Ancient building label system construction method based on multi-dimensional feature fusion

Abstract

The invention discloses a multi-dimensional feature fusion-based ancient building label system construction method, which comprises the following steps of: firstly, converting an abstract aesthetics principle into computable, quantifiable and systematic indexes by constructing a four-dimensional feature space comprising a model, a color, an ornament and a layout; by introducing culture significance factors (integrating building propagation degree, expert score and endangered degree) as an optimization weight formula, optimization adjustment of different measurement dimensions is realized so as to calculate an importance index; by establishing a'building component-symbol-semantics' association map, the cultural connotation of the building is deeply excavated and understood, so that the overall cultural semantic density of the ancient building is calculated; and finally, dividing three-level annotation precision based on the importance index of the component and the cultural semantic density. The method provides a complete set of decoding scheme for historical building digital primitive protection and keeping innovation, and is suitable for database construction, digital twin modeling and cultural memory inheritance.

Description

Technical Field

[0001] The present invention relates to the field of digital cultural heritage protection technology, and in particular to a method for constructing an ancient building labeling system that combines the multi-dimensional features of computer vision, natural language processing, and knowledge graphs, for achieving systematic analysis and structured annotation of ancient buildings. Background Art

[0002] As important carriers of cultural heritage, the preservation and inheritance of ancient architecture rely on precise annotation. Traditional ancient architecture annotation systems have several limitations: First, they are relatively single-dimensional, often focusing solely on basic information like physical attributes, such as dimensions and materials, while ignoring the rich cultural meaning inherent in these buildings, such as their historical value, artistic characteristics, and cultural connotations. Second, architectural aesthetic characteristics, such as structural rhythm and the harmony of virtuality and reality, lack quantifiable metrics, relying instead on manual judgment and subjectivity, resulting in inconsistent and inefficient aesthetic assessments. Third, traditional annotation methods employ a unified standard and fail to dynamically adjust annotation levels based on component importance or cultural value. This makes it difficult to address the diverse characteristics of different building types (e.g., palaces and residential buildings), resulting in a lack of flexibility and dynamic adaptability in annotation results. Fourth, existing technologies also have significant shortcomings in cultural semantic mining. The extraction and expression of the cultural connotations of ancient architecture is not thorough and systematic, making it difficult to effectively link and integrate architectural components with symbolic and semantic information, thus limiting the full display and dissemination of the cultural value of ancient architecture.

[0003] While existing technologies have made some progress in the digital archiving and preservation of ancient architecture, many challenges remain, including the integrity of feature dimensions, the scientific nature of aesthetic quantification, the relevance of multi-source data, and the flexibility of annotation strategies. Therefore, a method for constructing an ancient architecture labeling system that can integrate multi-dimensional features, dynamically adjust annotation levels, and achieve in-depth cultural semantic analysis is urgently needed. Summary of the Invention

[0004] The present invention aims to solve the problems of single dimension and lack of cultural semantics in traditional labeling system, and provides a method for constructing ancient building labeling system based on four-dimensional feature space and dynamic weight, so as to realize the full-chain decoding of building physical properties and deep cultural semantics.

[0005] The present invention provides a method for constructing a multi-dimensional feature-fused ancient building label system, comprising the following steps:

[0006] Step 1: Construct a four-dimensional feature space

[0007] Integrate multi-source data including 3D point cloud data, historical image data, and local chronicles to construct a four-dimensional feature space containing shape features, color features, pattern features, and layout features;

[0008] Step 2: Decompose the sub-features of the four-dimensional features and quantify the aesthetics

[0009] Step 3: Calculate the dynamic weight of cultural significance of each dimension

[0010] The cultural significance factor is introduced, which comprehensively considers the architectural spread, expert ratings, and endangered status. The dynamic weight of cultural significance of each dimension is calculated based on the eigenvector and the corresponding cultural significance factor.

[0011] Step 4: Construct a correlation graph and calculate the architectural cultural semantic density D

[0012] Using image-text separation technology, we separated the images and text in books into image-text pairs. Using the BERT model, we converted the text describing ancient buildings into machine-processable text sequences. Based on the Pierce semiotic ternary relationship, we established an "architectural component-symbol-semantic" association map. Starting from architectural components, we linked the symbolic references of the components in specific contexts to semantic information, and comprehensively calculated the overall cultural semantic density (D) of ancient buildings, enabling in-depth exploration and understanding of the cultural connotations of the buildings.

[0013] Step 5: Divide the annotation accuracy

[0014] The importance index I of the building is calculated using the feature vectors and weight coefficients of each dimension, and the annotation accuracy is divided into three levels based on the importance index I and the cultural semantic density D: L1 is fine annotation, L2 is intermediate annotation, and L3 is basic annotation.

[0015] This method, developed for constructing a labeling system for ancient Chinese architecture, uses multi-dimensional feature quantification and integrates multi-layered data, including architectural appearance, historical context, regional culture, and structural characteristics, to create a comprehensive and accurate description of ancient buildings. This method provides strong technical support for the digital preservation, intelligent management, and academic research of ancient architecture, and offers new insights into the inheritance and innovation of cultural heritage, with broad application prospects and profound social value. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is a flow chart of a method for constructing an ancient building labeling system based on multi-dimensional feature fusion in an embodiment of the invention. DETAILED DESCRIPTION

[0018] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] like Figure 1 As shown, the implementation process includes feature extraction and aesthetic quantification, knowledge association and integration, calculation of four-dimensional feature weights based on cultural significance factors, and finally identification and calculation of the annotation level of ancient Chinese buildings, which are then annotated. Starting from multi-source data input, this process transforms aesthetic features into computable, quantifiable, and institutionalizable indicators, achieving a complete chain decoding of the physical properties of buildings and their underlying cultural meanings.

[0020] S01 Constructing a four-dimensional feature space: Integrating multi-source data such as the building's three-dimensional point cloud data, historical image data, and local chronicles and literature data to construct a four-dimensional feature space that includes shape, color, pattern, and layout.

[0021] S02 breaks down the sub-features of the four-dimensional features and quantitatively expresses the aesthetics: the form feature is a feature vector composed of the type and grade of the ancient building's roof, the number of bracket levels, and the number of bays; the color feature is a feature vector composed of the type and grade of the colored paintings and the proportion of the colored paintings; the ornamentation feature is a feature vector composed of the grade of the ornamentation and the richness of the ornamentation types; the layout feature is a feature vector composed of the dynamic rhythm index of the building complex where the single building is located and the measure of the mutual generation of virtuality and reality.

[0022] In this step, except for the layout feature vector which needs to be obtained through calculation, the shape, color and decoration features of the ancient buildings all follow the established type grade standards. The classification is based on the various feature type grades recorded in the "Regulations on Engineering Practice of the Ministry of Works of the Qing Dynasty" and the "Construction Code". The type grades are converted into feature vectors according to the principle of "the higher the type grade, the larger the corresponding value". If the building to be marked has no type grade, the value is the smallest.

[0023] Among them, the layout characteristics are quantified by the building complex dynamic rhythm index and virtual-real interaction measurement. In view of the layout characteristics, the building complex dynamic rhythm index Q is calculated based on the curvature change rate of the building lines and the building height difference. The calculation formula is:

[0024] Q=λ1·σ H +λ2·ρ HL +λ3·C contour

[0025] Among them, σ H is the standard deviation of building height, ρ HLis the height-volume correlation coefficient, C contour It represents the complexity of the edge contour line of the building complex, and λ is the weight coefficient λ1+λ2+λ3=1. This indicator is used to measure the dynamic beauty and rhythmic sense of the building layout.

[0026] For the layout characteristics, the calculation formula of the virtual-real mutual generation metric R is:

[0027]

[0028] Where A 实 is the physical building area, A 总 is the total building area, is the golden ratio, A 庭院 is the courtyard area, which is used to evaluate the coordination between virtual and real spaces in building layout.

[0029] The layout feature vector needs to be obtained through calculation. The three characteristics of ancient buildings, namely color, shape and decoration, all follow the established type grade standards. The classification is based on the type grades of each feature recorded in the "Regulations on Engineering Practice of the Ministry of Works of the Qing Dynasty" and the "Construction Code". The type grade is converted into a feature vector according to the principle of "the higher the type grade, the larger the corresponding value". If the building to be marked has no type grade, the value will be the smallest.

[0030] S03 introduces the cultural significance factor δ to calculate the dynamic weight of each dimension: this factor comprehensively considers the building's popularity, expert ratings, and endangered status. Based on the relevant eigenvalues and significance factors, the weight of each dimension in the four-dimensional feature space is calculated to achieve dynamic adjustment of the priority of each dimension. The calculation method of the cultural significance factor is:

[0031]

[0032] Among them, δ k represents the dynamic weight of cultural significance of the k-th dimension, the building communication degree is the communication degree of the building, and the expert score k It means that the experts score the kth dimension, and the degree of endangerment k represents the assessment of the actual preservation status and inheritance risk of the kth dimension, where k = 1, 2, …, 4, representing the four dimensions of form, color, decoration, and layout, respectively. Specifically, in this embodiment, the expert rating is the average score given by three or more ancient architecture experts on the cultural value, artistic value, historical significance, architectural craftsmanship, and representativeness within the ancient architecture system for each of these four dimensions. The degree of endangerment is scored based on the degree of damage, inheritance, or public awareness.

[0033] Before calculating the weights of each dimension, the original feature vectors of different dimensions need to be normalized to map them to a unified interval, and multiple sub-features under the same dimension are combined into a single value. The formula is as follows:

[0034]

[0035] V dim The final feature index of the target dimension, that is, the single numerical result after integrating multiple sub-features under the same dimension; a i Represents the weight of the i-th sub-feature. The sum of all sub-feature weights is 1. The sub-feature weight is obtained according to the expert score. is the value of the i-th sub-feature after normalization; It represents the weighted summation of n sub-features under the same dimension, integrating the information of multiple sub-features, and n represents the number of sub-features.

[0036] The improved and optimized dynamic weight of cultural salience (DW-CSA) is calculated by the following formula:

[0037]

[0038] Among them, V k is the feature vector of the kth dimension (V1 represents the shape feature vector, V2 represents the color feature vector, V3 represents the pattern feature vector, and V4 represents the layout feature vector); δ k is the significance factor of the kth dimension; the weight satisfies the normalization condition: ∑w k =1.

[0039] S04 Knowledge Association and Fusion: Mining the "implicit knowledge" of architectural culture from text, images, and other data. Using image-text separation technology, we split the images and text in books into image-text pairs. Using the BERT model, we convert text describing ancient buildings into machine-processable text sequences. Based on the Pierce semiotic ternary relationship, we establish an "architectural component-symbol-semantic" association graph. Starting from architectural components, we connect the symbolic references of the components in specific contexts to semantic information, comprehensively calculating the overall cultural semantic density (D) of ancient buildings, achieving in-depth exploration and understanding of the architectural cultural connotations.

[0040] The cultural semantic density D, which is used to measure the richness of cultural semantics carried by ancient architectural components, is calculated using the following formula:

[0041]

[0042] Where m is the total number of building components per unit area; S i Represents the set of “symbolic” nodes directly associated with the i-th component; |M s| is the number of "semantic" nodes associated with the symbolic node s.

[0043] S05 Classification of Annotation Accuracy: This classification is based on the component importance index I and cultural semantic density D. The three levels of annotation accuracy are based on the component importance index I and cultural semantic density D. The component importance index I is a key indicator for measuring the comprehensive value of ancient architectural components. It is determined by the four-dimensional feature space (form, color, pattern, and layout) and the cultural significance factor. The four-dimensional feature space reflects the objective aesthetic characteristics of the component, while the cultural significance factor reflects the communication and influence of the cultural semantics carried by the component.

[0044] Cultural semantic density D is an important indicator for measuring the richness of cultural semantics carried by ancient architectural components. It is determined by the number of "symbolic" nodes directly associated with the components in the association graph and the number of "semantic" nodes associated with each symbolic node.

[0045] The three levels of annotation accuracy are as follows:

[0046] Level 1 fine annotation includes full five-dimensional information such as detailed symbols and historical allusions, level 2 intermediate annotation includes symbol type, historical period and basic process information, and level 3 only records basic geometric properties. The calculation formula of component importance index I is:

[0047] I=w1·V1+w2·V2+w3·V3+w4·V4

[0048] V1 represents the type feature vector, including the building roof, brackets, and width features; V2 is the color feature vector, including the type of painting and the proportion of painting; V3 is the pattern feature vector, including the pattern level and the richness of pattern types; V4 represents the layout feature vector, including the building dynamic rhythm index D and the virtual-real mutual generation degree R, and w1, w2, w3, and w4 represent the corresponding weight coefficients respectively.

[0049] Cultural semantic density D is used to measure the richness of cultural semantics carried by ancient architectural components. The calculation formula is as follows:

[0050]

[0051] Where m is the total number of building components per unit area; S i Represents the set of “symbolic” nodes directly associated with the i-th component in the association graph; |M s |The number of "semantic" nodes associated with each symbolic node s.

[0052] The annotation levels are divided into three levels: L1, L2, and L3 based on factors such as the historical value, cultural significance, and preservation status of ancient buildings. The higher the importance index I and cultural semantic density D, the greater the value of the component and the richer the cultural semantics it carries, and the higher the corresponding annotation level, as follows:

[0053] L1: I≥0.8&D≥0.8

[0054] L2: 0.5≤I<0.8&0.5≤D<0.8

[0055] L3: I<0.5&D<0.5

[0056] Among them, level L1 is fine annotation, which is suitable for ancient buildings with extremely high historical value, cultural significance and artistic value, such as royal palaces, tombs, temples, gardens, etc. The annotation includes symbol details, historical events and the above-mentioned full five-dimensional information: geometric information, material information, structural information, environmental information, and historical information.

[0057] The Hall of Supreme Harmony in the Forbidden City (a World Cultural Heritage) is a royal palace. The calculated I value is 0.92 and the D value is 0.95. The annotation content includes:

[0058] Symbolic Details: Detailed records of the ornamentation, carvings, and paintings on architectural components, including pattern styles, color combinations, and craftsmanship. Historical Events: Detailed records of historical events, personal stories, and cultural background related to the building. Full Five-Dimensional Information: Completely records the building's three-dimensional geometry, material information, structural information, environmental information, and historical information.

[0059] Level L2 is an intermediate level of annotation, suitable for ancient buildings with certain historical, cultural, and artistic value, such as folk ancestral halls, guild halls, and residential buildings. It records symbol type, historical period, and basic craftsmanship information. Chengzhi Hall in Hongcun, Anhui (provincial cultural relic), has an I value of 0.65 and a D value of 0.68. The annotation content includes:

[0060] Symbol type: records the main symbol types on building components, such as dragon patterns, phoenix patterns, floral patterns, etc.; historical period: records the construction year, historical evolution and other information of the building; basic process information: records the main construction technology, material usage and other information of the building; three to five categories of geometric information, material information, structural information, environmental information, and historical information.

[0061] Level L3, the basic annotation level, is suitable for historical buildings with relatively low historical, cultural, and artistic value, such as ordinary residences, warehouses, and workshops. It annotates basic geometric properties of components, such as length, width, height, and shape. For ordinary residences in Jiangnan (non-cultural relics), the calculated I value is 0.35 and the D value is 0.3. The annotation content includes the physical and geometric properties of the building itself, such as geometry, material information, and structural information.

[0062] Furthermore, in this embodiment, the classification is optimized and adjusted according to the following rules:

[0063] L1: A building must meet any of the following criteria: First, it must be listed as a World Cultural Heritage Site or a National Key Cultural Relics Protection Site; Second, it must possess unique historical, cultural, or artistic value, such as representing the highest level of architectural achievement of a particular period; and it must be in excellent condition, fully reflecting its original historical appearance.

[0064] Level L2 requires any one of the following conditions: first, it is listed as a provincial or municipal cultural relic protection unit; it has certain historical value, cultural significance or artistic value, such as works representing the architectural characteristics of a certain region or nation; it is in good condition and can basically reflect the original historical appearance.

[0065] Level L3, ancient buildings that do not belong to levels L1 and L2.

[0066] In addition to the above embodiments, the present invention may also have other implementations. Any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of protection required by the present invention.

Claims

1. A method for constructing an ancient building label system with multi-dimensional feature fusion, comprising the following steps: Step 1: Construct a four-dimensional feature space Integrate multi-source data including 3D point cloud data, historical image data, and local chronicles to construct a four-dimensional feature space containing shape features, color features, pattern features, and layout features; Step 2: Decompose the sub-features of the four-dimensional features and quantify the aesthetics Step 3: Calculate the dynamic weight of cultural significance of each dimension The cultural significance factor is introduced, which comprehensively considers the architectural spread, expert ratings, and endangered status. The dynamic weight of cultural significance of each dimension is calculated based on the eigenvector and the corresponding cultural significance factor. Step 4: Construct a correlation graph and calculate the architectural cultural semantic density D Using image-text separation technology, we separated the images and text in books into image-text pairs. Using the BERT model, we converted the text describing ancient buildings into machine-processable text sequences. Based on the Pierce semiotic ternary relationship, we established an "architectural component-symbol-semantic" association map. Starting from architectural components, we linked the symbolic references of the components in specific contexts to semantic information, and comprehensively calculated the overall cultural semantic density (D) of ancient buildings, enabling in-depth exploration and understanding of the cultural connotations of the buildings. Step 5: Divide the annotation accuracy The importance index I of the building is calculated using the feature vectors and weight coefficients of each dimension, and the annotation accuracy is divided into three levels based on the importance index I and the cultural semantic density D: L1 is fine annotation, L2 is intermediate annotation, and L3 is basic annotation.

2. The method for constructing a labeling system for ancient buildings according to claim 1, characterized in that: In step 2, the type feature is a feature vector composed of the type level of the ancient building roof, the number of bracket levels, and the number of widths; The color feature is a feature vector composed of the level of the painting type and the proportion of the painting; the pattern feature is a feature vector composed of the level of the pattern and the richness of the pattern types; Layout characteristics are characteristic vectors composed of the dynamic rhythm index and virtual-real interaction of the building complex within which a single building resides. While the layout characteristic vectors must be calculated, the structural, color, and decorative features of ancient buildings all adhere to established typological hierarchies. These classifications are based on the typological hierarchies documented in the "Qing Gongbu Gongcheng Zuofa Zeli" and the "Yingzaofashi Fashi" (Building Code). The typological hierarchies are converted into characteristic vectors according to the principle of "the higher the typological hierarchies, the larger the corresponding values." If the building to be labeled does not have a typological hierarchical ...

3. The method for constructing a labeling system for ancient buildings according to claim 2, characterized in that: In step 2, the dynamic rhythm index Q of the building complex is calculated based on the layout characteristics and the curvature change rate of the building complex lines and the building height difference. The calculation formula is: Q=λ1·σ H +λ2·ρ HL +λ3·C contour Among them, σ H is the standard deviation of building height, ρ HL is the height-volume correlation coefficient, C contour It represents the complexity of the edge contour line of the building complex, and λ is the weight coefficient λ1+λ2+λ3=1. This indicator is used to measure the dynamic beauty and rhythmic sense of the building layout.

4. The method for constructing a labeling system for ancient buildings according to claim 1, characterized in that: In step 2, for the layout features, the calculation formula of the virtual-real mutual generation metric R is: Where A 实 is the physical building area, A 总 is the total building area, is the golden ratio, A 庭院 is the courtyard area, and the virtual-real mutual generation metric is used to evaluate the degree of coordination between virtual and real spaces in the architectural layout.

5. The method for constructing a labeling system for ancient buildings according to claim 1, characterized in that: In step 3, the calculation formula of the cultural significance factor is as follows: Among them, δ k represents the dynamic weight of cultural significance of the k-th dimension, the building communication degree is the communication degree of the building, and the expert score k It means that the experts score the kth dimension, and the degree of endangerment k It represents the evaluation value of the actual preservation status and inheritance risk of the kth dimension, where k = 1, 2,…, 4, representing the four dimensions of shape, color, decoration, and layout respectively.

6. The method for constructing a label system for ancient buildings according to claim 5, characterized in that: In step 3, before calculating the weights of each dimension, the original feature vectors of different dimensions need to be normalized to map them to a unified interval, and multiple sub-features under the same dimension are integrated into a single value. The formula is as follows: V dim The final feature index of the target dimension, that is, the single numerical result after integrating multiple sub-features under the same dimension; a i Represents the weight of the i-th sub-feature. The sum of all sub-feature weights is 1. The sub-feature weight is obtained according to the expert score. is the value of the i-th sub-feature after normalization; Represents the weighted sum of n sub-features under the same dimension, where n represents the number of sub-features; The dynamic weight of cultural significance is calculated using the following formula: Among them, V k is the characteristic index of the kth dimension; δ k is the significance factor of the kth dimension; the weight coefficient w k Satisfy the normalization condition: ∑w k =1.

7. The method for constructing a label system for ancient buildings according to claim 1, characterized in that: In step 4, a cultural semantic density D is constructed to measure the richness of cultural semantics carried by ancient architectural components. The calculation formula is as follows: Where m is the total number of building components per unit area; S i Represents the set of "symbolic" nodes directly associated with the i-th component; |M s | is the number of "semantic" nodes associated with the symbolic node s.

8. The method for constructing a label system for ancient buildings according to claim 1, characterized in that: In step 5, an importance index I is constructed to measure the comprehensive value of ancient building components. The calculation formula is as follows: I=w1·V1+w2·V2+w3·V3+w4·V4.

9. The method for constructing a label system for ancient buildings according to claim 1, characterized in that: In step 5, the annotation levels are divided into three levels: L1, L2, and L3 according to the following rules: L1: I≥0.8&D≥0.8 L2: 0.5≤I<0.8&0.5≤D<0.8 L3: I<0.5&D<0.5 Among them, level L1 is fine annotation, which is suitable for ancient buildings with extremely high historical value, cultural significance and artistic value. The annotation includes symbol details, historical events and the above-mentioned full five-dimensional information. The full five-dimensional information includes geometric information, material information, structural information, environmental information, and historical information; level L2 is intermediate annotation, which is suitable for ancient buildings with certain historical value, cultural significance and artistic value. The annotation includes symbol type, historical period and basic craft information; level L3 is basic annotation, which is suitable for ancient buildings with relatively low historical value, cultural significance and artistic value. Level L3 only includes the basic geometric properties of components.

10. The method for constructing a label system for ancient buildings according to claim 9, characterized in that: In step 5, the classification is optimized and adjusted according to the following rules: L1: Any of the following conditions is sufficient:

1. Listed as a World Cultural Heritage Site or a National Key Cultural Relics Protection Site; 2. Possessing unique historical, cultural, or artistic value; L2: Any of the following conditions is met:

1. Listed as a provincial or municipal cultural relic protection unit; 2. Possessing certain historical, cultural or artistic value; Level L3, ancient buildings that do not belong to levels L1 and L2.

Images