CN-122020549-A - Urban landscape element intelligent extraction method based on multi-source heterogeneous data fusion

CN122020549ACN 122020549 ACN122020549 ACN 122020549ACN-122020549-A

Abstract

The invention discloses an intelligent extraction method of urban feature elements fused by multi-source heterogeneous data, which comprises the steps of dividing a vein line, determining a vein influence point, calculating a vein influence degree, screening vein nodes, extracting multi-source heterogeneous urban information, preprocessing, mapping the vein, carrying out feature processing to obtain apparent urban features, clustering the apparent urban features, calculating feature differences to determine feature uniqueness, taking intersection sets of different apparent urban features, calculating association weights and vein fusion degrees among the apparent urban features, and carrying out weighted fusion to obtain urban vein features, and extracting integral urban feature elements of different periods of a target city according to the vein node order to obtain a city feature element set based on vein development. The method can provide a scientific, systematic and operable quantitative analysis tool for cognition, evaluation, protection and inheritance of urban landscapes, and effectively promote urban landscape management to go from experience leading to data driving and intelligent decision making.

Inventors

LI XIANGNING
YE YU
GAO CHANGJUN
ZHANG HUALI
WU JINGFEN
KANG SHANZHI
LI YAN
YUAN KUN

Assignees

同济大学

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (6)

1. The intelligent extraction method of the urban landscape elements fused by the multi-source heterogeneous data is characterized by comprising the following steps of: S1, dividing the venation lines according to the city development history, extracting venation influence events of a target city in each venation line to determine venation influence points, calculating venation influence degree of each venation influence point according to the venation influence events, and screening venation nodes; S2, extracting multi-source heterogeneous city information among the vein nodes, preprocessing, performing vein mapping to obtain vein feature semantic vectors, and performing feature processing on the vein feature semantic vectors to obtain apparent city features; S3, clustering apparent urban landscapes of the target city, calculating the landscape difference between each landscape cluster and the neighborhood landscape cluster, determining the uniqueness of the landscape and associating the landscape clusters with the corresponding landscape clusters; s4, taking intersections of different apparent urban landscapes in the same landscape cluster, calculating association degree weights among the apparent urban landscapes according to intersection results, determining the context fusion degree of the corresponding landscape cluster according to the association degree weights, and carrying out weighted fusion to obtain the urban context landscape; S5, urban feature elements of a feature cluster are formed by feature uniqueness, context fusion degree and urban context features, and the integral urban feature elements of the target cities in different time periods are extracted according to the context node sequence, so that a city feature element set based on context development is obtained; The venation lines comprise ancient venation lines, near modern venation lines and modern venation lines, and the venation influencing events comprise dynasties alternation, administrative area adjustment, population policy, urban construction policy and development planning; the multi-source heterogeneous city information comprises text information, image data, GIS data and sensor data; the apparent city features include climate features, building features and humane activity features; The integral city landscape elements consist of landscape elements of all landscape clusters in the same period.
2. The method for intelligently extracting urban feature elements through multi-source heterogeneous data fusion according to claim 1, wherein the method for screening out the vein nodes comprises the following steps: dividing the venation lines into ancient venation lines, near-modern venation lines and modern venation lines according to the city development history, and extracting venation influence events and corresponding event characteristics of a target city in each venation line, wherein the event characteristics comprise influence range grades, authority grades, duration, adjustment strength and associated item quantity; Taking a time point corresponding to the venation influence event as a venation influence point, and calculating venation influence degree corresponding to the venation influence point according to the event characteristics, wherein the expression is as follows: ; Wherein the method comprises the steps of In order to achieve the effect of the text pulse, 、、 As a result of the feature event weights, In order to influence the range class, For the authority level of the authority level, For the duration of time it is possible, In order to adjust the strength of the steel sheet, In order to correlate the number of items, As a coefficient of the decay in time, For the current point in time of the evaluation, Is the event occurrence time point; and screening the venation influence points with the venation influence degree larger than the influence degree threshold as venation nodes on the venation line.
3. The method for intelligently extracting urban feature elements through multi-source heterogeneous data fusion according to claim 1, wherein the method for obtaining the apparent urban feature comprises the following steps: extracting multi-source heterogeneous city information of a target city in a time period between the venation nodes and preprocessing; constructing a text vein knowledge graph according to the historical text vein information Wherein As a set of entities, The entity set comprises buildings, streets, public space entities, climates, terrains, life convenience entities and residents, wherein the relation set comprises space-time relations, derivative relations, influence relations and characterization relations; Inputting the pretreated multi-source heterogeneous city information into a cross-mode encoder for feature mapping, and splicing the multi-source heterogeneous city information with an embedded vector of a text vein knowledge map to obtain a text vein feature semantic vector; the text vein style semantic vector expression is: ; ; Wherein the method comprises the steps of Is the first Group sample No Class data source Is characterized by that the Chinese character and vein feature semantic vector, As a function of the encoder, For encoder parameters, obtained by multi-modal pre-training, Is an entity The context aware embedded vector of (1) is obtained by focusing on the information of the network aggregation neighbor entity, As a function of the non-linear activation, Is an entity Is a set of neighbor entities of (a), In order for the attention to be weighted, As a matrix of linear transformations that can be learned, Is the initial feature vector of the neighbor entity.
4. The method for intelligently extracting urban landscape elements through multi-source heterogeneous data fusion according to claim 1, wherein the method for determining the uniqueness of the landscape comprises the following steps: The feature processing comprises feature extraction, SVM feature classification and space-time matching; extracting urban basic map information of a period corresponding to the venation nodes, dividing and setting administrative region boundaries according to administrative regions corresponding to the period between the venation nodes, and acquiring administrative boundary vector data of street/village and town levels; dividing cities into different-scale landscape clusters by adopting a three-level clustering strategy, wherein the specific steps are (1) dividing administrative units according to administrative boundary vector data, (2) establishing regular grids in the administrative units with the area larger than a set threshold value, calculating the grid density of the regular grids, reducing the grid size of a high-density grid area and keeping the grid size of a low-density area, wherein the grid density is calculated according to the building density, population density and land mixing degree, (3) carrying out morphological similarity evaluation on adjacent grids, merging the highly similar regular grids, and finally outputting optimized grids, wherein all apparent urban landscape in the optimized grids is taken as the landscape clusters; Calculating climate environment feature differences of all the feature clusters and the neighborhood feature clusters by adopting a weighted Euclidean distance, calculating building feature differences of all the feature clusters and the neighborhood feature clusters by adopting a method of combining a Jacard distance and a cosine distance, measuring distribution differences of all the feature clusters and the neighborhood feature clusters by adopting KL divergence to determine human activity feature differences, and taking average weighted values of the climate environment feature differences, the building feature differences and the human activity feature differences as the feature differences of all the feature clusters and the neighborhood feature clusters; calculating the mean value of the feature differences of each feature cluster and all neighborhood feature clusters, and quantitatively obtaining the feature uniqueness of the feature clusters, wherein the expression is as follows: ; Wherein the method comprises the steps of Is that The uniqueness of the wind aspect clusters, Is that The mean value of the differences of the features of the feature cluster and all the neighborhood feature clusters, Is the amplification factor.
5. The intelligent extraction method for urban feature elements fused by multi-source heterogeneous data according to claim 1, wherein the method for calculating the association degree weight between the apparent urban features comprises the following steps: Taking intersection sets of different apparent city features in the same feature cluster, determining intersection degrees of the apparent city features according to intersection results, calculating mutual information entropy of the apparent city features, and determining association weights among the apparent city features according to the intersection degrees and the mutual information entropy, wherein the expression is as follows: ; ; ; Wherein the method comprises the steps of Is apparent to the urban landscape And (3) with Is used to determine the weight of the (c) in the (c), Is the mutual information entropy of apparent urban landscape, To show the degree of intersection of urban landscapes, As the degree of sensitivity coefficient of the intersection, 、 Is of the form of a wind And (3) with Is set of feature classes of (c) in the set, 、 Is of the form of a wind And (3) with Is a set of feature keywords of (1), Is of the form of a wind And (3) with The time difference of the main change of the intersection characteristic of (c), In order to be a time decay constant, Time series of two types of features And (3) with Is used for the correlation coefficient of (a), Is characterized by two kinds of characteristics And (3) with The area of the space that is co-existing, For the total spatial area in which the two types of features occur, The cross-ratios of the meshes are related to the two types of features, 、、、 As the degree of intersection weight, As a probability distribution of the degree of intersection, 、 Edge probability, which is the degree of intersection.
6. The method for intelligently extracting urban feature elements through multi-source heterogeneous data fusion according to claim 1, wherein the method for obtaining urban venation features comprises the following steps: Determining an association weight matrix according to the association weights among the apparent urban landscapes, and calculating the context fusion degree of the corresponding landscape cluster by the association weight matrix, wherein the expression is as follows: ; Wherein the method comprises the steps of As a cluster of features the degree of fusion of the text and the pulse, For the matrix trace, the overall intensity of the weights is reflected, In order to associate the weight matrix with the weight matrix, Frobenius norms for matrix asymmetry; Calculating attention coefficients of the target apparent city feature and the neighbor apparent city feature through the graph attention network and the associated weights, and weighting and aggregating the neighbor apparent city feature to the target apparent city feature according to the attention coefficients to obtain the city vein feature; And extracting the integral city feature elements of different periods of the target city according to the sequence of the character nodes to obtain a city feature element set based on character development.

Description

Urban landscape element intelligent extraction method based on multi-source heterogeneous data fusion Technical Field The invention relates to the technical field of urban landscape protection and planning, in particular to an intelligent urban landscape element extraction method based on multi-source heterogeneous data fusion. Background The urban feature is taken as the comprehensive expression of the spatial morphology and cultural connotation of urban substances, is an important carrier for the urban historical vein and the contemporary development, scientifically and accurately identifies and extracts the urban feature elements, has important significance for continuing urban features, guiding protection planning and implementing accurate management and control, and simultaneously provides a new direction for urban feature research through the development of remote sensing, geographic information systems, big data and artificial intelligence technologies. The method mainly comprises the following steps of (1) carrying out analysis on most methods based on current situation data, not carrying out time sequence deconstruction on cities in a history long river, leading to text vein de-embedding and difficult to distinguish the history lamination and the contemporary modeling of the wind feature, (2) carrying out data island problem, wherein the existing scheme has single data source, can not effectively fuse multi-source heterogeneous data such as text history materials, geographic information, image vision, real-time sensors and the like, and has insufficient integral and complexity characterization on the wind feature, (3) carrying out analysis on the dimension singleness, wherein the traditional method lacks systematic association analysis and comprehensive evaluation on the multi-dimensional wind feature such as a natural substrate, human activity and the like, and (4) evaluating subjectivity and statization, wherein the judgment of the wind feature and the value lacks a quantifiable and dynamic objective evaluation model, and the dynamic evolution process and the internal coordination relationship of the wind feature are difficult to reflect. Therefore, the invention provides an intelligent extraction method of urban feature elements fused by multi-source heterogeneous data, a time sequence skeleton is constructed by dividing vein lines and quantifying vein influence events, multi-source data are fused to perform cross-modal semantic alignment and feature extraction, and dual quantitative indexes of feature uniqueness and vein fusion degree are innovatively provided, so that a dynamic, structured and interpretable urban feature set is finally formed, a scientific, systematic and operable quantitative analysis tool can be provided for cognition, evaluation, protection and inheritance of urban features, and urban feature management is effectively promoted to move from experience leading to data driving and intelligent decision. Disclosure of Invention The invention aims to provide an intelligent extraction method for urban landscape elements by multi-source heterogeneous data fusion. In order to achieve the above purpose, the invention is implemented according to the following technical scheme: The invention comprises the following steps: dividing the venation lines according to the city development history, extracting venation influence events of a target city in each venation line to determine venation influence points, calculating venation influence degree of each venation influence point according to the venation influence events, and screening out venation nodes; extracting multi-source heterogeneous city information among the vein nodes, preprocessing, performing vein mapping to obtain vein feature semantic vectors, and performing feature processing on the vein feature semantic vectors to obtain apparent city features; clustering apparent urban landscapes of the target city, calculating the landscapes difference between each landscapes cluster and the neighborhood landscapes cluster, determining the uniqueness of the landscapes and associating the corresponding landscapes clusters; taking intersection sets of different apparent urban features in the same feature cluster, calculating association degree weights among the apparent urban features according to intersection results, determining the context fusion degree of the corresponding feature cluster according to the association degree weights, and carrying out weighted fusion to obtain the urban context features; Urban feature elements of the feature cluster are formed by feature uniqueness, context fusion degree and urban context features, and the integral urban feature elements of the target cities in different time periods are extracted according to the context node sequence, so that a city feature element set based on context development is obtained; The venation lines comprise ancient venation lines, near modern venation lines and modern ve