Search

CN-121996808-A - Building elevation style identification and generation method and system based on deep learning and semantic analysis

CN121996808ACN 121996808 ACN121996808 ACN 121996808ACN-121996808-A

Abstract

The invention discloses a building elevation style identification and generation method and a system based on deep learning and semantic analysis, belonging to the technical field of scene building elevation information identification, comprising the following steps of S10, acquiring a sample set of building graphics from a preset database, wherein the database comprises the building graphics corresponding to each building elevation style and decoration semantic; and S20, carrying out relevant processing on the building graph, and obtaining graph characteristics of different building graphs, wherein the graph characteristics comprise visual characteristics, structural characteristics, semantic characteristics and cultural characteristics. The invention improves the accuracy and efficiency of the identification and generation of the style of the building elevation through multidimensional feature extraction, zone location fine calculation, cis-position matching strategy and edge feature verification mechanism, and simultaneously enhances the applicability and robustness of the system, thereby having higher technical value and practical significance.

Inventors

  • WANG SHIHAN
  • LV MINGYANG

Assignees

  • 南京工业大学

Dates

Publication Date
20260508
Application Date
20251111

Claims (9)

  1. 1. The building facade style identification and generation method based on deep learning and semantic analysis is characterized by comprising the following steps of: S10, acquiring a sample set of building patterns from a preset database, wherein the database comprises building patterns corresponding to the styles and decoration semantics of all building facades; S20, performing correlation processing on the building graph, wherein the correlation processing comprises dividing the building graph into areas , ,..., , Representing the distinguished first Building patterns, and obtaining pattern features of different building patterns, wherein the pattern features comprise visual features, structural features, semantic features and cultural features; S30, performing relevant calculation in each graphic feature, wherein the relevant calculation comprises calculation based on the visual features, and the calculation mode comprises dividing the building graphic into areas and calculating the visual features in each divided area; S40, matching the style of the building elevation and the decoration semantics in the database according to the calculated result, acquiring 3-5 graphic features which are successfully matched with each style of the building elevation and the decoration semantics at the maximum times according to the matched result, and marking the graphic features as reference graphic features; s50, obtaining the location of the building graph corresponding to the graph feature with the largest matching success times from the reference graph feature, and marking the location as a first reference matching location; And S60, extracting the characteristics of the first reference matching location, wherein the characteristics comprise proportional characteristics, when the building graphic is matched with the building elevation style and the decoration semantics at the future time, taking the first reference matching location as a first cis-matching object, judging that the matching of the building graphic is successful if the first reference matching location is matched with the corresponding building elevation style and the decoration semantics, and generating the building graphic corresponding to the building elevation style and the decoration semantics, otherwise, judging that the matching is not performed.
  2. 2. The method for recognizing and generating a building facade style based on deep learning and semantic analysis according to claim 1, wherein in the step S20, the visual features comprise geometric shapes and proportions, the structural features comprise bearing systems, the semantic features comprise style keywords, and the cultural features comprise historical contexts, and wherein the geometric shapes and proportions comprise basic shapes, proportional relationships and symmetry.
  3. 3. The method for recognizing and generating a building facade style based on deep learning and semantic analysis according to claim 1, wherein in S30, the step of dividing the building graphics into areas comprises: Acquiring the area of a building graph, and dividing the building graph into at least 4 zone bits with equal proportion area based on the area; Dividing virtual location positions and real location positions in the divided locations, wherein the virtual location positions are hollowed-out parts in the building graph, and the real location positions are non-hollowed-out parts in the building graph; and calculating the occupied areas of the virtual location and the real location based on the areas, marking the occupied areas as reference areas, and calculating the proportion of the reference areas in the areas.
  4. 4. The method for recognizing and generating building facade style based on deep learning and semantic analysis according to claim 3, wherein the area occupied by the virtual location and the real location is calculated based on the area, and the area occupied by the virtual location is calculated according to the following formula: ; in the formula, Represent the first The total area of the portion of the dummy bit zone in each zone, Represent the first The total number of dummy bit regions within a single region, An index number indicating the dummy bit zone bit, =1,2,..., ; Represent the first The area of the individual dummy bit regions, For indicating function, for judging Whether or not the dummy bit zone belongs to the first Each location, when the virtual location area belongs to the location 1 When the time is equal to or 0 when the time is equal to or less than the time; First representing building graphics Each location.
  5. 5. The method for recognizing and generating building facade style based on deep learning and semantic analysis according to claim 4, wherein the area occupied by the real location is calculated according to the following formula: ; in the formula, Represent the first The total area of the portion of the real estate among the individual locations, Represent the first The total area of the individual zones is determined, Represent the first The total area of the portion of the dummy bit zone in each zone.
  6. 6. The method for recognizing and generating building elevation style based on deep learning and semantic analysis according to any one of claims 4 to 5, wherein the result of calculation of the area occupied by the virtual location area and the real location area is marked as virtual location result and real location result, the building elevation style and the decoration semantic corresponding to the calculated result are obtained in the building elevation style and the decoration semantic successfully matched with the building pattern, the building elevation style and the decoration semantic are marked as reference building elevation style and the decoration semantic, and the edge feature of the virtual location area is obtained according to the calculated area of the virtual location area, the edge feature comprises radian of the virtual location area, and the feature vector of the radian is learned through a convolutional neural network.
  7. 7. The method for identifying and generating a building facade style based on deep learning and semantic analysis according to claim 6, wherein the step of learning the feature vector of the radian through a convolutional neural network comprises the steps of: intercepting an edge of the virtual bit zone with an area which is one fifth of the area of the virtual bit zone as an edge zone, and marking the edge zone as a verification edge zone; acquiring the length of the verification edge region, and dividing the length equal proportion region into a left section length, a middle section length and a right section length; When the future time matches the building graph in the reference building elevation style and the decoration semantics, if the occupied area of the virtual position location and the real position location calculated for the building graph is the same as the virtual position result and the real position result, but the length of an edge area divided by one fifth of the area of the virtual position location in the virtual position location of the building graph is different from the verification edge area, judging that the building graph is not matched with the reference building elevation style and the decoration semantics, otherwise, judging that the building graph is not matched with the reference building elevation style and the decoration semantics.
  8. 8. The method for recognizing and generating a building elevation style based on deep learning and semantic analysis according to claim 7, wherein when it is determined that the building graphic does not match the reference building elevation style and decoration semantics, the same building graphic as the building graphic is obtained, and when the building graphic is matched with the building elevation style and decoration semantics at a future time, it is determined that the building graphic does not match the reference building elevation style and decoration semantics.
  9. 9. A system for applying a deep learning and semantic analysis based building facade style recognition and generation method as claimed in claim 1, comprising: the data acquisition module is used for acquiring a sample set of building graphics from a preset database, wherein the database comprises building graphics corresponding to each building facade style and decoration semantics; a data processing module for performing related processing on the building graph, wherein the related processing comprises dividing the building graph into areas , ,..., , Representing the distinguished first Building patterns, and obtaining pattern features of different building patterns, wherein the pattern features comprise visual features, structural features, semantic features and cultural features; The feature extraction module is used for carrying out related calculation in each graphic feature, wherein the related calculation comprises calculation based on the visual features, the calculation mode comprises dividing the building graphic into areas, and calculating the visual features in each divided area; the data fusion processing module comprises a matching unit, an acquisition unit and a judgment unit; The matching unit is used for matching the style of the building elevation and the decoration semantics in the database according to the calculated result, acquiring 3-5 graphic features which are successfully matched with each style of the building elevation and the decoration semantics for the largest times according to the matching result, and marking the graphic features as reference graphic features; The obtaining unit is used for obtaining the location of the building graph corresponding to one graph feature with the largest matching success times in the reference graph features, and marking the location as a first reference matching location; The judging unit is used for extracting the characteristics of the first reference matching location, wherein the characteristics comprise proportional characteristics, when the building graphic is matched with the building elevation style and the decoration semantics at the future time, the first reference matching location is used as a first cis-matching object, if the first reference matching location is matched with the corresponding building elevation style and the decoration semantics, the building graphic is successfully matched, the building graphic corresponding to the building elevation style and the decoration semantics is generated, and otherwise, the building graphic is not judged.

Description

Building elevation style identification and generation method and system based on deep learning and semantic analysis Technical Field The invention relates to the technical field of computer vision and artificial intelligence, in particular to a building elevation style identification and generation method and system based on deep learning and semantic analysis. Background The semantic and the function of the building are expressed through graphic elements in the field of building design, so that the aesthetic is concerned, and the cultural meaning and the use experience of the building are directly influenced. Building graphic elements (such as facades, planes or decorative elements) bear characteristics of specific styles, such as columns of classical styles, simple lines of modern styles or rough materials of industrial styles, are not only combinations of geometric shapes, but also expression of building languages, and can transmit historical, cultural or functional information. Therefore, how to identify and generate building patterns conforming to a specific style by an intelligent means becomes an important research direction in the design field. For the research on the aspect, the application document with the application number of CN202411596612.1 provides a point cloud generation method and a system for deep learning semantic segmentation of a building prefabricated part. According to the technical scheme, the document is divided into three parts of the prefabricated part category, the corresponding size and the geometric characteristic, so that the structured building prefabricated part text is obtained, the readability of building specifications is improved, the reader is facilitated to understand, and the size and quality detection of the building prefabricated part is served. The other application number is CN202210297755.7, and the technical scheme includes that S1, a texture mapping library of component maps containing different first semantics is constructed, S2, an original texture mapping of a three-dimensional model elevation is extracted, S3, a second semantic meaning of the original texture mapping is identified to obtain a corresponding second semantic label map, S4, the original texture mapping is cut through the second semantic label map to obtain a cutting mapping corresponding to specific semantics of different parts of the three-dimensional model, and S5, the texture mapping corresponding to the cutting mapping is searched in the texture mapping library. The technical scheme solves the problems of low efficiency and long time of manually repairing the three-dimensional reconstruction model mapping. However, the above-mentioned technical solution still has limitations that it is difficult to capture commonalities and disparities across styles when processing graphics features of different styles, resulting in neglecting the architectural semantic association behind the graphics, so that the generated graphics lack consistency in style. Therefore, how to accurately identify the graphic features of different styles and the semantics thereof when generating the building graphics and ensure that the generated results are consistent in style becomes a key problem in the field of design intelligence. Disclosure of Invention The present invention has been made in view of the above-mentioned problems occurring in the prior art of computer vision and artificial intelligence. Therefore, one of the purposes of the invention is to provide a building elevation style identification and generation method and a system thereof based on deep learning and semantic analysis, which improves the accuracy and efficiency of building elevation style identification and generation and enhances the applicability and robustness of the system through multidimensional feature extraction, zone-bit refined calculation, cis-position matching strategy and edge feature verification mechanism, thereby not only optimizing the building design flow, but also providing digital support for cultural heritage protection, and having higher technical value and practical significance. In order to solve the technical problems, the invention provides the following technical scheme: on one hand, the invention provides a building facade style identification and generation method based on deep learning and semantic analysis, which comprises the following steps: S10, acquiring a sample set of building patterns from a preset database, wherein the database comprises building patterns corresponding to the styles and decoration semantics of all building facades; S20, performing correlation processing on the building graph, wherein the correlation processing comprises dividing the building graph into areas ,,...,,Representing the distinguished firstBuilding patterns, and obtaining pattern features of different building patterns, wherein the pattern features comprise visual features, structural features, semantic features and cultural fea