Building facade style identification and generation method and system based on deep learning and semantic analysis

Abstract

The invention discloses a building facade style identification and generation method and system based on deep learning and semantic analysis, and belongs to the technical field of scene building facade information identification, and the method comprises the steps: S10, obtaining a sample set of building graphs in a preset database, the database comprising building graphs corresponding to building facade styles and decoration semantics; s20, related processing is carried out on the building graphs, graph features of different building graphs are obtained, and the graph features comprise visual features, structural features, semantic features and cultural features. According to the building facade style recognition and generation method, through multi-dimensional feature extraction, zone location fine calculation, a syn-position matching strategy and an edge feature verification mechanism, the accuracy and efficiency of building facade style recognition and generation are improved, meanwhile, the applicability and robustness of the system are enhanced, and the technical value and practical significance are high.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and artificial intelligence, and in particular to a method and system for architectural facade style recognition and generation based on deep learning and semantic analysis. Background Technology

[0002] In the field of architectural design, graphic elements are used to express the semantics and functions of buildings, which is not only related to aesthetics but also directly affects the cultural significance and user experience of buildings. Architectural graphic elements (such as facades, plans, or decorative components) carry the characteristics of specific styles, such as classical columns, minimalist lines in modern styles, or rugged materials in industrial styles. These elements are not merely combinations of geometric shapes but also contain expressions of architectural language, conveying historical, cultural, or functional information. Therefore, how to identify and generate architectural graphics that conform to specific styles through intelligent means has become an important research direction in the design field.

[0003] Regarding this research, application CN202411596612.1 provides a point cloud generation method and system for deep learning semantic segmentation of prefabricated building components. The technical solution mainly includes: creating structured text for generating rules of prefabricated building components; automatically generating standardized prefabricated building component models based on the Revit secondary development environment; and generating point cloud data using graphics methods. This technical solution obtains structured text for prefabricated building components by dividing the document into three parts: prefabricated component category, corresponding dimensions, and geometric features. This improves the readability of building codes, making them easier for readers to understand, and serves the dimensional and quality inspection of prefabricated building components.

[0004] Another application, CN202210297755.7, provides a method and system for rapid generation of building facade textures based on semantic understanding. This technical solution includes: S1: constructing a texture map library containing component textures with different first semantics; S2: extracting the original texture maps of the 3D model facade; S3: identifying the second semantics of the original texture maps to obtain corresponding second semantic tag maps; S4: cropping the original texture maps using the second semantic tag maps to obtain cropped texture maps corresponding to specific semantics of different parts of the 3D model; S5: retrieving the texture map corresponding to the cropped texture map from the texture map library. This technical solution solves the problems of low efficiency and long time required for manually repairing textures in 3D reconstructed models.

[0005] However, the aforementioned technical solutions still have limitations. When processing graphic features of different styles, they struggle to capture the commonalities and differences across styles, leading to the neglect of the architectural semantic relationships behind the graphics and resulting in a lack of stylistic consistency in the generated graphics. Therefore, how to accurately identify the graphic features and semantics of different styles while ensuring stylistic consistency in the generated results has become a key issue in the field of intelligent design. Summary of the Invention

[0006] In view of the problems existing in the fields of computer vision and artificial intelligence, the present invention is proposed.

[0007] Therefore, one of the objectives of this invention is to provide a method and system for identifying and generating architectural facade styles based on deep learning and semantic analysis. Through multi-dimensional feature extraction, refined location calculation, sequential matching strategy, and edge feature verification mechanism, it improves the accuracy and efficiency of architectural facade style identification and generation, while enhancing the applicability and robustness of the system. This not only optimizes the architectural design process but also provides digital support for cultural heritage protection, and has high technical value and practical significance.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] On the one hand, this invention provides a method for architectural facade style recognition and generation based on deep learning and semantic analysis, including the following steps:

[0010] S10: Obtain a sample set of architectural graphics from a preset database, wherein the database includes architectural graphics corresponding to the style and decorative semantics of each building facade;

[0011] S20: Perform relevant processing on the building graphics, the relevant processing including classifying the building graphics into... , ,..., , Indicates the distinction of the first The system collects architectural drawings and obtains the graphic features of different architectural drawings, including visual features, structural features, semantic features, and cultural features.

[0012] S30: Perform relevant calculations on each of the aforementioned graphic features. The relevant calculations include calculations based on the visual features. The calculation method includes dividing the building graphics into regions and performing visual feature calculations on each of the divided regions.

[0013] S40: Based on the calculation results, match the architectural facade style and decorative semantics in the database, and obtain the 3 to 5 graphic features that match each architectural facade style and decorative semantics the most times, and mark the graphic features as reference graphic features.

[0014] S50: Obtain the location of the building graphic corresponding to the graphic feature that has the most successful matches from the reference graphic features, and mark the location as the first reference matching location;

[0015] S60: Extract the features of the first reference matching location, the features including proportion features; when matching the architectural facade style and decorative semantics of the architectural graphic at a future time, take the first reference matching location as the first priority matching object. If the first reference matching location matches the corresponding architectural facade style and decorative semantics, it is determined that the matching of the architectural graphic is successful, and an architectural graphic corresponding to the architectural facade style and decorative semantics is generated; otherwise, no determination is made.

[0016] In a preferred embodiment of the present invention, in step S20, the visual features include geometric shape and proportion, the structural features include a load-bearing system, the semantic features include style keywords, and the cultural features include historical context; wherein, the geometric shape and proportion include basic shape, proportional relationship, and symmetry.

[0017] In a preferred embodiment of the present invention, in step S30, the architectural graphic is divided into zones, and the division steps include:

[0018] Obtain the area of ​​the building graphic, and divide the building graphic into at least 4 areas of equal size based on the area;

[0019] Within each of the defined locations, there are distinguishable virtual locations and real locations. The virtual locations are the hollowed-out parts in the architectural drawings, and the real locations are the non-hollowed-out parts in the architectural drawings.

[0020] Based on the area, calculate the area occupied by the virtual and real locations, mark the occupied area as the reference area, and calculate the proportion of the reference area in the total area.

[0021] In a preferred embodiment of the present invention, the area occupied by the dummy location and the real location is calculated based on the area, and the area occupied by the dummy location is calculated according to the following formula:

[0022] ;

[0023] In the formula, Indicates the first The total area of ​​the portion of the dummy location in each location. Indicates the first The total number of vacant locations within each location. The index number representing the virtual position. =1,2,..., ;

[0024] Indicates the first The area of ​​each vacant location, This is an indicator function used to determine the first... Does the dummy location belong to the first...? Each location, when the vacant area belongs to a location The value is 1 if it is true, and 0 otherwise. The first character representing the architectural drawing Location.

[0025] In a preferred embodiment of the present invention, the area occupied by the actual location is calculated according to the following formula:

[0026] ;

[0027] In the formula, Indicates the first The total area of ​​the portion of the actual location in each location. Indicates the first The total area of ​​each location Indicates the first The total area of ​​the portion of the virtual location in each location.

[0028] In a preferred embodiment of the present invention, the following steps are taken: the calculation results of the area occupied by the virtual and real locations are marked as virtual results and real results; the architectural facade style and decorative semantics corresponding to the calculated results are obtained from the architectural facade style and decorative semantics that are successfully matched with the architectural graphics; the architectural facade style and decorative semantics are marked as reference architectural facade style and decorative semantics; and the edge features of the virtual location are obtained according to the calculated area of ​​the virtual location, the edge features including the curvature of the virtual location, and the feature vector of the curvature is learned through a convolutional neural network.

[0029] In a preferred embodiment of the present invention, the step of learning the feature vector of the radian using a convolutional neural network includes:

[0030] At the edge of the dummy position, an area equal to one-fifth of the area of ​​the dummy position is cut off as an edge region, and the edge region is marked as the verification edge region;

[0031] Obtain the length of the verification edge region, and divide the length into left segment length, middle segment length, and right segment length in equal proportions;

[0032] When matching a building graphic with the reference building facade style and decorative semantics at a future time, if the area occupied by the virtual and real locations calculated for the building graphic is the same as the virtual and real results, but the length of the edge area divided by one-fifth of the area of ​​the virtual location in the building graphic is different from the verification edge area, then it is determined that the building graphic does not match the reference building facade style and decorative semantics; otherwise, no determination is made.

[0033] In a preferred embodiment of the present invention, when it is determined that the architectural graphic does not match the style and decorative semantics of the reference architectural facade, an architectural graphic identical to the architectural graphic is obtained; when the architectural graphic is matched with the architectural facade style and decorative semantics in the future, it is determined that the architectural graphic does not match the style and decorative semantics of the reference architectural facade.

[0034] On the other hand, the present invention provides a system for a building facade style recognition and generation method based on deep learning and semantic analysis as described above, comprising:

[0035] The data acquisition module is used to acquire a sample set of architectural graphics from a preset database, which includes architectural graphics corresponding to the style and decorative semantics of each building facade.

[0036] The data processing module is used to perform relevant processing on the building graphics, including classifying the building graphics into... , ,..., , Indicates the distinction of the first The system collects architectural drawings and obtains the graphic features of different architectural drawings, including visual features, structural features, semantic features, and cultural features.

[0037] The feature extraction module is used to perform relevant calculations on each of the graphic features. The relevant calculations include calculations based on the visual features. The calculation method includes dividing the building graphic into regions and performing visual feature calculations on each region.

[0038] A data fusion processing module, comprising a matching unit, an acquisition unit, and a judgment unit;

[0039] The matching unit is used to match the architectural facade style and decorative semantics in the database according to the calculation results, and to obtain the 3 to 5 graphic features that match each architectural facade style and decorative semantics the most times according to the matching results, and to mark the graphic features as reference graphic features.

[0040] The acquisition unit is used to acquire the location of the building graphic corresponding to the graphic feature that has been matched most successfully in the reference graphic features, and mark the location as the first reference matching location;

[0041] The judgment unit is used to extract the features of the first reference matching location, the features including proportion features; when matching the architectural facade style and decorative semantics of the architectural graphic in the future, the first reference matching location is used as the first priority matching object. If the first reference matching location matches the corresponding architectural facade style and decorative semantics, it is determined that the matching of the architectural graphic is successful, and an architectural graphic corresponding to the architectural facade style and decorative semantics is generated; otherwise, no judgment is made.

[0042] Beneficial effects:

[0043] 1. This invention achieves a comprehensive and in-depth analysis of building facades by extracting visual, structural, semantic, and cultural features of architectural graphics. This multi-dimensional feature fusion method improves the accuracy of style recognition and can more accurately capture the unique style of building facades. At the same time, by dividing the building facade into multiple areas of equal size and further distinguishing between virtual and real areas, and by calculating the proportion of these areas in the total area and the edge features of virtual areas, the system can capture the influence of local details on the overall style, thereby further improving recognition accuracy.

[0044] 2. During the recognition process, the system will match 3 to 5 graphic features that are closest to the target style from the database as references. In the generation stage, the first reference matching area is matched first to ensure that the generated result is highly consistent with the target style. This mechanism effectively avoids style deviation of the generated result and improves the generation quality.

[0045] 3. During the verification phase, by comparing the length of the edge area of ​​the virtual location of the building graphic with the verification edge area, mismatched candidate graphics are further eliminated. This dual verification mechanism significantly improves the reliability of the generated results and ensures the high quality of the generated building facade. Furthermore, by prioritizing the first reference matching location, if the match is successful, the result is generated directly without traversing the entire database. This strategy can greatly reduce the amount of computation, improve the system response speed, and optimize the generation efficiency. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the modular structure of the building facade style recognition and generation system based on deep learning and semantic analysis according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the process structure of an embodiment of the present invention; The numbers in the diagram are: 110 - Data acquisition module; 120 - Data processing module; 130 - Feature extraction module; 140 - Data fusion processing module; 1401 - Matching unit; 1402 - Acquisition unit; 1403 - Judgment unit. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0048] Because existing technologies struggle to capture the commonalities and differences across styles when processing graphic features of different styles, they neglect the architectural semantic relationships behind the graphics, resulting in a lack of stylistic consistency in the generated graphics.

[0049] Based on this, the present invention proposes a method and system for architectural facade style recognition and generation based on deep learning and semantic analysis. Through multi-dimensional feature extraction, refined location calculation, order matching strategy and edge feature verification mechanism, it improves the accuracy and efficiency of architectural facade style recognition and generation, while enhancing the applicability and robustness of the system.

[0050] The present solution will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0051] Reference Figures 1 to 2 This is one embodiment of the present invention, which provides a method for architectural facade style recognition and generation based on deep learning and semantic analysis, including the following steps:

[0052] S10: Obtain a sample set of architectural graphics from a preset database, which includes architectural graphics corresponding to the style and decorative semantics of each building facade;

[0053] S20: Perform relevant processing on the architectural graphics, including classifying the architectural graphics into... , ,..., , Indicates the distinction of the first Each architectural graphic is analyzed, and its graphic features are obtained, including visual features, structural features, semantic features, and cultural features.

[0054] Visual features include geometric shapes and proportions, structural features include load-bearing systems, semantic features include style keywords, and cultural features include historical context; among them, geometric shapes and proportions include basic shapes, proportional relationships, and symmetry.

[0055] In this embodiment, by combining S10 and S20, the comprehensiveness and diversity of the data are ensured through a sample set covering multiple styles and semantics, providing a rich foundation for subsequent recognition and generation. Furthermore, by combining visual, structural, semantic, and cultural features, a multi-dimensional analysis of the building facade is achieved, improving the accuracy and depth of the recognition.

[0056] S30: Perform relevant calculations on each graphic feature. These calculations include calculations based on visual features. The calculation method includes dividing the building graphic into locations and calculating visual features within each of these locations. The steps for dividing the building graphic into locations include:

[0057] Obtain the area of ​​the building graphic and divide the building graphic into at least 4 proportionally sized areas based on the area;

[0058] Within each of the defined locations, distinguish between virtual locations and real locations. Virtual locations are the openwork parts in the architectural design, while real locations are the non-openwork parts in the architectural design.

[0059] The area occupied by the virtual and real locations is calculated based on the area, and the occupied area is marked as the reference area. The proportion of the reference area in the total area is then calculated.

[0060] The area occupied by the virtual and real locations is calculated based on the area. The area occupied by the virtual location is calculated using the following formula:

[0061] ;

[0062] In the formula, Indicates the first The total area of ​​the vacant locations in each location. Indicates the first The total number of vacant locations within each location. The index number representing the virtual position. =1,2,..., ;

[0063] Indicates the first The area of ​​each vacant location, This is an indicator function used to determine the first... Does the dummy location belong to the first...? Each location, when the vacant area belongs to a location The value is 1 if it is true, and 0 otherwise. The first character representing the architectural drawing Location;

[0064] The area occupied by the actual location is calculated using the following formula:

[0065] ;

[0066] In the formula, Indicates the first The total area of ​​the actual location portion of each location. Indicates the first The total area of ​​each location Indicates the first The total area of ​​the vacant locations in each location;

[0067] It should be noted that by dividing the area and calculating the area, the local details of the building facade are captured, which improves the accuracy of the identification; while the edge features of the virtual area (such as curvature) provide additional identification information, which helps to distinguish building facades with similar styles but different details.

[0068] Explanation of the relationship between the two formulas above:

[0069] The two formulas above are used to calculate the area of ​​the virtual and real areas in architectural drawings, respectively. These two formulas are closely related in principle and together constitute a comprehensive analytical framework for the spatial occupancy of architectural drawings.

[0070] The first formula iterates through all the virtual locations and determines whether they belong to the current location. The area is summed up to obtain the total area of ​​all empty spaces in the location. This calculation aims to quantify the space occupied by the hollow parts in the architectural drawing.

[0071] The second formula obtains the total area of ​​the real location by subtracting the total area of ​​the virtual location from the total area of ​​the location. This calculation aims to quantify the space occupied by the non-hollowed-out parts (i.e., solid parts) in the architectural drawing.

[0072] The two formulas calculate the area of ​​the virtual and real areas in the building drawing, respectively. The sum of the two equals the total area of ​​the entire area, reflecting the integrity of the space occupation.

[0073] The two formulas work together to provide quantitative analysis of the spatial structure of architectural graphics, offering crucial data support for subsequent architectural facade style identification and generation. Through these two formulas, the system can more accurately understand the spatial layout of architectural graphics, thereby improving the accuracy of style identification and the quality of the generated results.

[0074] S40: Based on the calculation results, match the architectural facade style and decorative semantics in the database, and obtain the 3 to 5 graphic features that match each architectural facade style and decorative semantics the most times, and mark the graphic features as reference graphic features.

[0075] It should be noted that by selecting high-frequency matching graphic features as references, the reliability and accuracy of the recognition results are ensured, while limiting the number of reference graphic features reduces the computational load of subsequent processing and improves system efficiency.

[0076] S50: Obtain the location of the building graphic corresponding to the graphic feature that has the most successful matches from the reference graphic features, and mark the location as the first reference matching location;

[0077] S60: Extract the features of the first reference matching location, including proportion features; when matching the architectural facade style and decorative semantics of the architectural graphic in the future, the first reference matching location is used as the first priority matching object. If the first reference matching location matches the corresponding architectural facade style and decorative semantics, the matching of the architectural graphic is determined to be successful, and an architectural graphic corresponding to the architectural facade style and decorative semantics is generated; otherwise, no determination is made.

[0078] In this embodiment, by combining S50 and S60, the location of the building graphic corresponding to the graphic feature with the most successful matching times is selected as the first reference matching location from the reference graphic features. The proportional features of this location are extracted. The first reference matching location is used as the priority matching object, which speeds up the matching speed and improves the system response efficiency. Furthermore, by extracting the proportional features of key locations, the key information of the recognition process is further focused, which improves the targeting of the recognition.

[0079] This approach, which prioritizes matching key locations, ensures the accuracy and style consistency of the generated results. Furthermore, the strict matching judgment mechanism avoids misjudgments and style deviations, thereby improving the reliability of the generated results.

[0080] The calculated areas of the virtual and real locations are labeled as virtual and real results, respectively. The architectural facade style and decorative semantics corresponding to the calculated results are obtained from the architectural facade style and decorative semantics that are successfully matched with the architectural graphics. The architectural facade style and decorative semantics are labeled as reference architectural facade style and decorative semantics. The edge features of the virtual locations are obtained based on the calculated area of ​​the virtual locations. The edge features include the curvature of the virtual locations. The feature vector of the curvature is learned through a convolutional neural network.

[0081] Based on the above, the feature vector of radians is learned through a convolutional neural network. The steps include:

[0082] At the edge of the dummy region, an area equal to one-fifth of the area of ​​the dummy region is cut off as the edge region, and the edge region is marked as the verification edge region;

[0083] Obtain the length of the verification edge region, and divide the length into left segment length, middle segment length, and right segment length proportionally;

[0084] When matching architectural graphics with reference architectural facade style and decorative semantics in future moments, if the area occupied by the virtual and real locations calculated for the architectural graphics is the same as the virtual and real results, but the length of the edge area divided by one-fifth of the area of ​​the virtual location in the architectural graphics is different from the verification edge area, then the architectural graphics are determined to be mismatched with the reference architectural facade style and decorative semantics; otherwise, no determination is made.

[0085] In this embodiment, edge feature verification further eliminates misjudgments caused by local detail differences, improving the robustness of the system; while strict control over the edge features of the virtual location ensures the sensitivity of the recognition process to details, improving the accuracy of recognition.

[0086] Based on the above, further, when it is determined that the architectural graphic does not match the facade style and decorative semantics of the reference building, an architectural graphic identical to the architectural graphic is obtained. When the architectural graphic is matched with the facade style and decorative semantics in the future, it is determined that the architectural graphic does not match the facade style and decorative semantics of the reference building.

[0087] Based on the above, this application improves the accuracy and efficiency of building facade style recognition and generation through multi-dimensional feature extraction, refined location calculation, sequential matching strategy and edge feature verification mechanism, while enhancing the applicability and robustness of the system.

[0088] This embodiment, in conjunction with the above-mentioned method for architectural facade style recognition and generation based on deep learning and semantic analysis, also proposes a working system applied to this method, as follows:

[0089] The data acquisition module 110 is used to acquire a sample set of architectural graphics from a preset database, which includes architectural graphics corresponding to the style and decorative semantics of each building facade.

[0090] Data processing module 120 is used to perform relevant processing on architectural graphics, including classifying architectural graphics into... , ,..., , Indicates the distinction of the first Each architectural graphic is analyzed, and its graphic features are obtained, including visual features, structural features, semantic features, and cultural features.

[0091] The feature extraction module 130 is used to perform relevant calculations on various graphic features. The relevant calculations include calculations based on visual features. The calculation method includes dividing the building graphic into regions and performing visual feature calculations on each region.

[0092] The data fusion processing module 140 includes a matching unit 1401, an acquisition unit 1402, and a judgment unit 1403.

[0093] The matching unit 1401 is used to match the architectural facade style and decorative semantics in the database according to the calculation results, and to obtain the 3 to 5 graphic features that match each architectural facade style and decorative semantics the most times according to the matching results, and to mark the graphic features as reference graphic features.

[0094] The acquisition unit 1402 is used to acquire the location of the building graphic corresponding to the graphic feature that has been matched most successfully in the reference graphic features, and mark the location as the first reference matching location;

[0095] The judgment unit 1403 is used to extract the features of the first reference matching location, including the proportion feature. When matching the architectural facade style and decorative semantics of the architectural graphic in the future, the first reference matching location is used as the first priority matching object. If the first reference matching location matches the corresponding architectural facade style and decorative semantics, the matching of the architectural graphic is determined to be successful, and an architectural graphic corresponding to the architectural facade style and decorative semantics is generated; otherwise, no judgment is made.

[0096] In summary, this application improves the accuracy and efficiency of building facade style identification and generation through multi-dimensional feature extraction, refined location calculation, order matching strategy, and edge feature verification mechanism. At the same time, it enhances the applicability and robustness of the system. This can not only optimize the architectural design process, but also provide digital support for cultural heritage protection, and has high technical value and practical significance.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for architectural facade style recognition and generation based on deep learning and semantic analysis, characterized in that, Includes the following steps: S10: Obtain a sample set of architectural graphics from a preset database, wherein the database includes architectural graphics corresponding to the style and decorative semantics of each building facade; S20: Perform relevant processing on the building graphics, the relevant processing including classifying the building graphics into... , ,..., , Indicates the distinguished first The system collects architectural drawings and obtains the graphic features of different architectural drawings, including visual features, structural features, semantic features, and cultural features. S30: Perform relevant calculations on each of the aforementioned graphic features. The relevant calculations include calculations based on the visual features. The calculation method includes dividing the building graphics into regions and performing visual feature calculations on each of the divided regions. S40: Based on the calculation results, match the architectural facade style and decorative semantics in the database, and obtain the 3 to 5 graphic features that match each architectural facade style and decorative semantics the most times, and mark the graphic features as reference graphic features. S50: Obtain the location of the building graphic corresponding to the graphic feature that has the most successful matches from the reference graphic features, and mark the location as the first reference matching location; S60: Extract the features of the first reference matching location, the features including proportion features; when matching the architectural facade style and decorative semantics of the architectural graphic in the future, take the first reference matching location as the first priority matching object. If the first reference matching location matches the corresponding architectural facade style and decorative semantics, it is determined that the matching of the architectural graphic is successful, and an architectural graphic corresponding to the architectural facade style and decorative semantics is generated. Conversely, no judgment is made.

2. The architectural facade style recognition and generation method based on deep learning and semantic analysis as described in claim 1, characterized in that, In S20, the visual features include geometric shape and proportion, the structural features include load-bearing system, the semantic features include style keywords, and the cultural features include historical context; wherein, the geometric shape and proportion include basic shape, proportional relationship and symmetry.

3. The method for architectural facade style recognition and generation based on deep learning and semantic analysis as described in claim 1, characterized in that, In step S30, the building graphic is divided into zones, and the division steps include: Obtain the area of ​​the building graphic, and divide the building graphic into at least 4 areas of equal size based on the area; Within each of the defined locations, there are distinguishable virtual locations and real locations. The virtual locations are the hollowed-out parts in the architectural drawings, and the real locations are the non-hollowed-out parts in the architectural drawings. Based on the area, calculate the area occupied by the virtual and real locations, mark the occupied area as the reference area, and calculate the proportion of the reference area in the total area.

4. The method for architectural facade style recognition and generation based on deep learning and semantic analysis as described in claim 3, characterized in that, The areas occupied by the dummy and real locations are calculated based on the aforementioned areas. The area occupied by the dummy location is calculated using the following formula: ； In the formula, Indicates the first The total area of ​​the portion of the dummy location in each location. Indicates the first The total number of vacant locations within each location. The index number representing the virtual position. =1,2,..., ; Indicates the first The area of ​​each dummy location, This is an indicator function used to determine the first... Does the dummy location belong to the first...? Each location, when the vacant area belongs to a location The value is 1 if it is true, and 0 otherwise. The first character representing the architectural drawing Location.

5. The method for architectural facade style recognition and generation based on deep learning and semantic analysis as described in claim 4, characterized in that, The area occupied by the actual location is calculated using the following formula: ； In the formula, Indicates the first The total area of ​​the portion of the actual location in each location. Indicates the first The total area of ​​each location Indicates the first The total area of ​​the portion of the virtual location in each location.

6. A method for architectural facade style recognition and generation based on deep learning and semantic analysis as described in any one of claims 4 to 5, characterized in that, The results of calculating the area occupied by the virtual and real locations are marked as virtual results and real results, respectively. The architectural facade style and decorative semantics corresponding to the calculated results are obtained from the architectural facade style and decorative semantics that are successfully matched with the architectural graphics. The architectural facade style and decorative semantics are marked as reference architectural facade style and decorative semantics. The edge features of the virtual location are obtained according to the calculated area of ​​the virtual location. The edge features include the curvature of the virtual location. The feature vector of the curvature is learned through a convolutional neural network.

7. The method for architectural facade style recognition and generation based on deep learning and semantic analysis as described in claim 6, characterized in that, The steps for learning the feature vector of the radian using a convolutional neural network include: At the edge of the dummy position, an area equal to one-fifth of the area of ​​the dummy position is cut off as an edge region, and the edge region is marked as the verification edge region; Obtain the length of the verification edge region, and divide the length into left segment length, middle segment length, and right segment length in equal proportions; When matching a building graphic with the reference building facade style and decorative semantics at a future time, if the area occupied by the virtual and real locations calculated for the building graphic is the same as the virtual and real results, but the length of the edge area divided by one-fifth of the area of ​​the virtual location in the building graphic is different from the verification edge area, then it is determined that the building graphic does not match the reference building facade style and decorative semantics; otherwise, no determination is made.

8. The method for architectural facade style recognition and generation based on deep learning and semantic analysis as described in claim 7, characterized in that, When it is determined that the architectural graphic does not match the facade style and decorative semantics of the reference building, an architectural graphic identical to the architectural graphic is obtained. When the architectural graphic is matched for facade style and decorative semantics in the future, it is determined that the architectural graphic does not match the facade style and decorative semantics of the reference building.

9. A system applied to the architectural facade style recognition and generation method based on deep learning and semantic analysis as described in claim 1, characterized in that, include: The data acquisition module is used to acquire a sample set of architectural graphics from a preset database, which includes architectural graphics corresponding to the style and decorative semantics of each building facade. The data processing module is used to perform relevant processing on the building graphics, including classifying the building graphics into... , ,..., , Indicates the distinguished first The system collects architectural drawings and obtains the graphic features of different architectural drawings, including visual features, structural features, semantic features, and cultural features. The feature extraction module is used to perform relevant calculations on each of the graphic features. The relevant calculations include calculations based on the visual features. The calculation method includes dividing the building graphic into regions and performing visual feature calculations on each region. A data fusion processing module, comprising a matching unit, an acquisition unit, and a judgment unit; The matching unit is used to match the architectural facade style and decorative semantics in the database according to the calculation results, and to obtain the 3 to 5 graphic features that match each architectural facade style and decorative semantics the most times according to the matching results, and to mark the graphic features as reference graphic features. The acquisition unit is used to acquire the location of the building graphic corresponding to the graphic feature that has been matched most successfully in the reference graphic features, and mark the location as the first reference matching location; The judgment unit is used to extract the features of the first reference matching location, the features including proportion features; when matching the architectural facade style and decorative semantics of the architectural graphic in the future, the first reference matching location is used as the first priority matching object. If the first reference matching location matches the corresponding architectural facade style and decorative semantics, it is determined that the matching of the architectural graphic is successful, and an architectural graphic corresponding to the architectural facade style and decorative semantics is generated. Conversely, no judgment is made.

Images