Geographic scene-based demographic data spatialization method

Abstract

The invention relates to a demographic data spatialization method based on geographic scenes, and belongs to the technical field of data processing. In order to solve the problems that current spatial population distribution is difficult to align with objective geographic entities and population distribution weights are not accurate enough, a geographic scene level model of a multi-level space structure is constructed based on more and more complete expression feature data and a spatial clustering method; according to the method, block scene units are divided and classified from a geographic space, a feature expression system of a geographic scene hierarchical structure model is formed in combination with expression features and multi-level space structure division, and the system comprises four categories of'internal features-self features-neighborhood features-regional features'; the multi-level and systematic feature expression mode covers the attributes of the block scene units and the relation between the block scene units and the surrounding environment and the surrounding area, the machine learning model can comprehensively and deeply understand complex factors affecting population distribution, and the accuracy of population density prediction is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a population statistical data spatialization method based on geographic scenes. BACKGROUND

[0002] Population data spatialization is a key issue in the field of digital earth research, which involves converting population statistical data into a form with geographic spatial information for understanding and application. Geographic scenes cover land use types at the scale of urban blocks, such as residential areas, commercial areas, industrial areas, park green spaces, etc., which is a key spatial scale modeling expression of the obvious differences in urban social, economic, ecological and other functions. These different geographic scenes carry their own specific functions and are key data bases for urban planning and management. By taking geographic scenes as spatial units and spatializing and mapping population statistical data, the population distribution in each different geographic scene area can be more clearly understood, so that corresponding disaster prevention and mitigation strategies and measures can be more targeted, and population evacuation, material arrangement, etc. can be better arranged; targeted emergency response strategies can be specified, and urban facility planning can be performed.

[0003] Current population spatialization takes regular spatial grids as units, and spatializes population statistical data by using related geographic elements or data within the grid and the distance of adjacent geographic elements as potential allocation characteristics. There are mainly three problems with this: first, the population distribution is difficult to align with objective geographic entities, and the mixing effect is significant; for example, a regular spatial grid may cover both residential and commercial areas, making it difficult to accurately correspond to objective and real geographic entity scenes. Second, the complex internal structure, topological adjacency, spatial distance and regional characteristics of irregular blocks between geographic scenes are difficult to express through existing population distribution models, resulting in inaccurate population allocation weights. Third, fine geographic scene type attributes that have an absolute impact on population distribution are missing in the expression of population distribution allocation weights; for example, attributes such as whether an area is an industrial or commercial area directly restrict population capacity and population distribution, but if these factors are not fully considered in population allocation weights, there will be deviations between the actual population distribution and the population distribution.

[0004] Therefore, it is necessary to provide an improved technical solution to address the above-mentioned deficiencies of the prior art. SUMMARY

[0005] The purpose of the present application is to overcome the deficiencies of the prior art and provide a population statistical data spatialization method based on geographic scenes.

[0006] In order to achieve the above object, the present application provides the following technical solutions:

[0007] The population statistical data spatialization method based on geographical scenes comprises the following steps:

[0008] S1, geographical scene level modeling;

[0009] The street scene unit is divided in space according to the existing method and data; the expression features of each street scene unit are extracted from the earth observation data and social perception data; the street scene unit is classified by using a supervised machine learning classification method, different geographical scene types are obtained; and a multi-level spatial structure division is performed on the street scene unit by using a spatial clustering method; and a geographical scene level structure model is formed by integrating the classification result of the street scene unit and the multi-level spatial structure division result, which is used for subsequent population density spatialization;

[0010] S2, level structure feature expression;

[0011] Based on the modeling result of the geographical scene level, the features of each street scene unit in the four categories of "internal feature, self feature, neighborhood feature and regional feature" are extracted;

[0012] S3, population density machine learning; based on the four categories of features obtained in S2, the population density value of each street scene unit is predicted by machine learning.

[0013] Further, in step S1, the street scene unit is divided in space according to the existing multi-source geographical information thematic data and the existing street scale land use classification method;

[0014] The expression features of each street scene unit are extracted from the earth observation data and social perception data;

[0015] Street scene unit samples are selected, and the street scene unit is classified by using a supervised machine learning classification method, different geographical scene types are obtained;

[0016] On the basis of the classification result of the street scene unit, a multi-level spatial structure division is performed on the street scene unit by using a spatial clustering method, and a multi-level spatial structure division of the street scene is obtained;

[0017] The classification result of the street geographical scene and the multi-level spatial structure division result are taken as a geographical scene level model, which supports the subsequent level structure feature expression and population density spatialization modeling.

[0018] Further, in step S1, the multi-source geographic information thematic data includes OSM road network, river system, administrative boundary, land cover, building height; the earth observation data includes high-resolution optical / SAR remote sensing image, night / micro-light remote sensing image, multi-spectral / hyper-spectral image; the social perception data includes POI data, mobile signaling data, satellite positioning trajectory data, social media data, public transportation data.

[0019] Further, in step S2, the internal features include volume rate, proportion and arrangement of internal land cover types, night / micro-light image intensity, POI type and density, social media data density and dynamic change amount;

[0020] The self features include type, size, direction and other attribute features of the block scene unit;

[0021] The neighborhood features include the adjacency topological relationship and distance of roads, rivers, administrative centers, transportation hubs, park green spaces, educational facilities, medical facilities, commercial facilities, cultural facilities, industrial areas, and ecological protection areas;

[0022] The regional features refer to the statistical features of the upper-level region containing the current block scene unit to be extracted in the geographical scene hierarchical structure, including the size, direction, shape, arrangement, and combination features of different geographical scene types.

[0023] Further, in step S3, before machine learning prediction, cross-scale training samples are prepared for machine learning model training; the cross-scale training samples use population statistical data above the block scale, and these population statistical data are related to the population statistics of township scale or other administrative units including block scale.

[0024] Further, in step S3, before machine learning prediction, cross-scale training samples are prepared for machine learning model training; the cross-scale training samples use population statistical data above the block scale, and these population statistical data are related to the population statistics of township scale or other administrative units including block scale.

[0025] Further, in step S3, the machine learning model training process based on cross-scale training samples is as follows:

[0026] The four categories of features extracted in step S2 are used as independent variables, and the corresponding block scene unit population density is used as the dependent variable; the predicted density of the block scene unit is multiplied by the area of the block scene unit to calculate the predicted total population, and the predicted total population of the township administrative unit where the block scene unit is located is further calculated.

[0027] A loss function is constructed based on the difference between the predicted total population and the actual statistical population data, which is used to feedback and train the machine learning model, optimize the parameters in the machine learning model, until the preset accuracy is met or the predetermined feedback iteration number is reached.

[0028] Further, in step S3, the machine learning model training process based on the cross-scale training samples is:

[0029] Using the four categories of features extracted in step S2 as independent variables, and the corresponding block scene unit population density as the dependent variable; multiplying the predicted density of the block scene unit by the area of the block scene unit to obtain the predicted total population, and further calculating the predicted total population of the township administrative unit where the block scene unit is located;

[0030] The least squares algorithm is used to solve the regression coefficients of the machine learning model, and the spatial distribution modeling of the block scale population density is realized to predict the block scene unit.

[0031] Further, the population statistical data spatialization method based on geographic scenes further comprises step S4, population density scale correction;

[0032] The cross-scale correction formula is used to further improve the prediction results, and the correction formula is as follows:

[0033]

[0034] In the formula, C i represents the predicted value P i of the population density of the i-th block scene unit, i G j represents the actual statistical population of the township where the i-th block scene unit is located, P j represents the population density prediction value and the area of the j-th block scene in the township where the i-th block scene is located.

[0035] The beneficial effects of the present application are:

[0036] Compared with the current artificial spatialization method using regular spatial grid as the unit, the present application constructs a multi-level spatial structure geographic scene hierarchical model based on more and more complete feature data and spatial clustering method, divides and classifies the block scene unit from the geographic space, and combines the feature expression and multi-level spatial structure division to form a feature expression system of the geographic scene hierarchical structure model, which provides a more accurate basis for subsequent population density spatialization. The feature expression of the geographic scene hierarchical model includes four categories of "internal features, self features, neighborhood features and regional features". This multi-level and systematic feature expression method covers the properties of the block scene unit itself and the relationship with the surrounding environment and the surrounding area, so that the machine learning model can more comprehensively and deeply understand the complex factors affecting the population distribution, and the accuracy of the model is improved;

[0037] The block unit population density prediction result obtained by the method can more finely reflect the distribution of population in geographical space, and provide more accurate and detailed population distribution information for government departments, city planners and related researchers, and help them to make more scientific and reasonable city planning management scheme, disaster prevention and mitigation scheme and the like. BRIEF DESCRIPTION OF DRAWINGS

[0038] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application, and are incorporated herein by reference. The embodiments depicted herein are provided by way of example only, and together with the specification serve to explain the application. In the drawings:

[0039] Figure 1 The flowchart of the present application.

[0040] Figure 2 The block scene unit division schematic diagram of the embodiment of the present application.

[0041] Figure 3 The result schematic diagram of the population spatialization of part of the block scene unit. DETAILED DESCRIPTION

[0042] In order to facilitate the understanding of the technical solutions provided in the present application, first, the related concepts are explained.

[0043] Block scene unit: the block scene unit (neighborhood or block scene unit) is a commonly used analysis unit in the fields of city planning, geographic information system (GIS), environmental science and the like, but its specific definition and boundary division can vary according to the research purpose, data availability and methodology. The present application is to divide the urban area into block scene units using roads, rivers, streets, administrative boundaries and the like after determining the city boundary, and the area of each block scene unit is not uniform. The type of block scene unit is divided according to the standard GB / T 21010-2017 "Land Use Status Classification".

[0044] Geographical scene type: the first level classification system of the third national land survey classification system or the first level class in the national standard GB / T 21010-2017 "Land Use Status Classification" can be referred to, such as the classification of: arable land, garden land, forest land, grassland, commercial and service land, industrial and mining storage land, residential land, public management and public service land, special land, transportation land, water area and water conservancy facilities land, and other land.

[0045] Expression features: the expression features of each block scene unit extracted in S1 are indexes or variables extracted from the earth observation data and social perception data, which can reflect various attributes and characteristics of the block scene unit, and are used to describe and distinguish different block scene units, and provide the basis for subsequent supervised machine learning classification. The expression features at least include volume rate, internal land cover type proportion and arrangement, night / micro-light image intensity, POI type and density, social media data density and dynamic change amount, and can also include road density, bus station number, municipal facility distribution, noise level, cultural relic site protection area, etc.

[0046] Internal land cover type arrangement: the internal land cover type mainly includes farmland, forest, shrub, grassland, water, construction land, bare land, etc. There is no specific standard, and existing classification systems can be referred to for classification, such as Land Use Status Classification (GB / T 21010-2017), Land Space Investigation, Planning, and Land and Sea Use Classification Guide (Trial Implementation) (Natural Resources Administration

[2020] No. 51), and Land Cover Classification System of Chinese Academy of Sciences Remote Sensing Application Institute.

[0047] Internal land cover type arrangement: it is a description and quantification of the distribution and combination of different land cover types in each block scene unit, which is used to reflect the spatial pattern and characteristics of land use in the block. The proximity index, diversity index or aggregation index in the landscape index can be selected. The proximity index can adopt nearest neighbor distance, average nearest distance, average proximity index, and similar proximity. The diversity index can adopt landscape richness, Shannon diversity index (SHDI), Shannon evenness (SHEI) or Simpson diversity index, etc. The aggregation index can adopt aggregation index, cohesion (COHESION) or contagion index (CONTAG), etc.

[0048] Machine learning: the machine learning method is used to classify the block scene in the application. The machine learning model can adopt support vector machine (SVM), decision tree, random forest, neural network, etc.

[0049] Spatial clustering method: the partition clustering in the geographic spatial clustering method is adopted, such as K-means clustering (K-Means Clustering), K-medoids clustering (K-medoids clustering / PAM), CLARA algorithm, K-means++ algorithm (K-Means++ Clustering), etc.

[0050] The multi-level spatial structure division of the block scene: is automatically clustered by a spatial clustering method, and presents a multi-level or hierarchical structure, which is different from the different levels of administrative divisions artificially divided, and is different from the specific name or type of the conventional administrative unit (for example, country-province-city-county-town-village).

[0051] POI type and density: the POI type refers to the POI classification of map software such as Gaode Map and Baidu Map.

[0052] Next, taking the population distribution of a city area as an example, the population statistical data spatialization method based on the geographical scene will be introduced in detail in combination with the drawings.

[0053] As shown in Figure 1 The population statistical data spatialization method based on the geographical scene includes the following steps:

[0054] S1, geographical scene hierarchical modeling;

[0055] The block scene units are divided from the geographical space according to the existing methods and data, Figure 2 The block scene units are divided from the geographical space according to the existing methods and data, Figure 2 There is no river, and each block scene unit is divided according to the road; the expression features of each block scene unit are extracted from the observation data and social perception data; a supervised machine learning classification method is used to classify the block scene units to obtain different geographical scene types; a multi-level spatial structure division is performed on the block scene units by using a spatial clustering method; a geographical scene hierarchical structure model is formed by integrating the classification results of the block scene units and the multi-level spatial structure division results, which is used for subsequent population density spatialization; the maximum first-level region of the multi-level spatial structure of the block scene is the research area range. S2, hierarchical structure feature expression;

[0056] Based on the geographical scene hierarchical structure modeling, the four categories of features, i.e., internal features, self features, neighborhood features, and regional features, are extracted, which can be understood as the classification results of the expression features of each block scene unit.

[0057] S3, population density machine learning;

[0058] Based on the four categories of features obtained in S2, the population density value of each block scene unit is predicted by machine learning to establish the mapping relationship between the features (four categories of features) and the population density.

[0059] Machine learning predicts the population density value of each block scene unit, which needs to make samples for model training or parameter calibration; here, there are two strategies for model training. The first strategy is to collect the real population of the selected sample block unit, and the number of training samples should not be less than the dimension number of the four category features in step S2, to prevent the number of samples from being too small and the learning effect being poor; that is, block scene scale population density sample collection. The second strategy is cross-scale model training, and the cross-scale training sample uses population statistical data above the block scale, for example, using the township scale population statistical data close to the block scale for training, that is, township scale population density statistical data collection.

[0060] Preferably, cross-scale model training is used, and the specific implementation is to use the four categories of features extracted in step S2 as independent variables, and the corresponding block scene unit population density as dependent variables; the predicted density of the block scene unit is multiplied by the area of the block scene unit to calculate the predicted total population, and the predicted total population of the township administrative unit where the block scene unit is located is further calculated. There are two methods for optimizing the machine learning model, the first method is to use the difference between the total population and the actual statistical data as the loss function, train the model and optimize the parameters through error feedback until the accuracy requirement is met or the predetermined feedback iteration number is reached, and this method belongs to supervised training; the second method is to use a weighted regression model to solve the regression coefficients by using weighted least squares method and other algorithms to optimize the fitting ability of the model to different samples. Preferably, the first method of optimizing the machine learning model is used.

[0061] Figure 3 The results of the population spatialization of part of the block scene unit are shown in the schematic diagram, Figure 3 Different colors represent different population densities.

[0062] Further, under the premise of using cross-scale model training, the geographic scene-based population statistical data spatialization method further comprises step S4, population density scale correction; the purpose is to further improve the prediction results, and the main principle is that the prediction value and the true value of the block scale and the township scale are assumed to be the same; the correction formula is as follows:

[0063]

[0064] In the formula, C i represents the population density prediction value P i of the i-th block scene unit, G i represents the actual statistical population of the township where the i-th block scene unit is located, P j represents the population density prediction value and the area of the j-th block scene of the i-th block scene.

[0065] In addition, if it is necessary to subdivide the population by gender and age, it is only necessary to replace the corresponding population density prediction values in steps S3 and S4 with population density values corresponding to attributes such as gender and age, and collect corresponding statistical data to perform model training and cross-scale correction.

[0066] Further, in step S1, the street scene unit is divided in space according to existing multi-source geographic information thematic data and an existing street scale land use classification method; the multi-source geographic information thematic data includes OSM road network, river system, administrative boundary, land cover, building height, etc.; the earth observation data includes high-resolution optical / SAR remote sensing image, night / micro-light remote sensing image, multispectral / hyperspectral image, etc.; the social perception data includes POI data, mobile signaling data, satellite positioning trajectory data, social media data, public transportation data, etc.

[0067] The expression features of each street scene unit are extracted from the earth observation data and the social perception data.

[0068] The street scene unit samples are selected, and a supervised machine learning classification method is used to classify the street scene units, to obtain different geographic scene types.

[0069] On the basis of the classification results of the street scene units, a spatial clustering method is used to divide the street scene units into a multi-level spatial structure, to obtain a multi-level spatial structure above the street scene; the classification results of the street geographic scene and the multi-level spatial structure division results are taken as a geographic scene hierarchical model, to support subsequent hierarchical structure feature expression and population density spatialization modeling. The “multi-level spatial structure above the street scene” is a system with hierarchical nesting relationship constructed by a spatial clustering algorithm, and provides multi-scale feature support for population statistical data spatialization and a basis for the hierarchical structure feature expression in step S2.

[0070] Further, in step S2, the internal features include indicators such as volume rate, proportion and arrangement of internal land cover types, night / micro-light image intensity, POI type and density, social media data density and dynamic change amount, etc.

[0071] The self features include attribute features such as the type, size, direction, etc. of the street scene unit.

[0072] The neighborhood features include the adjacency topological relationship and distance of roads, rivers, administrative centers, transportation hubs, park green spaces, educational facilities, medical facilities, commercial facilities, entertainment facilities, industrial zones, ecological protection zones, etc.

[0073] The regional features refer to the statistical features of the upper-level region containing the current street scene unit to be extracted in the geographic scene hierarchical structure, including features such as the size, direction, shape, arrangement, combination, etc. of different geographic scene types.

[0074] The present application constructs a geographical scene hierarchical model based on more complete expression feature data and spatial clustering methods, divides and classifies the block scene units from the geographical space, and forms a feature expression system of the geographical scene hierarchical structure model in combination with the expression features and the multi-level spatial structure division, so as to provide a more accurate basis for subsequent population density spatialization. The feature expression of the geographical scene hierarchical model includes four categories of 'internal features, self features, neighborhood features and regional features'. This multi-level and systematic feature expression method covers the attributes of the block scene units themselves and the relationship with the surrounding environment and the surrounding region, so that the machine learning model can more comprehensively and deeply understand the complex factors influencing the population distribution, and the accuracy of the model is improved.

[0075] The method of the present application can be used in various intelligent electronic devices, including but not limited to mobile phones, tablet computers, notebook computers, personal computers, smart televisions, smart screens and the like.

[0076] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application are within the protection scope of the claims of the present application.

Claims

1. A spatialization method for demographic data based on geographic scenarios, characterized in that, Includes the following steps: S1, Geographic Scene Hierarchy Modeling; Based on existing methods and data, street scene units are divided geographically; Extracting expressive features of scene units in each neighborhood from Earth observation data and social perception data; Supervised machine learning classification methods are used to classify the street scene units to obtain different geographic scene types; spatial clustering methods are used to divide the street scene units into multi-level spatial structures; the geographic scene hierarchy structure model formed by integrating the street scene unit classification results and the multi-level spatial structure division results is used for subsequent population density spatialization. S2, Hierarchical structure characteristics; Based on the results of geographic scene hierarchy modeling, features of each block scene unit are extracted in four categories: "internal features - self-features - neighborhood features - regional features". S3, Population Density Machine Learning: Based on the features of the four categories obtained in S2, the population density value of each block scene unit is predicted through machine learning.

2. The spatialization method for demographic data based on geographic scenarios according to claim 1, characterized in that: In step S1, street scene units are spatially divided based on existing multi-source geographic information thematic data and existing street-scale land use classification methods; Extracting expressive features of scene units in each neighborhood from Earth observation data and social perception data; Street scene unit samples are selected, and supervised machine learning classification methods are used to classify the street scene units to obtain different geographical scene types; Based on the classification results of street scene units, spatial clustering method is used to divide the street scene units into multi-level spatial structures, and obtain multi-level spatial structure division of street scene. The results of the street geographic scene classification and the multi-level spatial structure division are used as geographic scene hierarchy models to support the subsequent hierarchical structure feature expression and spatial modeling of population density.

3. The spatialization method for demographic data based on geographic scenarios according to claim 2, characterized in that: In step S1, the multi-source geographic information thematic data includes OSM road network, river system, administrative boundaries, land cover, and building height; the Earth observation data includes high-resolution optical / SAR remote sensing images, night light / low light remote sensing images, and multispectral / hyperspectral images; and the social perception data includes POI data, mobile phone signaling data, satellite positioning trajectory data, social media data, and public transportation data.

4. The spatialization method for demographic data based on geographic scenarios according to claim 1, characterized in that: In step S2, the internal features include indicators such as plot ratio, proportion and arrangement of internal land cover types, intensity of nighttime / low-light imagery, POI type and density, and social media data density and dynamic changes. Its own characteristics include attributes such as the type, size, and orientation of the street scene unit; Neighborhood characteristics include the adjacency topology and distances of roads, rivers, administrative centers, transportation hubs, parks and green spaces, educational facilities, medical facilities, commercial facilities, cultural and recreational facilities, industrial zones, and ecological protection zones; Regional features refer to the statistical characteristics of the upper-level regions in the geographic scene hierarchy that contain the current street scene unit to be extracted, including the size, direction, shape, arrangement, and combination features of different geographic scene types.

5. The spatialization method for demographic data based on geographic scenarios according to claim 1, characterized in that: In step S3, before machine learning prediction, training samples are first created to train the machine learning model. The training samples are collected and statistically analyzed to determine the actual population of the selected sample street scene units; the number of training samples should not be less than the number of dimensions of the four category features in step S2.

6. The spatialization method for demographic data based on geographic scenarios according to claim 1, characterized in that: In step S3, before machine learning prediction, cross-scale training samples are first created to train the machine learning model; the cross-scale training samples use population statistics at the block scale or above, which are similar to the population statistics of townships or other administrative units at the block scale.

7. The spatialization method for demographic data based on geographic scenarios according to claim 6, characterized in that: In step S3, the machine learning model training process based on cross-scale training samples is as follows: The features of the four categories extracted in step S2 are used as independent variables, and the population density of the corresponding street scene unit is used as the dependent variable. The predicted density of the street scene unit is multiplied by the area of ​​the street scene unit to calculate the predicted total population, and the predicted total population of the township administrative unit where the street scene unit is located is further calculated. A loss function is constructed based on the difference between the predicted total population and the actual population data. This function is used to train the machine learning model and optimize the various parameters in the machine learning model until the preset accuracy is met or the predetermined number of feedback iterations is reached.

8. The spatialization method for demographic data based on geographic scenarios according to claim 6, characterized in that: In step S3, the machine learning model training process based on cross-scale training samples is as follows: The features of the four categories extracted in step S2 are used as independent variables, and the population density of the corresponding street scene unit is used as the dependent variable. The predicted density of the street scene unit is multiplied by the area of ​​the street scene unit to calculate the predicted total population, and the predicted total population of the township administrative unit where the street scene unit is located is further calculated. The least squares algorithm is used to solve the regression coefficients of the machine learning model to realize the spatial distribution model of population density at the street scale, so as to make predictions for street scene units.

9. The method for spatializing demographic data based on geographic scenarios according to any one of claims 6 to 8, characterized in that: It also includes step S4, population density scale correction; The prediction results are further improved by using a cross-scale correction formula, which is as follows: ； In the formula, Indicates the first Population density prediction values ​​for street scene units Population density after cross-scale correction Indicates the first The actual population of the town / town where the street scene unit is located. Indicates the first The town where the street scene is located j Population density predictions and area for each neighborhood scene.

Images