KR102516588B1 - A measurement device, method and program that provides measured and estimated values of fine dust concentration through satellite image analysis using an artificial intelligence model - Google Patents

Abstract

The present invention relates to a measurement apparatus, method, and program for providing measured and estimated values of fine dust concentration through satellite image analysis using an artificial intelligence model. A fine dust concentration measurement apparatus according to the present invention comprises: at least one processor; and a memory storing instructions that instruct the at least one processor to perform at least one step. According to the present invention, errors that may occur due to regional differences can be reduced.

Description

Measuring device, method and program that provide measured and estimated values of fine dust concentration through satellite image analysis using artificial intelligence model ARTIFICIAL INTELLIGENCE MODEL}

The present invention relates to a measuring device, method and program for providing a fine dust concentration measurement value and estimated value through satellite image analysis using an artificial intelligence model.

Unless otherwise indicated herein, material described in this section is not prior art to the claims in this application, and inclusion in this section is not an admission that it is prior art.

As the adverse effects of fine dust on the human body emerge, interest in the concentration of fine dust in the air is increasing.

Accordingly, the number of installed fine dust observatories is increasing, and a technology for providing the concentration of fine dust in the air using the value measured at the fine dust observatories is being developed. In addition, fine dust concentration measuring devices in the form that users can personally own are being developed.

However, there is a problem in that it is difficult to accurately provide the concentration of fine dust in an area far away from the observatory.

In addition, in the case of a personal fine dust concentration measuring device, there is a problem that the price of the measuring device is excessively increased for accurate measurement, and a measurement device with an appropriate price causes a large error between the measured value and the actual fine dust concentration. there is

Korean Registered Patent No. 10-1810211 (registered on December 12, 2017)

According to one embodiment of the present invention, an apparatus for measuring the concentration of fine dust through a satellite image of the ground surface is provided.

In addition, the fine dust concentration measuring device, at least one processor (processor); and a memory for storing instructions instructing the at least one processor to perform at least one step.

In the at least one step, the satellite image is divided into a plurality of predetermined areas, and the areas are classified into an observable area and an unobservable area based on a ratio of an area occupied by clouds to a total area of the area. doing; obtaining a concentration of fine dust in the observable region through a pre-learned concentration measurement model; obtaining a predicted value of the fine dust concentration in the unobservable region using a pre-learned concentration estimation model and geographic information of the unobservable region, and correcting the predicted value based on the fine dust concentration in the observable region; And calculating a residual corresponding to the predicted value through a kriging analysis of the fine dust concentration and the predicted value of the observable region, and calculating the fine dust concentration of the unobservable region based on the predicted value and the residual steps may be included.

In addition, the concentration measurement model is generated through machine learning for learning data generated by labeling a learning satellite image with a fine dust concentration measurement value corresponding to the learning satellite image, and a plurality of feature vectors from the observable region. and calculates the concentration of fine dust in the observable region based on a plurality of weather vectors generated based on the weather information of the observable region and the plurality of feature vectors, and the concentration estimation model, the geography for learning It is generated through machine learning for learning data generated by labeling information with the fine dust concentration measurement value corresponding to the geographic information for learning, and calculating the predicted value through regression analysis on at least one independent variable determined from the geographic information. can do.

In addition, the plurality of feature vectors are extracted through a convolutional neural network (CNN), and the fine dust concentration corresponding to the observable region may be calculated through a random forest regressor.

In addition, in the at least one step, based on the geographic information corresponding to the satellite image, determining an area in which an expected value of fine dust emission is equal to or greater than a predetermined standard among the entire areas of the satellite image as a monitoring area, and The method may further include generating a time-sequential fine dust concentration pattern based on the concentration of fine dust in the monitoring area for a period of time.

In addition, the step of generating a time-sequential fine dust concentration pattern based on the fine dust concentration of the monitoring area for the preset period of time includes the ratio of the area of the residential area to the area of the entire area, the ratio of the area of the industrial area to the area of the entire area, the total Calculating the expected value based on the commercial area area ratio to the area, the length and area of the road, the average number of buildings, the average building area, the average floor area ratio, and environmental pollution facilities.

According to an embodiment of the present invention, the fine dust concentration is calculated by a fine dust concentration measurement model and a fine dust concentration estimation model that match the region where the satellite image is collected. Through this, errors that may occur due to regional variations can be reduced.

According to another embodiment of the present invention, the concentration of fine dust is calculated through a satellite image, geographic information corresponding to the satellite image, and meteorological information corresponding to the satellite image. Through this, the accuracy of the fine dust concentration calculated for an area without an observatory can be improved.

According to another embodiment of the present invention, it is possible to continuously monitor a monitoring area for an area with a high expected value for fine dust emission.

1 is a schematic diagram of a fine dust concentration measurement system according to an embodiment.
2 is a block diagram showing functional modules of the fine dust concentration measuring device according to FIG. 1 by way of example.
3 is a diagram conceptually illustrating a process of collecting learning data for each of a plurality of regions.
4 is a flowchart illustrating a process of generating a concentration measurement model by the concentration measurement model generation unit according to FIG. 2 .
5 is a diagram conceptually showing a convolutional neural network (CNN) and a random forest regressor used to generate a concentration measurement model.
6 is a flowchart illustrating a process of generating a concentration estimation model by the concentration estimation model generation unit according to FIG. 2 .
FIG. 7 is a diagram conceptually illustrating a process of classifying a region of a satellite image by the region classification unit according to FIG. 2 .
FIG. 8 is a flowchart illustrating a process of obtaining the fine dust concentration corresponding to the observable area of the received satellite image by the concentration measurement unit according to FIG. 2 .
FIG. 9 is a flowchart illustrating a process of obtaining the concentration of fine dust corresponding to an unobservable region of a received satellite image by the concentration estimator according to FIG. 2 .
10 is a flowchart illustrating a process of determining whether an abnormal pattern of fine dust concentration is generated by the abnormal pattern determination unit according to FIG. 2 .
FIG. 11 is a diagram showing the hardware configuration of the device according to FIG. 1 by way of example.

Since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. Like reference numerals have been used for like elements throughout the description of each figure.

Terms such as first, second, A, and B may be used to describe various components, but the components should not be limited by the terms. These terms are only used for the purpose of distinguishing one component from another. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element, without departing from the scope of the present invention. The terms and/or include any combination of a plurality of related recited items or any of a plurality of related recited items.

It is understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, but other elements may exist in the middle. It should be. On the other hand, when an element is referred to as “directly connected” or “directly connected” to another element, it should be understood that no other element exists in the middle.

Terms used in this application are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, the terms "include" or "have" are intended to designate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in the present application, they should not be interpreted in an ideal or excessively formal meaning. don't

The term 'fine dust' used below means fine dust having a predetermined particle diameter or less. For example, fine dust may mean PM10, which is a particle having a diameter of 10 μm or less. For example, fine dust may mean PM2.5 having a particle diameter of 2.5 μm or less.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a schematic diagram of a fine dust concentration measurement system according to an embodiment.

Referring to FIG. 1, the fine dust concentration measurement system includes a device 100, a user terminal 200, a satellite server 300, an observatory 400, a weather server 500, and a Geographic Information System (GIS) server 600. ) may be included.

The user terminal 200 is a terminal of a user who wants to use the measured fine dust concentration providing service, registers user information in the device 100, and uses various functions of the fine dust concentration providing service through the device 100. can

The artificial satellite server 300 stores a satellite image captured by an artificial satellite operating away from the earth's surface. In one embodiment, the artificial satellite may be an artificial satellite operating at a predetermined interval from the surface of the earth. In one embodiment, the satellite server 300 may receive satellite images from the satellite at preset time intervals. In one embodiment, the artificial satellite image may be an image of a preset resolution or a preset size.

The satellite server 300 transmits the satellite image received from the satellite to the device 100 .

The observatory 400 measures the concentration of fine dust at the installed location. A plurality of observation stations 400 may be installed in each of a plurality of regions. In one embodiment, the fine dust concentration observation station 400 may be installed in different numbers for each region. The fine dust concentration observation station 400 can measure the concentrations of ultrafine dust (PM2.5), fine dust (PM10), ozone (O3), nitrogen dioxide (NO2), carbon monoxide (CO), and sulfur dioxide (SO2). there is.

The plurality of observatories 400 transmit the measured concentration of fine dust to the device 100. In one embodiment, when there is a separate observatory server (not shown) that receives all of the fine dust concentrations measured at the plurality of observatories 400, the device 100 measures the fine dust concentration from the observatory server (not shown). can receive

The weather server 500 is a server that collects weather information and is connected to sensors that measure weather information in an information communication manner. In one embodiment, the meteorological information includes cloud formation area, initial precipitation, atmospheric instability, atmospheric motion vector, aerosol optical thickness, yellow dust optical thickness, aerosol particle size, temperature and humidity profile, cloudiness, cloud top, cloud top temperature, atmospheric pressure, It may include altitude, sea surface temperature, aerosol visibility, cloud optical thickness, cloud particle effective radius, shortwave radiation, longwave radiation, etc.

The weather server 500 transmits weather information to the device 100 .

The GIS server 600 stores geographic information about a preset area. The preset area may be divided into a plurality of subregions (Grid) having a preset size, and geographic information may be stored for each of the plurality of subregions. In one embodiment, the geographic information, for the detailed area, the ratio of the area of the residential area to the total area, the ratio of the area of the industrial area to the total area, the ratio of the area of the commercial area to the total area, the length and area of roads, the average number of buildings, the average Information on building area, average floor area ratio, type and number of environmental polluting facilities, etc. may be included. In one embodiment, the geographic information may include a ratio of a barren area to a total area, a ratio of a grassland area to a total area, and the like. Geographical information is not limited to the above examples, and various pieces of information related to fine dust generation may be included in the geographic information.

The GIS server 500 transmits geographic information to the device 100.

Examples of the user terminal 200, the satellite server 300, the observatory 400, the weather server 500, and the GIS server 600 include a desktop computer, a laptop computer, and a notebook computer capable of communicating. (notebook), smart phone, tablet PC, mobile phone, smart watch, smart glass, e-book reader, PMP (portable multimedia player) , portable game console, navigation device, digital camera, digital multimedia broadcasting (DMB) player, digital audio recorder, digital audio player, digital video recorder ), a digital video player, a personal digital assistant (PDA), and the like. The observatory 400 will be configured to measure the concentrations of ultrafine dust (PM2.5), fine dust (PM10), ozone (O3), nitrogen dioxide (NO2), carbon monoxide (CO), and sulfur dioxide (SO2). can

The apparatus 100 calculates the fine dust concentration based on the information received from the satellite server 300, the observatory 400, the weather server 500, and the GIS server 600, and calculates the fine dust concentration to the user terminal ( 200) may be at least one server.

The device 100 may provide the fine dust concentration to the user terminal 200 by operating a fine dust concentration providing service.

The device 100 classifies a satellite image of the ground surface corresponding to the observatory 400 among a plurality of satellite images received from the satellite server 300 as a satellite image for learning, and a satellite image for learning and a satellite image for learning corresponding to the satellite image for learning. A concentration measurement model may be generated based on the fine dust concentration and satellite images for learning and corresponding meteorological information.

The device 100 classifies geographic information corresponding to the location of the observatory 400 among a plurality of geographic information received from the GIS server 600 as geographic information for learning, and geographic information for learning and fine details corresponding to the geographic information for learning. A concentration estimation model may be generated based on the dust concentration.

The apparatus 100 may divide each of the plurality of satellite images received from the satellite server 300 into a plurality of regions, and classify the plurality of regions into an observable region and an unobservable region. The apparatus 100 may classify an area in which the ratio of the cloud area to the total area is equal to or greater than a preset standard among the plurality of areas as an unobservable area.

The device 100 may receive a satellite image from the satellite server 300 and obtain the concentration of fine dust from an observable region of the satellite image through a concentration measurement model.

The apparatus 100 may receive a satellite image from the satellite server 300 and obtain a predicted value of the concentration of fine dust from an unobservable region of the satellite image through a concentration estimation model.

The apparatus 100 may calculate a residual for the predicted value through a Kriging analysis of the predicted value of the fine dust concentration in the unobservable area and the fine dust concentration in the observable area.

The apparatus 100 may correct the predicted value of the fine dust concentration for the unobservable region based on the predicted value and the residual.

The apparatus 100 may generate a distribution map of the fine dust concentration for the entire area where the satellite image is received, based on the fine dust concentration of the observable area and the predicted value of the fine dust concentration of the unobservable area.

The apparatus 100 may set an area corresponding to the geographic information as a monitoring area when the expected value of fine dust emission based on the geographic information is equal to or greater than a preset standard.

The apparatus 100 may generate a time-sequential fine dust concentration pattern based on the fine dust concentration in the monitoring area collected during a preset period.

The apparatus 100 may determine that an abnormal pattern has occurred when the fine dust concentration corresponding to the monitoring area deviates from the time-sequential fine dust concentration pattern.

Fig. 2 is a block diagram illustrating functional modules of the device according to Fig. 1 by way of example;

Referring to FIG. 2, the apparatus 100 includes a model generator 101 for generating a concentration measurement model and a concentration estimation model, and an area classification unit 102 for classifying a received satellite image into an observable region and an unobservable region. , the concentration measurement unit 103 for measuring the concentration of fine dust in the observable region through the concentration measurement model, the predicted value of the fine dust concentration in the unobservable region through the concentration estimation model, and the estimated predicted value using the Kriging method A concentration estimator 104 corrected through a concentration distribution map generator 105 that generates a distribution map of fine dust concentrations based on the fine dust concentrations of the observable area and the non-observable area, and detects the occurrence of abnormal patterns in the monitoring area. An abnormal pattern determining unit 106 may be included.

The model generation unit 101 includes a concentration measurement model generation module 101a for generating a concentration measurement model based on satellite images for learning and a concentration estimation model generation module 101b for generating a concentration estimation model based on geographic information for learning. ).

In one embodiment, the model generation unit 101 may generate a plurality of clusters through clustering of a plurality of satellite images for learning, and may generate a density measurement model corresponding to each of the plurality of clusters. . As clustering methods, K-Means Clustering, Mean-Shift Clustering, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), Expectation-Maximization (EM) Clustering using Gaussian Mixture Models (GMM), Agglomerative Hierarchical Clustering, and the like can be used. .

In one embodiment, the model generator 101 may generate a plurality of clusters through clustering of a plurality of geographic information for learning, and may generate a concentration estimation model corresponding to each of the plurality of clusters. .

In one embodiment, clustering may be performed on datasets grouped by region. For example, a plurality of satellite images for learning or a plurality of geographic information for learning may be grouped by region to generate regional datasets, and clustering may be performed on the regional datasets.

In one embodiment, clustering may be performed on datasets grouped by season. For example, a plurality of satellite images for learning or a plurality of geographic information for learning may be grouped by season to create seasonal datasets, and clustering may be performed on the seasonal datasets.

In one embodiment, clustering may be performed on datasets grouped by season and region. For example, a plurality of satellite images for learning or a plurality of geographic information for learning are grouped by season and region to create datasets grouped by season and region, and clustering is performed on the datasets grouped by season and region. It can be.

3 is a diagram conceptually illustrating a process of collecting learning data for each of a plurality of regions.

Referring to FIG. 3 , satellite images and geographic information on the ground surface corresponding to the observatory 400 are used as learning data.

The model generating unit 101 determines the received satellite image of the ground surface at a position corresponding to the plurality of observatories 400 as a satellite image for learning, and determines the determined satellite image for learning and the fine dust concentration measurement value corresponding to the image for learning. Labeling creates training data. In addition, the model generating unit 101 may generate a concentration measurement model through machine learning on a plurality of learning data.

In addition, the model generating unit 101 determines the received geographic information of a plurality of observatories 400 and corresponding positions as geographic information for learning, and determines the determined geographic information for learning and the fine dust concentration measurement value corresponding to the geographic information for learning. Labeling creates training data. In addition, the model generating unit 101 may generate a concentration estimation model through machine learning on a plurality of learning data.

In an embodiment, the model generation unit 101 may determine that a training satellite image having a ratio of an area of a cloud region to an area of the entire region of the satellite images for learning is equal to or greater than a predetermined standard as noise and may remove the satellite image for learning.

The model generation unit 101 groups a plurality of training data by region and season to generate regional and seasonal datasets, generates a plurality of clusters through clustering of the regional and seasonal datasets, and generates a plurality of clusters. A concentration measurement model and a concentration estimation model corresponding to each may be created.

Since a plurality of concentration measurement models and concentration estimation models corresponding to each of the clusters are generated, regional characteristics and seasonal characteristics can be well reflected in the concentration measurement models and concentration estimation models. Through this, when calculating the fine dust concentration, it is possible to reduce errors that may occur due to regional and seasonal deviations.

The number of observatories 400 installed to measure the concentration of fine dust is different for each region. Since the satellite image and geographic information of the ground surface corresponding to the observatory 400 are used as learning data, the data amount of the learning data varies by region.

The model generation unit 101 classifies a plurality of satellite images for training to correspond to each of a plurality of regions, generates a plurality of datasets corresponding to each of a plurality of regions, and generates the minimum data among the plurality of datasets. Under sampling may be performed on a plurality of datasets based on the data amount of the dataset having the amount. The model generator 101 may generate a plurality of clusters by clustering a plurality of undersampled datasets. The model generator 101 may generate a concentration measurement model and a concentration estimation model corresponding to each of the plurality of clusters. Since the amount of data in the datasets is uniformed through undersampling, the characteristics of the dataset in the region where the amount of data is small can be reflected in the concentration measurement model and the concentration estimation model.

FIG. 4 is a flowchart illustrating a process of generating a concentration measurement model by the concentration measurement model generation unit 101 according to FIG. 2 .

First, the concentration measurement model generation module 101a receives a plurality of satellite images for learning including a ground surface image of a position corresponding to each of the plurality of observatories 400 (S110).

In addition, the concentration measurement model generation module 101a classifies a plurality of training images to correspond to each of a plurality of regions, and generates a plurality of datasets corresponding to each of a plurality of regions (S120).

In addition, the concentration measurement model generation module 101a removes a satellite image for learning that is determined to be noise among a plurality of satellite images for learning included in each of a plurality of datasets (S130).

In one embodiment, the concentration measurement model generation module 101a may determine that a satellite image for learning having a ratio of a cloud area to a total area of a plurality of satellite images for learning is equal to or greater than a preset standard as noise and may remove it.

In addition, the concentration measurement model generation module 101a performs undersampling on a plurality of datasets based on the data amount of the dataset having the minimum data amount among the plurality of datasets from which noise has been removed (S140). ).

In addition, the concentration measurement model generating module 101a generates a plurality of clusters by clustering the plurality of undersampled datasets (S150).

In addition, the concentration measurement model generation module 101a generates a concentration measurement model corresponding to each of a plurality of clusters (S160).

5 is a diagram conceptually showing a convolutional neural network (CNN) and a random forest regressor used to generate a concentration measurement model.

The concentration measurement model generation module 101a outputs a plurality of feature vectors from satellite images for training through a convolutional neural network (CNN).

In one embodiment, a convolutional neural network (CNN) may include 10 convolutional layers and 4 max pooling layers. Feature maps output from the first convolutional layer and the second convolutional layer are downsampled by the first Max Pooling layer. The downsampled feature map may be output through the third convolution layer and the fourth convolution layer, and the output feature map is downsampled by the second Max Pooling layer. The downsampled feature map may be output through the fifth convolution layer, the sixth convolution layer, and the seventh convolution layer, and the output feature map is downsampled by the third max pooling layer. The downsampled feature map may be output through the eighth convolution layer, the eighth convolution layer, and the tenth convolution layer, and the output feature map is downsampled by the fourth Max Pooling layer. The downsampled feature map is output through the 11th convolution layer, the 12th convolution layer, and the 13th convolution layer. The output feature map is downsampled by the Global Max Pooling layer. The downsampled feature map is output through a fully connected layer. A fully connected layer may consist of a plurality of layers, and the number of initially input feature vectors may be 512, and the number of finally output feature vectors may consist of 128.

In addition, the concentration measurement model generating module 101a may concatenate a plurality of feature vectors output through a convolutional neural network (CNN) and a plurality of weather vectors. The concentration measurement model generating module 101a may vectorize weather information corresponding to a satellite image for learning among weather information received from the weather server 500 . In one embodiment, vectors in Temperature (T), Relative Humidity (RH), Wind Speed (WS), and Sea Level Pressure (SLP) may be used as weather vectors.

In addition, the concentration measurement model generation module 101a learns the concentration measurement model so that the concentration of fine dust is calculated from a plurality of connected feature vectors and a plurality of meteorological vectors through a random forest regressor. A random forest regressor is a type of ensemble learning method used in regression analysis, and outputs an average prediction value from a plurality of decision trees constructed in the learning process.

In addition, the concentration measurement model generation module 101a may evaluate the learned concentration measurement model through a test satellite image, and adjust hyperparameters of the random forest based on the evaluation result.

6 is a flowchart illustrating a process of generating a concentration estimation model by the concentration estimation model generation unit according to FIG. 2 .

First, the concentration estimation model generating module 101b determines geographic information of a position corresponding to each of a plurality of observatories among a plurality of geographic information as geographic information for learning (S210).

In addition, the concentration estimation model generation module 101b selects at least one independent variable from the geographic information for learning, and selects the concentration of fine dust measured at an observatory corresponding to the geographic information for learning as a dependent variable (S220).

In addition, the concentration estimation model generation module 101b generates a learning data set based on at least one independent variable and a dependent variable (S230).

In addition, the concentration estimation model generation module 101b generates a concentration estimation model through regression analysis on the generated learning data set (S240).

In one embodiment, multiple regression analysis may be used for regression analysis.

The generated concentration estimation model can be defined by Equation 1 below.

In Equation 1, s0 means the target position to estimate the concentration, ZR (s0) means the concentration estimated at s0, qk (s0) means independent variables in s0, βk is independent variables It means the regression coefficient for , and p-1 means the number of independent variables. q0(s0) is 1.

Various information included in the geographic information for learning can be set as independent variables, and independent variables that have a relatively large influence on the dependent variable can be determined through regression analysis on the learning dataset.

FIG. 7 is a diagram conceptually illustrating a process of classifying a region of a satellite image by the region classification unit 102 according to FIG. 2 .

When the satellite image is received from the artificial satellite server 300, the region classification unit 102 divides the received satellite image into a plurality of regions, and classifies the divided regions into an observable region and an unobservable region. The region classifying unit 102 may classify a region in which a ratio of a cloud area to a total area is equal to or greater than a predetermined standard among a plurality of regions as an unobservable region.

In an embodiment, the area classification unit 102 may classify a plurality of areas into an observable area and an unobservable area based on weather information received from the weather server 500 . Weather information may include information about areas created by clouds.

In an embodiment, the area classification unit 102 may classify a plurality of areas into an observable area and an unobservable area through a previously learned area classification model to classify cloud areas. The image classification model may be generated through supervised learning using a convolutional neural network (CNN).

The region classification unit 102 transmits the classified observable region to the concentration measuring unit 103 .

The region classification unit 102 transmits the classified unobservable region to the density estimation unit 104 .

FIG. 8 is a flowchart illustrating a process of obtaining the fine dust concentration corresponding to the observable area of the received satellite image by the concentration measurement unit according to FIG. 2 .

First, when the observable region is received, the concentration measuring unit 103 acquires the fine dust concentration corresponding to the observable region through the pre-learned concentration measurement model.

First, the concentration measurement model determines a plurality of feature vectors from the observable region (S210). Specifically, a plurality of feature vectors are extracted through a convolutional neural network (CNN) included in the density measurement model.

In addition, the concentration measurement unit 103 generates a plurality of weather vectors based on the weather information corresponding to the observable region (S220). Specifically, the concentration measurer 103 may generate weather vectors by converting weather information corresponding to an observable region among weather information received from the weather server 500 into a plurality of vectors. In one embodiment, weather vectors may be defined as feature vectors according to temperature (T), relative humidity (RH), wind speed (WS) and sea level pressure (SLP). can

In addition, the concentration measurement unit 103 concatenates a plurality of feature vectors and a plurality of weather vectors (S230).

In addition, the concentration measurement model calculates the fine dust concentration corresponding to the observable region based on the plurality of feature vectors and the plurality of meteorological vectors connected to each other (S240). Specifically, the fine dust concentration is calculated from a plurality of feature vectors and a plurality of meteorological vectors connected through a random forest regressor included in the concentration measurement model.

The concentration measurement unit 103 obtains the concentration of fine dust from the concentration measurement model.

FIG. 9 is a flowchart illustrating a process of obtaining the concentration of fine dust corresponding to an unobservable region of a received satellite image by the concentration estimator according to FIG. 2 .

First, when the unobservable area is received, the concentration estimation unit 104 obtains a predicted value for the concentration of fine dust in the unobservable area from geographic information corresponding to the unobservable area through the previously learned concentration estimation model (S310). ).

The concentration estimator 104 extracts independent variables having a relatively strong correlation with the concentration of fine dust, determined in the process of learning the concentration estimation model, from geographic information corresponding to the unobservable region. In addition, the concentration estimation unit 104 inputs the extracted independent variables to the concentration estimation model, and obtains a predicted value for the fine dust concentration of the unobservable region output from the concentration estimation model.

In addition, the concentration estimator 104 calculates a residual corresponding to the predicted value through kriging analysis.

The residual may be defined by Equation 2 below.

In Equation 2, s0 means a target position to estimate the concentration, e(s0) means a residual at s0, and wi(s0) is a kriging weight corresponding to each of a plurality of sample positions (Kriging weight) , and e(si) means a regression residual corresponding to each of a plurality of sample positions. n means the number of sample positions.

In one embodiment, a plurality of contiguous regions adjacent to the unobservable region may be used as the plurality of sample locations.

A kriging weight corresponding to each of the plurality of sample positions may be determined based on a variogram function. The variogram function is a measure of the similarity of data at a certain distance.

The empirical variogram function may be defined by Equation 3 below.

In Equation 3, γ (h) means a variogram function, Z (x _i ) means the concentration of fine dust at sample locations, and h means the distance between sample locations. n means the number of sample location pairs. In one embodiment, a distance between sample locations may be set based on a size of a preset observable area.

A theoretical variogram function may be modeled from an empirical variogram function, and a kriging weight may be determined through covariance derived from the theoretical variogram function. As the theoretical variogram function, a spherical function, an exponential function, a Gaussian function, or the like may be used.

When the residual is calculated, the concentration estimator 104 calculates the fine dust concentration of the unobservable region based on the predicted value and the residual (S330).

The concentration of fine dust in the unobservable region can be defined by Equation 4 below.

In Equation 4, s0 means the target position for which the concentration is to be estimated, ZRK(s0) means the estimated concentration at s0, qk(s0) means independent variables at s0, and βk is independent variables It means the regression coefficient for , and p-1 means the number of independent variables. q0(s0) is 1. wi(s0) means a kriging weight corresponding to each of a plurality of sample positions, and e(si) means a regression residual corresponding to each of a plurality of sample positions. n means the number of sample positions.

When the observable region and the unobservable region are classified in the region classification unit 102, the concentration measurement unit 103 calculates the fine dust concentration corresponding to the observable region, and the concentration estimation unit 104 corresponds to the unobservable region. The fine dust concentration is calculated. Through this, the fine dust concentration for the entire area of the satellite image received from the satellite server 300 can be calculated.

The concentration distribution map generation unit 105 may generate a distribution map of the concentration of fine dust for the entire region where the satellite image is received, based on the concentration of fine dust in the side region and the concentration of fine dust in the unobservable region.

FIG. 10 is a flowchart illustrating a process in which the abnormal pattern determination unit 106 according to FIG. 2 determines whether an abnormal pattern of fine dust concentration has occurred.

First, the abnormal pattern determining unit 106 determines a monitoring area based on geographic information received from the GIS server 600 (S410).

Specifically, the abnormal pattern determining unit 106 determines, as a monitoring area, an area in which an expected value of fine dust emission is equal to or greater than a preset standard among all areas of the satellite image. The abnormal pattern determination unit 106 calculates an expected value based on the satellite image and the corresponding geographic information, and the geographic information is the ratio of the area of the residential area to the area of the entire area, the ratio of the area of the industrial area to the area of the entire area, and the total area It may include the ratio of area to commercial area, length and area of road, average number of buildings, average building area, average floor area ratio, type and number of environmental polluting facilities, etc.

In addition, the abnormal pattern determination unit 106 generates a time-sequential fine dust concentration pattern based on the fine dust concentration measured for a preset period of time in the monitoring area (S420). For example, based on the concentration of fine dust for 20 years, it is possible to generate a concentration pattern of fine dust that changes over time in spring, summer, autumn, and winter.

In addition, the abnormal pattern determination unit 106 may determine that an abnormal pattern has occurred when the fine dust concentration corresponding to the monitoring area deviates from the time-sequential fine dust concentration pattern (S430). For example, when an excessively high fine dust concentration is calculated compared to the average fine dust concentration in March of the monitoring area, it may be determined that an abnormal pattern has occurred.

Through this, an area with a high possibility of fine dust generation (eg, a dense factory area) is set as a monitoring area, and abnormal patterns (eg, pollutant emissions exceeding the standard) generated in the monitoring area can be monitored. there is.

FIG. 11 is a diagram showing the hardware configuration of the device according to FIG. 1 by way of example.

Referring to FIG. 11 , the apparatus 100 includes at least one processor 110 and a memory storing instructions instructing the at least one processor 110 to perform at least one operation. (memory).

The at least one operation may include the components 101 to 106 of the above-described device 100 or other functions or operation methods.

Here, the at least one processor 110 may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor for performing methods according to embodiments of the present invention. can Each of the memory 120 and the storage device 160 may include at least one of a volatile storage medium and a non-volatile storage medium.

For example, the memory 120 may be one of a read only memory (ROM) and a random access memory (RAM), and the storage device 160 may be a flash-memory. , a hard disk drive (HDD), a solid state drive (SSD), or various memory cards (eg, a micro SD card).

In addition, the device 100 may include a transceiver 130 that performs communication through a wireless network. In addition, the device 100 may further include an input interface device 140 , an output interface device 150 , a storage device 160 , and the like. Each component included in the device 100 may be connected by a bus 170 to communicate with each other.

For example, the device 100 may include a communicable desktop computer, a laptop computer, a notebook, a smart phone, a tablet PC, and a mobile phone. , smart watch, smart glass, e-book reader, PMP (portable multimedia player), portable game device, navigation device, digital camera, DMB (digital multimedia broadcasting) player , a digital audio recorder, a digital audio player, a digital video recorder, a digital video player, a personal digital assistant (PDA), and the like.

The methods according to the present invention may be implemented in the form of program instructions that can be executed by various computer means and recorded on a computer readable medium. Computer readable media may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on a computer readable medium may be specially designed and configured for the present invention or may be known and usable to those skilled in computer software.

Examples of computer readable media may include hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions may include not only machine language codes generated by a compiler but also high-level language codes that can be executed by a computer using an interpreter and the like. The hardware device described above may be configured to operate with at least one software module to perform the operations of the present invention, and vice versa.

In addition, the above-described method or device may be implemented by combining all or some of its components or functions, or may be implemented separately.

Although the above has been described with reference to preferred embodiments of the present invention, those skilled in the art will variously modify and change the present invention within the scope not departing from the spirit and scope of the present invention described in the claims below. You will understand that it can be done.

Claims (1)

As a device for measuring the concentration of fine dust through satellite images of the ground surface,
at least one processor; and
A memory for storing instructions instructing the at least one processor to perform at least one step;
At least one step is
dividing the satellite image into a plurality of predetermined areas and classifying the areas into an observable area and an unobservable area based on a ratio of an area occupied by clouds to a total area of the area;
obtaining a concentration of fine dust in the observable region through a pre-learned concentration measurement model;
obtaining a predicted value of the fine dust concentration in the unobservable region using a pre-learned concentration estimation model and geographic information of the unobservable region, and correcting the predicted value based on the fine dust concentration in the observable region;
Calculating a residual corresponding to the predicted value through a kriging analysis of the fine dust concentration and the predicted value of the observable region, and calculating the fine dust concentration of the unobservable region based on the predicted value and the residual ; and
Generating a distribution map of the concentration of fine dust over the entire area where the satellite image is received, using the concentration of fine dust in the observable area and the concentration of fine dust in the unobservable area,
The concentration measurement model,
It is generated through machine learning on the learning data generated by labeling the learning satellite image with the fine dust concentration measurement value corresponding to the learning satellite image,
determining a plurality of feature vectors from the observable region;
calculating a concentration of fine dust in the observable region based on a plurality of weather vectors generated based on weather information of the observable region and the plurality of feature vectors;
The concentration estimation model,
It is generated through machine learning for learning data generated by labeling the learning geographic information with the fine dust concentration measurement value corresponding to the learning geographic information,
Calculating the predicted value through regression analysis on at least one independent variable determined from the geographic information,
Fine dust concentration measuring device.

Images