CN-121982393-A - Method and system for constructing and classifying traditional house decoration value factors

CN121982393ACN 121982393 ACN121982393 ACN 121982393ACN-121982393-A

Abstract

The invention discloses a method and a system for constructing and classifying a traditional folk house decoration value factor, wherein decoration images and point cloud data are acquired through multi-source data acquisition, a decoration element data set is generated after semantic segmentation and standardization processing, a multi-level feature construction function dependency graph is extracted, filtering and reordering optimization are performed by using a large language model, a Text2SQL expansion framework is constructed based on the optimized function dependency graph, multi-dimensional value factors of decoration elements are extracted, a multi-granularity space-time flow model is constructed through space-time mode enumeration and optimization, a decoration value factor system is formed, the system is imported into a map database, and a two-way mapping relation is established, so that multi-dimensional classification and association retrieval of the decoration elements are realized. The invention solves the problems of strong subjectivity and single classification system of the traditional decoration value evaluation, and realizes systematic, standardized construction and intelligent retrieval of the decoration value.

Inventors

HU CHENGKAI
CHEN CHEN
WU GE
CHEN ZHAOXUAN
LI JIAPENG
YE CHENXI

Assignees

广东工业大学

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. A method for constructing and classifying a traditional folk house decoration value factor, comprising the steps of: acquiring original data containing image data and point cloud data based on multi-source data acquisition equipment for acquiring traditional house decoration; performing semantic segmentation and size standardization processing on the original data to generate a standardized decoration element data set; Extracting multi-level features based on the standardized decoration element data set, constructing a function dependency graph, and filtering and reordering the function dependency graph through a large language model to obtain an optimized function dependency graph; Constructing a Text2SQL expansion framework based on the optimized function dependency graph, carrying out semantic analysis and query conversion on the decoration element description, and extracting multidimensional value factors of the decoration elements; performing space-time pattern enumeration and pattern optimization based on the multidimensional value factors, and constructing a multi-granularity space-time stream pattern to form a decoration value factor system; and importing the decoration value factor system into a map database, establishing a bidirectional mapping relation between the decoration elements and the value factors, and realizing multidimensional classification and association retrieval of the decoration elements.
2. The method of claim 1, wherein performing semantic segmentation and size normalization processing on the raw data to generate a normalized decorative element dataset comprises: denoising and correcting the original data, and improving the image contrast by adopting a self-adaptive histogram equalization algorithm to obtain preprocessed image data; performing pixel-level semantic segmentation on the preprocessed image data, and extracting independent decorative element areas to obtain a segmented decorative element image; And carrying out super-resolution reconstruction and perspective transformation correction on the segmented decorative element image, uniformly converting the super-resolution reconstruction and perspective transformation correction into standard size and resolution, and generating the standardized decorative element data set.
3. The method of claim 1, wherein extracting multi-level features based on the standardized decorative element dataset, constructing a functional dependency graph, comprises: performing feature extraction on the standardized decorative element data set to obtain low-level visual features comprising shape descriptors, texture features and color distribution; Extracting scale invariant features and direction gradient histogram features based on the standardized decoration element dataset to obtain local invariant feature descriptors; performing advanced semantic feature extraction on the standardized decorative element data set based on a pre-trained convolutional neural network, and fusing the standardized decorative element data set with the low-level visual features and the local invariant feature descriptors to obtain multi-level feature representation; and defining characteristic nodes and attribute nodes based on the multi-level characteristic representation, establishing dependency relationship edges among the nodes, and constructing the function dependency graph.
4. The method of claim 1, wherein filtering and reordering the functional dependency graph through a large language model to obtain an optimized functional dependency graph comprises: performing domain knowledge enhancement on the large language model based on a cultural knowledge base, and evaluating rationality of each dependency side in the function dependency graph to obtain a dependency evaluation result; Identifying abnormal modes in the function dependency graph based on the dependency relationship evaluation result, and filtering unreasonable association modes to obtain a filtered function dependency graph; And calculating importance scores of all nodes and credibility scores of all sides based on the filtered function dependency graph, and carrying out node reordering and edge weight updating on the filtered function dependency graph to obtain the optimized function dependency graph.
5. The method of claim 1, wherein constructing a Text2SQL extension framework based on the optimized function dependency graph, performing semantic parsing and query conversion on the decoration element description, and extracting multidimensional value factors of the decoration element, comprises: Performing term recognition on the natural language description of the decoration element, and mapping the recognized terms to standardized concepts to obtain standardized term representations; carrying out semantic deconstructment on the standardized term representation, and splitting the standardized term representation into attributes, relations and constraint conditions to obtain a structured query element; Mapping the structured query element to nodes and paths in the optimized function dependency graph, and constructing a multi-step reasoning chain to obtain a query path mapping result; and generating a structured query statement based on the query path mapping result, executing the structured query statement, and extracting the multidimensional value factors of the decoration elements.
6. The method of claim 1, wherein performing space-time pattern enumeration and pattern optimization based on the multi-dimensional value factors, constructing a multi-granularity space-time stream pattern, forming a decorative value factor system, comprises: Defining a basic value pattern of the decorative element based on the multi-dimensional value factor, wherein the basic value pattern comprises an artistic value pattern, a historical value pattern, a social culture value pattern and a technical value pattern; scanning and counting the basic value modes, and selecting a value factor meeting a minimum support threshold as a seed mode to obtain a seed mode set; Combining the seed modes in the seed mode set in pairs, calculating the co-occurrence frequency of each combination, and reserving the combination meeting the minimum support requirement to obtain a composite value mode; Adding time constraint and space constraint to the composite value mode, analyzing the distribution rule of the value factor combination on a time axis and a geographic space, and generating a space-time flow mode; Pruning is carried out on the space-time stream modes, the lifting degree and the confidence degree of each space-time stream mode are calculated, and the space-time stream modes with the interestingness meeting the threshold requirement are reserved, so that the multi-granularity space-time stream mode is obtained.
7. The method of claim 6, further comprising, after obtaining the multi-granularity spatial-temporal-flow pattern: applying constraint conditions to the multi-granularity space-time stream mode based on application scene requirements, wherein the constraint conditions comprise region limitation, time span limitation and cultural attribute limitation; Performing constraint propagation analysis on the constraint conditions, detecting contradictions among the constraints, and adjusting to generate a limited mode meeting the constraint conditions; calculating a comprehensive significance score for the multi-granularity space-time stream mode, wherein the comprehensive significance score is calculated by weighting based on four dimensions of representativeness, distinguishing degree, cultural value and application value; selecting candidate modes with highest scores based on the comprehensive significance scores, calculating the similarity among the candidate modes, and reserving a representative mode with highest significance for a similar mode cluster to obtain a top K mode; And layering and integrating the basic value mode, the multi-granularity space-time flow mode, the limited mode and the top K mode, and establishing a correlation network among modes to form the decorative value factor system.
8. The method of claim 1, wherein importing the decoration value factor system into a map database, establishing a bidirectional mapping relationship between decoration elements and value factors, and realizing multidimensional classification and association retrieval of the decoration elements, comprises: defining entity types of the knowledge graph based on the decoration value factor system, wherein the entity types comprise decoration element entities, value factor entities, region entities and dynasty entities; defining relationship types among the entity types, wherein the relationship types comprise belonging relationship, influence relationship and evolution relationship, and setting attribute specifications for each relationship type; establishing a forward mapping from the decoration element to the value factors, and supporting rapid extraction of the value factors contained in the decoration element from the decoration element; Establishing reverse mapping of the value factors to the decoration elements, and supporting retrieval of a decoration element set with specific value characteristics from the value factors; and calculating weighted relevance scores of the mapping relations, optimizing an index structure of a map database based on the weighted relevance scores, and realizing multidimensional classification and associated retrieval of the decoration elements.
9. The method of claim 1, wherein the multi-dimensional value factor comprises: an artistic value factor comprising a modeling value index, a composition value index, and a color value index; a historical value factor comprising a chronological value index, a genre value index, and an evolving value index; a social culture value factor comprising a symbolic meaning index, a folk-custom activity association index and a social hierarchy index; A technical value factor comprising a manufacturing process index, a material characteristic index and an innovation index; and carrying out quantization scoring on each value factor, establishing an association intensity matrix among the value factors, and expressing the mutual influence relation among different value factors.
10. A system for constructing and classifying a traditional civil decorative value factor, comprising: The data acquisition module is used for acquiring traditional house decoration based on multi-source data acquisition equipment to acquire original data comprising image data and point cloud data; The preprocessing module is used for carrying out semantic segmentation and size standardization processing on the original data to generate a standardized decoration element data set; the function dependency graph construction module is used for extracting multi-level features based on the standardized decoration element data set, constructing a function dependency graph, and filtering and reordering the function dependency graph through a large language model to obtain an optimized function dependency graph; The value factor extraction module is used for constructing a Text2SQL expansion frame based on the optimized function dependency graph, carrying out semantic analysis and query conversion on the decoration element description, and extracting the multidimensional value factors of the decoration elements; the space-time pattern construction module is used for carrying out space-time pattern enumeration and pattern optimization based on the multi-dimensional value factors, constructing a multi-granularity space-time stream pattern and forming a decoration value factor system; And the classification mapping module is used for importing the decoration value factor system into a map database, establishing a bidirectional mapping relation between the decoration elements and the value factors, and realizing multidimensional classification and association retrieval of the decoration elements.

Description

Method and system for constructing and classifying traditional house decoration value factors Technical Field The invention relates to the technical field of cultural heritage digital protection, in particular to a value evaluation and classification technology of traditional folk house decoration, and particularly relates to a method and a system for constructing and classifying a value factor of traditional folk house decoration. Background The traditional folk house decoration is taken as an important substance cultural heritage, and contains rich historic, artistic, aesthetic and social cultural values. These decorative elements are often in the form of various textures, engravings, coloured drawings on building elements, etc., carrying cultural genes and aesthetic ideas of specific areas, nations and times. Current research on traditional civilian decoration relies mainly on human experience for value assessment and classification. Researchers acquire decoration information through the modes of on-site investigation, photographic recording, manual drawing and the like, and then induction and arrangement are carried out on the form characteristics and cultural connotation of decoration elements based on professional knowledge. Another common method is to build a simple digital image library and store the collected traditional house decoration photos roughly classified according to regions or types. The most relevant prior art attempts to apply computer vision technology to the traditional decoration research field, extract the basic morphological features of the decoration through an image processing algorithm, and combine an expert knowledge system to perform preliminary classification and retrieval. The technology mainly adopts basic algorithms such as edge detection, contour extraction and the like to identify the geometric characteristics of the decorative pattern, and then classifies the identification result through a preset rule base. However, the prior art has the following technical problems: Firstly, the cultural value extraction of the decorative elements is too dependent on expert experience, and a systematic and standardized value factor construction method is lacked, so that the value evaluation result is strong in subjectivity and poor in comparability, and a unified evaluation standard is difficult to form. Secondly, the existing classification system often adopts a single dimension or simple hierarchical structure, complex association relation and multidimensional value attribute among traditional decoration elements cannot be effectively expressed, and deep mining and application of traditional household decoration resources are limited. Third, the traditional method lacks an effective semantic parsing and reasoning mechanism when processing the mapping of decoration features and cultural semantics, and cannot accurately extract the deep cultural value behind the decoration elements. Disclosure of Invention The invention aims to solve the technical problems of strong subjectivity, single classification system and inaccurate cultural semantic mapping of traditional folk house decoration value evaluation in the prior art, and provides a method and a system for constructing and classifying traditional folk house decoration value factors. In order to achieve the above purpose, the invention adopts the following technical scheme: A traditional folk-house decoration value factor construction and classification method comprises the steps of acquiring original data comprising image data and point cloud data based on multi-source data acquisition equipment, conducting semantic segmentation and size standardization processing on the original data to generate a standardized decoration element data set, extracting multi-level features based on the standardized decoration element data set, constructing a function dependency graph, filtering and reordering the function dependency graph through a large language model to obtain an optimized function dependency graph, constructing a Text2SQL expansion frame based on the optimized function dependency graph, conducting semantic analysis and query conversion on decoration element description, extracting multi-dimensional value factors of decoration elements, conducting space-time mode enumeration and mode optimization based on the multi-dimensional value factors to construct a multi-granularity space-time flow mode to form a decoration value factor system, importing the decoration value factor system into a database, establishing a bidirectional mapping relation between the decoration elements and the value factors, and achieving multi-dimensional classification and association retrieval of the decoration elements. Further, performing semantic segmentation and size standardization processing on the original data to generate a standardized decoration element data set, performing denoising and correction processing on the original data, improving image cont