Large-scale resident welfare multi-source remote sensing space mapping method

Abstract

The invention discloses a large-scale resident welfare multi-source remote sensing spatial mapping method, which is based on multi-source remote sensing data, adopts a remote sensing information mining method to extract remote sensing spatial characteristics such as night light radiation intensity, impervious surface density, vegetation coverage index, road density and the like, and extracts remote sensing recessive characteristics related to resident welfare at the same time. Constructing a resident welfare remote sensing feature set; the resident welfare remote sensing features and the resident welfare values calculated by county scale social statistical yearbook data are combined, a space simulation model of resident welfare and remote sensing feature indexes is constructed, a downscaling method is further adopted, resident welfare space continuous simulation is achieved, an improved space self-correlation analysis method is adopted, and the resident welfare analysis efficiency is improved. And large-scale evolution space mapping of resident welfare is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing intelligent assessment of resident well-being, and in particular relates to a large-scale multi-remote sensing spatial mapping method for resident well-being. Background Art

[0002] Residents' well-being is an important indicator for measuring social development and people's quality of life, but it is difficult to conduct large-scale well-being surveys and assessments. Large-scale resident well-being assessment mapping requires a large amount of integration of multi-source data such as statistical data and survey data, and the use of technical methods such as geographic information systems and spatial analysis to draw a spatial distribution map of residents' well-being. However, it still faces challenges in data accuracy, mapping model construction, and large-scale comparability.

[0003] Residents' well-being is affected by multiple factors such as geographical environment, economic development level, social culture, etc., and there is significant spatial heterogeneity. Large-scale resident well-being assessment involves a wide range of areas, scattered data sources, inconsistent data standards and formats in different departments and regions, resulting in difficulties in data acquisition and integration. At the same time, residents' well-being is a dynamically changing concept. Most of the current assessment methods focus on static analysis and lack the concept of dynamic changes in residents' well-being. Most of the current assessment methods focus on static analysis and lack the ability to effectively monitor and predict the dynamic changes in residents' well-being. It is difficult to timely reflect the temporal and spatial evolution trend of well-being levels, which is not conducive to the timely adjustment and optimization of policies. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the large-scale multi-remote sensing spatial mapping method of resident well-being provided by the present invention solves the problem that existing related methods are difficult to achieve rapid spatial assessment and mapping of large-scale resident well-being.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a large-scale multi-source remote sensing spatial mapping method for resident welfare, comprising the following steps:

[0006] S1. Collect multi-source data on residents’ well-being space at the county level and pre-process them;

[0007] S2, extracting remote sensing features related to residents’ well-being based on the preprocessed data;

[0008] The remote sensing features include night light intensity features, impervious surface index features, vegetation index features, road density index features, point of interest data features and remote sensing implicit features;

[0009] S3. Collect national statistical yearbook data at the county level, perform weighted statistics on them, and quantitatively determine the well-being value of residents;

[0010] S4. Combining remote sensing features and resident well-being values, a random forest model is constructed to predict resident well-being values. The predicted results of resident well-being values ​​are mapped from the county scale to the pixel scale through the downscaling function, and combined with remote sensing feature sequences of different periods, the evaluation results of resident well-being values ​​of different periods at a large scale are obtained.

[0011] S5. Based on the evaluation results of residents’ well-being, calculate the spatial autocorrelation index of different regions, and draw a spatial distribution map of residents’ well-being based on it to show the spatial aggregation and correlation patterns of residents’ well-being.

[0012] Further, in step S1, the multi-source data includes multi-source optical remote sensing data, nighttime light data, point of interest data, and statistical yearbook data;

[0013] The pre-processing comprises:

[0014] The preprocessing of multi-source optical remote sensing data includes radiometric calibration, atmospheric correction, geometric correction, band registration, multi-band image synthesis and image fusion.

[0015] The preprocessing of night light data includes radiation calibration, atmospheric correction, geometric correction, and data clipping and splicing;

[0016] The preprocessing of POI data includes duplicate data removal, data cleaning, coordinate system unification and data classification.

[0017] Further, in step S2, the night light intensity characteristics include the total value of night light intensity, the mean value of night light intensity and the standard deviation of night light intensity;

[0018] The impervious surface index characteristic is a normalized difference impervious surface index obtained by calculating the reflectivity of the short-wave infrared band, the near-infrared band, the green band and the mid-infrared band;

[0019] The vegetation index features include normalized vegetation index and enhanced vegetation index;

[0020] The road density index characteristic is the ratio of road length per unit area;

[0021] The point of interest data feature is the proportion of each type of point of interest in the area;

[0022] The remote sensing latent features are characteristic information extracted from remote sensing images that represent intrinsic attributes related to residents' well-being, physical processes or human activities.

[0023] Furthermore, the method for extracting the remote sensing implicit remote sensing feature is specifically as follows:

[0024] The remote sensing images in the study area are cropped and normalized, and the processed input feature maps are input into the trained convolutional neural network to obtain the remote sensing hidden features;

[0025] The convolutional neural network comprises an input layer, a plurality of convolutional layers, a pooling layer and an output layer connected in sequence;

[0026] The output feature map O of the lth convolutional layer l (i,j,c) is:

[0027]

[0028] In the formula, w l (m,n,d,c) represents the weight of the cth convolution kernel at position (m,n) corresponding to the dth input channel in the lth convolution layer, b l (c) represents the bias of the cth convolution kernel in the lth convolution layer, D l Represents the input feature map I l The number of channels, i, j are the coordinates of the output feature map, s l represents the convolution step size, k l ×k l represents the convolution kernel size in the lth convolution layer, D l Indicates the number of input channels;

[0029] The output feature map p of the pooling layer l (i,j,c) is:

[0030]

[0031] In the formula, p l ×p l represents the pooling window, represents the pooling step size;

[0032] The output layer is a fully connected layer, which outputs n remote sensing implicit features C n It is expressed as:

[0033] C n =W fc ×vec(F)+b fc

[0034] Where W fc represents the weight matrix of the fully connected layer, b fc represents the bias vector of the fully connected layer, vec(F) represents the vector obtained by flattening the output feature map F of the pooling layer, and its dimension is hwd, where h, w and d represent the height, width and depth of the output feature map respectively.

[0035] Furthermore, the step S3 includes the following sub-steps:

[0036] S31. Collect statistical yearbook data at the county level and select statistical indicators related to residents’ well-being;

[0037] S32. Standardize the selected statistical indicators by using the range standardization method;

[0038] S33, determining the weights of each statistical indicator after standardization by entropy weight method;

[0039] S34. Quantitatively calculate the resident welfare value of the study area based on the weights of each statistical indicator and statistical yearbook data.

[0040] Furthermore, the step S4 includes the following sub-steps:

[0041] S41. Calculate the mean and variance of various remote sensing characteristics in the county as the remote sensing characteristic values ​​corresponding to the resident welfare value;

[0042] S42. Based on the remote sensing feature values ​​and resident welfare values ​​at the county scale, a random forest model is constructed, and a downscaling function is used to map the resident welfare values ​​at the county scale to the pixel scale;

[0043] S43. Based on the predicted value of resident welfare value at pixel scale, remote sensing feature sequences of different periods are used to set quantitative indicators to explore the dynamic evolution law of resident welfare value over time in a large scale range, and the evaluation results of resident welfare value in different periods at a large scale are obtained.

[0044] Furthermore, the step S42 includes the following sub-steps:

[0045] S42-1. Construct dataset D based on the residents’ welfare values ​​and remote sensing feature values ​​at the county scale;

[0046] The data set D includes several subsets (D1, D2, ..., D t ), each subset represents the remote sensing feature values ​​and resident welfare values ​​corresponding to the resident welfare values ​​of different counties, D t = {Y t ,X t1 ,X t2 ,…,X ti ,…,X tn}; where Y t represents the welfare value of residents in the tth subset, X t1 ,X t2 ,…,X ti ,…,X tn Represents Y t The corresponding 1st to nth remote sensing feature values;

[0047] S42-2. For each subset, construct a corresponding decision tree;

[0048] S42-3, the k decision trees constructed are combined into a random forest model, the remote sensing feature values ​​in the simulated area are input into the random forest model, and the prediction results corresponding to each decision tree are regressed to obtain the predicted value of the resident welfare value;

[0049] S42-4, constructing a downscaling function f(x) based on the correlation between the remote sensing feature values ​​at the county scale and the pixel scale;

[0050] S42-5. Use the downscaling function f(x) to map the predicted values ​​of residents’ welfare at the county scale to the pixel scale.

[0051] Furthermore, in step S42-2, during the decision tree construction process, p remote sensing feature values ​​are randomly selected from the n remote sensing feature values ​​of the subset at each node, and the corresponding information gain Gain (D, A) is:

[0052]

[0053] In the formula, information gain Gain(D,A) is used to measure the contribution of remote sensing feature values ​​to the data set D, Entropy(D v ) represents the entropy of the data set, which is used to measure the degree of disorder of the data set. A represents the selected remote sensing feature value. Values(A) represents the set of all possible values ​​of remote sensing feature A. v represents the sample subset in the dataset whose remote sensing feature A takes the value v, It indicates the proportion of the sample subset whose remote sensing feature A takes value v in the dataset D;

[0054] In step S42-5, the predicted value Y of the resident welfare at the pixel scale under the jth pixel pixel (j) is expressed as:

[0055] Y pixel (j) = f(Y countt ,x 1j ,x 2j ,…,x ij ,…x nj )

[0056]

[0057] Where f(·) represents the downscaling function, x 1j ,x 2j ,……x nj represents the remote sensing feature value of the jth pixel, Y countyrepresents the well-being value of residents at the county level, β0(j) represents the regression coefficient associated with pixel j, and x ij Represents the i-th remote sensing feature value corresponding to the j-th pixel position.

[0058] Furthermore, the step S5 includes the following sub-steps:

[0059] S51. Set the distance-based spatial weight matrix W d and the weight matrix W based on the strength of economic ties e ;

[0060] S52, the matrix W d and W e Compound, and obtain the composite spatial weight matrix W;

[0061] S53. Calculate the spatial autocorrelation index I between the welfare values ​​of residents in different regions based on the composite spatial weight matrix W. i mprove d ;

[0062] S54, divide the study area into grids according to geographic coordinates, and render the grids in different levels according to the size of the residents' well-being values;

[0063] S55, according to the spatial autocorrelation index I improved , mark / connect areas with significant positive / negative correlation, show the spatial aggregation and correlation patterns of residents' well-being, and obtain the results of spatial mapping of residents' well-being.

[0064] Furthermore, in step S51, the distance-based spatial weight matrix W d Elements in It represents the degree of association between region i and region j based on geographical distance, which is expressed as:

[0065]

[0066] Where θ represents the distance attenuation parameter, d ij represents the geographical distance between region i and region j;

[0067] The weight matrix W based on the strength of economic ties d It is expressed as:

[0068]

[0069] In the formula, represents the trade volume indicator between region i and region j, represents the investment flow index between region i and region j, represents the similarity index of industrial structure between region i and region j, α, β, γ represent the weight coefficients of each index;

[0070] In step S53, the spatial autocorrelation index I improved It is expressed as:

[0071]

[0072] In the formula, x i and x j Represent the resident welfare index values ​​of region i and region j respectively, represents the mean of the resident welfare index, W ij is the element in the composite spatial weight matrix W;

[0073] Among them, I improved >0 indicates that residents’ well-being is spatially positively correlated, that is, high-value areas tend to be adjacent to high-value areas, and low-value areas tend to be adjacent to low-value areas. improved <0 indicates negative spatial correlation, that is, high-value areas are adjacent to low-value areas. improved =0 indicates random distribution in space.

[0074] The beneficial effects of the present invention are:

[0075] (1) The present invention adopts a convolutional neural network to mine hidden characteristic factors of resident welfare remote sensing and realize factor optimization, thus overcoming the technical difficulty of quantitative extraction of original resident welfare remote sensing characteristics;

[0076] (2) The present invention uses statistical data to construct multi-indicator resident welfare data, adopts entropy weight analysis method to determine the weight of each indicator, and realizes quantitative calculation of resident welfare;

[0077] (3) The present invention adopts a random forest model to construct a spatial simulation model of residents’ well-being and remote sensing characteristic indicators, and further adopts a downscaling method to achieve spatial simulation and mapping of residents’ well-being.

[0078] (4) This paper adopts an improved spatial autocorrelation analysis method to explore the evolution characteristics and spatial mapping of large-scale resident well-being. This method comprehensively considers the impact of geographic proximity and economic connections on the spatial distribution of resident well-being. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 A flow chart of the large-scale multi-remote sensing spatial mapping method for resident well-being provided by the present invention. DETAILED DESCRIPTION

[0080] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0081] The embodiment of the present invention provides a large-scale multi-source remote sensing spatial mapping method for resident welfare, such as Figure 1 As shown, the following steps are included:

[0082] S1. Collect multi-source data on residents’ well-being space at the county level and pre-process them;

[0083] S2, extracting remote sensing features related to residents’ well-being based on the preprocessed data;

[0084] The remote sensing features include night light intensity features, impervious surface index features, vegetation index features, road density index features, point of interest data features and remote sensing implicit features;

[0085] S3. Collect national statistical yearbook data at the county level, perform weighted statistics on them, and quantitatively determine the well-being value of residents;

[0086] S4. Combining remote sensing features and resident well-being values, a random forest model is constructed to predict resident well-being values. The predicted results of resident well-being values ​​are mapped from the county scale to the pixel scale through the downscaling function, and combined with remote sensing feature sequences of different periods, the evaluation results of resident well-being values ​​of different periods at a large scale are obtained.

[0087] S5. Based on the evaluation results of residents’ well-being, calculate the spatial autocorrelation index of different regions, and draw a spatial distribution map of residents’ well-being based on it to show the spatial aggregation and correlation patterns of residents’ well-being.

[0088] In step S1 of the embodiment of the present invention, the multi-source data includes multi-source optical remote sensing data, nighttime light data, located point of interest data, and statistical yearbook data;

[0089] The pre-processing comprises:

[0090] The preprocessing of multi-source optical remote sensing data includes radiometric calibration, atmospheric correction, geometric correction, band registration, multi-band image synthesis and image fusion.

[0091] The preprocessing of night light data includes radiation calibration, atmospheric correction, geometric correction, and data clipping and splicing;

[0092] The preprocessing of the located POI data includes duplicate data removal, data cleaning, coordinate system unification and data classification.

[0093] Specifically, in this embodiment, for multi-source optical remote sensing data, optical remote sensing data such as Landsat 8, Gaofen-1, Gaofen-6 and Sentinel-2 at the county scale are obtained, and the optical remote sensing data selects cloudless, ice-free and snow-free data. Optical remote sensing data such as Landsat8, Gaofen-1, Gaofen-6 and Sentinel-2 include 15-meter multispectral images, 2-meter panchromatic\8-meter multispectral, 10-meter, 20-meter, 60-meter and other multispectral images; after the image is acquired, radiation calibration, atmospheric correction, geometric correction, band registration, multi-band image synthesis, image fusion and other pre-processing are performed.

[0094] For night light data, obtain night light data at the county level. SDGSAT-1 is equipped with a low-light and multi-spectral imager, with 1 panchromatic band and 3 color bands, including night light images with a resolution of 30 meters; after the image is acquired, it undergoes pre-processing such as radiation calibration, atmospheric correction, geometric correction, data cropping and splicing.

[0095] For the data of the points of interest located, POI (points of interest) data are collected through online map data platforms and mobile device positioning, including data information of various commercial facilities, public service facilities, industrial areas and township service points in the county. The acquired data is pre-processed by removing duplicate data, data cleaning, coordinate system one, data classification statistics, etc.

[0096] For statistical yearbook data, the China County Statistical Yearbook (county and city volume) was used, which includes collecting county-scale social statistical yearbook data, collecting information such as registered population, regional GDP, local public budget revenue, household savings deposits, the number of enterprises above a certain scale, and the number of students in primary and secondary schools, and screening out counties and cities with missing data.

[0097] In step S2 of the embodiment of the present invention, among the remote sensing features related to residents' well-being:

[0098] The night light intensity characteristics in this embodiment include the total value of night light intensity, the mean value of night light intensity, and the standard deviation of night light intensity;

[0099] Among them, the night light intensity values ​​of all valid pixels of the preprocessed SDGSAT-1 night light remote sensing data in the study county are accumulated to obtain the total night light intensity (TNL):

[0100]

[0101] Where, DN i is the i-th gray level, C iis the number of pixels corresponding to the gray level, and n is the maximum gray level.

[0102] Divide the total value of nighttime light intensity in the county by the total number of valid pixels to obtain the nighttime light intensity mean (NLM):

[0103]

[0104] In the formula, C i is the number of pixels corresponding to the gray level, and n is the maximum gray level.

[0105] The night light intensity standard deviation NTSD represents the difference in the spatial distribution of night light. The square of the difference between the night light intensity value of each pixel and the mean night light intensity is calculated. The square of the difference of all pixels is accumulated and then divided by the total number of valid pixels to obtain the variance. The square root of the variance is taken to obtain the night light intensity standard deviation (NTSD):

[0106]

[0107] Where, DN i is the i-th gray level, and NTM is the mean value of night light.

[0108] The impervious surface index feature in this embodiment is a normalized difference impervious surface index obtained by calculating the reflectivity of the shortwave infrared band, the near infrared band, the green band and the mid-infrared band; specifically, the impervious surface index feature of the county scale is extracted based on the pre-processed high-resolution satellite images and Landsat series satellite data containing visible light, near infrared and other bands, and the normalized difference impervious surface index (NDISI) is used. By calculating the reflectivity of the shortwave infrared band (SWIR), the near infrared band (NIR), the green band (Green) and the mid-infrared band 1 (MIR1), the reflection difference of different bands on different objects is cleverly used to highlight the characteristics of the impervious surface, thereby realizing the effective extraction of the impervious surface, which is expressed as:

[0109]

[0110] Among them, SWIR is the reflectivity of the shortwave infrared band, NIR is the reflectivity of the near infrared band, Green is the reflectivity of the green band, and MIR1 is the reflectivity of the mid-infrared band 1.

[0111] In this embodiment, the vegetation index features include the normalized vegetation index and the enhanced vegetation index; specifically, the county scale features are extracted based on the preprocessed Landsat series satellite data containing visible light, near infrared and other bands and the Sentinel-2 satellite data obtained above, mainly including the normalized vegetation index (NDVI) and the enhanced vegetation index (EVI).

[0112] Among them, the Normalized Difference Vegetation Index (NDVI) is calculated by dividing the difference between the near-infrared band reflectance (NIR) and the red band reflectance (R) by the sum of the two. By taking advantage of the high reflection of vegetation in the near-infrared band and the strong absorption in the red band, the difference between vegetation in these two bands is highlighted, thus effectively reflecting the growth status and coverage of vegetation, which is expressed as:

[0113] NVDI=(NIR-R) / (NIR+R)

[0114] Where NIR represents the reflectivity of the near-infrared band, and R represents the reflectivity of the red band.

[0115] In addition to considering the near-infrared and red bands, the Enhanced Vegetation Index (EVI) also introduces the blue band reflectance (B). The numerator amplifies the difference between vegetation in the near-infrared and red bands through a coefficient of 2.5, making the vegetation signal more obvious. The 6R term emphasizes the sensitivity of the red band to vegetation chlorophyll absorption, and the -7.5B term is used to correct the influence of the blue band to reduce interference from factors such as atmospheric scattering. Finally, 1 is added to avoid the denominator being zero. EVI has a larger dynamic range in areas with high vegetation coverage and low vegetation coverage, and can better reflect subtle changes in vegetation. The calculation formula is as follows:

[0116] EVI=2.5×(NIR-R) / (NIR+6R-7.5B+1)

[0117] Among them, NIR represents the reflectivity of the near-infrared band, R represents the reflectivity of the red band, and B represents the reflectivity of the blue band.

[0118] The road density index feature in this embodiment is the length ratio of roads per unit area. Specifically, the road density index (RDI) is obtained by collecting high-resolution remote sensing images and related geographic information data and preprocessing them, and then extracting road information using image classification, edge detection or object-oriented methods to obtain the length ratio of roads per unit area. This can intuitively show the density of roads in a county. It is expressed as:

[0119] RDI = total length of roads / area of ​​study area

[0120] The point of interest data feature in this embodiment is the proportion of the number of points of interest of each type in the region; specifically, the number of POIs such as companies, tourist attractions, hotels, and restaurants is counted at the county scale, and the distribution uniformity is calculated, which is expressed as:

[0121] POI = Statistical number / Research area

[0122] The remote sensing implicit features in this embodiment are feature information extracted from remote sensing images that represent intrinsic properties related to residents' well-being, physical processes, or human activities. Feature information is extracted from remote sensing images using advanced data processing techniques (such as machine learning, deep learning, or statistical analysis).

[0123] The method for extracting the remote sensing hidden features in this embodiment is specifically as follows:

[0124] The remote sensing images in the study area are cropped and normalized, and the processed input feature map is input into the trained convolutional neural network to obtain the remote sensing hidden features.

[0125] Specifically, the county-scale remote sensing image data is cropped and normalized to make it suitable for input into the convolutional neural network. Suppose the input remote sensing image is I(x, y), where x and y are the pixel coordinates of the image. The normalization operation can be expressed as:

[0126]

[0127] Where μ is the mean of all pixel values ​​in the image and σ is the standard deviation.

[0128] The convolutional neural network consists of an input layer, several convolutional layers, a pooling layer, and an output layer connected in sequence;

[0129] Among them, the output feature map O of the lth convolutional layer l (i,j,c) is:

[0130]

[0131] In the formula, w l (m,n,d,c) represents the weight of the cth convolution kernel at position (m,n) corresponding to the dth input channel in the lth convolution layer, b l (c) represents the bias of the cth convolution kernel in the lth convolution layer, D l Represents the input feature map I l The number of channels, i, j are the coordinates of the output feature map, s l represents the convolution step size, k l ×k l represents the convolution kernel size in the lth convolution layer, D l Indicates the number of input channels;

[0132] The output feature map p of the pooling layer l (i,j,c) is:

[0133]

[0134] In the formula, p l ×pl represents the pooling window, represents the pooling step size;

[0135] After a series of convolutional and pooling layers, a feature map tensor F is obtained, whose dimensions are h×w×d (height h, width w and depth d). It is flattened into a vector vec(F) with dimensions hwd; a fully connected layer is designed as the output layer, and the weight matrix is ​​W fc , the dimension is n×hwd (n is the number of implicit socio-economic features to be extracted), and the n remote sensing implicit features C output by the output layer are obtained. n It is expressed as:

[0136] C n =W fc ×vec(F)+b fc

[0137] Where W fc represents the weight matrix of the fully connected layer, b fc represents the bias vector of the fully connected layer, vec(F) represents the vector obtained by flattening the output feature map F of the pooling layer, and its dimension is hwd, where h, w and d represent the height, width and depth of the output feature map respectively.

[0138] Step S3 in the embodiment of the present invention includes the following sub-steps:

[0139] S31. Collect statistical yearbook data at the county level and select statistical indicators related to residents’ well-being;

[0140] S32. Standardize the selected statistical indicators by using the range standardization method;

[0141] S33, determining the weights of each statistical indicator after standardization by entropy weight method;

[0142] S34. Quantitatively calculate the resident welfare value of the study area based on the weights of each statistical indicator and statistical yearbook data.

[0143] In step S31 of this embodiment, the selected statistical indicators related to residents' welfare include registered population, regional gross domestic product, number of enterprises above a certain size, number of students in primary and secondary schools, and other statistical indicators.

[0144] In step S32 of this embodiment, the above statistical indicators are standardized, and the jth indicator of the i-th county is represented by x ij , the standardized formula is:

[0145]

[0146] In the formula, y ij It is the index value after range standardization.

[0147] In step S33 of this embodiment, the entropy weight method is an objective weighting method, which determines the weight according to the information entropy of the indicator and calculates the entropy value e according to the indicator value of the range standardization method. j , and then calculate the coefficient of variation g j , determine the weight w of each indicator j The smaller the information entropy, the greater the amount of information provided by the indicator, the greater the difference coefficient, and the greater the weight.

[0148] The entropy value is:

[0149]

[0150] e j =-k∑p ij ln(p ij )

[0151]

[0152] Among them, y ij is the index value after range standardization, n is the number of counties where resident welfare assessment is conducted, and k is a constant related to the number of counties n.

[0153] The coefficient of variation is:

[0154] g j =1-e j

[0155] Calculate the weight of each indicator:

[0156]

[0157] In the formula, m is the number of indicators.

[0158] In step S34 of this embodiment, the resident welfare value at the county level is calculated as:

[0159] WB i =x i,1 ×ω1+x i,2 ×ω2……x i,j ×ω j

[0160] Where WB i is the welfare value of residents in the ith county, x ij is the jth index value of the i-th county, ω j is the weight value of the jth indicator.

[0161] Step S4 of the embodiment of the present invention includes the following sub-steps:

[0162] S41. Calculate the mean and variance of various remote sensing characteristics in the county as the remote sensing characteristic values ​​corresponding to the resident welfare value;

[0163] S42. Based on the remote sensing feature values ​​and resident welfare values ​​at the county scale, a random forest model is constructed, and a downscaling function is used to map the resident welfare values ​​at the county scale to the pixel scale;

[0164] S43. Based on the predicted value of resident welfare value at pixel scale, remote sensing feature sequences of different periods are used to set quantitative indicators to explore the dynamic evolution law of resident welfare value over time in a large scale range, and the evaluation results of resident welfare value in different periods at a large scale are obtained.

[0165] In step S41 of this embodiment, the mean of each remote sensing feature is calculated. and variance s i 2 for:

[0166]

[0167] In the formula, x ij It represents the value of the i-th remote sensing characteristic index at the j-th pixel, and m is the number of pixels in the county.

[0168] Step S42 of this embodiment includes the following sub-steps:

[0169] S42-1. Construct dataset D based on the residents’ welfare values ​​and remote sensing feature values ​​at the county scale;

[0170] The data set D contains several subsets (D1, D2, ..., D t ), each subset represents the remote sensing feature values ​​and resident welfare values ​​corresponding to the resident welfare values ​​of different counties, D t = {Y t ,X t1 ,X t2 ,…,X ti ,…,X tn}; where Y t represents the welfare value of residents in the tth subset, X t1 ,X t2 ,…,X ti ,…,X tn Represents Y t The corresponding 1st to nth remote sensing feature values;

[0171] S42-2. For each subset, construct a corresponding decision tree;

[0172] In the process of decision tree construction, p remote sensing feature values ​​are randomly selected from the n remote sensing feature values ​​of the subset at each node, and the corresponding information gain Gain (D, A) is:

[0173]

[0174] In the formula, information gain Gain(D,A) is used to measure the contribution of remote sensing feature values ​​to the data set D, Entropy(D v ) represents the entropy of the data set, which is used to measure the degree of disorder of the data set. A represents the selected remote sensing feature value. Values(A) represents the set of all possible values ​​of remote sensing feature A. v represents the sample subset in the dataset whose remote sensing feature A takes the value v, It indicates the proportion of the sample subset whose remote sensing feature A takes value v in the dataset D;

[0175] in;

[0176]

[0177] In the formula, C is the set of sample categories, p c Represents the proportion of each type of sample in C;

[0178] S42-3, the k decision trees constructed are combined into a random forest model, the remote sensing feature values ​​in the simulated area are input into the random forest model, and the prediction results corresponding to each decision tree are regressed to obtain the predicted value of the resident welfare value;

[0179] After the decision tree is built, k decision trees are combined into a random forest model. For a new data point x, its predicted value It is determined by the prediction results of each decision tree in the random forest. For regression problems, the following is used:

[0180]

[0181] Among them, k represents the number of sample subsets, that is, the number of decision trees; y t is the prediction result of the t-th decision tree for the new data point S42-4. Based on the association between the remote sensing feature values ​​at the county scale and the pixel scale, a downscaling function f(x) is constructed;

[0182] Among them, the construction of the downscaling function fully considers the intrinsic correlation between the county scale results and the remote sensing characteristic indicator values ​​at the pixel scale, which can be determined with the help of advanced technical means such as statistical relationships and spatial interpolation. For example, the downscaling method based on geographically weighted regression can achieve accurate conversion from county to pixel scale by estimating the regression coefficient related to location, thereby achieving a high-resolution assessment of the residents' welfare space at the pixel scale.

[0183] S42-5, using the downscaling function f(x), the predicted values ​​of residents’ welfare at the county scale are mapped to the pixel scale;

[0184] Among them, the predicted value of resident welfare at pixel scale under the jth pixel is Y pixel (j) is expressed as:

[0185] Y pixel (j) = f(Y county ,x 1j ,x 2j ,…,x ij ,…x nj )

[0186]

[0187] Where f(·) represents the downscaling function, x 1j ,x 2j ,……x nj represents the remote sensing feature value of the jth pixel, Y county represents the well-being value of residents at the county level, β0(j) represents the regression coefficient associated with pixel j, and x ij Represents the i-th remote sensing feature value corresponding to the j-th pixel position.

[0188] In step S43 of this embodiment, for remote sensing data sequences of different periods, the resident welfare assessment results corresponding to each period are obtained according to the random forest model, and the dynamic evolution of resident welfare over time is further explored. By calculating quantitative indicators such as the rate of change, the time series change trend of resident welfare in a large scale is intuitively presented, providing scientific data support and decision-making basis for long-term planning and policy formulation. For remote sensing data of period t, the above steps are repeated to obtain the resident welfare assessment results y of different periods t , calculate the rate of change r t :

[0189]

[0190] In step S5 of the embodiment of the present invention, an improved spatial autocorrelation analysis method is used to explore the evolution characteristics and spatial mapping of large-scale resident well-being. This method comprehensively considers the impact of geographic spatial proximity and economic connections on the spatial distribution of resident well-being; considering the complexity and multi-source of resident well-being data, not only the classic distance-based spatial weight matrix is ​​introduced, but also a weight matrix based on the intensity of economic connections is combined.

[0191] Based on this, step S5 of the embodiment of the present invention includes the following sub-steps:

[0192] S51. Set the distance-based spatial weight matrix W d and the weight matrix W based on the strength of economic ties e ;

[0193] Among them, the distance-based spatial weight matrix W d Elements in It represents the degree of association between region i and region j based on geographical distance, which is expressed as:

[0194]

[0195] Where θ represents the distance attenuation parameter, d ij represents the geographical distance between region i and region j;

[0196] The weight matrix W based on the strength of economic ties d It is expressed as:

[0197]

[0198] In the formula, represents the trade volume indicator between region i and region j, represents the investment flow index between region i and region j, represents the similarity index of industrial structure between region i and region j, α, β, γ represent the weight coefficients of each index;

[0199] S52, the matrix W d and W e Compound, get the composite spatial weight matrix W = W d +W e ;

[0200] S53. Calculate the spatial autocorrelation index I between the welfare values ​​of residents in different regions based on the composite spatial weight matrix W. i mprove d ;

[0201] Among them, the spatial autocorrelation index I improved It is expressed as:

[0202]

[0203] In the formula, x i and x j Represent the resident welfare index values ​​of region i and region j respectively, represents the mean of the resident welfare index, W ij is the element in the composite spatial weight matrix W;

[0204] Among them, I improved >0 indicates that residents’ well-being is spatially positively correlated, that is, high-value areas tend to be adjacent to high-value areas, and low-value areas tend to be adjacent to low-value areas. improved <0 indicates negative spatial correlation, that is, high-value areas are adjacent to low-value areas. improved=0 means random distribution in space;

[0205] S54, divide the study area into grids according to geographic coordinates, and render the grids in different levels according to the size of the residents' well-being values;

[0206] S55, according to the spatial autocorrelation index I improved , mark / connect areas with significant positive / negative correlation, show the spatial aggregation and correlation patterns of residents' well-being, and obtain the results of resident well-being spatial mapping;

[0207] Specifically, the grids are rendered in grades according to the values ​​of the residents' well-being index. For example, the values ​​of the residents' well-being index are divided into three levels: high, medium and low, and displayed in different colors (such as red for high values, yellow for medium values, and blue for low values). Using the results of the spatial autocorrelation index, areas with significant positive or negative correlations are specially marked or connected with lines to intuitively show the spatial aggregation and correlation patterns of residents' well-being.

[0208] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0209] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A large-scale multi-source remote sensing spatial mapping method for resident welfare, characterized in that: The following steps are involved: S1. Collect multi-source data on residents’ well-being space at the county level and pre-process them; S2, extracting remote sensing features related to residents’ well-being based on the preprocessed data; The remote sensing features include night light intensity features, impervious surface index features, vegetation index features, road density index features, point of interest data features and remote sensing implicit features; S3. Collect national statistical yearbook data at the county level, perform weighted statistics on them, and quantitatively determine the well-being value of residents; S4. Combining remote sensing features and resident well-being values, a random forest model is constructed to predict resident well-being values. The predicted results of resident well-being values ​​are mapped from the county scale to the pixel scale through the downscaling function, and combined with remote sensing feature sequences of different periods, the evaluation results of resident well-being values ​​of different periods at a large scale are obtained. S5. Based on the evaluation results of residents’ well-being, calculate the spatial autocorrelation index of different regions, and draw a spatial distribution map of residents’ well-being based on it to show the spatial aggregation and correlation patterns of residents’ well-being.

2. The large-scale multi-source remote sensing spatial mapping method for resident well-being according to claim 1 is characterized in that: In step S1, the multi-source data includes multi-source optical remote sensing data, nighttime light data, point of interest data, and statistical yearbook data; The pre-processing comprises: The preprocessing of multi-source optical remote sensing data includes radiometric calibration, atmospheric correction, geometric correction, band registration, multi-band image synthesis and image fusion. The preprocessing of night light data includes radiation calibration, atmospheric correction, geometric correction, and data clipping and splicing; The preprocessing of POI data includes duplicate data removal, data cleaning, coordinate system unification and data classification.

3. The large-scale multi-source remote sensing spatial mapping method for resident well-being according to claim 1 is characterized in that: In step S2, the night light intensity characteristics include the total value of night light intensity, the mean value of night light intensity and the standard deviation of night light intensity; The impervious surface index characteristic is a normalized difference impervious surface index obtained by calculating the reflectivity of the short-wave infrared band, the near-infrared band, the green band and the mid-infrared band; The vegetation index features include normalized vegetation index and enhanced vegetation index; The road density index characteristic is the ratio of road length per unit area; The point of interest data feature is the proportion of each type of point of interest in the area; The remote sensing latent features are characteristic information extracted from remote sensing images that represent intrinsic attributes related to residents' well-being, physical processes or human activities.

4. The large-scale multi-source remote sensing spatial mapping method for resident welfare according to claim 3 is characterized in that: The method for extracting the remote sensing implicit remote sensing feature is specifically as follows: The remote sensing images in the study area are cropped and normalized, and the processed input feature maps are input into the trained convolutional neural network to obtain the remote sensing hidden features; The convolutional neural network comprises an input layer, a plurality of convolutional layers, a pooling layer and an output layer connected in sequence; The output feature map O of the lth convolutional layer l (i,j,c) is: In the formula, w l (m,n,d,c) represents the weight of the cth convolution kernel at position (m,n) corresponding to the dth input channel in the lth convolution layer, b l (c) represents the bias of the cth convolution kernel in the lth convolution layer, D l Represents the input feature map I l The number of channels, i, j are the coordinates of the output feature map, s l represents the convolution step size, k l ×k l represents the convolution kernel size in the lth convolution layer, D l Indicates the number of input channels; The output feature map p of the pooling layer l (i,j,c) is: In the formula, p l ×p l represents the pooling window, represents the pooling step size; The output layer is a fully connected layer, which outputs n remote sensing implicit features C n It is expressed as: C n =W fc ×vec(F)+b fc Where W fc represents the weight matrix of the fully connected layer, b fc represents the bias vector of the fully connected layer, Vec(F) represents the vector obtained by flattening the output feature map F of the pooling layer, and its dimension is hwd, where h, w and d represent the height, width and depth of the output feature map respectively.

5. The large-scale multi-source remote sensing spatial mapping method for resident well-being according to claim 1 is characterized in that: The step S3 comprises the following sub-steps: S31. Collect statistical yearbook data at the county level and select statistical indicators related to residents’ well-being; S32. Standardize the selected statistical indicators by using the range standardization method; S33, determining the weights of each statistical indicator after standardization by entropy weight method; S34. Quantitatively calculate the resident welfare value of the study area based on the weights of each statistical indicator and statistical yearbook data.

6. The large-scale multi-source remote sensing spatial mapping method for resident well-being according to claim 1 is characterized in that: The step S4 comprises the following sub-steps: S41. Calculate the mean and variance of various remote sensing characteristics in the county as the remote sensing characteristic values ​​corresponding to the resident welfare value; S42. Based on the remote sensing feature values ​​and resident welfare values ​​at the county scale, a random forest model is constructed, and a downscaling function is used to map the resident welfare values ​​at the county scale to the pixel scale; S43. Based on the predicted value of resident welfare value at pixel scale, remote sensing feature sequences of different periods are used to set quantitative indicators to explore the dynamic evolution law of resident welfare value over time in a large scale range, and the evaluation results of resident welfare value in different periods at a large scale are obtained.

7. The large-scale multi-source remote sensing spatial mapping method for resident well-being according to claim 6 is characterized in that: The step S42 includes the following sub-steps: S42-1. Construct dataset D based on the residents’ welfare values ​​and remote sensing feature values ​​at the county scale; The data set D includes several subsets (D1, D2, ..., D t ), each subset represents the remote sensing feature values ​​and resident welfare values ​​corresponding to the resident welfare values ​​of different counties, d t = {Y t ,X t1 ,X t2 ,…,X ti ,…,X tn }; where Y t represents the welfare value of residents in the tth subset, X t1 ,X t2 ,…,X ti ,…,X tn Represents Y t The corresponding 1st to nth remote sensing feature values; S42-2. For each subset, construct a corresponding decision tree; S42-3, the k decision trees constructed are combined into a random forest model, the remote sensing feature values ​​in the simulated area are input into the random forest model, and the prediction results corresponding to each decision tree are regressed to obtain the predicted value of the resident welfare value; S42-4, constructing a downscaling function f(x) based on the correlation between the remote sensing feature values ​​at the county scale and the pixel scale; S42-5. Use the downscaling function f(x) to map the predicted values ​​of residents’ welfare at the county scale to the pixel scale.

8. The large-scale multi-source remote sensing spatial mapping method for resident well-being according to claim 7 is characterized in that: In step S42-2, during the decision tree construction process, p remote sensing feature values ​​are randomly selected from the n remote sensing feature values ​​in the subset at each node, and the corresponding information gain Gain (D, A) is: In the formula, information gain Gain(D,A) is used to measure the contribution of remote sensing feature values ​​to the data set D, Entropy(D v ) represents the entropy of the data set, which is used to measure the degree of disorder of the data set. A represents the selected remote sensing feature value. Values(A) represents the set of all possible values ​​of remote sensing feature A. v represents the sample subset in the dataset whose remote sensing feature A takes the value v, It indicates the proportion of the sample subset whose remote sensing feature A takes value v in the dataset D; In step S42-5, the predicted value Y of the resident welfare at the pixel scale under the jth pixel pixel (j) is expressed as: Y pixel (j)=f(Y county ,x 1j ,x 2j ,…,x ij ,…x nj ) Where f(·) represents the downscaling function, x 1j ,x 2j ,……x nj represents the remote sensing feature value of the jth pixel, Y county represents the well-being value of residents at the county level, β0(j) represents the regression coefficient associated with pixel j, and x ij Represents the i-th remote sensing feature value corresponding to the j-th pixel position.

9. The large-scale multi-source remote sensing spatial mapping method for resident well-being according to claim 1 is characterized in that: The step S5 comprises the following sub-steps: S51. Set the distance-based spatial weight matrix W d and the weight matrix W based on the strength of economic ties e ; S52, the matrix W d and W e Compound, and obtain the composite spatial weight matrix W; S53. Calculate the spatial autocorrelation index I between the welfare values ​​of residents in different regions based on the composite spatial weight matrix W. i mprove d ; S54, divide the study area into grids according to geographic coordinates, and render the grids in different levels according to the size of the residents' well-being values; S55, according to the spatial autocorrelation index I improved , mark / connect areas with significant positive / negative correlation, show the spatial aggregation and correlation patterns of residents' well-being, and obtain the results of spatial mapping of residents' well-being.

10. The large-scale multi-source remote sensing spatial mapping method for resident well-being according to claim 9 is characterized in that: In step S51, the distance-based spatial weight matrix W d Elements in It represents the degree of association between region i and region j based on geographical distance, which is expressed as: Where θ represents the distance attenuation parameter, d ij represents the geographical distance between region i and region j; The weight matrix W based on the strength of economic ties d It is expressed as: In the formula, represents the trade volume indicator between region i and region j, represents the investment flow index between region i and region j, represents the similarity index of industrial structure between region i and region j, α, β, γ represent the weight coefficients of each index; In step S53, the spatial autocorrelation index I improved It is expressed as: In the formula, x i and x j Represent the resident welfare index values ​​of region i and region j respectively, represents the mean of the resident welfare index, W ij is the element in the composite spatial weight matrix W; Among them, I improved >0 indicates that residents’ well-being is spatially positively correlated, that is, high-value areas tend to be adjacent to high-value areas, and low-value areas tend to be adjacent to low-value areas. improved <0 indicates negative spatial correlation, that is, high-value areas are adjacent to low-value areas. improved =0 indicates random distribution in space.

Images