Architectural drawing element intelligent identification and labeling method based on graphic feature extraction

Abstract

The invention relates to the technical field of drawing recognition, in particular to an architectural drawing element intelligent recognition and labeling method based on graphic feature extraction, comprising: acquiring an architectural drawing file to be processed, the architectural drawing file to be processed comprising a two-dimensional CAD drawing and a three-dimensional building information model drawing; format analysis is conducted on the to-be-processed building drawing file, graph geometric data and layer description information are extracted, the graph geometric data comprise point coordinates, line coordinates and plane coordinates forming a building graph, and the layer description information comprises name information and layer priority information of building drawing layers; and building a building drawing feature database according to the graphic geometric data and the layer description information. According to the method, through deep fusion analysis of the graphic geometric data and the layer description information, and through the structured drawing feature database and the feature fusion module, various elements in the architectural drawing can be accurately identified and labeled, and the accuracy of drawing labeling can be effectively improved through the accurate identification.

Description

Technical Field

[0001] This invention relates to the field of drawing recognition technology, specifically to a method for intelligent recognition and annotation of architectural drawing elements based on graphic feature extraction. Background Technology

[0002] With the acceleration of urbanization, the complexity and scale of architectural designs are increasing year by year. Traditional manual drawing recognition methods can no longer adapt to this change. For example, drawings for complex projects such as high-rise buildings and large commercial complexes often involve hundreds or thousands of drawing elements. Manual recognition and annotation are not only inefficient but also prone to errors. This situation urgently requires the use of automation technology to improve work efficiency and accuracy.

[0003] Currently, traditional methods rely on manual identification, annotation, and modification of drawing elements. This is not only time-consuming but also prone to errors, especially in complex architectural drawings. Manual annotation often consumes a lot of time and effort, and due to human factors, traditional methods are prone to identification errors or omissions. In particular, in drawings with complex details, inaccurate annotations or omissions may occur. Such errors may lead to increased costs and time delays in the subsequent project implementation phase.

[0004] Furthermore, traditional drawing analysis relies on manual item-by-item inspection, which is inefficient and cannot meet the rapidly evolving needs of architectural design. Every time drawings are updated or modified, manual inspection and annotation are required again, leading to repetitive work and wasted resources. In addition, traditional methods usually rely on fixed standard templates or manual coding, which are difficult to adapt to different types of architectural drawings or different file formats. When the drawing format changes or needs to be adjusted according to project requirements, traditional methods cannot respond quickly and often require a lot of manual intervention and adjustment. Moreover, the ability to identify complex graphic elements is limited, especially for drawings with complex details and varied structures. Manual judgment and processing often have to rely on experience, which not only increases the workload but also increases the risk of errors. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent identification and annotation of architectural drawing elements based on graphic feature extraction, comprising: Obtain architectural drawing files to be processed, including two-dimensional CAD drawings and three-dimensional building information model drawings; The architectural drawing file to be processed is parsed to extract graphic geometric data and layer description information. The graphic geometric data includes point coordinates, line coordinates and surface coordinates that constitute the architectural graphics. The layer description information includes the name information and layer priority information of the architectural drawing layers. Based on the graphic geometric data and layer description information, a building drawing feature database is constructed, wherein the building drawing feature database includes graphic topology association data, layer element mapping data, historical identification and annotation records, and mesh data corresponding to closed graphics; Based on the graphic topology association data and mesh data in the architectural drawing feature database, closed graphics in the architectural drawings are identified to obtain an initial set of closed graphics. The mesh is closed structural attribute data generated and stored in the front-end database during parsing and rendering, and is not displayed in the initial state. The system retrieves layer element mapping data from the architectural drawing feature database and, based on the graphic geometry data of the initial closed graphic set, determines the element type of each closed graphic in the initial closed graphic set to generate graphic element annotation data. The graphic element annotation data is stored in a cache format and includes element category, point set data, direction information, and name information. The graphic element annotation data is embedded into the architectural drawing file to be processed to generate an annotated architectural drawing file. The ray-click function is used to determine whether the ray penetration position is on the target mesh. After clicking, the selected mesh is displayed, and the annotated architectural drawing file is displayed on the display interface.

[0006] Preferably, the architectural drawing file to be processed is obtained, including: Call the original drawing files in the architectural drawing database and distinguish whether the original drawing file format is 2D CAD format or 3D building information model format; The process involves separating graphic and layer data from 2D CAD files to obtain original 2D graphic data and original 2D layer data; and separating spatial graphic data and hierarchical layer data from 3D building information model files to obtain original 3D graphic data and original 3D layer data. The original data of two-dimensional graphics and three-dimensional graphics are uniformly divided into several graphics data blocks, and the original data of two-dimensional layers and three-dimensional layers are uniformly divided into several layer data blocks. Obtain the planar or spatial coordinate position of each graphic data block in the original drawing to obtain the graphic block position information; obtain the layer position of each layer data block in the original drawing to obtain the layer block position information.

[0007] Preferably, the architectural drawing file to be processed is parsed to extract graphic geometric data and layer description information, including: The architectural drawing file to be processed is input into a pre-trained drawing parsing model to obtain graphic geometric data and layer description information. The drawing parsing model includes a graphic encoding module, a layer encoding module, and a feature fusion module. The process involves inputting the architectural drawing file to be processed into a pre-trained drawing analysis model to obtain graphic geometric data and layer description information, including: The graphic encoding module extracts features from the graphic structural information in the architectural drawing file to be processed, and obtains graphic feature vectors. The layer encoding module performs feature processing on the initial layer information in the architectural drawing file to be processed, and obtains the layer attribute vector. The graphic feature vector and the layer attribute vector are input into the feature fusion module, and the feature fusion module performs weight allocation and information integration to obtain the fused feature matrix; The geometric data representing the graphic structure and the layer description information representing the layer characteristics are extracted from the fused feature matrix.

[0008] Preferably, the graphic encoding module extracts features from the graphic structural information in the architectural drawing file to be processed, obtaining a graphic feature vector, including: The graphic data block and the corresponding graphic block position information are input into the graphic encoding module, wherein the graphic encoding module includes a linear transformation layer and a position encoding layer; Each graphic data block is input into the linear transformation layer, and the graphic data blocks are dimension-mapped through the linear transformation layer to output the graphic embedding vector of each graphic data block. The position information of each graphic block is input into the position encoding layer. The position encoding layer performs feature transformation on the coordinate position or spatial position and outputs the position embedding vector of each graphic data block. The graphic embedding vector and the position embedding vector corresponding to each graphic data block are added element-wise to obtain the local graphic features of each graphic data block. By stitching together the local graphic features of all graphic data blocks in the order of their positions in the original drawing, a global graphic feature vector is obtained.

[0009] Preferably, the layer encoding module performs feature processing on the initial layer information in the architectural drawing file to be processed to obtain a layer attribute vector, including: Extract initial layer information from the layer data block, wherein the initial layer information includes layer name text and layer priority identifier; Input the layer name text into the text embedding layer of the layer feature encoding module, perform semantic feature transformation on the name text through the text embedding layer, and output the layer name embedding vector; The layer priority identifier is matched with a preset priority-value mapping table, the priority identifier is converted into the corresponding priority value, the priority value is transformed through a value embedding layer, and the layer priority embedding vector is output. The layer name embedding vector corresponding to each layer data block is concatenated with the layer priority embedding vector to obtain the local layer features of each layer data block; By concatenating the local layer features of all layer data blocks according to the hierarchical order in the original drawing, a global layer attribute vector is obtained.

[0010] Preferably, the graphic feature vector and the layer attribute vector are input into the feature fusion module, and the feature fusion module performs weight allocation and information integration to obtain a fused feature matrix, including: The graphic feature vector is input into the first attention submodule of the feature fusion module to calculate the importance weight of each dimension of the graphic feature vector to the geometric extraction of the graphic, and thus obtain the graphic feature weight vector. The layer attribute vector is input into the second attention submodule of the feature fusion module to calculate the importance weight of each dimension feature in the layer attribute vector to the layer attribute extraction, and the layer feature weight vector is obtained. The graphic feature vector and the graphic feature weight vector are multiplied element-wise to obtain the weighted graphic feature vector; the layer attribute vector and the layer feature weight vector are multiplied element-wise to obtain the weighted layer attribute vector. The weighted graphic feature vector and the weighted layer attribute vector are stacked column-wise to form a two-dimensional fusion feature matrix.

[0011] Preferably, based on the graphic geometric data and layer description information, a building drawing feature database is constructed, including: Perform topological relationship calculations on graphic geometric data to determine the connection relationships between points, lines, and surfaces, including nesting relationships and adjacency distribution relationships, and generate graphic topological relationship data; Perform element mapping analysis on layer description information, establish the correspondence between layer identifiers and building element types, match the relationship between layer priority and element recognition weight, and generate layer element mapping data; The graphic topology association data, layer element mapping data, pre-imported historical identification and annotation records, and closed graphic mesh data generated by parsing and rendering are structured and stored to form a building drawing feature database.

[0012] Preferably, based on the graphic topology association data and mesh data in the architectural drawing feature database, closed graphics in the architectural drawings are identified to obtain an initial set of closed graphics. The mesh is closed structural attribute data generated during parsing and rendering and stored in the front-end database; it is not initially displayed and includes: The topological association data of the graphic and the pre-stored mesh data are retrieved from the architectural drawing feature database to determine the boundary contour dataset of the graphic to be identified. The ray method is used to make a preliminary judgment on the closure of the boundary contour dataset. Candidate closed shapes that meet the closure conditions are filtered by judging whether the ray penetration position is on the corresponding mesh. A contour tracking algorithm is used to verify the boundary integrity of candidate closed shapes, and shapes with incomplete boundaries are eliminated to obtain an initial set of closed shapes. The initial set of closed shapes is used for element type determination.

[0013] Preferably, for each closed shape in the initial set of closed shapes, the element type is determined, and graphic element annotation data is generated, including: Retrieve layer element mapping data from the architectural drawing feature database to determine the candidate range of element types corresponding to the layer to which each initial closed graphic belongs; Extract the point set data and basic parameters calculated based on the point set for each initial closed shape, determine the element attribute parameters, and determine the element position parameters based on the coordinate information in the point set data, generating graphic element annotation data that includes element category, point set data, direction, and name information.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention can automatically identify elements in architectural drawings, reduce the complexity of manual operation and human error, and improve work efficiency. Through in-depth fusion analysis of graphic geometric data and layer description information, and through a structured drawing feature database and feature fusion module, it can accurately identify and annotate various elements in architectural drawings and accurately determine their type, attributes and location. This accurate identification can effectively improve the accuracy of drawing annotation. This invention can be easily expanded and upgraded through independent modules such as a graphic encoding module and a layer encoding module, so that it can adapt to different types of architectural drawings and different formats of drawings. It can also be adjusted according to specific project requirements through graphic features and layer attribute features, and make up for the deficiencies in graphic element recognition through historical recognition and annotation records. This invention, through detailed layer analysis and feature fusion, can more accurately describe the attributes, priorities, and hierarchical structure of elements in drawings, which helps to make more detailed construction and modifications during the design phase. By automating annotation and reducing the need for manual review, it can greatly reduce the time and cost in the project implementation process, while improving productivity. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1, please refer to Figure 1 This invention provides a technical solution: an intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction, comprising: S1. Obtain the architectural drawing files to be processed, which include 2D CAD drawings and 3D building information model drawings. S2. Parse the format of the architectural drawing file to be processed, and extract the graphic geometric data and layer description information. The graphic geometric data includes the point coordinates, line coordinates and surface coordinates that make up the architectural graphics, and the layer description information includes the name information and layer priority information of the architectural drawing layers. S3. Based on the geometric data and layer description information, build a feature database for architectural drawings. The feature database for architectural drawings includes graphic topology association data, layer element mapping data, historical identification and annotation records, and mesh data corresponding to closed graphics. S4. Based on the graphic topology association data and mesh data in the architectural drawing feature database, identify the closed graphics in the architectural drawings to obtain an initial set of closed graphics. The mesh is the closed structure attribute data generated and stored in the front-end database during parsing and rendering, and is not displayed in the initial state. S5. Retrieve layer element mapping data from the architectural drawing feature database, and determine the element type of each closed shape in the initial closed shape set based on the graphic geometry data of the initial closed shape set, and generate graphic element annotation data. The graphic element annotation data is stored in a cache format, including element category, point set data, direction information and name information. S6. Embed the graphic element annotation data into the architectural drawing file to be processed to generate an annotated architectural drawing file. Use ray clicking to determine whether the ray penetration position is on the target mesh. After clicking, the selected mesh is displayed, and the annotated architectural drawing file is displayed on the display interface.

[0018] It should be noted that the architectural drawing files to be processed include 2D CAD drawings and 3D Building Information Model (BIM) drawings. CAD drawings are generally traditional floor plans, while 3D BIM drawings are more complex 3D models, including information such as the building's structure, piping, and electrical systems. Example: Suppose we have a CAD floor plan of a building and a BIM 3D model of the same building; we want to extract architectural elements from these two and annotate them. Mesh data is mainly used for the following two aspects: Closed structure attribute data: This data includes various geometric information of building drawings, such as the boundaries and shapes of walls, doors, windows, floors, etc. It is usually generated by 3D modeling software and stored in the mesh data storage. Front-end database storage: In the front-end database, the mesh data is stored in a digital format so that it can be used in later visualization displays, such as displaying the appearance of the building through rendering technology. The system parses architectural drawing files into a machine-readable format, extracting geometric data and layer description information. Graphical geometric data includes the coordinates of points, lines, and surfaces that make up the graphics in the architectural drawing; for example, the start and end coordinates of walls, the positions of windows, and the dimensions of doors. Each architectural drawing typically includes multiple layers, each representing different architectural elements (such as walls, wiring, windows, and doors). Each layer has a different priority, determining its display order. Example: For a CAD architectural floor plan, the system extracts the following data: Point coordinates: wall start point (10,20) and end point (10,60); Line coordinates: wall edge line from (10,20) to (10,60); Layer information: wall layer priority 1, wiring layer priority 2. The architectural drawing feature database stores various information extracted from architectural drawings to facilitate subsequent querying and processing. This includes: graphic topology association data representing the spatial relationships between graphic elements; for example, the relative positions of walls, doors, and windows; records of mapping relationships between different layer elements and actual building elements, such as the wall layer corresponding to wall elements in the building; and records of previously identified and annotated results for subsequent identification and improvement. Example: For a building's drawings, the database might store the following information: Graphic topology: the relationship between walls and doors, and the relationship between the door opening direction and the wall; Layer mapping: the electrical layer corresponding to electrical wiring and conduits in the building, and the wall layer corresponding to the exterior walls of the building; Historical annotations: previously annotated window positions, door sizes, etc. Based on the topological association data of the graphics, the system identifies closed shapes (such as walls, room outlines, etc.) in architectural drawings. For closed shapes, it identifies their shape, size, and relative position. For example, suppose there is a rectangular area in the drawing that represents a room. By identifying the closed shape of this area, the system will extract the boundary information of the room and identify it as a room element. Based on layer information and geometric data, the system determines which element the initially identified closed shape belongs to (e.g., wall, window, door, room, etc.). Simultaneously, the system generates corresponding annotation data for each element, including its category, attribute parameters (e.g., size, material), and location coordinates. Example: After identifying a closed rectangle, the system determines it is a room and generates annotation data: Element category: Room; Identification attributes: Area, length, width; Location coordinates: Coordinates of the upper left corner of the room (50, 100). Embedding Annotation Data: The generated graphic element annotation data is embedded into the original architectural drawing file; in this way, the original architectural drawing file becomes a drawing with annotation information, displaying the attributes of each element; Displaying Annotated Architectural Drawings: The annotated architectural drawings are displayed on the interface, and users can view detailed information about each element in the drawing; Example: On the architectural drawing, there may originally only be the outline of the room and the lines of the walls. After using this method, the annotations of the room (such as room 1, area 20 square meters) and the annotations of the doors and windows (such as window 2, size 1.5x2 meters) will be displayed on the drawing.

[0019] In an optional embodiment, obtaining the architectural drawing file to be processed includes: Call the original drawing files in the architectural drawing database and distinguish whether the original drawing file format is 2D CAD format or 3D building information model format; The process involves separating graphic and layer data from 2D CAD files to obtain original 2D graphic data and original 2D layer data; and separating spatial graphic data and hierarchical layer data from 3D building information model files to obtain original 3D graphic data and original 3D layer data. The original data of two-dimensional graphics and three-dimensional graphics are uniformly divided into several graphics data blocks, and the original data of two-dimensional layers and three-dimensional layers are uniformly divided into several layer data blocks. Obtain the planar or spatial coordinate position of each graphic data block in the original drawing to obtain the graphic block position information; obtain the layer position of each layer data block in the original drawing to obtain the layer block position information.

[0020] It should be noted that the original drawing files are retrieved from the architectural drawing database, and the file formats must be distinguished. There are two common formats for architectural drawings: usually floor plans, which include two-dimensional planar data such as walls, doors, windows, furniture, and facilities; and three-dimensional building information model (BIM) formats, which include more detailed three-dimensional spatial data, representing the complete structure of the building, including walls, floors, ceilings, pipes, electrical systems, etc. For example, there are two files: one is a two-dimensional CAD floor plan of the building (such as plan views of doors, windows, and walls), and the other is a three-dimensional BIM file (including three-dimensional structural information such as columns, beams, pipes, ceilings, etc.). The system separates graphic data (such as elements like walls, doors, and windows) from layer data (such as the classification and priority of different elements) in 2D CAD drawings. Graphic data represents the actual geometric data in architectural drawings, while layer data describes the grouping and order of various elements in the drawing. 2D graphic raw data includes the geometric data of points, lines, and surfaces in the drawing, representing the shape and position of architectural elements. 2D layer raw data describes the layer information of the drawing, such as which elements belong to the wall layer and which belong to the electrical layer. Example: In a 2D CAD drawing, walls may be located on the wall layer, while electrical installations may be located on the electrical layer. The system separates graphic data (such as the coordinates of walls and the boundary lines of windows) from layer data (such as layer names like wall and electrical) for subsequent processing. 3D graphic raw data describes the spatial data of a building, such as the 3D coordinates of walls, the spatial location of rooms, and the geometric information of elements such as three-dimensional doors and windows. 3D layer raw data is similar to the layer data in 2D CAD, but in BIM, these layers may also involve the spatial hierarchy of the building, such as the hierarchical information of floors, areas, pipes, structures, etc. For example, a BIM file may have a layer that includes information about two layers of walls, and the 3D coordinates of the walls describe the actual location, shape, and thickness of the walls. Based on the geometry in the drawings, the graphic data is divided into several blocks, each block including a set of related geometric elements; for example, a graphic data block may include the coordinate data of the walls, doors and windows of a room, or the three-dimensional coordinate data of a section of pipe; for example, for a room, the graphic data block may include the geometric data of the four walls, doors and windows of the room; for a piping system, the graphic data block may include the three-dimensional coordinate information of all pipes and connection points. The elements of each layer are divided according to their attributes and functions, and each layer includes a description of a type of architectural element. For CAD drawings and BIM drawings, layer data blocks may have different content and structures, but their function is the same—to represent the grouping of different types of elements in the drawing. For example, in a 2D CAD drawing, there may be a wall layer data block, which includes the graphic data of all walls; in a BIM drawing, an electrical layer data block may include 3D data related to the electrical system. For each graphic data block, the system needs to determine its position on the drawing. For two-dimensional drawings, the position is usually represented by planar coordinates (such as X, Y coordinates); for three-dimensional drawings, the position is represented by spatial coordinates (such as X, Y, Z coordinates). Example: Suppose there is a data block of a room. The system can determine the position of the room on the floor plan by extracting the coordinates of the four walls of the room (for example, the coordinates of the top left corner are (50, 100)). For layer data blocks, the system identifies the hierarchical position of each layer and determines their display order or hierarchical relationship in the overall drawing; this helps to correctly display the priority of different layers when displaying them; for example, in a BIM drawing, there may be ground layer and roof layer layers, which have different hierarchical positions; the ground layer layer may have a higher priority, while the roof layer layer has a lower priority.

[0021] In an optional embodiment, the architectural drawing file to be processed is parsed to extract graphic geometric data and layer description information, including: The architectural drawing file to be processed is input into a pre-trained drawing parsing model to obtain graphic geometric data and layer description information. The drawing parsing model includes a graphic encoding module, a layer encoding module, and a feature fusion module. The process involves inputting the architectural drawing file to be processed into a pre-trained drawing analysis model to obtain graphic geometric data and layer description information, including: The graphic encoding module extracts features from the graphic structural information in the architectural drawing file to be processed, and obtains graphic feature vectors. The layer encoding module performs feature processing on the initial layer information in the architectural drawing file to be processed, and obtains the layer attribute vector. The graphic feature vector and the layer attribute vector are input into the feature fusion module. The feature fusion module performs weight allocation and information integration to obtain the fused feature matrix. Extract graphic geometric data representing the graphic structure and layer description information representing the layer characteristics from the fused feature matrix.

[0022] It's important to note the graphic encoding module: This module is responsible for analyzing the geometric structure information in the drawings. It extracts graphic features from the drawings using techniques such as Convolutional Neural Networks (CNNs), including geometric information such as lines, angles, shapes, and sizes, and converts these features into digital vector representations. After feature extraction, the graphic is converted into a digital vector that includes various features of the graphic structure (such as position, shape, size, and orientation). These feature vectors effectively represent specific elements in the drawing. For example, if there is a rectangular wall in the drawing, the graphic encoding module will extract the rectangle's coordinates, side length, angles, etc., and represent it as a feature vector. For instance, this vector might include the rectangle's center coordinates, aspect ratio, and angles. The layer encoding module focuses on layer information in drawings (such as wall layers, electrical layers, pipe layers, etc.). Elements in drawings are usually organized into different layers according to function and type. The layer encoding module extracts the feature information of these layers. The layer encoding module generates a layer attribute vector, including information about the layer, such as the layer name, level, priority, and type (such as wall, pipe, electrical, etc.). This information will help subsequent processing distinguish the characteristics of different layers. Example: Suppose there is a wall layer and an electrical layer. The layer encoding module will extract the attribute information of these layers. For example, the wall layer may include the wall thickness, material, color, etc., while the electrical layer may include the type, direction, and wiring location of the wires, etc. The function of the feature fusion module is to integrate feature vectors from the graphic encoding module and the layer encoding module to form a more comprehensive feature matrix. This module obtains the final feature matrix by weighting and fusing different features, reflecting the geometric information of the graphic and the attribute information of the layers. The feature fusion module may assign different weights to graphic features and layer attribute features, depending on their importance in the architectural drawing. Through weighted fusion, the final fused feature matrix can more comprehensively represent the content of the drawing. Example: Suppose a drawing contains a wall (represented by a graphic feature vector) and an electrical facility (represented by a layer feature vector). The feature fusion module combines this information and may assign different weights based on the relationship between the wall and the electrical facility, thus obtaining a more comprehensive fused feature vector. For example, the electrical facility may have a greater weight on the electrical layer and a smaller weight on the wall layer. Geometric data of the graphics is extracted from the fusion feature matrix; this data describes the spatial coordinates, shape, size, etc. of the building elements; descriptive information of the layers is extracted from the fusion feature matrix, which is usually related to the layer's attributes, functions, priorities, types, etc.; Example: After feature fusion, the system can extract the geometric data of a room from the fusion matrix, such as the room's size, location, shape, etc.; at the same time, it can also extract the layer information of the room, such as whether the room belongs to the wall layer or the interior decoration layer.

[0023] In an optional embodiment, a graphic encoding module extracts features from the graphic structural information in the architectural drawing file to be processed, obtaining a graphic feature vector, including: The graphic data block and the corresponding graphic block position information are input into the graphic encoding module, which includes a linear transformation layer and a position encoding layer. Each graphic data block is input into a linear transformation layer, which performs dimension mapping on the graphic data blocks and outputs the graphic embedding vector of each graphic data block. The position information of each graphic block is input into the position encoding layer. The position encoding layer performs feature transformation on the coordinate position or spatial position and outputs the position embedding vector of each graphic data block. The graphic embedding vector and the position embedding vector corresponding to each graphic data block are added element-wise to obtain the local graphic features of each graphic data block. By stitching together the local graphic features of all graphic data blocks in the order of their positions in the original drawing, a global graphic feature vector is obtained.

[0024] It's important to note that each element in an architectural drawing (such as walls, windows, and doors) can be considered a graphic data block. These data blocks may include different types of information, such as the size, shape, and boundaries of the graphic. Each graphic data block also has corresponding location information, such as coordinates (X and Y coordinates in a 2D drawing). This location information helps determine the spatial distribution of the graphic data blocks in the drawing. For example, suppose there is an architectural floor plan that includes a rectangular wall (graphic data block), and this wall is located at a certain position on the drawing (location coordinates). The size and coordinates of this rectangular wall are a combination of the graphic data block and its location information. The graphic data block is first mapped to dimensions through a linear transformation layer. This means mapping the graphic's attributes (such as size, shape, etc.) to a unified spatial representation (i.e., embedding space), which is usually a high-dimensional space to facilitate model processing. After the linear transformation, the graphic data block is represented as a vector that includes the graphic's features. These features can include information such as shape, size, material, and color. Example: Suppose the wall on the drawing is a rectangle. The linear transformation layer converts its size and shape into a high-dimensional vector, such as [5,3,0.7], representing the wall's length, width, and height. The role of the positional encoding layer is to convert the positional information (such as coordinates) of graphic blocks into a corresponding vector. This vector helps the model understand the spatial relationship of the element. Through the positional encoding layer, the model converts coordinate information (such as X and Y coordinates) into a vector representation, enabling the model to better understand the spatial relationship between graphic elements. Example: Suppose the position of the wall is on the drawing (X=10, Y=5). The positional encoding layer converts this coordinate information into a vector, such as [10, 5], representing the specific position of the wall on the drawing. Adding the graphic embedding vector to the position embedding vector yields the local graphic features of each graphic data block. This operation combines the geometric and positional features of the graphic to create a more comprehensive graphic description. The resulting vector includes the shape, size, and spatial position of the graphic, forming the comprehensive features of the graphic data block. Example: Suppose the graphic embedding vector is [5,3,0.7] and the position embedding vector is [10,5]. The resulting local graphic feature after adding them is [15,8,0.7], which includes the wall's dimensions and its position on the drawing. By stitching together the local graphic features of all graphic data blocks in the drawing according to their position in the drawing, a complete global graphic feature vector is obtained. This vector describes the geometric information and spatial layout of the entire architectural drawing. For example, if there are multiple graphic data blocks (such as walls, windows, doors, etc.) in the drawing, their local graphic features are as follows: [15,8,0.7] (wall), [8,4,0.4] (window), [2,3,0.5] (door). After stitching these features together in order, the final global graphic feature vector may be: [15,8,0.7,8,4,0.4,2,3,0.5].

[0025] In an optional embodiment, the layer encoding module performs feature processing on the initial layer information in the architectural drawing file to be processed to obtain a layer attribute vector, including: Extract initial layer information from the layer data block, which includes the layer name text and the layer priority identifier; Input the layer name text into the text embedding layer of the layer feature encoding module, perform semantic feature transformation on the name text through the text embedding layer, and output the layer name embedding vector; The layer priority identifier is matched with a preset priority-value mapping table, the priority identifier is converted into the corresponding priority value, the priority value is transformed through a value embedding layer, and the layer priority embedding vector is output. The layer name embedding vector corresponding to each layer data block is concatenated with the layer priority embedding vector to obtain the local layer features of each layer data block; By concatenating the local layer features of all layer data blocks according to the hierarchical order in the original drawing, a global layer attribute vector is obtained.

[0026] It should be noted that each layer (such as walls, electrical wiring, etc.) has a name, and the model needs to understand the meaning of these names; each layer also has a priority indicator, which indicates the importance of the layer in the design or drawing (for example, walls may have a higher priority, while electrical wiring has a lower priority); Example: Suppose there is a layer named wall on the drawing with a priority of 1, while a layer named electrical wiring has a priority of 3. The names of layers (such as walls, electrical wiring) will undergo semantic transformation through a text embedding layer. This process uses natural language processing techniques (such as word vectors, BERT, etc.) to convert the name text into corresponding vectors, capturing the semantic information of the name. Through text embedding, the layer name is converted into a high-dimensional vector, representing the semantic features of the layer. For example, the layer name "wall" is converted into a vector through the embedding layer, which might be: [0.45, 0.23, 0.89], representing the semantic information of the wall, while "electrical wiring" is converted into another vector, such as: [0.11, 0.57, 0.42]. Each layer's priority identifier (such as 1, 2, 3, etc.) is converted into a corresponding numerical value through a mapping table. The priority value helps the model understand the importance of the layer in the overall structure. The priority value is input into the embedding layer to obtain a corresponding vector representation, capturing the priority features. For example, the priority of the wall layer is 1. After priority mapping, it may be converted into the numerical value 1, and then a vector is obtained through the embedding layer, such as [0.7, 0.1]. The priority of the electrical wiring layer is 3, and the vector might be [0.2, 0.4]. Local layer features are used to concatenate the embedding vector of the layer name with the embedding vector of the priority to form a comprehensive feature for each layer. For example, for the wall layer, the name embedding vector is [0.45, 0.23, 0.89], and the priority embedding vector is [0.7, 0.1]. Concatenating them yields: [0.45, 0.23, 0.89, 0.7, 0.1]. For the electrical wiring layer, the name embedding vector is [0.11, 0.57, 0.42], and the priority embedding vector is [0.2, 0.4]. Concatenating them yields: [0.11, 0.57, 0.42, 0.2, 0.4]. By concatenating the local layer features of all layers according to the hierarchical order of the drawing, a global layer attribute vector is obtained. This vector represents the layer features of the entire architectural drawing, including the name, priority, and relative position of each layer. For example, assuming there are multiple layers on the drawing, including walls ([0.45,0.23,0.89,0.7,0.1]) and electrical wiring ([0.11,0.57,0.42,0.2,0.4]), the final concatenated global layer attribute vector might be: [0.45,0.23,0.89,0.7,0.1,0.11,0.57,0.42,0.2,0.4]. This global vector can be used as a mathematical representation of the layers in the architectural drawing for further analysis, automated design, or intelligent recognition.

[0027] In an optional embodiment, the graphic feature vector and the layer attribute vector are input into the feature fusion module. The feature fusion module performs weight allocation and information integration to obtain a fused feature matrix, including: The graphic feature vector is input into the first attention submodule of the feature fusion module to calculate the importance weight of each dimension of the graphic feature vector to the geometric extraction of the graphic, and thus obtain the graphic feature weight vector. The layer attribute vector is input into the second attention submodule of the feature fusion module to calculate the importance weight of each dimension feature in the layer attribute vector to the layer attribute extraction, and the layer feature weight vector is obtained. The graphic feature vector and the graphic feature weight vector are multiplied element-wise to obtain the weighted graphic feature vector; the layer attribute vector and the layer feature weight vector are multiplied element-wise to obtain the weighted layer attribute vector. The weighted graphic feature vector and the weighted layer attribute vector are stacked column-wise to form a two-dimensional fusion feature matrix.

[0028] It should be noted that, assuming there is a graphic feature vector, which represents the features of a graphic, including multi-dimensional information such as position, size, and shape, the graphic feature vector will be input into the first attention submodule. The task of this module is to calculate the importance weights of each dimension of the graphic feature vector. Through the attention mechanism, the model can focus on which features of the graphic are more important to the final task. For example, assuming the graphic feature vector is [0.5, 0.3, 0.8, 0.4], the first attention submodule calculates the importance weights of each dimension, for example: [0.6, 0.2, 0.9, 0.3]. The layer attribute vector includes different attributes of the layer (such as layer name, priority, position, etc.). Through the layer attribute vector, the model can understand the characteristics of the layer and its role in the overall structure. The layer attribute vector will be input into the second attention submodule to calculate the importance weight of each dimension feature for layer attribute extraction. Example: Assuming the layer attribute vector is [0.7, 0.4, 0.9], the second attention submodule calculates the weight of each dimension, for example: [0.8, 0.5, 1.0]. The graphic feature vector is multiplied element-wise by the graphic feature weight vector (Hadamard product), with each feature value weighted according to its corresponding weight. The weighted graphic feature vector better reflects which features are more important for graphic recognition or extraction. Example: Graphic feature vector: [0.5, 0.3, 0.8, 0.4], graphic feature weight vector: [0.6, 0.2, 0.9, 0.3], weighted graphic feature vector: [0.5, 0.6, 0.3, 0.2, 0.8, 0.9, ...]. [0.4 * 0.3] = [0.3, 0.06, 0.72, 0.12], multiplies the layer attribute vector element-wise with the layer feature weight vector; the weighted layer attribute vector highlights more critical information in the layer attributes; Example: Layer attribute vector: [0.7, 0.4, 0.9], layer feature weight vector: [0.8, 0.5, 1.0], the weighted layer attribute vector is: [0.7 * 0.8, 0.4 * 0.5, 0.9 * 1.0] = [0.56, 0.2, 0.9]; The weighted graphic feature vector and the weighted layer attribute vector are stacked column-wise to form a two-dimensional fusion feature matrix. This matrix integrates information from graphic features and layer attributes and can be used for subsequent feature extraction or decision-making processes.

[0029] In an optional embodiment, a building drawing feature database is constructed based on graphic geometric data and layer description information, including: Perform topological relationship calculations on graphic geometric data to determine the connection relationships between points, lines, and surfaces, including nesting relationships and adjacency distribution relationships, and generate graphic topological relationship data; Perform element mapping analysis on layer description information, establish the correspondence between layer identifiers and building element types, match the relationship between layer priority and element recognition weight, and generate layer element mapping data; The topological association data of the graphics, the mapping data of layer elements, the pre-imported historical identification and annotation records, and the closed graphic mesh data generated by parsing and rendering are stored in a structured manner to form a feature database of architectural drawings.

[0030] It should be noted that graphic features in architectural drawings, such as points, lines, and surfaces, represent the geometric structure of a building; for example, a point represents a location, a line represents a boundary, and a surface represents a region. Topological relationships refer to the connections and relative relationships between different geometric elements. In architectural drawings, the following types of topological relationships may exist: Connection relationships refer to which points, lines, and surfaces are directly connected; for example, the relationship between a wall (line) and a door (surface) in a room, or two intersecting lines; Nesting relationships refer to whether one element is contained within another element; for example, a small room (surface) may be nested within another element. Within a large room (face); adjacency distribution relationships refer to which elements are adjacent but not directly connected; for example, two adjacent walls, or the relationship between one floor and another floor; Example: Suppose there is an architectural drawing that includes a room (face) and a door (face), with the door connected to a wall (line); after topological relationship calculation, we get: Connection relationship: the door and the wall are connected; Nesting relationship: the room includes the door; Adjacency distribution relationship: the door and other walls of the room are adjacent; through these topological relationships, we can generate topological association data of the graph, recording the connection, nesting, and adjacency relationships between each element; Architectural drawings are typically divided into multiple layers, each containing different architectural elements (such as walls, windows, and doors). Each layer includes descriptive information, such as the layer name and priority. Elements within a layer need to be matched with actual architectural elements. For example, there's the correspondence between layer identifiers and architectural element types: one layer might represent wall elements, while another might represent door and window elements. There's also the matching relationship between layer priority and element recognition weight: some layers may have higher priorities, meaning elements in these layers have higher weight during the recognition process. Example: Suppose there are two layers: Layer 1: identified as walls, priority 1, including all wall geometry; Layer 2: identified as doors and windows, priority 2, including all door and window geometry. Through element mapping analysis, the system will identify the elements on these layers and associate them with architectural element types (such as walls, doors, and windows). Based on the above analysis, the system generates mapping data between layers and building elements, recording the relationship between each layer and the building elements it contains. This mapping data is an important part of the layer description information, which helps with subsequent analysis and processing. For example, Layer 1 (wall layer) may include multiple wall elements: Wall A, Wall B, Wall C, etc.; Layer 2 (door and window layer) may include multiple door and window elements: Door 1, Door 2, Window 1, etc. The mapping data will record all wall elements included in Layer 1, and all door and window elements included in Layer 2. The structured storage step stores all relevant data in a structured manner in the database. Structured storage means that data will be saved in an ordered form such as tables and fields, which facilitates subsequent querying and operation. In architectural drawing analysis, historical identification and annotation records may include elements that were automatically identified or manually annotated in the past; for example, whether a room is marked as a bedroom, whether certain walls have load labels, etc. Examples: Graphic topology association data stores the connection, nesting, and adjacency relationships between geometric elements such as rooms, walls, doors, and windows; layer element mapping data stores the mapping relationships between each layer and its included architectural elements (such as walls, doors, and windows); and records past annotation information for architectural elements, such as which walls were marked as load-bearing walls and which windows were marked as emergency exits.

[0031] In an optional embodiment, based on graphic topology association data and mesh data in the architectural drawing feature database, closed graphics in the architectural drawings are identified to obtain an initial set of closed graphics. The mesh is closed structural attribute data generated during parsing and rendering and stored in the front-end database; it is not initially displayed and includes: The topological association data of the graphics and the pre-stored mesh data are retrieved from the architectural drawing feature database to determine the boundary contour dataset of the graphics to be identified. The ray method is used to make a preliminary judgment on the closure of the boundary contour dataset. Candidate closed shapes that meet the closure conditions are filtered by judging whether the ray penetration position is on the corresponding mesh. A contour tracking algorithm is used to verify the boundary integrity of candidate closed figures, and figures with incomplete boundaries are eliminated to obtain an initial set of closed figures. The initial set of closed figures is used for element type determination.

[0032] It should be noted that the system extracts graphic topology data from the database, especially the boundary contour datasets of the graphics. These boundary contour datasets include the external outlines of building elements, such as the exterior walls of rooms, the outlines of doors and windows, and the boundaries of stairs. By analyzing this boundary data, the system can clarify the spatial relationships and geometric structures of building elements. For example, suppose that in architectural drawings, the exterior walls of a room have multiple connected line segments. Through graphic topology association data, the system will extract the coordinate data of these line segments to form a closed contour dataset, which represents the external boundaries of the room. The ray casting method is a common technique for determining whether a figure is closed. In this method, a ray originates from a point and extends in a specific direction. If the number of intersections between the ray and the boundary of the figure is odd, the figure is closed; if the number of intersections is even, the figure is open. This method allows for the determination of the closure of figures and the selection of candidate closed figures. Example: Suppose the exterior wall of a room is composed of several line segments. The ray casting method can detect whether these line segments form a closed region. If the figure is closed, the ray will intersect the boundary line with an odd number of intersections, and the exterior wall of the room is considered a closed figure. After the initial screening using the ray tracing method, the boundaries of the candidate closed shapes need further verification to ensure they are complete and without any breaks or missing parts. The contour tracing algorithm tracks the boundaries of the shapes, checking for any missing boundary sections. If incomplete boundaries are found, these incomplete shapes will be discarded, leaving only complete closed shapes. Example: For the shape of the exterior wall of a room, the contour tracing algorithm will scan along the line segment boundaries of the wall to check for any missing parts. If the wall shape has missing parts, the algorithm will discard the shape; if the wall shape has complete boundaries, it will be retained as a complete closed shape. After verification using the ray tracing and contour tracking algorithms, a preliminary set of closed shapes will be obtained. These shapes are considered complete and closed. This set is used for further building element type determination, that is, to determine which building element (such as room, wall, window, door, etc.) these closed shapes correspond to. Example: Suppose that through these steps, multiple closed shapes have been identified, such as the exterior wall of a room, the outline of a window, the outline of a door, etc. The system can then determine the building element type to which these closed shapes belong based on their shape, size, and other characteristics.

[0033] In an optional embodiment, the element type of each closed shape in the initial set of closed shapes is determined, and graphic element annotation data is generated, including: Retrieve layer element mapping data from the architectural drawing feature database to determine the candidate range of element types corresponding to the layer to which each initial closed graphic belongs; Extract the point set data and basic parameters calculated based on the point set for each initial closed shape, determine the element attribute parameters, and determine the element position parameters based on the coordinate information in the point set data, generating graphic element annotation data that includes element category, point set data, direction, and name information.

[0034] It's important to note that in architectural design, architectural drawings typically contain multiple layers, each representing different types of elements, such as walls, doors, windows, and furniture. Layers are a crucial management method in architectural drawings, used to distinguish different categories of elements. For example, one layer might represent the structural layer, another the electrical layer, and yet another the piping layer. For instance, suppose a building floor plan contains two layers: a wall layer and a door / window layer. In the drawing, a closed shape on the wall layer might represent the outline of a wall, while a closed shape on the door / window layer might represent the outline of a door or window. By accessing layer element mapping data in the architectural drawing feature database, each shape is categorized into its corresponding layer. For example, a closed shape on the wall layer is identified as a "wall," while a closed shape on the door / window layer is identified as a "door" or "window." After each closed graphic is assigned to a specific layer, the system determines a candidate element type based on the layer type; for example, the graphic in the wall layer may be a wall, column, window opening, etc.; the graphic in the door and window layer may be a door, window, curtain, etc.; for example, the candidate range of closed graphics in the door and window layer may include: door, window, curtain; the candidate range of closed graphics in the wall layer may include: exterior wall, interior wall, load-bearing wall. Point set data refers to the coordinate information of all points that make up a closed shape (e.g., the coordinates of the four corners of a rectangle). This point set data not only includes the coordinate information of the shape, but can also be used to calculate other basic parameters of the shape, such as area, side length, angle, etc. For example, suppose there is a closed rectangular shape with the coordinates of four points as follows: P1(0,0), P2(0,5), P3(10,5), P4(10,0). Based on these point sets, the area of ​​the rectangle (10*5=50 square meters) and the side lengths (side lengths are 10 and 5 respectively) can be calculated. If this is a shape in a wall layer, the system can further calculate that it is a wall and determine its size and location. Element attribute parameters include material, thickness, and function; these attributes are usually associated with layers and graphic shapes; while position parameters refer to the specific location of the element on the architectural drawing, which is usually determined by the coordinates of the graphic. For example, for a closed graphic of a wall, if the coordinates on the drawing indicate the location of the wall, its position in the building can be calculated using point set data (e.g., it is a wall in the living room, located at a specific position on the architectural drawing); if the closed graphic belongs to a wall layer, the system may automatically label its attribute parameters, such as material (e.g., concrete wall) and thickness (e.g., 20 cm). Finally, annotation data for the graphic elements will be generated. This annotation data contains all the information about the element, such as: element category: wall, door, window, etc.; point set data: the coordinates of each point that makes up the graphic; direction: such as the orientation of the wall or the opening and closing direction of the door; name information: such as the wall number, door type, etc. For example, for a rectangular wall graphic, the annotation data generated by the system may include: element category: wall, point set data: P1(0,0), P2(0,5), P3(10,5), P4(10,0), direction: the orientation of the wall may be horizontal or an angle value may be given; name information: such as "wall-001", indicating that this is the interior wall of the living room.

[0035] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for intelligent identification and annotation of architectural drawing elements based on graphic feature extraction, characterized in that, include: Obtain architectural drawing files to be processed, including two-dimensional CAD drawings and three-dimensional building information model drawings; The architectural drawing file to be processed is parsed to extract graphic geometric data and layer description information. The graphic geometric data includes point coordinates, line coordinates and surface coordinates that constitute the architectural graphics. The layer description information includes the name information and layer priority information of the architectural drawing layers. Based on the graphic geometric data and layer description information, a building drawing feature database is constructed, wherein the building drawing feature database includes graphic topology association data, layer element mapping data, historical identification and annotation records, and mesh data corresponding to closed graphics; Based on the graphic topology association data and mesh data in the architectural drawing feature database, closed graphics in the architectural drawings are identified to obtain an initial set of closed graphics. The mesh is closed structural attribute data generated and stored in the front-end database during parsing and rendering, and is not displayed in the initial state. The system retrieves layer element mapping data from the architectural drawing feature database and, based on the graphic geometry data of the initial closed graphic set, determines the element type of each closed graphic in the initial closed graphic set to generate graphic element annotation data. The graphic element annotation data is stored in a cache format and includes element category, point set data, direction information, and name information. The graphic element annotation data is embedded into the architectural drawing file to be processed to generate an annotated architectural drawing file. The ray-click function is used to determine whether the ray penetration position is on the target mesh. After clicking, the selected mesh is displayed, and the annotated architectural drawing file is displayed on the display interface.

2. The intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction according to claim 1, characterized in that, Obtain the architectural drawing files to be processed, including: Call the original drawing files in the architectural drawing database and distinguish whether the original drawing file format is 2D CAD format or 3D building information model format; The process involves separating graphic and layer data from 2D CAD files to obtain original 2D graphic data and original 2D layer data; and separating spatial graphic data and hierarchical layer data from 3D building information model files to obtain original 3D graphic data and original 3D layer data. The original data of two-dimensional graphics and three-dimensional graphics are uniformly divided into several graphics data blocks, and the original data of two-dimensional layers and three-dimensional layers are uniformly divided into several layer data blocks. Obtain the planar or spatial coordinate position of each graphic data block in the original drawing to obtain the graphic block position information; obtain the layer position of each layer data block in the original drawing to obtain the layer block position information.

3. The intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction according to claim 2, characterized in that, The architectural drawing file to be processed is parsed to extract graphic geometric data and layer description information, including: The architectural drawing file to be processed is input into a pre-trained drawing parsing model to obtain graphic geometric data and layer description information. The drawing parsing model includes a graphic encoding module, a layer encoding module, and a feature fusion module. The process involves inputting the architectural drawing file to be processed into a pre-trained drawing analysis model to obtain graphic geometric data and layer description information, including: The graphic encoding module extracts features from the graphic structural information in the architectural drawing file to be processed, and obtains graphic feature vectors. The layer encoding module performs feature processing on the initial layer information in the architectural drawing file to be processed, and obtains the layer attribute vector. The graphic feature vector and the layer attribute vector are input into the feature fusion module, and the feature fusion module performs weight allocation and information integration to obtain the fused feature matrix; The geometric data representing the graphic structure and the layer description information representing the layer characteristics are extracted from the fused feature matrix.

4. The intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction according to claim 3, characterized in that, The graphic encoding module extracts features from the graphic structural information in the architectural drawing file to be processed, resulting in a graphic feature vector, including: The graphic data block and the corresponding graphic block position information are input into the graphic encoding module, wherein the graphic encoding module includes a linear transformation layer and a position encoding layer; Each graphic data block is input into the linear transformation layer, and the graphic data blocks are dimension-mapped through the linear transformation layer to output the graphic embedding vector of each graphic data block. The position information of each graphic block is input into the position encoding layer. The position encoding layer performs feature transformation on the coordinate position or spatial position and outputs the position embedding vector of each graphic data block. The graphic embedding vector and the position embedding vector corresponding to each graphic data block are added element-wise to obtain the local graphic features of each graphic data block. By stitching together the local graphic features of all graphic data blocks in the order of their positions in the original drawing, a global graphic feature vector is obtained.

5. The intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction according to claim 4, characterized in that, The layer encoding module performs feature processing on the initial layer information in the architectural drawing file to be processed, resulting in a layer attribute vector, including: Extract initial layer information from the layer data block, wherein the initial layer information includes layer name text and layer priority identifier; Input the layer name text into the text embedding layer of the layer feature encoding module, perform semantic feature transformation on the name text through the text embedding layer, and output the layer name embedding vector; The layer priority identifier is matched with a preset priority-value mapping table, the priority identifier is converted into the corresponding priority value, the priority value is transformed through a value embedding layer, and the layer priority embedding vector is output. The layer name embedding vector corresponding to each layer data block is concatenated with the layer priority embedding vector to obtain the local layer features of each layer data block; By concatenating the local layer features of all layer data blocks according to the hierarchical order in the original drawing, a global layer attribute vector is obtained.

6. The intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction according to claim 5, characterized in that, The graphic feature vector and the layer attribute vector are input into the feature fusion module. The feature fusion module performs weight allocation and information integration to obtain a fused feature matrix, including: The graphic feature vector is input into the first attention submodule of the feature fusion module to calculate the importance weight of each dimension of the graphic feature vector to the geometric extraction of the graphic, and thus obtain the graphic feature weight vector. The layer attribute vector is input into the second attention submodule of the feature fusion module to calculate the importance weight of each dimension feature in the layer attribute vector to the layer attribute extraction, and the layer feature weight vector is obtained. The graphic feature vector and the graphic feature weight vector are multiplied element-wise to obtain the weighted graphic feature vector; the layer attribute vector and the layer feature weight vector are multiplied element-wise to obtain the weighted layer attribute vector. The weighted graphic feature vector and the weighted layer attribute vector are stacked column-wise to form a two-dimensional fusion feature matrix.

7. The intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction according to claim 6, characterized in that, Based on the aforementioned geometric data and layer description information, a feature database for architectural drawings is constructed, including: Perform topological relationship calculations on graphic geometric data to determine the connection relationships between points, lines, and surfaces, including nesting relationships and adjacency distribution relationships, and generate graphic topological relationship data; Perform element mapping analysis on layer description information, establish the correspondence between layer identifiers and building element types, match the relationship between layer priority and element recognition weight, and generate layer element mapping data; The graphic topology association data, layer element mapping data, pre-imported historical identification and annotation records, and closed graphic mesh data generated by parsing and rendering are structured and stored to form a building drawing feature database.

8. The intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction according to claim 7, characterized in that, Based on the graphic topology association data and mesh data in the architectural drawing feature database, closed graphics in the architectural drawings are identified to obtain an initial set of closed graphics, including: The topological association data of the graphic and the pre-stored mesh data are retrieved from the architectural drawing feature database to determine the boundary contour dataset of the graphic to be identified. The ray method is used to make a preliminary judgment on the closure of the boundary contour dataset. Candidate closed shapes that meet the closure conditions are filtered by judging whether the ray penetration position is on the corresponding mesh. A contour tracking algorithm is used to verify the boundary integrity of candidate closed shapes, and shapes with incomplete boundaries are eliminated to obtain an initial set of closed shapes. The initial set of closed shapes is used for element type determination.

9. The intelligent identification and annotation method for architectural drawing elements based on graphic feature extraction according to claim 8, characterized in that, For each closed shape in the initial set of closed shapes, the element type is determined, and graphic element annotation data is generated, including: Retrieve layer element mapping data from the architectural drawing feature database to determine the candidate range of element types corresponding to the layer to which each initial closed graphic belongs; Extract the point set data and basic parameters calculated based on the point set for each initial closed shape, determine the element attribute parameters, and determine the element position parameters based on the coordinate information in the point set data, generating graphic element annotation data that includes element category, point set data, direction, and name information.

Images