CN117541928B - Urban building material stock estimation method and system based on convolutional neural network - Google Patents
Urban building material stock estimation method and system based on convolutional neural network Download PDFInfo
- Publication number
- CN117541928B CN117541928B CN202410027285.1A CN202410027285A CN117541928B CN 117541928 B CN117541928 B CN 117541928B CN 202410027285 A CN202410027285 A CN 202410027285A CN 117541928 B CN117541928 B CN 117541928B
- Authority
- CN
- China
- Prior art keywords
- building
- area
- total
- urban
- grid
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000004566 building material Substances 0.000 title claims abstract description 106
- 238000000034 method Methods 0.000 title claims abstract description 71
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 26
- 238000011160 research Methods 0.000 claims abstract description 74
- 239000011159 matrix material Substances 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims description 23
- 239000000178 monomer Substances 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 16
- 238000005286 illumination Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 6
- 230000007704 transition Effects 0.000 claims description 5
- 239000004567 concrete Substances 0.000 claims description 2
- 239000000463 material Substances 0.000 description 26
- 238000009826 distribution Methods 0.000 description 16
- 238000011161 development Methods 0.000 description 10
- 230000000694 effects Effects 0.000 description 9
- 238000012545 processing Methods 0.000 description 7
- 230000008901 benefit Effects 0.000 description 5
- 238000011176 pooling Methods 0.000 description 5
- 238000004064 recycling Methods 0.000 description 5
- 241000282414 Homo sapiens Species 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 230000001737 promoting effect Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 229910052782 aluminium Inorganic materials 0.000 description 1
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 239000011449 brick Substances 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 239000004035 construction material Substances 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 239000010949 copper Substances 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 125000001475 halogen functional group Chemical group 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 229910052751 metal Inorganic materials 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000012916 structural analysis Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Abstract
The invention discloses a city building material stock estimating method and system based on a convolutional neural network, comprising the following steps: the night light remote sensing image of the current research area is utilized to be input into a corresponding available total building area estimation model, and a total building area estimation value of the current research area is obtained; obtaining the urban land utilization type in the current research area, overlapping the use intensity coefficients of various building materials corresponding to the urban land utilization type to an estimation grid, and establishing a space reference matrix of the use intensity coefficients of the building materials; and obtaining the accounting result of the building material stock of the research area based on the building material using the spatial reference matrix of the intensity coefficient and the estimated value of the total building area of the current research area.
Description
Technical Field
The invention belongs to the technical field of night light remote sensing, and particularly relates to a city building material stock estimation method and system based on a convolutional neural network.
Background
In recent decades, the total amount of global natural resource exploitation rises rapidly, about 60% of the consumption belongs to the building industry, and more than half of the total amount of the global natural resource exploitation is accumulated in urban systems as building material stock, wherein the building material stock refers to the accumulation amount of various building materials in the buildings in the boundaries of the urban systems. With the continued depth of the concept of sustainable development, it is appreciated that the vast body of building materials and the long life cycle will have a tremendous, widespread and profound impact on the environment. Accordingly, research on building material inventory has been increasingly paid attention in recent years, mainly around structural analysis, carbon emission accounting, energy demand evaluation, and the like. More importantly, the building material stock is also a potential huge secondary resource, and the effective recovery and utilization of the recoverable resources in the building materials can bring remarkable environmental, social and economic benefits, and has positive significance for exploring environment-friendly urban development roads and management modes, promoting the development of recycling economy and realizing sustainable development targets (SDG). The key premise to achieve these goals is to fully calculate and evaluate the size, composition and distribution of the current building material inventory.
Existing building material stock accounting methods are mainly divided into two categories: one type is a "Top-down" method based on macroscopic socioeconomic statistics, which allows accounting of inventory by calculating the difference between the inflow and outflow of material in the system over a period of time. However, this approach is often limited by the availability, macroscopicity and hysteresis of the statistics, which are difficult to effectively apply in cities and below, while the result of the accounting does not reveal the spatial distribution characteristics of the building material inventory. In order to cope with the above-mentioned drawbacks, in recent years, a "Bottom-up" method has been proposed by the academy, which proceeds from building monomers stepwise upward for accounting for building material inventory. The method combines building contour three-dimensional Geographic Information System (GIS) data reflecting the building space morphology with building material use intensity coefficients to account for building material inventory, so that the method has the capability of finely characterizing the spatial distribution characteristics of the building material inventory on the urban scale. However, it should be noted that this method relies on three-dimensional GIS data of the building contour, which generally covers only partially developed areas and has a high acquisition cost, thus limiting its applicability to a large extent.
In recent years, night lamplight remote sensing technology has become an effective means for quantitatively analyzing the activity intensity of human beings and the economic development level of areas by virtue of high space-time and large-range observation advantages. Research shows that the intensity of night light has strong correlation with the total building area and the material stock, and the total building area is the sum of all building areas above the ground of all buildings in the construction land range in the urban system. Based on this, researchers build an inversion model using known total building area or material inventory data and night light brightness to estimate the total building area or material inventory of an unknown area. The remote sensing method becomes a potential method, not only effectively overcomes the dependence limit of the 'bottom-up' method on three-dimensional GIS data of the building outline, but also provides space distribution information for regional building material inventory assessment. However, the existing remote sensing model mostly adopts a traditional mathematical experience model to complete inversion tasks, has limitations in terms of processing massive data and mining deep data relations, and particularly has poor performance in terms of processing nonlinear relations. In addition, the night lamplight data has an overflow effect and a saturation effect, wherein the overflow effect refers to the phenomenon that the detected lamplight distribution range is larger than an actual area due to the effect of the light halo, and the saturation effect refers to the phenomenon that the brightness is not increased after exceeding a threshold value due to the limited monitoring capability of an early remote sensing detector on high-intensity night lamplight radiation; these factors may cause large errors in the mass inventory estimation. Therefore, in order to improve the accuracy of building material inventory assessment, a more accurate and reliable model estimation method needs to be deeply explored.
Deep learning methods are receiving increasing attention due to their significant advantages in terms of efficient processing of large data and accurate mapping of nonlinear relationships. However, in the current field of building inventory remote sensing estimation, there are few models based on deep learning methods. And current building inventory estimation studies often do not adequately account for differences in urban internal regional structures when constructing night light based material inventory inversion models, which can lead to large errors in the models. In particular, in the case of similar building stock, the luminance characteristics of the urban core area are obviously different from those of the edge area, that is, there is a remarkable spatial heterogeneity in the correlation between the night light luminance value and the building material stock, which is caused by the differences in building density and lighting requirements of different attributes and functional areas. Therefore, it is necessary to build inversion models for different regional structures in the city, and apply these models in a targeted manner in the prediction process, so as to obtain more reasonable and accurate estimation results.
Disclosure of Invention
The invention aims to: aiming at the limitation of the existing urban building material stock estimation method, the invention discloses an urban building material stock estimation method and system based on a convolutional neural network.
The technical scheme is as follows: a city building material stock estimating method based on convolutional neural network is characterized in that: the method comprises the following steps:
Step 1: acquiring a historical night light remote sensing image of a research area and a city building total area of the research area matched with the time of the historical night light remote sensing image, and establishing a basic data set taking the night light remote sensing image as a characteristic image and the city building total area as a label;
Step 2: dividing a basic data set according to different types of urban regional structures to obtain sub-data sets corresponding to various types of urban regional structures, and dividing the sub-data sets corresponding to any type of urban regional structures into a training set and a testing set; the urban area structure comprises: a core region structure, a transition region structure, and an edge region structure;
Step 3: constructing a total building area estimation model, for any type of urban regional structure, training the total building area estimation model by adopting a training set corresponding to the type of urban regional structure, and evaluating the trained total building area estimation model by adopting a test set corresponding to the type of urban regional structure to obtain an available total building area estimation model of the type of urban regional structure;
Step 4: the night light remote sensing image of the current research area is utilized to be input into a corresponding available total building area estimation model, and a total building area estimation value of the current research area is obtained;
Step 5: obtaining a calculation result of building material stock in the research area based on the using intensity coefficient of various building materials in the current research area and the total building area estimated value of the current research area;
The total building area estimation model is constructed based on a convolutional neural network.
Further, the specific operation of step 1 includes:
acquiring a historical night light remote sensing image of a research area;
acquiring historical building contour vector data of a study area, the building contour vector data comprising: geographic location of the individual building, floor area of the individual building, and floor number of the individual building;
Drawing estimation grids with the same spatial resolution as that of the historical night light remote sensing image, and calculating the total building area of all monomer buildings in each estimation grid according to the following formula ：
；
In the method, in the process of the invention,To estimate grid/>Interior (th)/>Floor area of monomer building,/>To estimate grid/>Interior (th)/>Floor number of single building,/>To estimate grid/>The number of individual buildings in;
Based on the estimated grid, a sliding window is selected, and the historical night lamplight remote sensing image is cut by utilizing the sliding window, so that a plurality of lamplight image slices are generated;
For any lamplight image slice, taking the lamplight image slice as a characteristic image of the total building area of all the monomer buildings in the estimated grid in the center of the lamplight image slice, taking the total building area of all the monomer buildings in the estimated grid in the center of the lamplight image slice as a label, and constructing to obtain a basic data set.
Further, in step 2, the dividing the basic data set according to different types of urban regional structures to obtain sub data sets corresponding to different types of urban regional structures, which specifically includes:
The illumination index of each light image slice is calculated according to the following formula ：
；
In the method, in the process of the invention,Estimation of grid/>, for within-slice light imagesIs a luminance value of (1);
Illumination index based on each light image slice And dividing the basic data set to obtain sub data sets corresponding to various urban region structures.
Further, step4 specifically includes:
drawing an estimated grid with the same spatial resolution as that of the night light remote sensing image of the current research area;
based on the estimated grid, a sliding window is selected, and a night lamplight remote sensing image of a current research area is cut by utilizing the sliding window, so that a plurality of lamplight image slices are generated;
and inputting the lamplight image slice into a corresponding available total building area estimation model to obtain a total building area estimation value of the current research area.
Further, the step5 specifically includes:
obtaining the urban land utilization type in the current research area, overlapping the use intensity coefficients of various building materials corresponding to the urban land utilization type to an estimation grid, and establishing a space reference matrix of the use intensity coefficients of the building materials; the building material uses any one element in a spatial reference matrix of intensity coefficients, expressed as:
；
In the method, in the process of the invention, Represents the/>Any one element in space reference matrix of building-like material,/>Representing the element at the/>Position in a spatial reference matrix of building-like material,/>Represents the/>Type of building/>The use strength coefficient of the building-like material;
and obtaining the accounting result of the building material stock of the research area based on the building material using the spatial reference matrix of the intensity coefficient and the estimated value of the total building area of the current research area.
Further, the method for obtaining the accounting result of the building material stock in the research area based on the building material using the spatial reference matrix of the intensity coefficient and the estimated value of the total building area in the current research area comprises the following specific operations:
For each estimation grid, the calculation formula of the accounting result is as follows:
；
In the method, in the process of the invention, Representing an estimated grid/>Middle/>Building-like material inventory,/>Representing an estimated grid/>Estimated value of total building area in/>Represents the/>Type of building/>The use strength coefficient of the building-like material;
Based on each estimated building material inventory of the grid, a building material inventory accounting result of the current study area is obtained.
The invention also discloses a city building material stock estimating system based on the convolutional neural network, which comprises:
The total building area estimation module is used for inputting night light remote sensing images of the current research area into the corresponding available total building area estimation model to obtain a total building area estimation value of the current research area;
the urban building material stock estimation module is used for obtaining a building material stock accounting result of the research area based on the using intensity coefficient of various building materials in the current research area and the total building area estimation value of the current research area;
The available total building area estimation model is obtained according to the following steps:
Acquiring a historical night light remote sensing image of a research area and a city building total area of the research area matched with the time of the historical night light remote sensing image, and establishing a basic data set taking the night light remote sensing image as a characteristic image and the city building total area as a label;
dividing a basic data set according to different types of urban regional structures to obtain sub-data sets corresponding to various types of urban regional structures, and dividing the sub-data sets corresponding to any type of urban regional structures into a training set and a testing set; the urban area structure comprises: a core region structure, a transition region structure, and an edge region structure;
Constructing a total building area estimation model, for any type of urban regional structure, training the total building area estimation model by adopting a training set corresponding to the type of urban regional structure, and evaluating the trained total building area estimation model by adopting a test set corresponding to the type of urban regional structure to obtain an available total building area estimation model of the type of urban regional structure;
The total building area estimation model is constructed based on a convolutional neural network.
Further, the method for acquiring the historical night light remote sensing image of the research area and the total area of the urban building of the research area matched with the time of the historical night light remote sensing image of the research area to establish a basic data set taking the night light remote sensing image as a characteristic image and the total area of the urban building as a label specifically comprises the following steps:
acquiring a historical night light remote sensing image of a research area;
acquiring historical building contour vector data of a study area, the building contour vector data comprising: geographic location of the individual building, floor area of the individual building, and floor number of the individual building;
Drawing estimation grids with the same spatial resolution as that of the historical night light remote sensing image, and calculating the total building area of all monomer buildings in each estimation grid according to the following formula ：
；
In the method, in the process of the invention,To estimate grid/>Interior (th)/>Floor area of monomer building,/>To estimate grid/>Interior (th)/>Floor number of single building,/>To estimate grid/>The number of individual buildings in;
Based on the estimated grid, a sliding window is selected, and the historical night lamplight remote sensing image is cut by utilizing the sliding window, so that a plurality of lamplight image slices are generated;
For any lamplight image slice, taking the lamplight image slice as a characteristic image of the total building area of all the monomer buildings in the estimated grid in the center of the lamplight image slice, taking the total building area of all the monomer buildings in the estimated grid in the center of the lamplight image slice as a label, and constructing to obtain a basic data set.
Further, the dividing the basic data set according to different types of urban regional structures to obtain sub data sets corresponding to different types of urban regional structures, and the specific operations include:
The illumination index of each light image slice is calculated according to the following formula ：
；
In the method, in the process of the invention,Estimation of grid/>, for within-slice light imagesIs a luminance value of (1);
Illumination index based on each light image slice And dividing the basic data set to obtain sub data sets corresponding to various urban region structures.
Further, in the total building area estimation module, the following steps are performed:
drawing an estimated grid with the same spatial resolution as that of the night light remote sensing image of the current research area;
Based on the estimated grid, a sliding window is selected, and a night lamplight remote sensing image of a current research area is cut by utilizing the sliding window, so that a plurality of lamplight image slices are generated;
inputting the light image slice into a corresponding available total building area estimation model to obtain a total building area estimation value of the current research area;
in the city building inventory estimation module, the following steps are performed:
Obtaining the urban land utilization type in the current research area, overlapping the use intensity coefficients of various building materials corresponding to the urban land utilization type to an estimation grid, and establishing a space reference matrix of the use intensity coefficients of the building materials; the building material uses any one element in a spatial reference matrix of intensity coefficients, expressed as:
；
In the method, in the process of the invention, Represents the/>Any one element in space reference matrix of building-like material,/>Representing the element at the/>Position in a spatial reference matrix of building-like material,/>Represents the/>Type of building/>The use strength coefficient of the building-like material;
for each estimated grid, the calculation formula of the building material inventory accounting result of each estimated grid is as follows:
；
In the method, in the process of the invention, Representing an estimated grid/>Middle/>Building-like material inventory,/>Representing an estimated grid/>Estimated value of total building area in/>Represents the/>Type of building/>The use strength coefficient of the building-like material;
Based on each estimated building material inventory of the grid, a building material inventory accounting result of the current study area is obtained.
The beneficial effects are that: compared with the prior art, the invention has the following advantages:
(1) The method/system respectively establishes inversion models aiming at different regional structures in the city, effectively solves the problem of spatial heterogeneity of the correlation relationship between building stock and light brightness, avoids the limitation of a single model facing complex characteristic relationship, improves model accuracy and robustness, and is also beneficial to popularization and application of building material stock estimation in other city regions;
(2) Compared with the traditional mathematical model, the method/system of the invention can better describe the complex mapping relation between night light and the total building area, fully consider the proximity relevance of the target object, effectively exert the image processing advantage of the convolutional neural network and better mine the spatial characteristics of pixel distribution in the image slice;
(3) The method/system establishes the total building area estimation model based on the convolutional neural network, outputs the estimated value of the total building area of the grid scale of the estimated area by directly inputting the night lamplight remote sensing image of the estimated area, and combines the strength coefficient of the material to realize the estimation of the storage quantity of the building materials of the high-resolution grid scale, thereby providing scientific support for recycling the urban building resources and promoting the stable promotion of the circular economy development and the sustainable development targets.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The technical scheme of the invention is further described with reference to the accompanying drawings and the embodiments.
The accurate grasp of the scale, structure and space pattern of the building material stock is an important premise that sustainable development researches such as urban building resource recycling and circular economy are effectively developed, and aiming at the limitation of the existing urban building material stock estimation method, the embodiment discloses an urban building material stock estimation method based on a convolutional neural network, taking the urban mass stock estimation of Yangtze river delta city group in 2020 as an example, and referring to fig. 1, the estimation method of the embodiment is described in detail, and comprises the following steps:
Step 1: and acquiring a night lamplight remote sensing image of the research area and a total area data set of the urban building, and constructing a basic data set through space matching. The specific operation comprises the following steps:
acquiring 2020 night lamplight remote sensing images through a national earth system science data center, wherein the spatial resolution is 500m; and obtaining partial city building contour vector data of triangular city groups such as Nanjing, suzhou and Shanghai through a Gaode map, wherein the partial city building contour vector data comprises building geographic positions, occupied areas and floor numbers.
Drawing estimation grids with the same spatial resolution as night lamplight remote sensing images in ArcGIS software, and calculating the total building area of all building monomers in each estimation grid：
；
In the method, in the process of the invention,To estimate grid/>Interior (th)/>Floor area of monomer building,/>To estimate grid/>Interior (th)/>Floor number of single building,/>To estimate grid/>The number of individual buildings in;
Cutting night lamplight remote sensing images by adopting sliding windows with 18 estimated grids at the side length and 2 estimated grids at the step length to generate a plurality of lamplight image slices, and regarding any lamplight image slice as a characteristic image of the total building area of all single buildings in the central 2×2 estimated grid range of the corresponding area of the lamplight image slice, fully considering the proximity correlation of geographic objects; for cities with total building areas in Nanjing, suzhou, shanghai and the like, namely areas with lamplight image slices and corresponding total building areas exist at the same time, taking the lamplight image slices as characteristic images and taking the total building areas as labels; constructing a basic data set required by model training and testing; in this embodiment, the light image slice with a larger range predicts the total building area of all building units in a smaller range corresponding to the central area. In this embodiment, the total building area of all building units in the grid is estimated by using the light image slice prediction 2×2 with the 18×18 sliding window, and the model trained by the data set has better effect although the prediction result loses one time of spatial resolution. The prediction task of estimating the total building area of all building monomers in the grid by adopting the lamplight image slices with odd side lengths in the central area 1 multiplied by 1 can be completed, so that the loss of spatial resolution is avoided, but the correlation between the lamplight brightness and the building area is weakened under the scale, and the model obtained by training is difficult to ensure higher accuracy.
In this embodiment, the reason why the light image slice is used as the feature image of the total building area of all the individual buildings in the grid range estimated by 2×2 in the center of the area corresponding to the light image slice is that: the total building area estimation model framework used in the subsequent step is a CNN model, and the total building area estimation model framework starts from the grid unit scale, and for the light value and the total building area value which are positioned at the same geographic position, the corresponding total building area is estimated based on the light image slice.
Step 2: the basic data set is divided into sub data sets of different types of urban regional structures, and the training set and the testing set are respectively divided. The specific operation comprises the following steps:
Due to the difference of human activity intensity, different types of regional structures exist in cities, such as urban core areas with high building density and lighting requirements, urban border areas with low building density and lighting requirements, and certain differences exist in the mapping relationship between the light brightness and the total building area in the different types of urban regional structures. The remarkable correlation of night light brightness and human activity intensity provides powerful support for the method of identifying different regional structures such as city cores, city edges and the like through night light images, and the existing students accurately divide the city regional structures based on the light image brightness characteristics of different regions. Based on the above, in order to effectively identify samples belonging to different types of urban regional structures, and further establish a total building area estimation model for the different types of urban regional structures, the embodiment introduces an illuminance index The brightness characteristics of the light image slices are expressed and are used as the basis of sample division:
；
In the method, in the process of the invention, Estimation of grid/>, for within-slice light imagesIs a luminance value of (a).
Drawing illumination index of each light image slice in basic data setThe frequency distribution histogram of the sample can find that the curve form corresponding to the overall sample distribution histogram is the normal distribution curve superposition of different types of samples. By identifying a plurality of peaks and valleys of the overall sample distribution curve and combining expertise, the basic data set is divided into A, B, C different city region structure sample data sets in the embodiment.
And (3) splitting three Gaussian distributions by applying a Gaussian mixture model (Gaussian Mixture Model) to the overall sample distribution condition, and completing the identification and division of different types of samples. And carrying out visual operation on the dividing result on the map, and finding that the distribution pattern of the divided sample set accords with the ring layer distribution structure of the core region, the transition region and the edge region in the urban regional structure. Next, the training set and the test set are divided for each type of sample set, respectively.
Step 3: the method comprises the steps of constructing a total building area estimation model frame based on a convolutional neural network, training a total building area estimation model by utilizing a training set aiming at different types of urban regional structures, and evaluating the performance of the model by utilizing a testing set. The specific operation comprises the following steps: and constructing a total building area estimation model frame based on the convolutional neural network, and adopting a grid search strategy to adjust and optimize the network structure. The total building area estimation model framework constructed in this embodiment includes three convolution layers, one max-pooling layer and three fully connected layers. The concrete structure is as follows: 32 3×3 convolution kernels, a first convolution layer with a step length of 1, 64 3×3 convolution kernels, a second convolution layer with a step length of 1, a 2×2 pooling window, a maximum pooling layer with a step length of 2, 32 3×3 convolution kernels, a third convolution layer with a step length of 1, and three fully connected layers with neuron numbers of 128, 64, and 1, respectively; an Adam optimizer and an average absolute error MAE are adopted as loss functions, and L2 regularization is introduced to improve the robustness of the model. The total building area estimation model frame constructed in the embodiment has better long triangle performance, and can adjust the structure of the total building area estimation model frame (namely, the number and combination modes of CNN basic layers such as a convolution layer, a full connection layer and the like, and key super parameters such as the number of convolution kernels, the learning rate and the like) according to different tasks and data sets, and is not limited to the structure.
The embodiment adds a characteristic weighting processing module based on CBAM module improvement after the second convolution layer, and introduces a spatial attention mechanism for the total building area estimation model. The feature weighting processing module firstly performs the feature weighting processing on the feature of the size of the objectInput feature map/> (height x width x number of channels)Completing global maximum pooling operation and global average pooling operation at a channel level to obtain a size of/>Feature map/>And/>; Then/>And/>Spliced according to channels to formIs obtained by a convolution operation of a 3 x 3 convolution kernel with a size/>Is a feature image of (1); finally, a sigmoid activation function is input to obtain a spatial attention diagram/>And input feature map/>Multiplication to obtain weighted feature map/>. Therefore, the anti-interference capability of the model on irrelevant noise in the multidimensional features and the extraction capability of effective information are improved. According to the embodiment, a spatial attention mechanism is introduced, so that the model can identify important features in the image slice, certain irrelevant features and noise caused by overlarge selection of a neighborhood range are restrained, and the scientificity and estimation performance of modeling are improved.
After the total building area estimation model structure is determined, training and test evaluation of the corresponding category model are respectively completed based on the divided A, B, C types of sample data sets of different urban regional structures. For each type of model, firstly, optimizing key super parameters such as learning rate, batch size and the like by adopting a Bayesian optimization algorithm and adopting a random search strategy so as to determine the optimal super parameter configuration of the model; secondly, inputting a training sample set for training, continuously iterating parameters in a network in the training process of the model, and ending the training of the model when the loss function reaches the minimum and is stable; and finally, testing the model precision by adopting a test sample set. Through inspection, the overall error of the comprehensive prediction result of each model in the embodiment on the test set is less than 5%, and the determination coefficients of the true value and the prediction value of a single sampleThe value is more than 0.87, and the model has good performance.
In the embodiment, the convolutional neural network is utilized, the difference of urban region structures is considered, a total building area estimation model utilizing night light remote sensing images is established aiming at different urban region structure types, and accurate estimation of the total building area of the city on a high-resolution grid scale is realized.
Step 4: and estimating the total building area of the research area by using the night light images and the established total building area estimation model. The specific operation comprises the following steps:
calculating illumination index of estimated grid in investigation region Dividing according to the urban region structure dividing method of different types determined in the step2, inputting the divided different types of prediction samples into a total building area estimation model of corresponding types, and estimating to obtain the total building area of the research area.
Step 5: and constructing a building material use intensity coefficient space reference based on the urban land use type data and the material use intensity coefficient data.
In this embodiment, a method for distinguishing building types based on residential building and non-residential building is adopted, and different building types are estimated by using different building material use intensity coefficients, as shown in table 1.
TABLE 1 building Material use Strength coefficient (Unit: kg/m 2)
Acquiring urban land utilization type data of an estimated area through an Open street map (Open STREET MAP, OSM), according to different land types (residential areas and non-residential areas), overlapping various material utilization intensity coefficients of corresponding building types to a grid unit, and establishing a space reference matrix of the building material utilization intensity coefficients：
；
In the method, in the process of the invention,Represents the/>Space reference matrix of building-like material,/>Represents the/>Any one element in space reference matrix of building-like material,/>Representing the element at the/>Position in a spatial reference matrix of building-like material,/>Represents the/>Type of building/>Use strength coefficient of building-like material. Based on which spatial properties are added to the building material using the intensity coefficients.
Step 6: the spatial reference matrix of the intensity coefficient and the estimated total building area are combined with building materials to calculate the building material stock of a research area, the spatial pattern of the material stock is displayed from a high spatial resolution grid scale, and the distribution characteristics of the material stock are revealed, so that the effective grasp of building resources and the stable promotion of urban sustainable development targets are facilitated. The specific operation comprises the following steps: and (3) based on the total building area obtained in the step (4), calculating by combining the building materials and using a space reference matrix of the intensity coefficient to obtain a final grid scale building material stock accounting result. For each grid, the calculation formula of the accounting result is as follows:
；
In the method, in the process of the invention, Representing an estimated grid/>Middle/>Building-like material inventory,/>Representing an estimated grid/>Estimated value of total building area in/>Represents the/>Type of building/>The use strength coefficient of the building-like material; finally, a grid scale Yangtze river delta city group building material stock accounting result with the spatial resolution of 1km is obtained.
The accounting result of this embodiment shows that: the total stock of building materials in the Yangtze river delta city group in 2020 is 19473.94Tg (million tons); in the metal substances with higher recovery values, the total steel stock is 536.71Tg, the total copper stock is 392.59Tg, and the total aluminum stock is 20.91Tg; in other building materials, the total cement inventory was 2728.59Tg, the total gravel inventory was 6682.06Tg, the total sand inventory was 5193.05Tg, the total brick inventory was 1752.07Tg, and their total amount was 83.9% of the total building material inventory. From the urban unit point of view: the total stock of construction materials in Shanghai, suzhou and Nanjing is maximum, 4119.25Tg, 2526.78Tg and 1747.02Tg respectively; the total stock of building materials in pool, boat and mountain and Xuancheng was minimal, 11.08Tg, 48.00Tg and 48.75Tg respectively.
The method integrates multi-source data (night light remote sensing images, three-dimensional GIS data of building outlines and urban land utilization type data), establishes an independent convolutional neural network model by adopting a partition strategy aiming at the total building area of different regional structures in the city, and completes estimation of urban building material storage with high spatial resolution and large scale range by combining building materials to use intensity coefficients. The method is applied to the Yangtze river delta city group, realizes fine estimation of the building material stock of the city group, and reveals the spatial distribution characteristics of the city group. The invention opens up a new idea and a new way for the application of the deep learning method in the field of building material stock estimation, improves the accuracy of building material mass estimation, provides powerful support for city planning and building material resource recycling by integrating various data sources and adopting a partition model construction strategy, and is beneficial to promoting the sustainable development of cities and the optimization of resource recycling.
Claims (6)
1. A city building material stock estimating method based on convolutional neural network is characterized in that: the method comprises the following steps:
Step 1: acquiring a historical night light remote sensing image of a research area and a city building total area of the research area matched with the time of the historical night light remote sensing image, and establishing a basic data set taking the night light remote sensing image as a characteristic image and the city building total area as a label;
Step 2: dividing a basic data set according to different types of urban regional structures to obtain sub-data sets corresponding to various types of urban regional structures, and dividing the sub-data sets corresponding to any type of urban regional structures into a training set and a testing set; the urban area structure comprises: a core region structure, a transition region structure, and an edge region structure;
Step 3: constructing a total building area estimation model, for any type of urban regional structure, training the total building area estimation model by adopting a training set corresponding to the type of urban regional structure, and evaluating the trained total building area estimation model by adopting a test set corresponding to the type of urban regional structure to obtain an available total building area estimation model of the type of urban regional structure;
Step 4: the night light remote sensing image of the current research area is utilized to be input into a corresponding available total building area estimation model, and a total building area estimation value of the current research area is obtained;
Step 5: obtaining a calculation result of building material stock in the research area based on the using intensity coefficient of various building materials in the current research area and the total building area estimated value of the current research area;
the total building area estimation model is constructed based on a convolutional neural network;
The specific operation of the step1 comprises the following steps:
acquiring a historical night light remote sensing image of a research area;
acquiring historical building contour vector data of a study area, the building contour vector data comprising: geographic location of the individual building, floor area of the individual building, and floor number of the individual building;
The estimated grids with the same spatial resolution as the historical night light remote sensing image are drawn, and the total building area MFA i of all the single buildings in each estimated grid is calculated according to the following formula:
Wherein s j is the floor area of the jth single building in the estimated grid i, f j is the floor number of the jth single building in the estimated grid i, and n is the number of single buildings in the estimated grid i;
Based on the estimated grid, a sliding window is selected, and the historical night lamplight remote sensing image is cut by utilizing the sliding window, so that a plurality of lamplight image slices are generated;
For any lamplight image slice, taking the lamplight image slice as a characteristic image of the total building area of all the monomer buildings in the estimated grid in the center of the lamplight image slice, taking the total building area of all the monomer buildings in the estimated grid in the center of the lamplight image slice as a label, and constructing to obtain a basic data set;
In step 2, the basic data set is divided according to different types of urban regional structures to obtain sub data sets corresponding to different types of urban regional structures, and the specific operations include:
the illumination index L of each light image slice is calculated according to the following formula:
L＝Σli；
Wherein, l i is the brightness value of the estimated grid i in the light image slice;
and dividing the basic data set based on the illumination index L of each light image slice to obtain sub data sets corresponding to various urban regional structures.
2. The urban building mass storage estimation method based on convolutional neural network according to claim 1, wherein: the step 4 specifically comprises the following steps:
drawing an estimated grid with the same spatial resolution as that of the night light remote sensing image of the current research area;
based on the estimated grid, a sliding window is selected, and a night lamplight remote sensing image of a current research area is cut by utilizing the sliding window, so that a plurality of lamplight image slices are generated;
and inputting the lamplight image slice into a corresponding available total building area estimation model to obtain a total building area estimation value of the current research area.
3. The urban building mass storage estimation method based on convolutional neural network according to claim 2, wherein: the step 5 specifically comprises the following steps:
obtaining the urban land utilization type in the current research area, overlapping the use intensity coefficients of various building materials corresponding to the urban land utilization type to an estimation grid, and establishing a space reference matrix of the use intensity coefficients of the building materials; the building material uses any one element in a spatial reference matrix of intensity coefficients, expressed as:
Wherein R k (a, b) represents any one element in the spatial reference matrix of the kth class of building material, (a, b) represents the position of the element in the spatial reference matrix of the kth class of building material, A use strength coefficient of a kth building material representing a first building type;
and obtaining the accounting result of the building material stock of the research area based on the building material using the spatial reference matrix of the intensity coefficient and the estimated value of the total building area of the current research area.
4. A method for estimating urban building mass storage based on convolutional neural network according to claim 3, wherein: the building material based spatial reference matrix of the intensity coefficient and the total building area estimated value of the current research area are used for obtaining the accounting result of the building material stock of the research area, and the concrete operation comprises the following steps:
For each estimation grid, the calculation formula of the accounting result is as follows:
In the method, in the process of the invention, Representing the inventory of kth building material in the estimated grid i, MFA i represents the total building area estimate in the estimated grid i,/>A use strength coefficient of a kth class building material representing a first building type;
Based on each estimated building material inventory of the grid, a building material inventory accounting result of the current study area is obtained.
5. An urban building material stock estimation system based on a convolutional neural network, which is characterized in that: comprising the following steps:
The total building area estimation module is used for inputting night light remote sensing images of the current research area into the corresponding available total building area estimation model to obtain a total building area estimation value of the current research area;
the urban building material stock estimation module is used for obtaining a building material stock accounting result of the research area based on the using intensity coefficient of various building materials in the current research area and the total building area estimation value of the current research area;
The available total building area estimation model is obtained according to the following steps:
Acquiring a historical night light remote sensing image of a research area and a city building total area of the research area matched with the time of the historical night light remote sensing image, and establishing a basic data set taking the night light remote sensing image as a characteristic image and the city building total area as a label;
dividing a basic data set according to different types of urban regional structures to obtain sub-data sets corresponding to various types of urban regional structures, and dividing the sub-data sets corresponding to any type of urban regional structures into a training set and a testing set; the urban area structure comprises: a core region structure, a transition region structure, and an edge region structure;
Constructing a total building area estimation model, for any type of urban regional structure, training the total building area estimation model by adopting a training set corresponding to the type of urban regional structure, and evaluating the trained total building area estimation model by adopting a test set corresponding to the type of urban regional structure to obtain an available total building area estimation model of the type of urban regional structure;
the total building area estimation model is constructed based on a convolutional neural network;
The method for acquiring the historical night light remote sensing image of the research area and the urban building total area of the research area matched with the time of the historical night light remote sensing image of the research area establishes a basic data set taking the night light remote sensing image as a characteristic image and the urban building total area as a label, and specifically comprises the following steps:
acquiring a historical night light remote sensing image of a research area;
acquiring historical building contour vector data of a study area, the building contour vector data comprising: geographic location of the individual building, floor area of the individual building, and floor number of the individual building;
The estimated grids with the same spatial resolution as the historical night light remote sensing image are drawn, and the total building area MFA i of all the single buildings in each estimated grid is calculated according to the following formula:
Wherein s j is the floor area of the jth single building in the estimated grid i, f j is the floor number of the jth single building in the estimated grid i, and n is the number of single buildings in the estimated grid i;
Based on the estimated grid, a sliding window is selected, and the historical night lamplight remote sensing image is cut by utilizing the sliding window, so that a plurality of lamplight image slices are generated;
For any lamplight image slice, taking the lamplight image slice as a characteristic image of the total building area of all the monomer buildings in the estimated grid in the center of the lamplight image slice, taking the total building area of all the monomer buildings in the estimated grid in the center of the lamplight image slice as a label, and constructing to obtain a basic data set;
Dividing the basic data set according to different types of urban regional structures to obtain sub data sets corresponding to the types of urban regional structures, wherein the specific operation comprises the following steps:
the illumination index L of each light image slice is calculated according to the following formula:
L＝Σli；
Wherein, l i is the brightness value of the estimated grid i in the light image slice;
and dividing the basic data set based on the illumination index L of each light image slice to obtain sub data sets corresponding to various urban regional structures.
6. The urban building mass storage estimation system based on convolutional neural network according to claim 5, wherein: in the total building area estimation module, the following steps are performed:
drawing an estimated grid with the same spatial resolution as that of the night light remote sensing image of the current research area;
Based on the estimated grid, a sliding window is selected, and a night lamplight remote sensing image of a current research area is cut by utilizing the sliding window, so that a plurality of lamplight image slices are generated;
inputting the light image slice into a corresponding available total building area estimation model to obtain a total building area estimation value of the current research area;
in the city building inventory estimation module, the following steps are performed:
Obtaining the urban land utilization type in the current research area, overlapping the use intensity coefficients of various building materials corresponding to the urban land utilization type to an estimation grid, and establishing a space reference matrix of the use intensity coefficients of the building materials; the building material uses any one element in a spatial reference matrix of intensity coefficients, expressed as:
Wherein R k (a, b) represents any one element in the spatial reference matrix of the kth class of building material, (a, b) represents the position of the element in the spatial reference matrix of the kth class of building material, A use strength coefficient of a kth building material representing a first building type;
for each estimated grid, the calculation formula of the building material inventory accounting result of each estimated grid is as follows:
In the method, in the process of the invention, Representing the inventory of kth building material in the estimated grid i, MFA i represents the total building area estimate in the estimated grid i,/>A use strength coefficient of a kth class building material representing a first building type;
Based on each estimated building material inventory of the grid, a building material inventory accounting result of the current study area is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410027285.1A CN117541928B (en) | 2024-01-09 | 2024-01-09 | Urban building material stock estimation method and system based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410027285.1A CN117541928B (en) | 2024-01-09 | 2024-01-09 | Urban building material stock estimation method and system based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117541928A CN117541928A (en) | 2024-02-09 |
| CN117541928B true CN117541928B (en) | 2024-04-19 |
Family
ID=89788439
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410027285.1A Active CN117541928B (en) | 2024-01-09 | 2024-01-09 | Urban building material stock estimation method and system based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117541928B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101728137B1 (en) * | 2016-02-04 | 2017-04-19 | (주)한라지리정보 | Method for land-cover item images classification by using satellite picture and GIS |
| CN107229913A (en) * | 2017-05-23 | 2017-10-03 | 国家地理空间信息中心 | Density of population analysis system based on high score satellite remote sensing date combination building height |
| CN112149594A (en) * | 2020-09-29 | 2020-12-29 | 同济大学 | Urban construction assessment method based on deep learning and high-resolution satellite images |
| CN114612800A (en) * | 2022-04-07 | 2022-06-10 | 南京信息工程大学 | Method and system for accounting urban building material stock and space-time change |
| CN115630567A (en) * | 2022-09-29 | 2023-01-20 | 天津大学 | Coastal zone soil organic carbon reserve simulation and prediction method |
| CN117011109A (en) * | 2023-08-17 | 2023-11-07 | 云南大学 | Urban road material stock accounting method |
| CN117078925A (en) * | 2023-07-27 | 2023-11-17 | 北京工业大学 | Building rubbish annual output accurate calculation method based on RDSA-deep LabV3+ network |
-
2024
- 2024-01-09 CN CN202410027285.1A patent/CN117541928B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101728137B1 (en) * | 2016-02-04 | 2017-04-19 | (주)한라지리정보 | Method for land-cover item images classification by using satellite picture and GIS |
| CN107229913A (en) * | 2017-05-23 | 2017-10-03 | 国家地理空间信息中心 | Density of population analysis system based on high score satellite remote sensing date combination building height |
| CN112149594A (en) * | 2020-09-29 | 2020-12-29 | 同济大学 | Urban construction assessment method based on deep learning and high-resolution satellite images |
| CN114612800A (en) * | 2022-04-07 | 2022-06-10 | 南京信息工程大学 | Method and system for accounting urban building material stock and space-time change |
| CN115630567A (en) * | 2022-09-29 | 2023-01-20 | 天津大学 | Coastal zone soil organic carbon reserve simulation and prediction method |
| CN117078925A (en) * | 2023-07-27 | 2023-11-17 | 北京工业大学 | Building rubbish annual output accurate calculation method based on RDSA-deep LabV3+ network |
| CN117011109A (en) * | 2023-08-17 | 2023-11-07 | 云南大学 | Urban road material stock accounting method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117541928A (en) | 2024-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shi et al. | Exploring spatiotemporal patterns of electric power consumption in countries along the Belt and Road | |
| CN107703564B (en) | Rainfall prediction method and system and electronic equipment | |
| CN102332210B (en) | Method for extracting real-time urban road traffic flow data based on mobile phone positioning data | |
| CN106228808B (en) | City expressway travel time prediction method based on Floating Car space-time grid data | |
| CN108416686A (en) | A kind of Eco-Geo-Environment Type division method based on Coal Resource Development | |
| CN109359166B (en) | Space growth dynamic simulation and driving force factor contribution degree synchronous calculation method | |
| CN103336894B (en) | A kind of city plot ratio automatic Mesh Partition Method | |
| CN102184423B (en) | Full-automatic method for precisely extracting regional impervious surface remote sensing information | |
| CN111160622A (en) | Scenic spot passenger flow prediction method and device based on hybrid neural network model | |
| CN113505521B (en) | Urban waterlogging rapid forecasting method based on neural network-numerical simulation | |
| CN108898096B (en) | High-resolution image-oriented information rapid and accurate extraction method | |
| Zhang et al. | Using street view images to identify road noise barriers with ensemble classification model and geospatial analysis | |
| Tian et al. | Suburban identification based on multi-source data and landscape analysis of its construction land: A case study of Jiangsu Province, China | |
| CN109784228A (en) | A kind of photovoltaic plant identifying system and method based on image recognition technology | |
| Li et al. | Exploring multiple crowdsourced data to learn deep convolutional neural networks for road extraction | |
| Huang et al. | Comprehensive urban space representation with varying numbers of street-level images | |
| CN109064750A (en) | Urban road network traffic estimation method and system | |
| CN109034073A (en) | Predict method, system, equipment and the medium of building demolition waste yield | |
| CN117541928B (en) | Urban building material stock estimation method and system based on convolutional neural network | |
| CN115497006B (en) | Urban remote sensing image change depth monitoring method and system based on dynamic mixing strategy | |
| CN117151499A (en) | Monitoring and evaluating method and system for homeland space planning | |
| CN114612800A (en) | Method and system for accounting urban building material stock and space-time change | |
| He et al. | Modeling multi-type urban landscape dynamics along the horizontal and vertical dimensions | |
| Zhu et al. | Rural road network planning based on 5g and traffic big data | |
| CN113743789A (en) | Shared bicycle spatial distribution quantity estimation method based on urban built-up environmental factors |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |