A historical and culturally oriented method and system for spatial classification and regeneration

Abstract

This invention relates to the field of urban planning technology, and more particularly to a historical and cultural-oriented spatial classification and regeneration method and system. The method includes the following steps: acquiring historical and cultural data, and performing multimodal perception on the historical and cultural data to obtain historical and cultural spatial data; extracting deep features from the historical and cultural spatial data to obtain historical and cultural element data; constructing a spatial semantic map from the historical and cultural element data to obtain spatial semantic map data; classifying the spatial semantic map data according to its chronological evolution trajectory to obtain historical and cultural spatial classification data; generating spatial regeneration data based on the historical and cultural spatial classification data to obtain spatial regeneration data; acquiring urban planning demand data, and integrating the urban planning demand data and the spatial regeneration data to obtain integrated spatial regeneration data. This invention combines urban planning with spatial regeneration, effectively achieving a dynamic balance between historical and cultural preservation and modern urban development.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban planning, and in particular to a historical and cultural oriented space classification and regeneration method and system. BACKGROUND

[0002] With the acceleration of urbanization and the rising demand for cultural heritage protection, traditional blocks, traditional building groups, and traditional style areas, etc. typical historical and cultural spaces are facing the real challenges of complex spatial structure and diverse functional reuse needs in urban renewal. However, the identification and management of such spaces in the prior art rely on manual research and template rules, and lack the ability of multi-modal perception and unified structured expression of heterogeneous historical data such as images, texts, and GIS, making it difficult to automatically extract and model the spatial form, functional semantics, and cultural evolution characteristics. Artificial intelligence, abbreviated as AI, is a new technology science that studies, develops, and applies systems for simulating, extending, and expanding human intelligence. Artificial intelligence is a branch of computer science that aims to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. The problem is how to combine artificial intelligence with the use of historical and cultural space generation and intelligent reasoning of space regeneration strategies. SUMMARY

[0003] To solve the above technical problems, the present application provides a historical and cultural oriented space classification and regeneration method and system to solve at least one of the above technical problems.

[0004] The present application provides a historical and cultural oriented space classification and regeneration method, comprising the following steps:

[0005] Step S1: Obtain historical and cultural data, and perform multi-modal perception on the historical and cultural data to obtain historical and cultural space data; perform deep feature extraction on the historical and cultural space data to obtain historical and cultural element data;

[0006] Step S2: Construct a space semantic graph for the historical and cultural element data to obtain space semantic graph data; classify the space semantic graph data according to the chronological evolution trajectory to obtain historical and cultural space classification data;

[0007] Step S3: Generate space regeneration according to the historical and cultural space classification data to obtain space regeneration data;

[0008] Step S4: Obtain city planning demand data, and perform space regeneration fusion according to the city planning demand data and the space regeneration data to obtain space regeneration fusion data.

[0009] In the application, through multi-modal perception and deep feature extraction, fine identification of historical and cultural space and accurate extraction of cultural elements are realized, compared with traditional single mode perception technology, the identification accuracy and element retention integrity of cultural space are significantly improved. Through space semantic graph construction and age evolution track classification, not only the time sequence characteristics of space evolution can be revealed, but also the space groups with different cultural value evolution modes can be automatically classified, providing scientific basis for space regeneration. Combined with urban planning demand data for space regeneration fusion, the dynamic balance of historical and cultural protection and modern city development is effectively realized, avoiding the problem that the regeneration scheme is out of touch with the actual planning demand in the existing method.

[0010] Preferably, step S1 is specifically:

[0011] Obtaining historical and cultural data;

[0012] Multi-modal feature extraction is performed on the historical and cultural data to obtain multi-modal feature data;

[0013] Feature splicing is performed on the multi-modal feature data to obtain feature splicing data;

[0014] Self-attention coding is performed on the feature splicing data to obtain feature attention data;

[0015] Fusion reasoning is performed on the feature attention data to obtain historical and cultural space data;

[0016] Deep feature extraction is performed on the historical and cultural space data to obtain deep feature data;

[0017] According to the deep feature data, the historical and cultural space data is marked with cultural attributes to obtain historical and cultural element data.

[0018] In the application, in the historical and cultural data processing stage, multi-modal feature extraction and feature splicing are used to fuse multi-source information such as images, texts and spatial structures, effectively improving the integrity and information richness of historical and cultural space identification. The self-attention coding mechanism is used to assign importance weight to the multi-modal features, and the extraction ability of key cultural information is strengthened. Through the fusion reasoning process, the deep correlation and semantic consistency inference of different modal features are realized, which significantly improves the representation quality of historical and cultural space data. On this basis, deep feature extraction and cultural attribute marking can realize fine-grained identification and accurate classification of historical and cultural elements.

[0019] Preferably, the fusion reasoning is specifically:

[0020] Cross-modal correlation processing is performed on the feature attention data to obtain cross-modal correlation data;

[0021] The cross-modal correlation data is subjected to modal consistency optimization processing to obtain feature modal optimization data;

[0022] The feature modal optimization data is subjected to semantic completion reasoning to obtain feature modal completion data;

[0023] The feature modal completion data is subjected to spatial structure consistency fusion to obtain historical and cultural space data.

[0024] In the application, by performing cross-modal correlation processing on the feature attention data, the deep connection between different modal features can be accurately modeled, and the accuracy of multi-source feature fusion of historical and cultural data is improved. Through modal consistency optimization processing, the information redundancy and conflict between modalities are further eliminated, and the consistency and coordination of the overall feature expression are enhanced. Combined with semantic completion reasoning, the problem of missing or incomplete semantics of part of the modal data is effectively solved, and the completeness of spatial semantic reasoning is improved. Through spatial structure consistency fusion, the rational coherence of the fusion result in the spatial distribution and cultural structure logic is ensured.

[0025] Preferably, step S2 is specifically:

[0026] The historical and cultural element data is subjected to element node construction and spatial relationship edge construction to obtain element node data and spatial relationship edge data, respectively;

[0027] The spatial semantic graph is generated according to the element node data and the spatial relationship edge data to obtain spatial semantic graph data;

[0028] The chronological evolution trajectory data is obtained by extracting the chronological evolution trajectory according to the spatial semantic graph data;

[0029] The trajectory clustering data is obtained by clustering the chronological evolution trajectory data;

[0030] The historical and cultural space classification data is obtained by classifying the chronological evolution trajectory data according to the trajectory clustering data.

[0031] In the application, by constructing element nodes and spatial relationship edges on historical and cultural element data, the core elements and their spatial associations in historical and cultural space are systematically extracted, and the structured expression ability of cultural information organization is improved. The spatial semantic graph generated based on this not only realizes the bidirectional association of cultural elements and spatial positions, but also provides rich spatio-temporal evolution clues for subsequent evolution trajectory extraction. By clustering the chronological evolution trajectory data, different cultural space change groups under different historical evolution modes can be effectively identified, avoiding the local one-sidedness caused by traditional single time axis analysis. Finally, based on the trajectory clustering result, the spatial type and cultural evolution characteristics are accurately divided, greatly improving the scientificity of historical and cultural space management and regeneration decision-making.

[0032] Preferably, the spatial semantic graph data includes first spatial semantic graph data and second spatial semantic graph data, the time parameter corresponding to the first spatial semantic graph data is not greater than the time parameter corresponding to the second spatial semantic graph data, and the age evolution track extraction is specifically:

[0033] The first spatial semantic graph data and the second spatial semantic graph data are time-constrained node association pairing to obtain time-constrained node pair data;

[0034] The time-constrained node pair data is subjected to node pair evolution relationship determination to obtain node evolution data;

[0035] The node evolution data is subjected to spatial connectivity concatenation to obtain age evolution track data.

[0036] In the application, the node association pairing based on the time parameter constraint effectively ensures the logical continuity of the historical and cultural space nodes in the time sequence, and avoids the problems of random node association and evolution chain breakage in the prior art. Through node pair evolution relationship determination, the evolution mode between different time nodes can be identified in detail, such as function continuation, cultural transformation or space disappearance, which significantly improves the accuracy and explanatory power of the cultural evolution track modeling. Through spatial connectivity concatenation, not only the evolution path between nodes is reconstructed, but also the age evolution track data with spatial consistency and cultural logical continuity is formed.

[0037] Preferably, the track clustering is specifically:

[0038] The age evolution track data is subjected to spatial node extraction and cultural semantic node extraction to obtain spatial node data and cultural semantic node data;

[0039] The spatial node data and the cultural semantic node data are subjected to node pair pairing to obtain node pair data;

[0040] The node pair data is subjected to evolution track generation to obtain evolution track data;

[0041] The evolution track data is subjected to track local repair to obtain track repair data;

[0042] The track repair data is subjected to spatial path similarity matrix construction and time span similarity matrix construction to obtain spatial similarity matrix data and time span similarity matrix data, respectively;

[0043] The track repair data is subjected to hierarchical clustering according to the spatial similarity matrix data and the time span similarity matrix data to obtain track clustering data.

[0044] In the application, the spatial node extraction and cultural semantic node extraction of the chronological evolution trajectory data fully retain the spatial evolution characteristics and cultural evolution semantics in the trajectory, and improve the integrity and multidimensional information expression capability of the trajectory data. Through the node pair matching and evolution trajectory generation, the trajectory chain based on the continuity of time, space and semantics is reconstructed, and the evolution fault problem caused by the node isolation in the traditional trajectory analysis is avoided. In view of the noise or abnormal jump in the generated trajectory, the consistency and usability of the trajectory data are improved through the local repair processing of the trajectory. Through the construction of the spatial path similarity matrix and the time span similarity matrix, the dual similarity of the trajectory in the spatial dimension and the time dimension is realized, and the multi-scale recognition capability of the trajectory clustering is enhanced. Through the hierarchical clustering based on the double matrix, the cultural space groups in different evolution modes are accurately distinguished.

[0045] Preferably, step S3 is specifically:

[0046] According to the historical cultural space classification data, the cultural evolution characteristics are extracted to obtain cultural evolution characteristic data, wherein the cultural evolution characteristic data includes spatial function evolution characteristic data, building form change characteristic data, cultural symbol evolution characteristic data and time evolution trajectory characteristic data;

[0047] A preset historical cultural space evolution generation model is used for multi-path natural evolution sample generation of the cultural evolution characteristic data to obtain candidate space regeneration data, wherein the preset historical cultural space evolution generation model is obtained through the generation adversarial training of the historical cultural space classification experience data and the space regeneration label data pre-stored in the local;

[0048] The spatial coherence of the candidate space regeneration data is tested to obtain the space regeneration data.

[0049] In the application, the cultural evolution characteristics are extracted based on the historical cultural space classification data, the multidimensional characteristics such as spatial function evolution, building form change, cultural symbol evolution and time trajectory change are systematically identified, and the evolution expression capability of the cultural space is greatly enriched. The preset historical cultural space evolution generation model is used for multi-path natural evolution sample generation, the possible regeneration paths of the space in different evolution situations are effectively simulated, and the diversity and rationality of the space regeneration scheme are improved. The generation model is based on the adversarial training of the local historical experience data and the space regeneration label data, so that the generation result has historical coherence and real adaptability. Further, through the spatial coherence test, the candidate samples with spatial logic break or function conflict are eliminated, and finally the high-quality space regeneration data meeting the spatial evolution rule and the cultural protection requirement are output.

[0050] Preferably, the construction steps of the preset historical cultural space evolution generation model include the following steps:

[0051] The pre-stored historical and cultural space classification experience data is subjected to a feature expansion layer processing to obtain feature expansion data;

[0052] The feature expansion data is subjected to a deep feature transformation processing to obtain deep feature data;

[0053] The deep feature data is subjected to a structured feature generation to obtain structured feature data;

[0054] The structured feature data is subjected to a feature up-sampling processing to obtain feature up-sampling data;

[0055] The feature up-sampling data is subjected to a convolution processing to obtain candidate space regeneration sample data;

[0056] The candidate space regeneration sample data and space regeneration label data are subjected to a convolution coding block processing to obtain convolution coding block data;

[0057] The convolution coding block data is subjected to a feature tightening layer processing to obtain feature tightening layer data;

[0058] The feature tightening layer data is subjected to a reality scoring layer processing to obtain sample reality scoring data;

[0059] The candidate space regeneration sample data is subjected to an iterative training according to the sample reality scoring data to obtain a historical and cultural space evolution generation model.

[0060] In the application, the feature expansion layer processing and the deep feature transformation are performed on the pre-stored historical and cultural space classification experience data, the hidden high-order feature relationship in the space evolution process is fully mined, and the expression ability of the evolution feature learning is improved. Through the structured feature generation and the feature up-sampling, the multi-dimensional feature tensor with spatial semantics is reconstructed, laying a foundation for the space layout reasoning. The candidate space regeneration sample data is generated through the convolution processing, the spatial local feature and the overall cultural evolution trend are effectively modeled. Further, the semantic difference between the space regeneration sample and the real sample is accurately captured through the hierarchical processing of the convolution coding block, the feature tightening layer and the reality scoring layer. Through the adversarial iterative training based on the sample reality scoring, the generation model can adaptively optimize the generation strategy, and a historical and cultural space evolution generation model with historical and cultural rationality and spatial evolution naturalness is obtained.

[0061] Preferably, the step S4 specifically comprises:

[0062] obtaining city planning demand data;

[0063] performing attribute mapping according to the city planning demand data and the space regeneration data to obtain space planning mapping data;

[0064] Conflict detection is performed on the space planning mapping data to obtain space conflict detection data.

[0065] Collaborative adjustment is performed on the space conflict detection data to obtain space adjustment data.

[0066] Space layout fusion is performed on the space adjustment data to obtain space regeneration fusion data.

[0067] In the present application, by obtaining city planning demand data, the dynamic requirements of city development on space layout can be captured in real time, and the historical and cultural space regeneration process is closely linked with the current city function planning. Through attribute mapping processing based on city planning demand data and space regeneration data, the semantic correspondence and transformation of historical and cultural space function and modern city use are effectively realized. Through space conflict detection, potential functional contradictions, layout conflicts or resource overlap problems can be identified in time, and the pre-position risk perception ability of the space regeneration scheme is improved. Combined with collaborative adjustment processing, the adaptability between historical and cultural space and city planning target is dynamically optimized. Through space layout fusion operation, the collaborative unity of historical and cultural protection target and city space updating demand is realized.

[0068] Preferably, the present application also provides a historical and cultural oriented space classification and regeneration system for executing the historical and cultural oriented space classification and regeneration method as described above, which comprises:

[0069] A historical and cultural data processing module is configured to obtain historical and cultural data, perform multi-modal perception on the historical and cultural data to obtain historical and cultural space data, and perform deep feature extraction on the historical and cultural space data to obtain historical and cultural element data.

[0070] A space semantic modeling and trajectory classification module is configured to construct a space semantic graph based on the historical and cultural element data to obtain space semantic graph data, and classify the space semantic graph data according to the evolution trajectory of the ages to obtain historical and cultural space classification data.

[0071] A space regeneration generation module is configured to generate space regeneration based on the historical and cultural space classification data to obtain space regeneration data.

[0072] A space regeneration fusion module is configured to obtain city planning demand data, and perform space regeneration fusion based on the city planning demand data and the space regeneration data to obtain space regeneration fusion data.

[0073] The application has the beneficial effects that: by multi-modal perception and deep feature extraction of historical and cultural data in the initial stage, multi-source information such as images, texts and spatial structures are systematically fused, and the integrity and fine-grained expression ability of historical and cultural space recognition are greatly improved. By constructing a spatial semantic graph and performing chronological evolution trajectory classification, not only the dynamic modeling of historical and cultural elements in the space-time dimension is realized, but also the context and mode of spatial evolution are revealed in detail, and the spatial units of different cultural attributes and evolution characteristics are accurately divided. Based on the spatial classification results, spatial regeneration is generated, effectively simulating multi-path natural evolution samples that conform to the historical evolution law, avoiding the limitations of traditional spatial regeneration methods that break the historical continuity. Combined with urban planning demand data for attribute mapping, conflict detection and collaborative adjustment, the high coordination and unity between historical and cultural protection goals and modern urban development needs are realized. BRIEF DESCRIPTION OF DRAWINGS

[0074] Other characteristics, objects and advantages of the present application will become more apparent from the following detailed description of non-restrictive embodiments, made with reference to the attached drawings:

[0075] Figure 1 A step flow chart of a historical and cultural oriented space classification and regeneration method of an embodiment is shown;

[0076] Figure 2 A step flow chart of a historical and cultural data processing method of an embodiment is shown;

[0077] Figure 3 A step flow chart of a spatial semantic modeling and trajectory classification method of an embodiment is shown;

[0078] Figure 4 A step flow chart of a spatial regeneration generation method of an embodiment is shown;

[0079] Figure 5 A step flow chart of a spatial regeneration fusion method of an embodiment is shown. DETAILED DESCRIPTION

[0080] The technical method of the present application will be described in detail below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0081] In addition, the accompanying drawings are included to provide a further understanding of the application, and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments of the application and, together with the description, serve to explain the principles of the application. In the drawings:

[0082] It should be understood that, although the terms "first", "second" and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and / or" as used herein encompasses any and all combinations of one or more of the associated listed items.

[0083] Referring now to the drawings Figures 1 to 5 The present application provides a historical and cultural oriented space classification and regeneration method, comprising the following steps:

[0084] Step S1: Obtain historical and cultural data, and perform multi-modal perception on the historical and cultural data to obtain historical and cultural space data; perform deep feature extraction on the historical and cultural space data to obtain historical and cultural element data;

[0085] In an embodiment, data collection processing is performed. The sources of the historical and cultural data include, but are not limited to, historical literature archives, geographic information system (GIS) data, image data (such as satellite image maps, historical maps), local chronicle literature, and oral history data. The data collection methods include manual input, batch scanning combined with optical character recognition (OCR) technology, satellite remote sensing-based image collection, and on-site street view image shooting. After the data collection is completed, the system performs in-depth perception processing on the historical and cultural data based on multi-modal perception technology, including the following aspects, such as spatial perception, for data with geographic coordinate attributes, through spatial registration processing, using affine transformation and other methods based on landmark points to perform spatial alignment of historical maps and existing maps; text perception, for historical literature archives, applying natural language processing (NLP) technology, performing keyword extraction, named entity recognition (NER), and other methods to identify place names, building names, and other geographic or cultural entity information in the literature; image perception, for historical image data, using image segmentation technology (such as a segmentation model based on a U-Net neural network) to process the images and extract historical buildings, street boundaries, and natural landscapes and other visual elements. Based on the multi-modal fusion perception of the historical and cultural spatial data, deep feature extraction is performed, including spatial feature extraction, for each cultural space unit, calculating its basic spatial geometric properties, including plot area, shape complexity (defined as the ratio of perimeter to area), and plot center point coordinate information; semantic feature extraction, according to the extracted text, image, and spatial features, classifying the types of historical and cultural elements, such as building sites, commercial areas, residential areas, traditional buildings, and transportation nodes; time feature extraction, based on literature records, image timestamps, or morphological evolution inference methods, labeling the time attributes of each historical and cultural element. Based on the above feature extraction results, a structured historical and cultural spatial data set is formed, where each data record at least includes an element identifier (ID), an element type, a spatial range description, a corresponding age annotation, and detailed feature descriptions.

[0086] Step S2: constructing a spatial semantic graph for the historical and cultural element data to obtain spatial semantic graph data; classifying the spatial semantic graph data according to the age evolution track to obtain historical and cultural spatial classification data;

[0087] In an embodiment, a spatial semantic graph construction process is performed. Each historical and cultural element is defined as a node in the graph, and each node corresponds to a specific cultural element entity, such as an architectural site, a residential area, a traditional building, etc. For the association relationship between nodes, a spatial relationship edge is constructed. When two nodes have a close relationship in space (for example, the Euclidean distance between the centroids of the two nodes is less than 100 meters) and there is an association between the historical functions of the two nodes (for example, there is a historical functional connection between a temple and a market) (judged by a pre-set function table / parameter table, and the table is constructed with the aid of expert knowledge), an association edge is established between the two nodes. The node and the edge are respectively marked with attributes, including node attribute marking: type (such as traditional building, commercial facility, etc.); age; area (the unit can be unified as square meters); preservation state (marked as intact, partially damaged, or severely damaged). Edge attribute marking: relationship type (for example, proximity relationship, functional association relationship, evolution and inheritance relationship, etc.); weight value (determined comprehensively according to the spatial distance between nodes or the importance recorded in literature, wherein the closer the spatial distance or the more explicit the historical association, the higher the weight value). Through the above method, the spatial semantic graph of historical and cultural elements is constructed, and a complete graph structure including nodes, edges, and attribute information is formed.

[0088] After the spatial semantic graph is generated, an age evolution trajectory classification process is performed. Specifically, the following steps are included: based on the age attribute of the historical and cultural elements, all nodes are divided into corresponding time segments to form a time-layered data structure. According to the evolution trajectory extraction rule, the node evolution path is identified and constructed. If a historical and cultural element has a continuation or evolution relationship in different time segments (for example, a market gradually changes from a market to a modern commercial center), the corresponding age evolution trajectory is established. The evolution trajectory is recorded by a trajectory coding method, which connects the starting node, the intermediate change node, and the ending node in sequence to form a clear evolution chain. For example, the trajectory coding can represent: "market" evolves into "market", and transforms into "modern commercial center". Based on the trajectory features (such as node change category, time span length), the trajectory similarity is calculated, and the K-Means clustering algorithm or the density-based spatial clustering of applications with noise (DBSCAN) is used to automatically classify the trajectory data, and the representative cultural evolution mode group is extracted. The historical and cultural space classification data is output, which includes detailed description information of each typical evolution mode.

[0089] Step S3: generating spatial regeneration data according to the historical and cultural space classification data;

[0090] In an embodiment, a spatial regeneration generation rule is formulated to explicitly assign spatial functions and optimize spatial layout strategies. In terms of spatial function assignment, according to the trajectory information extracted from the historical and cultural space classification data, corresponding regeneration purposes are set for different types of historical and cultural elements. Specific assignment rules include but are not limited to setting the regeneration purpose of traditional building trajectories as a cultural exhibition hall to continue its spiritual and cultural functions and enhance public openness; setting the regeneration purpose of residential trajectories as a cultural and creative industry park to develop cultural and creative industries in combination with historical human settlement environment characteristics; and setting the regeneration purpose of commercial node trajectories as a characteristic commercial street to strengthen the original commercial atmosphere and focus on the combination of historical street style protection and commercial vitality. In terms of spatial layout optimization, according to the preservation state and spatial evolution characteristics reflected in the historical and cultural space classification data, the following regional categories are subdivided, including a protection priority area, which is set for areas with high historical evolution preservation (e.g., structure preservation rate exceeding 95%) and advocates the principle of minimum intervention, only necessary repair and environmental improvement; a redevelopment area, which is set for spatial units with low historical element preservation and located in areas with convenient transportation, encouraging function remodeling and spatial structure updating; and a transition coordination area, which is set for areas between the protection priority area and the redevelopment area, flexibly connecting the protection and development areas through the setting of urban green spaces, outdoor exhibition facilities, and other ways to alleviate the sense of spatial fragmentation. In the spatial regeneration generation process, spatial generation techniques are applied to support efficient layout optimization. Specifically, based on a geographic information system (GIS) platform, a regeneration land zoning map is generated according to the above function assignment results to clearly define the geographical scope and boundaries of each functional area; a generative design algorithm is used for spatial layout optimization, specifically, morphological transformation techniques are used to regularize and fine-tune the land form to improve land use efficiency; and a Voronoi diagram division algorithm is used to generate a reasonable building colony distribution pattern centered on historical nodes to ensure spatial coherence and visual order. Standardized spatial regeneration data is formed, with each spatial unit containing regeneration purpose labeling, geographical range definition, design guidance notes, and other information, in the format of { "Area ID": "A001", "Regeneration Purpose": "Cultural Park", "Area": "2.5 hectares", "Design Notes": {...}}.

[0091] Step S4: Obtain city planning demand data, and perform spatial regeneration fusion according to the city planning demand data and the spatial regeneration data to obtain spatial regeneration fusion data.

[0092] In an embodiment, urban planning demand data collection and processing is performed. The urban planning demand data is mainly derived from official planning data such as urban renewal plan documents, transportation hub layout maps, land use planning red line maps, and public service facility layout requirements published by the government. The system standardizes the collected raw planning data, including mapping text-based planning requirements into quantifiable specific indicators, such as a green space ratio of no less than 30%, a public facility service coverage radius of no more than 500 meters, a commercial functional area volume rate of no more than 2.0, and converting graphic data (such as land red lines and transportation layout) into a geographic information system (GIS) standard format to ensure consistency with the coordinate system of historical and cultural space data. After completing the standardization of planning requirements, spatial regeneration integration rules are developed, including priority control and spatial coordination adjustment. In terms of use priority control, if there is a conflict between historical and cultural regeneration use and urban planning use (for example, an original cultural use area is planned as future transportation hub land), the pre-set use priority determination standard is used for processing: if the regeneration area has high cultural value (such as reaching the cultural protection level 1 or 2 standard), the cultural use is prioritized; if the cultural value is low or there is a significant conflict between regeneration use and urban development needs, the urban planning use is prioritized, and cultural symbols (such as small monuments and cultural theme landmarks) are implanted to preserve some degree of historical memory. In terms of spatial coordination adjustment, the system adjusts according to the following specific rules, including adjusting the building density of the regeneration area according to the volume rate index set by the planning to ensure that the construction intensity meets the overall planning requirements; adding necessary public nodes such as squares, bus stops, and community service centers according to the layout requirements of urban public service facilities; optimizing the design of regeneration area traffic connection to ensure that the connectivity rate between the regeneration unit and the urban main traffic network reaches or exceeds 80% to improve accessibility and convenience. After completing the above integration and adjustment operations, the system generates spatial regeneration integration data. The integration data content includes updated regeneration unit attributes such as actual use categories, adjusted land area, configured public facility node information, and cultural preservation rate indicators under the integration scheme. For example, a fusion area can be recorded as: fusion area number F001, actual use is a mix of cultural exhibition and commercial street, green space ratio is 32%, and the main traffic connection point is subway line 3 exit C.

[0093] Preferably, step S1 is specifically:

[0094] Step S11: Obtain historical and cultural data;

[0095] In an embodiment, the historical and cultural data includes historical archives, local chronicles, cultural relic protection catalogues, oral materials; satellite images, historical maps, street view photos; GIS vector data (such as building points, road lines, land surfaces); timestamp data (such as construction year, repair record time). Public data interface grabbing (such as the API of the State Administration of Cultural Heritage, data of local government affairs platforms); OCR identification of old documents, with text correction and layout structure identification; remote sensing image downloading and spatial registration processing (such as landmark control point correction error <2 meters).

[0096] Step S12: multi-modal feature extraction is performed on the historical and cultural data to obtain multi-modal feature data;

[0097] In an embodiment, for the text modality, word segmentation is performed (such as Jieba, BERT word segmenter); key entities (such as building type, purpose, place name) are extracted; Embedding coding (such as Word2Vec / sentence vector SBERT, dimension unified to 256 dimensions) is performed. For the image modality, image segmentation (such as DeepLabV3+) is performed to extract building contours; visual feature extraction (ResNet, EfficientNet to extract feature vectors, dimension such as 512 dimensions) is performed; local descriptors (such as SIFT, ORB) are extracted. For the spatial modality, geographical attribute features such as area, perimeter, location center, shape compactness (C = A / P, where C is the shape compactness, A represents the area enclosed by the contour, and P represents the perimeter of the contour) are extracted; according to the spatial position relationship, the spatial structure characteristics such as proximity (such as shortest distance measurement) and connectivity (such as topological graph connected component analysis) are analyzed.

[0098] Step S13: feature concatenation is performed on the multi-modal feature data to obtain feature concatenated data;

[0099] In an embodiment, different modal features are uniformly normalized. Direct vector concatenation is performed in a fixed order, and the arrangement is: [text features | image features | spatial features]; low-rank mapping (such as reduction to a unified 512-dimensional feature space through linear transformation or sparse coding) is adopted. The dimension of the concatenated feature vector is consistent (for example, unified to 512 dimensions or 1024 dimensions). If there is a missing value (such as a missing image in some historical data), mean filling or zero vector filling is adopted.

[0100] Step S14: self-attention encoding is performed on the feature concatenated data to obtain feature attention data;

[0101] In an embodiment, the self-attention encoding process includes the construction of query vectors, key vectors, and value vectors, the calculation of attention weights, and the encoding and fusion of subspace features. For each concatenated feature vector, corresponding Q (query vector), K (key vector), and V (value vector) are generated through linear transformation:​ 、 、 is the learnable parameter matrix; for example, Q = concatenate (feature , K = concatenate (feature , V = concatenate (feature . Based on the similarity between the query vector and the key vector, the attention weight between each pair of features is calculated. The formula for calculating the attention weight is as follows , where is the square root of the dimension of the key vector. The features are divided into several subspaces, and self-attention is performed respectively, and then concatenated. High-weight features represent stronger historical and cultural relevance; weakly related features can be weakened. The feature vector set after feature attention encoding is obtained.

[0102] Step S15: fuse and infer the feature attention data to obtain historical and cultural spatial data;

[0103] In an embodiment, a fusion network (such as a feedforward neural network FFN) is used to perform nonlinear transformation (ReLU activation) on the attention features; or for cultural units with spatial relationships (such as building plots, block units), a spatial adjacency graph is constructed, and a graph neural network (referred to as GNN) is applied to perform feature propagation and fusion on the adjacency structure. In the inference process, the fused high-order feature vector is assigned to each cultural spatial unit, including building plots, block units, street segments, etc., so that each spatial unit has a deep representation of historical and cultural features. Based on the spatial unit features assigned by inference, the area groups belonging to the same cultural evolution chain in the space are identified, and the natural evolution sequence of the historical and cultural space is revealed; at the same time, based on the feature similarity, the spatial units are clustered (mapped to a pre-set spatial function label table based on the cluster center value) to divide into several spatial feature areas (such as traditional commercial blocks, modern industrial site groups, etc.).

[0104] Step S16: perform deep feature extraction on the historical and cultural spatial data to obtain deep feature data;

[0105] In an embodiment, a multi-layer neural network (such as a 3-layer MLP or ResNet Block) is designed to further extract the data after inference; the extracted information includes cultural time evolution characteristics (such as modern style, classical style); structural complexity (such as dense building area, open block); and feature (color, material, scale). After processing, feature dimension reduction processing (such as using PCA or t-SNE) is performed to retain more than 95% of the information. After dimension reduction, a set of deep feature representations with high discriminability is output, and each feature vector is attached with label information to identify the cultural evolution category, feature category, or structural property category to which it belongs.

[0106] Step S17: Label the historical and cultural spatial data with cultural attributes based on the depth feature data to obtain historical and cultural element data.

[0107] In one embodiment, cultural attribute labels are set according to a preset dictionary or expert knowledge base, such as type labels (residential / temple / commercial); style labels (B-type architectural style, C-type architectural style); preservation status (intact, partially damaged, abandoned); and protection level (national, provincial, municipal, general registration). Based on deep feature vectors, a classifier (such as XGBoost, LightGBM, or Transformer classifier head) automatically determines the confidence score: each label carries a confidence score (e.g., 90% confidence for "A-type style architecture"). If the confidence scores of multiple attributes are close (difference less than 5%), an expert system review decision is made (determined by preset knowledge rules or a manual review system) or a weighted merging of attributes is adopted (assigning a composite label or prioritizing the retention of the highest confidence category based on the confidence score ratio).

[0108] Preferably, the fusion reasoning specifically includes:

[0109] Cross-modal correlation processing is performed on the feature attention data to obtain cross-modal correlation data;

[0110] In one embodiment, the correlations between features of different modalities (text, image, space) are analyzed to establish cross-modal relationships. Attention features of each modality are mapped to the same dimension (e.g., unified to 512 dimensions) using linear transformation or a trainable projection matrix. Cross-modal similarity is calculated; in the unified feature space, the system calculates the cosine similarity between text-image, text-space, and image-space respectively. , for Feature similarity between This is the first modal feature vector. Let be the second modality feature vector. The first and second modality feature vectors can be one of the following: text-image, text-space, or image-space. Construct three sets of similarity matrices between the modalities, each matrix being of size . (n is the number of data points). For feature pairs with high similarity scores (based on a threshold, such as greater than 0.5), their feature weights are increased; for feature pairs with low similarity scores, their feature weights are decreased. Through dynamic weighted optimization, the correlation between features of different modalities is highlighted, thereby forming cross-modal correlation data that reflects cross-modal relationships.

[0111] Modality consistency optimization is performed on cross-modal correlation data to obtain characteristic mode optimized data;

[0112] In an embodiment, the system minimizes the statistical difference between the feature distributions of different modalities based on the maximum mean discrepancy method. The system calculates the square of the Euclidean norm of the difference between the mean vector of all sample feature vectors in the first modality feature set after being processed by the kernel mapping function and the mean vector of all sample feature vectors in the second modality feature set after being processed by the kernel mapping function. The system performs local consistency regularization processing. The system requires the consistency of feature vectors of different modalities in the local range to reach a preset standard within the spatially adjacent unit, that is, the system calculates the Euclidean distance between the cross-modality feature vectors corresponding to any adjacent unit and controls the distance to be lower than a set threshold (for example, less than 0.1). The system performs weight adjustment according to the importance index of each modality. Specifically, the system uses a modality confidence driven weight mechanism to set a fusion weight for each modality, and the weight dynamically changes according to the confidence of the modality. For example, if the text modality confidence , the image modality confidence , the weight is mapped by a preset weight mapping table to obtain , , and the rest is allocated to other modalities.

[0113] The system performs semantic completion reasoning on the feature modality optimization data to obtain feature modality completion data.

[0114] In an embodiment, the system detects missing semantic regions. The system compares and analyzes the feature vectors of each feature unit to detect units that lack key semantic dimensions (such as age information and use category labels). The judgment basis for feature missing is that there is no valid value in the related dimension or the confidence score is lower than a set threshold. The system performs multiple reasoning and completion methods, including a method based on neighbor reasoning. The system uses a k-nearest neighbor (KNN) strategy to retrieve the k nearest neighbor samples (for example, k is set to 5) of the feature unit with missing semantics. According to the mean or weighted average feature of the retrieved neighbor samples, the system infers the missing attribute information to complete the feature completion. The system uses a deep language model (such as BERT or T5) that is fine-tuned in a specific field to predict the missing semantic part according to the existing semantic context, for example, automatically inferring the historical use category or descriptive details, based on a language model-based completion method. The system constructs a graph structure based on a spatial semantic adjacency graph, where nodes represent spatial or object units and edges represent semantic similarity or spatial proximity. The system infers missing labels or attributes in the graph structure through a label propagation algorithm or a graph neural network (GNN) method to achieve cross-node semantic completion.

[0115] The system performs spatial structure consistency fusion on the feature modality completion data to obtain historical and cultural spatial data.

[0116] In an embodiment, the system detects the consistency relationship between each spatial unit (such as a block, a plot) in the dimensions of cultural type labels, age category labels, etc. If it is detected that adjacent units have obvious abnormal discontinuity in the above labels, for example, in a group of B-class architectural style buildings, a unit labeled as E-class architectural style appears, the system performs spatial smoothing processing, and corrects the abnormal label by neighborhood reclassification or local label correction method, and restores the overall spatial continuity. Spatial attribute weighted fusion processing is performed. The system considers the following three factors, spatial proximity score (denoted as ), which represents the proximity of the unit to the surrounding units in space; modal similarity score (denoted as ), which represents the similarity of the unit in multi-modal feature expression; historical relevance score (denoted as ), which represents the association strength of the unit with the surrounding historical and cultural elements. The system performs weighted fusion on the above score items according to the set weight, and calculates the fusion score. The fusion score is defined as the spatial proximity score of the unit multiplied by the weight coefficient 0.4, plus the modal similarity score multiplied by the weight coefficient 0.4, plus the historical relevance score multiplied by the weight coefficient 0.2. The attribute category with the highest score is assigned to the unit as its fusion label. Based on the fused spatial label, the system performs a connected region extraction operation. That is, the unit set with the same fusion label and spatially connected is extracted to form a continuous historical and cultural district, for example, a continuously distributed B-class architectural style block area. The extracted connected region is marked as a spatial regeneration unit.

[0117] Preferably, step S2 is specifically:

[0118] Step S21: element node construction and spatial relationship edge construction are performed on the historical and cultural element data, respectively, to obtain element node data and spatial relationship edge data;

[0119] In an embodiment, element nodes are constructed, such as each historical and cultural element (e.g., a single building, a block unit, a road node) as a node (Node); the node basic attributes include an ID (unique identifier, such as "Node_001"); a category (e.g., traditional architecture, folk house, public facility); an age; a spatial coordinate (a centroid point or a planar contour); and a preservation state (intact, partially damaged, or reconstructed). Spatial relationship edges are constructed, such as an edge (Edge) defined as an adjacent edge if the distance between the centroids of two nodes is less than a certain threshold (e.g., 150 meters), or a same-domain edge if the two nodes belong to the same plot / block, or a semantic enhanced edge if there is a documented functional association (e.g., "temple and market"). The edge attributes are defined as a distance (Euclidean distance); an association type (adjacent, functional association, age continuation); and an association weight (based on distance and functional correlation score, based on preset weight calculation, or based on preset rule engine for threshold mapping, for example, adjacent + functional correlation = high weight). Element node data and spatial relationship edge data are output.

[0120] Step S22: Spatial semantic graph generation is performed according to the element node data and the spatial relationship edge data, to obtain spatial semantic graph data.

[0121] In an embodiment, a standard graph data structure (e.g., an adjacency list or an adjacency matrix) is used to organize nodes and edges; nodes are connected by spatial relationship edges to form a complete spatial semantic graph. Each node includes an attribute dictionary (e.g., category, age, and protection level); each edge includes a weight, a distance, and an association description; and multiple edge connections between nodes are allowed (i.e., a node can have relationships with multiple nodes). In the adjacency list data structure, the spatial semantic graph is organized in the form of key-value pairs, where the key is the unique identifier of the element node; and the value is the list of target nodes connected to the node; each connection record includes the following fields: target node identifier; distance between nodes; and relationship type description.

[0122] In an embodiment, the system can perform time sequence grouping on the input data based on the time parameters (e.g., construction age, use period, etc.) carried in the element node data and the spatial relationship edge data, to construct spatial semantic graph data of different periods respectively. Alternatively, the system can first construct a complete spatial semantic graph including nodes and edges of all historical stages, and then perform subgraph extraction based on the time attributes of edges or nodes in subsequent processing. Specifically, by setting a time condition (e.g., the age field of a node or the time label of an edge), a subgraph within a specified time period can be extracted from the overall graph, which can be used as the data source of the first graph or the second graph.

[0123] Step S23: Age evolution trajectory extraction is performed according to the spatial semantic graph data, to obtain age evolution trajectory data.

[0124] In an embodiment, the evolution trajectory refers to a path from a historical node to another historical node via a change in spatial relationship or functional relationship; the time sequence requirement must be met between the nodes, i.e., the age of the starting node is earlier than the age of the ending node; the spatial connectivity or functional connectivity must be maintained between the nodes, i.e., there is an effective spatial relationship edge or functional relationship edge connection between the nodes. The extraction of the evolution trajectory is based on the traversal operation of the spatial semantic graph data, and can use the depth-first search (DFS) or breadth-first search (BFS) algorithm. The trajectory extraction rules include: selecting a node with an earlier age as the starting node; selecting a node with a later age as the ending node; all nodes on the path from the starting node to the ending node must maintain spatial or functional connectivity, and the associated weight of the connection edge between the nodes must be lower than a set threshold (for example, the edge weight is less than 0.5). For each extracted evolution trajectory, the following feature labels are performed, including: starting node identification; ending node identification; sequence of nodes passed through; trajectory length (step number); age span (time difference between the starting age and the ending age). The following constraint conditions are set in the trajectory extraction process: the number of steps of the trajectory is generally controlled between 3 and 10 steps to avoid dilution of the evolution meaning caused by too long trajectory; trajectories with obvious time span are extracted, such as the evolution chain from B-class building style to E-class building style; for trajectories with short span (such as B3-class building style to D1-class building style), trajectories with shorter links and clearer evolution relationship are preferentially extracted.

[0125] Step S24: trajectory clustering is performed on the age evolution trajectory data to obtain trajectory clustering data.

[0126] In an embodiment, the features of each trajectory are extracted, including: starting type; ending type; age span; trajectory length; trajectory path form (such as whether there is a cycle, jump, etc.). Trajectory similarity calculation (such as edit distance, dynamic time warping DTW) is performed; or the Euclidean distance is used to measure the similarity based on the trajectory feature vector. K-Means, DBSCAN or spectral clustering are used; the number of clusters is set according to the task requirements (such as k=5 classes: typical cultural evolution patterns). Each trajectory is assigned a category label according to the belonging cluster group (the data based on the group center is mapped through a preset label library), such as "traditional building evolution type", "folk house transformation type", "commercial regeneration type". The trajectory clustering data is output, and each trajectory has a cluster category.

[0127] Step S25: cultural space classification is performed on the age evolution trajectory data according to the trajectory clustering data to obtain historical cultural space classification data.

[0128] In an embodiment, the spatial classification rule defines that the space units covered by the start node, end node and all passing nodes of each trajectory are classified into the same cultural space category according to the cluster attribution category of each trajectory. Each cluster category corresponds to a cultural evolution theme, such as a traditional architectural style area, a folk house evolution area or a commercial prosperity area. In the classification process, spatial continuity checking is performed to ensure that the spatial connectivity rate between nodes in the same classification unit is not less than 80%. The spatial connectivity rate is defined as the number of node pairs that are directly or indirectly spatially related and connected between any two nodes within the classification unit, accounting for the proportion of all possible node pairs. If the spatial connectivity rate is detected to be less than 80%, local reclassification or node supplement operations are performed according to the actual connection of the nodes. For each cultural space classification unit, the following cultural attribute information is integrated and labeled: the dominant cultural category to which the space elements within the unit belong (based on the area ratio calculation of the preset electronic map, and the label of the space element with a proportion exceeding the threshold is taken as the dominant cultural category), such as traditional architecture, traditional folk house, commercial facilities, etc.; the dominant period of the node age data within the unit is extracted as the time marker of the unit; based on the trajectory evolution characteristics, the main change path of the space evolution within the unit is described, such as the evolution from traditional architecture to folk house, or the transformation from traditional commercial area to modern commercial complex.

[0129] Preferably, the spatial semantic graph data includes first spatial semantic graph data and second spatial semantic graph data, the time parameter corresponding to the first spatial semantic graph data is not greater than the time parameter corresponding to the second spatial semantic graph data, and the trajectory extraction of the chronological evolution is specifically:

[0130] The first spatial semantic graph data and the second spatial semantic graph data are time-constrained node association paired to obtain time-constrained node pair data;

[0131] In an embodiment, in the node pairing process, the following pairing constraint conditions are simultaneously satisfied, such as the Euclidean distance between the centroids of the nodes in the first spatial semantic graph and the nodes in the second spatial semantic graph does not exceed a preset threshold (for example, 100 meters), or the spatial overlap rate (i.e., the intersection over union IOU) between them reaches or exceeds a preset threshold (for example, 0.3). The category evolution relationship of the paired nodes should comply with the preset evolution logic (judged by an expert engine based on expert knowledge). For example, "folk house" can evolve into "shop", and "official office" can evolve into "modern office building"; if the category evolution logic is weak or not encouraged (such as "temple" evolving into "factory"), a low confidence label should be added to the pairing result. The time attribute of the node in the first spatial semantic graph needs to be earlier than or equal to the time attribute of the corresponding node in the second spatial semantic graph. For each node in the first spatial semantic graph, a candidate node set that satisfies all the pairing constraint conditions is searched in the second spatial semantic graph;

[0132] The time-constrained node pair data is subjected to node pair evolution relationship determination to obtain node evolution data;

[0133] In an embodiment, the system performs node pair evolution relationship determination processing on the time-constrained node pair data to generate node evolution data. Specifically, the system determines the evolution relationship type between each pair of nodes based on node attribute changes and records the corresponding change indicators. The evolution relationship type includes functional evolution, where the functional category of the node changes and the category change complies with the allowed evolution path in the preset evolution knowledge base mapping table, such as evolving from "residential building" to "commercial building". Morphological evolution, where the physical form characteristics (such as area, perimeter, and compactness) of the node change significantly, such as a significant increase in building area or an increase in architectural structural complexity. Functional and morphological dual evolution, where the node simultaneously changes in functional category and physical form. No significant evolution, where the node does not change significantly in functional category and physical form characteristics. The specific determination rules are as follows: to perform functional evolution determination, if the node category changes and the category change path complies with the allowed evolution relationship defined in the evolution knowledge base mapping table, it is determined to be functional evolution. To perform morphological evolution determination, the system compares the area, perimeter, and compactness of the node pair, calculates the change rate respectively, and the change rate is defined as: (post-node indicator value - pre-node indicator value) divided by pre-node indicator value. For example, if the area change rate exceeds the set threshold (e.g., 30%), it is determined to be morphological evolution. The perimeter and compactness can also be evaluated for significant changes according to similar methods. To perform comprehensive evolution determination, if the node pair complies with the above change criteria in both functional category and form characteristics, it is determined to be functional and morphological dual evolution. If the node pair does not meet the change determination threshold in both functional category and form characteristics, it is marked as no significant evolution.

[0134] According to the node evolution data, spatial connectivity is concatenated to obtain chronological evolution trajectory data.

[0135] In an embodiment, the geographical spaces corresponding to the plurality of evolution nodes can form a continuous path or a connected region, and the connections between the nodes satisfy the set spatial proximity and evolution consistency conditions. In a specific serial method, the system constructs a node evolution graph. The evolved nodes are used as node elements in the graph, and if the Euclidean distance between the centroids of any two nodes is less than or equal to a set threshold (for example, 100 meters), an edge is established between the two nodes. The weight of the edge is set according to the spatial distance between the nodes and the evolution similarity, wherein the evolution similarity can be quantitatively scored according to whether the functional category change of the nodes is consistent. The system extracts the chronological evolution trajectory based on the node evolution graph according to the following trajectory extraction rules: select a node with a high cultural weight or preservation value from the first spatial semantic graph (Graph1) as the starting point of the trajectory. The cultural weight can be determined by comprehensive scoring of the node area, the degree of age, and the historical and cultural level. Select the node corresponding to the starting point in the second spatial semantic graph (Graph2) as the end point of the trajectory. Use the shortest path algorithm (such as Dijkstra algorithm) to search for a connected path in the node evolution graph. The path must satisfy the evolution direction consistency, i.e., the time parameter of the nodes in the path shows an increasing trend. The Euclidean distance between any two consecutive nodes in the path must not exceed a set maximum jump threshold (for example, 100 meters). Perform trajectory validity check, such as the path length must be at least a set number of steps (such as at least 2 steps); the time span and the spatial span must satisfy a reasonable interval (such as a span of at least 30 years and a spatial range of at least 500 square meters). Output the chronological evolution trajectory data.

[0136] Preferably, wherein the trajectory clustering is specifically:

[0137] Perform spatial node extraction and cultural semantic node extraction on the chronological evolution trajectory data to obtain spatial node data and cultural semantic node data;

[0138] In an embodiment, the spatial node extraction traverses each chronological evolution trajectory to extract the spatial position attributes (centroid coordinates (x, y), area, contour shape, etc.) of each node in the trajectory; at the same time, the cultural semantic attributes (category, such as residence / temple / commercial building; age label; functional feature) of each node are extracted; the cultural labels are standardized (using a unified classification system, such as a building category standard code table); and cultural node data is obtained. Output a spatial node data list; and a cultural semantic node data list.

[0139] Perform node pair pairing on the spatial node data and the cultural semantic node data to obtain node pair data;

[0140] In an embodiment, based on the matching between the spatial nodes and the cultural semantic nodes, such as the spatial distance between the two nodes being less than a set threshold (such as 100 meters); the categories being similar or having an evolution association (such as folk house-folk house, traditional building-cultural space, the expert knowledge engine based on expert knowledge for judgment). For each node in the trajectory, the next node that meets the above conditions is found to form a node pair; the spatial distance, category change type, and age span of the node pair are recorded.

[0141] Evolution trajectory data is generated according to the node pair data;

[0142] In an embodiment, the evolution trajectory is constructed by concatenating the node pairs in chronological order to form a directed trajectory path; the trajectory format is node sequence (start point→ intermediate node→ end point). The time sequence must be unidirectionally increasing; the spatial jump control (the distance between consecutive nodes must not exceed a set threshold, such as 150 meters). The trajectory features are recorded as the total number of steps, the spatial span (Euclidean distance from the start point to the end point), the age span, and the evolution type chain (such as culture→ business). The output is a preliminary evolution trajectory data list.

[0143] Trajectory local repair is performed on the evolution trajectory data to obtain trajectory repair data;

[0144] In an embodiment, the trajectory local repair mainly corrects the interruption or mutation problems in the evolution trajectory. The trajectory abnormal scenarios include interruption, which refers to the missing nodes in the trajectory path, causing the trajectory chain to break and unable to form a complete evolution sequence; mutation, which refers to the large change in the node category in the trajectory path, making the evolution process unreasonable, such as the incoherent phenomenon of directly jumping from traditional function to industrial function. To address the above abnormalities, the system uses the following repair strategies: in the position where the trajectory chain is broken, the system searches for nodes that can supplement the connection within the adjacent time segments and spatial range; the K-nearest neighbor (KNN) method is used, combined with the time consistency filtering standard, to preferentially select nodes that are similar in cultural category to the previous and next nodes as the supplementary nodes; the spatial distance and age proximity are used as the priority search indicators for the nodes, and the category similarity is used as the main weight factor for the node completion selection. When the category of consecutive nodes in the trajectory changes abnormally (such as from traditional building to industrial plant), the system attempts to insert a reasonable intermediate transition category node between them; for example, by inserting a node with cultural function attributes, the evolution chain is corrected to “traditional→ cultural→ business”, to restore the evolution rationality and continuity of the trajectory; the selection of the inserted node is also based on the evaluation of spatial proximity, time continuity, and cultural category matching degree.

[0145] Spatial path similarity matrix construction and time span similarity matrix construction are performed on the trajectory repair data to obtain spatial similarity matrix data and time span similarity matrix data, respectively;

[0146] In one embodiment, during the construction of the spatial path similarity matrix, the system calculates the spatial path similarity by comparing the similarity of two trajectories in their spatial distribution. Trajectory Hausdorff distance or Dynamic Time Warping (DTW) is used; smaller trajectory distances indicate higher spatial path similarity; normalization to [0,1] is applied, with a similarity of 1 indicating complete identity. The time span similarity matrix compares the differences in the time span between two trajectories; the time span similarity is defined as: , For trajectory With trajectory Similarity in time span between them For trajectory The span of time, For trajectory The span of time, The maximum value among the two trajectory spans is also normalized to [0,1], where 1 indicates that the time spans are completely identical. Output the spatial similarity matrix and the time span similarity matrix.

[0147] Hierarchical clustering of trajectory repair data is performed based on spatial similarity matrix data and temporal span similarity matrix data to obtain trajectory clustering data.

[0148] In one embodiment, similarity is defined by combining spatial similarity and temporal similarity: ,in For trajectory With trajectory The fusion similarity between them This is the spatial similarity weighting coefficient. For trajectory With trajectory Similarity along spatial paths This is the time similarity weighting coefficient. For trajectory With trajectory Similarity over time. Initially, each trajectory is treated as an independent category. In each round of merging, the system selects the pair of trajectories or trajectory clusters with the highest similarity and performs the merging operation. The clustering process continues until preset clustering stopping conditions are met, including allowing merging only when the overall similarity between the trajectory groups to be merged is not lower than a set threshold (e.g., 0.8). The number of trajectories in each cluster group must not exceed the set maximum limit (e.g., 20 trajectories / cluster) to prevent excessive clustering and resulting in excessive heterogeneity within the groups. Through the above steps, the system outputs trajectory clustering data.

[0149] Preferably, step S3 specifically includes:

[0150] Step S31: Extracting cultural evolution features according to historical and cultural space classification data to obtain cultural evolution feature data, wherein the cultural evolution feature data includes space function evolution feature data, building form change feature data, cultural symbol evolution feature data, and time evolution trajectory feature data;

[0151] In an embodiment, the space function evolution feature data extraction is to extract the function type corresponding to each trajectory node to form a space function evolution sequence (such as: “dwelling”→“shop”→“office building”). For the function transition chain, a sparse coding vector can be constructed for each function category, or a function category Embedding method can be used to map the function category to a low-dimensional dense vector (such as a 64-dimensional embedding representation). The building form change feature data extraction is to calculate the following change rate indicators for the building form characteristics between trajectory nodes, including area change rate, contour compactness change rate (contour compactness at a later time minus contour compactness at an earlier time, divided by contour compactness at an earlier time), and height change rate. The cultural symbol evolution feature data extraction is to extract the change process of building style labels over time (such as: “B-class style building”→“C-class style building”). The symbol change feature is calculated, and is defined as follows: , wherein is the cultural symbol evolution feature data, is the style similarity, which can be inferred by a pre-trained model (such as a CLIP image-text matching model), and is the style retention rate (such as the degree of element detail retention, based on the ratio between elements at a later time and elements at an earlier time). The time evolution trajectory feature data extraction is to extract the following time features for the trajectory evolution process, time span, such as the latest node year minus the earliest node year; time continuity index, such as the average value of the interval between adjacent nodes in the trajectory; node density change trend, such as the change trend of the number of nodes per unit year, which can be fitted by simple linear regression to obtain the slope of the node density change.

[0152] Step S32: Generating a multi-path natural evolution sample of the cultural evolution feature data by using a preset historical and cultural space evolution generation model to obtain candidate space regeneration data, wherein the preset historical and cultural space evolution generation model is obtained by generative adversarial training of historical and cultural space classification experience data and space regeneration label data pre-stored locally;

[0153] In an embodiment, the historical and cultural space evolution generation model is a generative adversarial network (GAN) structure or its improved version (such as Conditional GAN, Diffusion-based Generator); the training data is the local pre-stored historical and cultural space classification experience data (real historical evolution samples) and space regeneration label data (actual reconstruction case labels, such as protection / update / replacement). The cultural evolution feature vector is taken as the input condition of the generator (G). The generator G receives the evolution feature and outputs a space regeneration candidate scheme (such as new function category, new building form, and new cultural symbol state). The discriminator D discriminates whether the output conforms to the historical evolution rationality and space regeneration experience rules (based on the experience learning of the training set). Random noise is introduced, and different path natural evolution samples are generated each time; each input evolution feature can generate N candidate samples (such as N=5 different evolution paths). The loss function includes the adversarial loss, the feature matching loss, and the path natural coherence loss. After the above generation process, each cultural evolution feature input can output N natural evolution paths, forming a candidate space regeneration data set.

[0154] Step S33: performing space coherence test on the candidate space regeneration data to obtain space regeneration data.

[0155] In an embodiment, the generated space regeneration scheme is guaranteed to be coherent and reasonable in geographical space distribution, cultural evolution logic, and form structure. The spatial adjacency between the trajectory nodes (the centroid Euclidean distance is continuously less than a set threshold, such as 100 meters); isolated nodes or large jump phenomena are not allowed (the jump rate is controlled within 5%). The function evolution chain logic is reasonable, and there is no sudden break (such as a residential building directly jumping to an industrial factory, and a commercial / office transition should be inserted). The area change, contour change, and building elevation change should be gentle (that is, the change rate is within a preset threshold range); a threshold is set, such as a single-step area growth change rate <50%. The building style transition is reasonable (such as B-class style→C-class style is acceptable, and B-class style→D-class style has low confidence and needs to be removed, which is judged based on a preset expert engine). The test is performed on each candidate space regeneration trajectory; the unqualified samples are discarded or repaired. The space regeneration data that conforms to all test standards is output as the formal space regeneration generation result.

[0156] Preferably, the construction steps of the preset historical and cultural space evolution generation model include the following steps:

[0157] The pre-stored historical and cultural space classification experience data is processed by a feature unfolding layer to obtain feature unfolding data;

[0158] In an embodiment, the system performs feature unfolding processing on the locally pre-stored historical and cultural space classification experience data to generate feature unfolding data. The historical and cultural space classification experience data is derived from a locally stored classification experience library, which records the functional category, age span, morphological change characteristics and cultural symbol change attributes of each space unit. The feature unfolding rules specifically include decomposing the building function change field recorded in the original classification data into two basic fields of starting functional category and ending functional category; decomposing the time evolution field into three basic fields of starting age, ending age and age span, wherein the age span is defined as the difference between the ending age and the starting age; decomposing the cultural symbol change field into two basic fields of starting style and ending style; decomposing the space morphological change field into three basic fields of area change rate, height change rate and compactness change rate, wherein the area change rate is defined as the ending area minus the starting area divided by the starting area, and the height change rate and the compactness change rate are also calculated in a similar manner. Numerical normalization is performed on continuous features to the [0, 1] interval. The output is structured and uniformly dimensioned feature unfolding data.

[0159] Deep feature transformation processing is performed on the feature unfolding data to obtain deep feature data;

[0160] In an embodiment, the system constructs a deep feature extraction structure by adopting a stacked multi-layer nonlinear mapping unit, and the nonlinear mapping unit adopts a multi-layer perceptron (MLP) structure. An activation function is connected after each mapping layer for nonlinear transformation, and the activation function selects one of a rectified linear unit (ReLU) or a Gaussian error linear unit (GELU). In terms of the number of layers, the system preferably sets the deep feature transformation network to 3 to 5 layers. The feature dimension is gradually compressed between layers, for example, from an initial 1024 dimensions to 512 dimensions, and then to 256 dimensions. A Dropout mechanism is introduced after the output of each hidden layer. The Dropout operation randomly discards part of the neuron outputs with a zeroing ratio of 0.2.

[0161] Structured feature generation is performed on the deep feature data to obtain structured feature data;

[0162] In an embodiment, the system rearranges (rearranges) a one-dimensional deep feature vector corresponding to each sample into a two-dimensional matrix structure. For example, when the original deep feature vector length is 256, it is rearranged into a 16x16 two-dimensional feature map, thereby forming a 16x16 feature unit distribution in space. In the rearrangement process, the system preferably maintains the local correlation between feature units, i.e., the feature content represented by adjacent units maintains continuity in semantics, for example, which can correspond to different attribute changes in the evolution process of the same building, such as area expansion, function migration, cultural symbol change, etc.

[0163] performing feature up-sampling processing according to the structured feature data to obtain feature up-sampling data;

[0164] In an embodiment, the system uses up-sampling operation to increase the size of the feature map. Preferably, the feature magnification is achieved by using de-convolution operation or sub-pixel convolution technology. The de-convolution operation expands the size of the feature map by modeling the standard convolution operation in reverse, and the sub-pixel convolution expands the size of the feature map by rearranging the sub-pixel structure. The magnification ratio is set to 2 or 4, for example, when the input feature map size is 16x16, the up-sampling processing can increase the spatial size to 32x32 or 64x64. During the up-sampling process, the system pays special attention to avoid the checkerboard effect caused by the up-sampling operation, i.e. the local artifact phenomenon caused by uneven up-sampling. To this end, the system introduces a smoothing convolution layer as a post-processing operation after up-sampling, which further smooths the feature map through small stride standard convolution to eliminate the structural noise.

[0165] performing convolution processing according to the feature up-sampling data to obtain candidate spatial regeneration sample data;

[0166] In an embodiment, the system uses a multi-layer stacked two-dimensional convolution (Conv2D) structure to extract and reconstruct features based on the input feature up-sampling data. Each convolution layer uses a 3x3 convolution kernel, the stride is set to 1, and the boundary padding (Padding) is set to 1. The convolution network is preferably stacked with 3 to 5 layers to gradually extract high-level spatial features. After each convolution operation, the system introduces batch normalization processing and then applies a rectified linear unit (ReLU) activation function. At the output end of the convolution network, the system generates multiple feature maps. Each feature map corresponds to a candidate spatial regeneration state, including but not limited to candidate spatial layout features, functional configuration features, or cultural symbol distribution features, etc. The feature maps together constitute a preliminary spatial evolution rationality candidate sample set.

[0167] performing convolution coding block processing on the candidate spatial regeneration sample data and the spatial regeneration label data to obtain convolution coding block data;

[0168] In an embodiment, the system adopts a convolutional encoder structure with shared weights to encode the candidate sample data and the real label data respectively. The so-called shared weights refer to the same set of convolutional encoder parameters being applied to the encoding process of the sample data and the label data, ensuring that the two types of data are expressed in the same feature space. The structure of the convolutional encoder includes the following processing units connected in series, performing two-dimensional convolution (Conv2D) operation with a convolution kernel size of 3x3, a stride of 1, and a boundary padding of 1; applying batch normalization processing to standardize the feature distribution; then applying a rectified linear unit (ReLU) activation function; performing a two-dimensional convolution operation again, with a convolution kernel size of 3x3, a stride of 1, and a boundary padding of 1; performing a flattening operation to flatten the two-dimensional feature map into a one-dimensional feature vector, and unifying the length of the encoded output feature vector, for example, to 512 dimensions.

[0169] According to the convolutional encoding block data, feature tightening layer processing is performed to obtain feature tightening layer data.

[0170] According to the convolutional encoding block data, feature tightening layer processing is performed to obtain feature tightening layer data.

[0171] In an embodiment, the system introduces one or more fully connected layers as a feature tightening unit. The input convolutional encoding feature vector (for example, 512 dimensions) is linearly transformed and compressed to a lower dimension, such as 128 dimensions or 64 dimensions. After the output of each fully connected layer, the system applies a nonlinear activation function processing, preferably using a rectified linear unit (ReLU) or a Gaussian error linear unit (GELU). The system can introduce a Dropout processing mechanism after the fully connected layer. Dropout randomly sets part of the neuron outputs to zero with a preset proportion during the training process.

[0172] According to the feature tightening layer data, authenticity scoring layer processing is performed to obtain sample authenticity scoring data.

[0173] In an embodiment, the system introduces a single-node fully connected output layer to generate the final authenticity score value. The output layer uses a Sigmoid activation function for nonlinear processing, which can compress the input features to the closed interval [0, 1], ensuring that the output score has a standardized probability interpretation. In the authenticity scoring result, if the score value is close to 1, it indicates that the generated sample is judged to be highly close to the real sample; if the score value is close to 0, it indicates that the generated sample is judged to be a fake or distorted sample. The system quantitatively evaluates the authenticity degree of the candidate sample based on the score value.

[0174] According to the sample authenticity scoring data, the candidate spatial regeneration sample data is iteratively trained to obtain a historical and cultural spatial evolution generation model.

[0175] In one embodiment, a dual-model structure is established, where the generator (G) aims to generate spatially reproduced samples that more closely resemble reality, maximizing the sample authenticity score; the discriminator (D) aims to correctly distinguish between real and generated samples, accurately determining the authenticity of the input data. During training, the system uses the standard GAN loss function as the optimization objective. Specifically, the generator G attempts to minimize the following loss function, while the discriminator D attempts to maximize it: ,in To counteract the impact of optimizing the target area based on game optimality, maximizing the discriminator, and minimizing the generator, The expected value of the logarithmic prediction when the true sample x is expected to follow the true data distribution is added to the expected value of the logarithmic inverse prediction of the noisy sample z after it has been generated by generator G. The expected value of the loss term for the true sample. To sample x from the real data distribution, The discriminator's predicted value for x. For the true data distribution, the sample The true probability distribution in which it is located. for Log probability value, To generate the expected value of the sample loss term, Let G be the "true" prediction probability of the discriminator for this generated sample. An alternating update strategy is adopted, that is, after the discriminator D is updated 5 times during training, the generator G is updated once to maintain a dynamic balance between generation and discrimination training. The optimizer used is the Adam optimizer, an adaptive learning rate optimization algorithm based on first-order moment estimation and second-order moment estimation, with the initial learning rate set to [value missing]. The training process monitors the mean score of the authenticity of the generated samples. When the mean score reaches or exceeds a set threshold (e.g., 0.85), the model is considered to have converged.

[0176] Preferably, step S4 specifically includes:

[0177] Step S41: Obtain urban planning needs data;

[0178] In an embodiment, the acquired urban planning requirement data includes the following: the latest urban planning text data, such as land use planning, functional zoning, traffic system arrangement, and public facility layout; vector format planning map data, such as land red line map and road system map; planning index requirements, such as green space rate not less than 30%, building height limit within 80 meters, and commercial land proportion requirement between 20% and 40%. The acquired urban planning requirement data is unified into a structured format, including attribute data table in CSV or JSON file format, and spatial data file in GeoJSON or Shapefile format. Each planning unit must be attached with the following necessary attribute fields, such as land use type, building height limit, volume rate range, and public service facility matching index.

[0179] Step S42: attribute mapping is performed according to the urban planning requirement data and the spatial regeneration data to obtain spatial planning mapping data.

[0180] In an embodiment, the mapping process aims to attribute corresponding and relationship hanging of each regeneration space unit and the corresponding urban planning unit. In the attribute mapping process, the following rules are followed: spatial position matching, the center point or main range of the regeneration space unit must completely fall within a certain urban planning land area, serving as the basis for spatial correspondence; attribute field matching, the main attribute fields of the corresponding planning unit, including land use type, building height limit, and volume rate range, are assigned and hung to the regeneration unit; function adaptation detection, the original function attribute of the regeneration unit is recorded and compared with the planning function requirement of the corresponding urban planning unit, and a matching state label is marked according to the comparison result, which is divided into two categories: matched and conflict. In the case of regeneration unit crossing multiple urban planning units, the following details are adopted: area proportion priority, the planning unit with the largest coverage area of the regeneration unit is taken as the main attribute source; secondary attribute recording, the secondary coverage area and the corresponding planning attribute are recorded as auxiliary reference.

[0181] Step S43: conflict detection is performed on the spatial planning mapping data to obtain spatial conflict detection data.

[0182] In an embodiment, field coincidence degree calculation is performed on the urban planning requirement data and the spatial regeneration data, and the data not existing in the urban planning requirement data in the spatial regeneration data is marked to obtain the spatial conflict detection data.

[0183] In an embodiment, the system maps the urban planning data and the regeneration data of each spatial region based on a preset electronic map template, and performs a field comparison operation on each mapped region. If there is an inconsistency between the key attribute fields (according to the field name, code or preset semantic standard), it is determined that there is a spatial conflict. The comparison fields and conflict triggering conditions are supported by a pre-defined attribute mapping rule table for configuration and adjustment.

[0184] In an embodiment, after attribute mapping, the system labels the regions that do not meet the planning conditions based on preset determination rules (such as functional compatibility standards, planning index thresholds, etc.). The labeling results include conflict types, conflict fields and recommended processing methods.

[0185] In an embodiment, the status attributes of the regeneration space unit and the planning attributes of the corresponding urban planning unit are inconsistent, or the regeneration unit violates the requirements of the urban planning index. The conflict detection content includes the following aspects, including functional conflict, which refers to the original function of the regeneration space unit and the preset function classification of the planning unit are different, and the function categories are incompatible, for example, the function between the historical street current use and the planning industrial land use is inconsistent, and the inconsistency / non-overlapping degree of the function field is mapped to [0, 1]; form conflict, which refers to the building height or volume rate of the regeneration space unit design exceeds the limit value set by the corresponding urban planning unit, for example, the volume rate is between 2.2-2.5, which is set as a slight conflict area, and the system outputs the conflict membership μ ∈ [0, 1]; layout conflict, which refers to the spatial layout of the regeneration space unit overlaps with the layout of the urban planning red line and the road system, or violates the planning boundary regulation, and the overlapping degree of the planning red line is output in numerical form. Conflict detection follows the following rule system, including functional compatibility detection, which is determined according to a preset functional compatibility table, for example, commercial functions can be compatible with historical streets, and industrial functions are not compatible; planning index hard constraint detection, which sets hard constraint conditions for building volume rate, building height and public supporting facility configuration index, and if it exceeds the upper limit, it is considered as a conflict, and is mapped to [0, 1] according to the degree of inconsistency. The conflict detection list includes the following contents, including conflict type (functional conflict, form conflict, layout conflict); conflict degree (classified according to conflict severity, such as slight conflict, moderate conflict and serious conflict); and the list of spatial units involved and corresponding attribute information.

[0186] Step S44: Cooperatively adjusting the spatial conflict detection data to obtain spatial adjustment data;

[0187] In one embodiment, the system constructs a structured representation for all conflicting spatial units, integrating their attribute vectors (such as area, building density, and land use records) with spatial adjacency relationships. This is then encoded using a graph neural network or spatial autoencoder network to obtain a low-dimensional representation vector of the conflict region. Based on the vector representation of the conflict region, unsupervised clustering (such as DBSCAN or Spectral Clustering) is used to automatically identify conflict pattern categories. For example, cluster labels are implicitly categorized into high-density conflict, functional mismatch conflict, and spatial boundary compression conflict (based on the ratio judgment between cluster centers generated by clustering, or based on a threshold judgment using preset expert knowledge, or relying on unsupervised combination analysis of inter-cluster distance, intra-cluster compactness, and attribute mean to form semantic labels such as high-density conflict, functional mismatch conflict, and spatial boundary compression conflict). The system constructs an optimization function based on spatial reachability, node connectivity, and attribute distribution balance, or based on one of the aforementioned specific indicators, to evaluate and rank candidate adjustment schemes for the conflict region. The optimization model uses an evolutionary search algorithm (such as genetic algorithm or particle swarm optimization) to automatically generate the optimal adjustment strategy vector from the candidate strategy set. Based on the optimized output, the spatial morphology, functional distribution, and boundary relationships of the conflict areas are structurally reconstructed, including adjustments to building volume, functional layout relocation, and boundary coordination. The system records all adjustment operations as an operation log and automatically generates a spatial adjustment dataset, including the original state, adjusted state, adjustment vector parameters, and affected area for each modified area.

[0188] Step S45: Perform spatial layout fusion on the spatial adjustment data to obtain spatial regeneration fusion data.

[0189] In one embodiment, the system constructs a spatial graph model based on adjusted spatial unit data, where nodes represent spatial units and edges represent spatial adjacency or functional relationships. Each node is accompanied by its attribute vector, including building scale, spatial type, connectivity weight, historical preservation identifier, etc., uniformly embedded as a vector space representation. Based on the spatial graph, the system calculates the reachability indices for each node pair, such as pedestrian connectivity and interaction path length. Shortest path search and graph structure diffusion algorithms are used to obtain the potential reachability matrix of the global space. The system constructs an evolution function (...). ,in For spatial layout objectives, The loss weight is optimized for local connectivity, with a value of 0.2. To optimize the loss for local connectivity, This represents the weight for the spatial functional distribution balance loss, with a value of 0.2. For the loss of spatial functional distribution balance, The weight for the continuity loss of building density is 0.5. for building density continuity loss, for semantic clustering consistency loss weight, taking 0.1 as a value, for semantic clustering consistency loss), automatically define the spatial layout target, and the evolution function parameters include a local connectivity enhancement target (optimizing the adjacent path), a spatial distribution balance (preventing functional polarization), building density continuity, and spatial semantic clustering consistency (derived from the self-supervised semantic clustering result). The optimization function is automatically learned through historical spatial sample distribution, self-supervised graph learning, or latent variable modeling. An evolutionary search algorithm (such as genetic algorithm, simulated annealing, and graph embedding perturbation optimization) is used to iteratively update the spatial node position and connection mode. The system dynamically selects the optimal fusion layout path by minimizing the deviation of the target function. In each round of evolution, edge movement, region merging, and function switching between nodes can occur to achieve self-organizing optimization of the layout. After the fusion is completed, the system outputs the spatial fusion graph corresponding to the final layout scheme, which contains the new spatial position, function label, and connection structure of all nodes. At the same time, the system automatically generates change annotation data to record the structural changes, evolution process, and fusion path from the initial layout to the fused layout.

[0190] Preferably, the application also provides a historical and cultural oriented space classification and regeneration system for performing the historical and cultural oriented space classification and regeneration method as described above, and the historical and cultural oriented space classification and regeneration system comprises:

[0191] a historical and cultural data processing module configured to acquire historical and cultural data, perform multi-modal perception on the historical and cultural data to obtain historical and cultural space data, and perform deep feature extraction on the historical and cultural space data to obtain historical and cultural element data;

[0192] a space semantic modeling and trajectory classification module configured to perform space semantic graph construction on the historical and cultural element data to obtain space semantic graph data, and perform chronological evolution trajectory classification on the space semantic graph data to obtain historical and cultural space classification data;

[0193] a space regeneration generation module configured to perform space regeneration generation according to the historical and cultural space classification data to obtain space regeneration data;

[0194] a space regeneration fusion module configured to acquire urban planning requirement data, and perform space regeneration fusion according to the urban planning requirement data and the space regeneration data to obtain space regeneration fusion data.

[0195] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the application is defined by the attached application file rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the application file are intended to be included in the application.

[0196] The foregoing is considered as illustrative only of the principles of the application. Numerous modifications and changes will readily occur to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the scope of the application is indicated by the appended claims rather than by the foregoing description, and all changes that come within the meaning and range of equivalents are intended to be embraced therein.

Claims

1. A historical and cultural oriented space classification and regeneration method, characterized in that, The method comprises the following steps: Step S1: obtaining historical and cultural data, and performing multi-modal perception on the historical and cultural data to obtain historical and cultural space data; performing deep feature extraction on the historical and cultural space data to obtain historical and cultural element data; Step S2: constructing element nodes and spatial relationship edges on the historical and cultural element data to obtain element node data and spatial relationship edge data respectively; generating a spatial semantic graph based on the element node data and the spatial relationship edge data to obtain spatial semantic graph data; extracting an age evolution track based on the spatial semantic graph data to obtain age evolution track data; clustering the age evolution track data to obtain track clustering data; classifying the age evolution track data based on the track clustering data to obtain historical and cultural space classification data; Step S3: extracting cultural evolution characteristics based on the historical and cultural space classification data to obtain cultural evolution characteristic data, wherein the cultural evolution characteristic data comprises spatial function evolution characteristic data, architectural form change characteristic data, cultural symbol evolution characteristic data, and time evolution track characteristic data; generating a multi-path natural evolution sample based on the cultural evolution characteristic data by using a preset historical and cultural space evolution generation model to obtain candidate space regeneration data, wherein the preset historical and cultural space evolution generation model is obtained by performing generative adversarial training based on historical and cultural space classification experience data and space regeneration label data pre-stored locally; performing spatial coherence testing on the candidate space regeneration data to obtain space regeneration data; Step S4: obtaining city planning requirement data, and performing space regeneration fusion based on the city planning requirement data and the space regeneration data to obtain space regeneration fusion data.

2. The method of claim 1, wherein, Step S1 specifically comprises: obtaining historical and cultural data; performing multi-modal feature extraction on the historical and cultural data to obtain multi-modal feature data; performing feature splicing on the multi-modal feature data to obtain feature splicing data; performing self-attention encoding on the feature splicing data to obtain feature attention data; performing fusion reasoning on the feature attention data to obtain historical and cultural space data; performing deep feature extraction on the historical and cultural space data to obtain deep feature data; annotating the historical and cultural space data based on the deep feature data to obtain historical and cultural element data.

3. The method of claim 2, wherein, The fusion reasoning specifically comprises: performing cross-modal correlation processing on the feature attention data to obtain cross-modal correlation data; performing modal consistency optimization processing on the cross-modal correlation data to obtain feature modal optimization data; performing semantic completion reasoning on the feature modal optimization data to obtain feature modal completion data; performing spatial structure consistency fusion on the feature modal completion data to obtain historical and cultural space data.

4. The method of claim 1, wherein, The spatial semantic graph data comprises first spatial semantic graph data and second spatial semantic graph data, the time parameter corresponding to the first spatial semantic graph data is not greater than the time parameter corresponding to the second spatial semantic graph data, and the age evolution track extraction specifically comprises: The first space semantic graph data and the second space semantic graph data are time-constrained node association and pairing to obtain time-constrained node pair data; The time-constrained node pair data is subjected to node pair evolution relationship judgment to obtain node evolution data; According to the node evolution data, the space connectivity is connected to obtain the age evolution trajectory data.

5. The method of claim 1, wherein, The trajectory clustering is specifically: The space node data and the cultural semantic node data are obtained by extracting the space node data and the cultural semantic node data from the age evolution trajectory data; The node pair data is obtained by pairing the space node data and the cultural semantic node data; According to the node pair data, the evolution trajectory data is generated to obtain the evolution trajectory data; The trajectory repair data is obtained by repairing the local trajectory of the evolution trajectory data; The space path similarity matrix and the time span similarity matrix are constructed to obtain the space similarity matrix data and the time span similarity matrix data, respectively; The trajectory clustering data is obtained by hierarchical clustering the trajectory repair data according to the space similarity matrix data and the time span similarity matrix data.

6. The method of claim 1, wherein, The construction steps of the preset historical and cultural space evolution generation model include the following steps: The feature expansion data is obtained by performing feature expansion layer processing on the historical and cultural space classification experience data pre-stored locally; The deep feature data is obtained by performing deep feature transformation processing on the feature expansion data; The structured feature data is obtained by performing structured feature generation on the deep feature data; The feature up-sampling data is obtained by performing feature up-sampling processing on the structured feature data; The candidate space regeneration sample data is obtained by performing convolution processing on the feature up-sampling data; The convolution coding block data is obtained by performing convolution coding block processing on the candidate space regeneration sample data and the space regeneration label data; The feature tight layer data is obtained by performing feature tight layer processing on the convolution coding block data; The sample authenticity score data is obtained by performing authenticity scoring layer processing on the feature tight layer data; The historical and cultural space evolution generation model is obtained by iteratively training the candidate space regeneration sample data according to the sample authenticity score data.

7. The method of claim 1, wherein, Step S4 is specifically: Obtain city planning requirement data; Attribute mapping is performed on the city planning requirement data and the space regeneration data to obtain space planning mapping data; The space planning mapping data is subjected to conflict detection to obtain space conflict detection data; The space conflict detection data is subjected to collaborative adjustment to obtain space adjustment data; The space adjustment data is subjected to space layout fusion to obtain space regeneration fusion data.

8. A historical and cultural oriented space classification and regeneration system, characterized in that, The historical and cultural oriented space classification and regeneration system for executing the method of claim 1 comprises: A historical and cultural data processing module is configured to obtain historical and cultural data, perform multi-modal perception on the historical and cultural data to obtain historical and cultural space data, and extract deep features from the historical and cultural space data to obtain historical and cultural element data; The space semantic modeling and trajectory classification module is configured to construct a space semantic graph based on the historical and cultural element data, to obtain space semantic graph data, and to classify the space semantic graph data according to time evolution trajectories, to obtain historical and cultural space classification data. The space regeneration generation module is configured to generate space regeneration based on the historical and cultural space classification data, to obtain space regeneration data. The space regeneration fusion module is configured to obtain city planning demand data, and to fuse the space regeneration based on the city planning demand data and the space regeneration data, to obtain space regeneration fusion data.

Images