconstruction method based on CAD two-dimensional drawings requires a lot of expertise and time to manually design feature extractors, and has poor robustness and generalization capabilities.

[0007] To this end, the present invention provides a system and method for reconstructing BIM models from 2D structural engineering drawings in a more efficient, accurate and economical manner. Summary of the invention

[0008] The purpose of the present invention is to provide a system and method for reconstructing models from 2D structural engineering drawings based on deep learning, so as to solve the technical problem of how to efficiently, accurately and economically reconstruct building information models from 2D structural engineering drawings raised in the above-mentioned background technology.

[0009] In order to achieve the above object, the present invention provides the following technical solutions:

[0010] A system for reconstructing a model of a 2D structural engineering drawing based on deep learning, comprising:

[0011] A user interface module is used to receive 2D structural engineering drawings input by users, display intermediate results in the process of processing the 2D structural engineering drawings, present the final generated 3D model, and perform model export operations;

[0012] The drawing processing module is used to analyze, extract and match the data in the 2D structural engineering drawings input by the user interface module to generate structured drawing information;

[0013] A data storage module is used to store the structured information generated by the drawing processing module in XML format, including component instance information, location information and attribute description information;

[0014] The three-dimensional model generation module is used to parse the XML data in the data storage module and create a three-dimensional model that complies with the building information model standard and contains geometry and attribute information based on the XML data; wherein,

[0015] The drawing processing module includes:

[0016] A drawing detection and recognition unit is used to perform object detection and recognition on components and annotations in 2D structural engineering drawings using a YOLOv5 target detection algorithm based on deep learning, and output an object list of detected components, wherein the object list includes object types and pixel coordinates of the objects;

[0017] The annotation extraction and matching unit is used to extract specific text content from the recognized annotation area based on the detection results of the drawing detection and recognition unit using the PaddleOCR text recognition algorithm, and match these annotations with the corresponding components to construct matching pairs consisting of components and annotations;

[0018] The coordinate system conversion unit is used to construct a modeling coordinate system and convert the pixel coordinates of the components detected by the drawing detection and recognition unit into drawing coordinates.

[0019] A method for reconstructing a model of a 2D structural engineering drawing based on deep learning, comprising:

[0020] S1, using the drawing detection and recognition unit to detect and recognize the 2D structural engineering drawings uploaded by the user in the user interface module using an object recognition algorithm, and the detection and recognition targets are entity object classes and annotation classes;

[0021] S2, extracting the annotation area from the data identified by the drawing detection and recognition unit by using a text recognition algorithm, and matching the annotation with the component to construct a matching pair consisting of the component and the annotation;

[0022] S3, constructing a modeling coordinate system through a coordinate system conversion unit, and converting the pixel coordinates of the object identified by the drawing detection and recognition unit into drawing coordinates in combination with dimension annotation;

[0023] S4, the extracted information is stored in XML format through the data storage module, and the three-dimensional model generation module generates a three-dimensional building information model based on the data stored in the data storage module.

[0024] Preferably, in S1, the object recognition algorithm adopts the YOLOv5 target detection algorithm based on deep learning.

[0025] Preferably, in S1, the entity object class is a component, and the component includes a beam and a column; the beam includes a horizontal beam, a vertical beam and an inclined beam.

[0026] Preferably, in S1, the annotation class is a component annotation, including a title, an axis grid, a dimension annotation, and an attribute annotation.

[0027] Preferably, S2 is specifically:

[0028] S21, cropping the image according to the detection area of ​​S1 to extract the valid annotation text area;

[0029] S22, using the PaddleOCR text recognition algorithm to perform text content recognition on the extracted annotation text area and extract the text content of the annotation;

[0030] S23, matching the attribute annotations and dimension annotations in the 2D structural engineering drawings with the corresponding beams and columns respectively, and obtaining the geometric information and semantic information of each beam and column.

[0031] Preferably, the geometric information is data related to the shape, size, position, angle and spatial relationship of the components.

[0032] Preferably, the semantic information is descriptive data related to the purpose, characteristics or functions of the component.

[0033] Preferably, the attribute annotation includes a centralized annotation and a position annotation, and the positions of the two annotations are located within a certain interval between two endpoints of the object; thus, the position constraint condition of the attribute annotation is obtained:

[0034] a. Position constraints for vertical attribute annotations:

[0035]

[0036] b. Position constraints for horizontal attribute annotations:

[0037]

[0038] Where:

[0039] (BBox i ) and BBox k : List of bounding boxes with attribute annotations in target detection results

[0040] {BBox1,BBox2,...,BBox n} in the element;

[0041] X PCS (BBox i ); In pixel coordinate system, BBox i and BBox k The X-axis coordinate of the center point;

[0042] Y PCS (BBox k ): BBox in pixel coordinate system i and BBox k The Y-axis coordinate of the center point;

[0043] D x : A list of distances between adjacent vertical grids;

[0044] D y : A list of distances between adjacent horizontal axis grids;

[0045] α: The position ratio between adjacent axis grids. The value of α varies depending on the attribute annotation type.

[0046] β: Maximum value of distance error.

[0047] Preferably, S3 is specifically:

[0048] S31, establish a modeling coordinate system with the intersection of the lowest horizontal axis grid and the leftmost vertical axis grid in the 2D structural engineering drawing as the origin;

[0049] S32, converting the pixel coordinate system into a modeling coordinate system, and calculating the specific position coordinates of the object in combination with the dimensioning.

[0050] Preferably, the conversion process between the modeling coordinate system point P (x, y) and the pixel coordinate system P (u, v) is expressed as:

[0051]

[0052] Where:

[0053] Scale x : Conversion coefficient between pixel coordinate system and modeling coordinate system X axis;

[0054] Scale y: Conversion coefficient between pixel coordinate system and modeling coordinate system Y axis;

[0055] u0 and v0 represent the coordinates of the origin of the modeling coordinate system in the pixel coordinate system.

[0056] Preferably, the calculation process of calculating the conversion coefficients: scx and scy is:

[0057]

[0058] Where:

[0059] P1(m,0): point on the X-axis of the modeling coordinate system;

[0060] (u1,v1): pixel coordinates corresponding to P1(m,0);

[0061] P2(0,n): point on the Y axis of the modeling coordinate system;

[0062] (u2,v2): The pixel coordinate system coordinate corresponding to P2(0,n).

[0063] Preferably, S4 is specifically:

[0064] S41, converting the extracted component information into structured data in XML format and storing the structured data in XML format, wherein the structured data in XML format includes component instance information, location information and attribute description information;

[0065] S42, parsing the structured data in XML format to create a data model including geometric and attribute information of building components; and generating a three-dimensional building information model based on the created data model.

[0066] Preferably, in S41, the attribute information of the component information includes a unique identifier, a name, a direction, a secondary beam and associated attributes, and the associated attributes include an ID and a type; the geometric information includes a width, a height, a length and a pixel coordinate system, and the pixel coordinate system includes an X-axis coordinate value and a Y-axis coordinate value; and also includes a modeling coordinate system, and the modeling coordinate system includes an X-axis coordinate value, a Y-axis coordinate value, a width and a height.

[0067] A computer-readable storage medium includes instructions and data, which, when executed, enables a computer to execute the above-mentioned method of reconstructing a model of a 2D structural engineering drawing based on deep learning.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] The present invention proposes a system and method for reconstructing models from 2D structural engineering drawings based on deep learning, which effectively solves the technical problems of accuracy, economy, efficiency and applicability in reconstructing BIM models from 2D structural engineering drawings in the prior art, and also solves the problem of low efficiency in data collection, organization and utilization in data management throughout the life cycle of buildings. Specifically, the present invention improves the accuracy and automation of the reconstruction process. By adopting the YOLOv5 target detection algorithm to automatically identify the components and annotations in the drawings, and combining the PaddleOCR text recognition technology to extract the annotation content, the manual intervention is greatly reduced, and the accuracy and consistency of recognition are improved; the processing ability of complex structures is enhanced. By combining technologies such as target detection, text recognition and coordinate conversion, the method can better handle non-standardized or special-structured buildings, and improves applicability; it also improves data management and utilization efficiency; by storing the recognition results in a structured XML format, it facilitates subsequent data retrieval, analysis and sharing, and improves the management and utilization efficiency of building data throughout the life cycle; it also promotes the consistency and real-time nature of information. The automated recognition and storage process ensures the consistency of the data, and also facilitates timely updates, solving the problem of information lag in traditional methods; it also reduces the cost of the reconstruction process. The present method directly uses existing 2D structural engineering drawings, without the need for expensive on-site scanning equipment or a large amount of manual CAD conversion work, significantly reducing the economic cost of the reconstruction process, and the application of deep learning models makes the feature extraction and recognition process efficient, which can quickly process a large number of 2D structural engineering drawings and greatly shorten the reconstruction cycle. Through these technical means, this application not only improves the efficiency and quality of BIM model reconstruction, but also provides a better foundation for data management throughout the life cycle of buildings, which is conducive to the promotion and application of BIM technology in existing buildings. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A schematic diagram of a process of a method in a preferred embodiment of the present invention;

[0071] Figure 2 This is a result diagram of target detection and recognition in a preferred embodiment of the present invention;

[0072] Figure 3 A schematic diagram of the conversion between the pixel coordinate system and the drawing coordinate system in a preferred embodiment of the present invention;

[0073] Figure 4 It is a schematic diagram of a three-dimensional model generated in Revit in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. All other embodiments obtained by ordinary technicians in the field without making creative work based on the embodiments of the present invention shall fall within the scope of protection of the present invention.

[0075] like Figure 1-Figure 4 As shown:

[0076] First preferred embodiment:

[0077] A system for reconstructing a model of a 2D structural engineering drawing based on deep learning, comprising:

[0078] A user interface module is used to receive 2D structural engineering drawings input by users, display intermediate results in the process of processing the 2D structural engineering drawings, present the final generated 3D model, and perform model export operations;

[0079] The drawing processing module is used to analyze, extract and match the data in the 2D structural engineering drawings input by the user interface module to generate structured drawing information;

[0080] A data storage module is used to store the structured information generated by the drawing processing module in XML format, including component instance information, location information and attribute description information;

[0081] The three-dimensional model generation module is used to parse the XML data in the data storage module and create a three-dimensional model that complies with the building information model standard and contains geometry and attribute information based on the XML data; wherein,

[0082] The drawing processing module includes:

[0083] A drawing detection and recognition unit is used to perform object detection and recognition on components and annotations in 2D structural engineering drawings using a YOLOv5 target detection algorithm based on deep learning, and output an object list of detected components, wherein the object list includes object types and pixel coordinates of the objects;

[0084] The annotation extraction and matching unit is used to extract specific text content from the recognized annotation area based on the detection results of the drawing detection and recognition unit using the PaddleOCR text recognition algorithm, and match these annotations with the corresponding components to construct matching pairs consisting of components and annotations;

[0085] The coordinate system conversion unit is used to construct a modeling coordinate system and convert the pixel coordinates of the components detected by the drawing detection and recognition unit into drawing coordinates.

[0086] Second preferred embodiment:

[0087] A method for reconstructing a model of a 2D structural engineering drawing based on deep learning, comprising:

[0088] S1, using the drawing detection and recognition unit to detect and recognize the 2D structural engineering drawings uploaded by the user in the user interface module using an object recognition algorithm, and the detection and recognition targets are entity object classes and annotation classes;

[0089] In S1, the object recognition algorithm adopts the YOLOv5 target detection algorithm based on deep learning;

[0090] The entity object class is a component, and the component includes a beam and a column; the beam includes a horizontal beam, a vertical beam and an inclined beam;

[0091] The annotation class is the annotation of the component, including title, axis grid, dimension annotation and attribute annotation.

[0092] S2, extracting the annotation area from the data identified by the drawing detection and recognition unit by using a text recognition algorithm, and matching the annotation with the component to construct a matching pair consisting of the component and the annotation;

[0093] S21, cropping the image according to the detection area of ​​S1 to extract the valid annotation text area;

[0094] S22, using the PaddleOCR text recognition algorithm to perform text content recognition on the extracted annotation text area and extract the text content of the annotation;

[0095] S23, matching the attribute annotations and dimension annotations in the 2D structural engineering drawings with the corresponding beams and columns respectively, to obtain geometric information and semantic information of each beam and column; the geometric information is data related to the shape, size, position, angle and spatial relationship of the component, and the semantic information is descriptive data related to the purpose, characteristics or function of the component;

[0096] The attribute annotation includes centralized annotation and position annotation. The positions of the two annotations are located within a certain interval between the two endpoints of the object. Therefore, the position constraint condition of the attribute annotation is obtained:

[0097] a. Position constraints for vertical attribute annotations:

[0098]

[0099] b. Position constraints for horizontal attribute annotations:

[0100]

[0101] Where:

[0102] (BBox i) and BBox k : List of bounding boxes with attribute annotations in target detection results

[0103] {BBox1,BBox2,...,BBox n} in the element;

[0104] X PCS (BBox i ); In pixel coordinate system, BBox i and BBox k The X-axis coordinate of the center point;

[0105] Y PCS (BBox k ): BBox in pixel coordinate system i and BBox k The Y-axis coordinate of the center point;

[0106] D x : A list of distances between adjacent vertical grids;

[0107] D y : A list of distances between adjacent horizontal axis grids;

[0108] α: The position ratio between adjacent axis grids. The value of α varies depending on the attribute annotation type.

[0109] β: Maximum value of distance error.

[0110] The above formula uses the pixel coordinate system (PCS) to locate the element and determines whether they match by comparing the difference between the marked position and the expected position, that is, checking whether the marked position is within a certain expected range between two adjacent axis grids; if the mark is within this range, then the mark is likely to belong to the component at this position. In the formula, α and β are parameters used to adjust the expected position, and γ is the allowable error range. If the calculated difference is less than γ, then the mark is considered to belong to the component at the said position.

[0111] S3, constructing a modeling coordinate system through a coordinate system conversion unit, and converting the pixel coordinates of the object identified by the drawing detection and recognition unit into drawing coordinates in combination with dimension annotation;

[0112] Since the object identified by target detection is a bounding box, the specific position of the object cannot be determined, and the position of the bounding box is the coordinate in the pixel coordinate system, and subsequent three-dimensional building generation cannot be performed. Therefore, the pixel coordinate system needs to be converted into a modeling coordinate system. S3 is specifically:

[0113] S31, establish a modeling coordinate system with the intersection of the lowest horizontal axis grid and the leftmost vertical axis grid in the 2D structural engineering drawing as the origin;

[0114] S32, converting the pixel coordinate system into a modeling coordinate system, and calculating the specific position coordinates of the object in combination with the dimensioning;

[0115] The conversion process between the modeling coordinate system point P (x, y) and the pixel coordinate system P (u, v) is expressed as:

[0116]

[0117] Where:

[0118] Scale x : Conversion coefficient between pixel coordinate system and modeling coordinate system X axis;

[0119] Scale y : Conversion coefficient between pixel coordinate system and modeling coordinate system Y axis;

[0120] u0 and v0 represent the coordinates of the origin of the modeling coordinate system in the pixel coordinate system.

[0121] Calculate the conversion coefficient: The calculation process of scx and scy is:

[0122]

[0123] Where:

[0124] P1(m,0): point on the X-axis of the modeling coordinate system;

[0125] (u1,v1): pixel coordinates corresponding to P1(m,0);

[0126] P2(0,n): point on the Y axis of the modeling coordinate system;

[0127] (u2,v2): The pixel coordinate system coordinate corresponding to P2(0,n).

[0128] S4, storing the extracted information in XML format through the data storage module, and the 3D model generation module generates a 3D building information model based on the data stored in the data storage module;

[0129] S41, converting the extracted component information into structured data in XML format and storing the structured data in XML format, wherein the structured data in XML format includes component instance information, location information and attribute description information;

[0130] Specifically, the attribute information of the component information includes a unique identifier, a name, a direction, a secondary beam, and associated attributes, wherein the associated attributes include an ID and a type; the geometric information includes a width, a height, a length, and a pixel coordinate system, wherein the pixel coordinate system includes an X-axis coordinate value and a Y-axis coordinate value; and further includes a modeling coordinate system, wherein the modeling coordinate system includes an X-axis coordinate value, a Y-axis coordinate value, a width, and a height;

[0131] S42, parsing the structured data in XML format to create a data model including geometric and attribute information of building components; and generating a three-dimensional building information model based on the created data model.

[0132] Third preferred embodiment:

[0133] A computer-readable storage medium includes instructions and data, which, when executed, enables a computer to execute the above-mentioned method of reconstructing a model of a 2D structural engineering drawing based on deep learning.

[0134] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When the use is implemented in whole or in part in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL) or wireless (e.g., infrared, wireless, microwave, etc.) mode) to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk SolidState Disk (SSD)), etc.

[0135] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A system for reconstructing a model of a 2D structural engineering drawing based on deep learning, characterized in that: include: A user interface module is used to receive 2D structural engineering drawings input by users, display intermediate results in the process of processing the 2D structural engineering drawings, present the final generated 3D model, and perform model export operations; The drawing processing module is used to analyze, extract and match the data in the 2D structural engineering drawings input by the user interface module to generate structured drawing information; A data storage module is used to store the structured information generated by the drawing processing module in XML format, including component instance information, location information and attribute description information; The three-dimensional model generation module is used to parse the XML data in the data storage module and create a three-dimensional model that complies with the building information model standard and contains geometry and attribute information based on the XML data; wherein, The drawing processing module includes: A drawing detection and recognition unit is used to detect and recognize components and annotations in 2D structural engineering drawings using a YOLOv5 target detection algorithm based on deep learning, and output an object list of detected components, wherein the object list includes object types and pixel coordinates of the objects; The annotation extraction and matching unit is used to extract specific text content from the recognized annotation area based on the detection results of the drawing detection and recognition unit using the PaddleOCR text recognition algorithm, and match these annotations with the corresponding components to construct matching pairs consisting of components and annotations; The coordinate system conversion unit is used to construct a modeling coordinate system and convert the pixel coordinates of the components detected by the drawing detection and recognition unit into drawing coordinates.

2. A method for reconstructing a model of a 2D structural engineering drawing based on deep learning, characterized in that: include: S1, using the drawing detection and recognition unit to detect and recognize the 2D structural engineering drawings uploaded by the user in the user interface module using an object recognition algorithm, and the detection and recognition targets are entity object classes and annotation classes; S2, extracting the annotation area from the data identified by the drawing detection and recognition unit by using a text recognition algorithm through the annotation extraction and matching unit, and matching the annotation with the component to construct a matching pair consisting of the component and the annotation; S3, constructing a modeling coordinate system through a coordinate system conversion unit, and converting the pixel coordinates of the object identified by the drawing detection and recognition unit into drawing coordinates in combination with dimension annotation; S4, the extracted component information is stored in XML format through the data storage module, and the three-dimensional model generation module generates a three-dimensional building information model based on the data stored in the data storage module.

3. The method for reconstructing a model of a 2D structural engineering drawing based on deep learning according to claim 2, characterized in that: In S1, the object recognition algorithm adopts the YOLOv5 target detection algorithm based on deep learning.

4. The method for reconstructing a model of a 2D structural engineering drawing based on deep learning according to claim 2, characterized in that: In S1, the entity object class is a component, and the component includes a beam and a column; the beam includes a horizontal beam, a vertical beam, and an inclined beam.

5. The method for reconstructing a model of a 2D structural engineering drawing based on deep learning according to claim 2, characterized in that: In S1, the annotation class is the annotation of the component, including title, axis grid, dimension annotation and attribute annotation.

6. The method for reconstructing a model of a 2D structural engineering drawing based on deep learning according to claim 2, characterized in that: S2 is specifically: S21, cropping the image according to the detection area of ​​S1 to extract the valid annotation text area; S22, using the PaddleOCR text recognition algorithm to perform text content recognition on the extracted annotation text area and extract the text content of the annotation; S23, matching the attribute annotations and dimension annotations in the 2D structural engineering drawings with the corresponding beams and columns respectively, and obtaining the geometric information and semantic information of each beam and column.

7. The method for reconstructing a model from a 2D structural engineering drawing based on deep learning according to claim 6, characterized in that: The geometric information is data related to the shape, size, position, angle and spatial relationship of components.

8. The method for reconstructing a model from a 2D structural engineering drawing based on deep learning according to claim 6, characterized in that: The semantic information is descriptive data related to the purpose, characteristics or functions of the component.

9. The method for reconstructing a model from a 2D structural engineering drawing based on deep learning according to claim 2, characterized in that: The attribute annotation includes centralized annotation and position annotation. The positions of the two annotations are located within a certain interval between the two endpoints of the object. Therefore, the position constraint condition of the attribute annotation is obtained: a. Position constraints for vertical attribute annotations: b. Position constraints for horizontal attribute annotations: Where: (BBox i ) and BBox k : List of bounding boxes with attribute annotations in target detection results {BBox1,BBox2,...,BBox n } in the element; X PCS (BBox i ); In pixel coordinate system, BBox i and BBox k The X-axis coordinate of the center point; Y PCS (BBox k ): BBox in pixel coordinate system i and BBox k The Y-axis coordinate of the center point; D x : A list of distances between adjacent vertical grids; D y : A list of distances between adjacent horizontal axis grids; α: The position ratio between adjacent axis grids. The value of α varies depending on the attribute annotation type. β: Maximum value of distance error.

10. The method for reconstructing a model of a 2D structural engineering drawing based on deep learning according to claim 2, characterized in that: S3 is specifically: S31, establish a modeling coordinate system with the intersection of the lowest horizontal axis grid and the leftmost vertical axis grid in the 2D structural engineering drawing as the origin; S32, converting the pixel coordinate system into a modeling coordinate system, and calculating the specific position coordinates of the object in combination with the dimensioning.

11. The method for reconstructing a model of a 2D structural engineering drawing based on deep learning according to claim 10, characterized in that: The conversion process between the modeling coordinate system point P(x, y) and the pixel coordinate system P(u, v) is expressed as: Where: Scale x : Conversion coefficient between pixel coordinate system and modeling coordinate system X axis; Scale y : Conversion coefficient between pixel coordinate system and modeling coordinate system Y axis; u0 and v0 represent the coordinates of the origin of the modeling coordinate system in the pixel coordinate system.

12. The method for reconstructing a model of a 2D structural engineering drawing based on deep learning according to claim 11, characterized in that: Calculate the conversion coefficient: The calculation process of scx and scy is: Where: P1(m,0): point on the X-axis of the modeling coordinate system; (u1,v1): pixel coordinates corresponding to P1(m,0); P2(0,n): point on the Y axis of the modeling coordinate system; (u2,v2): The pixel coordinate system coordinate corresponding to P2(0,n).

13. The method for reconstructing a 2D structural engineering drawing model based on deep learning according to claim 2, characterized in that: S4 is specifically: S41, converting the extracted component information into structured data in XML format and storing the structured data in XML format, wherein the structured data in XML format includes component instance information, location information and attribute description information; S42, parsing the structured data in XML format to create a data model including geometric and attribute information of building components; and generating a three-dimensional building information model based on the created data model.

14. The method for reconstructing a model from a 2D structural engineering drawing based on deep learning according to claim 13, characterized in that: In S41, the attribute information of the component information includes a unique identifier, a name, a direction, a secondary beam and associated attributes, and the associated attributes include an ID and a type; the geometric information includes a width, a height, a length and a pixel coordinate system, and the pixel coordinate system includes an X-axis coordinate value and a Y-axis coordinate value; and also includes a modeling coordinate system, and the modeling coordinate system includes an X-axis coordinate value, a Y-axis coordinate value, a width and a height.

15. A computer-readable storage medium, characterized in that: The method comprises instructions and data, which, when executed, enable a computer to execute the method for reconstructing a 2D structural engineering drawing model based on deep learning as described in claims 2 to 14.

Images