Method and system for estimating aerosol concentration

Abstract

To provide a method and a system for estimating the aerosol concentration which does not require a large-scale device.SOLUTION: Disclosed is a method for estimating the aerosol concentration of an object region using a computer, which includes the steps of: acquiring the physical quantity from a sample image obtained by photographing the object region; acquiring an estimated numerical value for estimating the aerosol concentration using the physical quantity; and estimating the aerosol concentration using the estimated numerical value.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method and system for estimating aerosol concentration. More specifically, the present invention relates to a method and system for estimating aerosol concentration based on captured images. [Background technology]

[0002] Japanese Patent No. 5046076 (Patent Document 1) describes a remote selective image measurement method for aerosols containing specific substances. This method detects only light of a specific wavelength emitted from a specific substance by optical excitation as an image.

[0003] This remote selective imaging method for aerosols requires large-scale equipment such as laser radar for optical excitation, which limits the locations where aerosol concentrations can be observed. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 5046076 Summary of the Invention [Problem to be solved by the invention]

[0005] An invention described in this specification aims to provide a method for estimating aerosol concentration without the need for large-scale equipment, and a system for implementing the method. [Means for solving the problem]

[0006] The above invention is basically based on the finding that aerosol concentration can be effectively estimated by analyzing images taken of the target area, obtaining physical quantities from sample images of the target area, and using these physical quantities to obtain numerical values ​​for estimation.

[0007] One invention relates to a method for estimating aerosol concentration in a target area using a computer, which includes a physical quantity acquisition step (S101), a numerical value acquisition step (S102) for estimation, and an aerosol concentration estimation step (S103). The physical quantity acquisition step (S101) is a step for obtaining physical quantities from a sample image obtained by photographing a target area. The estimation value acquisition step (S102) is a step for obtaining estimation values ​​for estimating aerosol concentration using the obtained physical quantities. The aerosol concentration estimation step (S103) is a step for estimating the aerosol concentration using the estimation value.

[0008] The above example method relates to using reflectance to estimate aerosol concentration in a region of interest. This method includes a luminance acquisition step (S101'), a reflectance acquisition step (S102'), and an aerosol concentration estimation step (S103'). The brightness acquisition step (S101') is a step for obtaining brightness I, which is a physical quantity of a sample image, from a sample image obtained by photographing a target area. The reflectance acquisition step (S102') is a step for obtaining the reflectance R of the target area using the reference illuminance L0 and the luminance I of a sample image obtained by photographing the target area. The aerosol concentration estimation step (S103') is a step for estimating the aerosol concentration using the reflectance R. A preferred example of the reference illuminance L0 is the illuminance obtained using a spatial image of the target area photographed at the same photographing location as the sample image. The sample image preferably includes partial area images of one or more of the upper sky area, the lower sky area, the horizon and boundary area, and the building area. Furthermore, a preferred example of the aerosol concentration estimation step (S103') is a step of estimating the aerosol concentration using the reflectance R by using machine learning. The reflectance R preferably includes the reflectance for each of a plurality of color components. Another example of the aerosol concentration estimation step is one that includes a step of estimating the aerosol concentration using the ratio of the reflectance for each of a plurality of color components. The reflectance of the target area obtained from the sample image of the target area is thought to be correlated with the wavelength dependence of light scattering by aerosols. Therefore, it is thought that the aerosol concentration can be effectively estimated by determining the reflectance of the target area.

[0009] The above example method relates to a method for estimating the aerosol concentration in a target area using the ratio of values ​​for each color of pixels. This method includes a color value acquisition step (S101''), a color ratio acquisition step (S102''), and an aerosol concentration estimation step (S103''). The color value acquisition step (S101'') is a step for obtaining a color value for each pixel of a sample image from a sample image obtained by photographing a target area. The color ratio acquisition step (S102'') is a step for determining the ratio of the values ​​for each color acquired in the color value acquisition step (S101''). The aerosol concentration estimation step (S103'') is a step for estimating the aerosol concentration by machine learning using the ratio of the values ​​for each color.

[0010] One invention relates to a computer-based aerosol concentration estimation system 1. This system 1 has an imaging unit 3, a physical quantity acquisition unit (e.g., brightness calculation unit) 5, a numerical value acquisition unit for estimation (e.g., reflectance acquisition unit) 7, and a (first) aerosol concentration estimation unit 9. The photographing unit 3 is an element for photographing a sample image. The estimation value acquisition unit 5 is an element for obtaining a predetermined physical quantity from a sample image. For example, the brightness calculation unit calculates the brightness I of the sample image. The estimation value acquisition unit 7 is an element for obtaining estimation values ​​for estimating aerosol concentration using physical quantities. For example, the reflectance acquisition unit uses the brightness I of the sample image and the reference illuminance L0 to obtain the reflectance R of the area where the sample image was captured. The aerosol concentration estimation unit 9 is an element for estimating the aerosol concentration using the estimation numerical value. [Effects of the Invention]

[0011] According to the above invention, the aerosol concentration in a target area can be determined simply by capturing a sample image. Therefore, this specification can provide a method for estimating aerosol concentration without requiring a large-scale device, and a system for implementing the method. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a flowchart showing an example of a method for estimating aerosol concentration in a target area. [Figure 2] FIG. 2 is a block diagram showing an example of a system for estimating aerosol concentration. [Figure 3] Figure 3 is a conceptual diagram showing an example of the configuration of an aerosol concentration estimation system. [Figure 4] FIG. 4 is a flowchart showing an example of a process for determining the aerosol concentration from the reflectance. [Figure 5] FIG. 5 is a conceptual diagram showing an example of pattern matching. [Figure 6] FIG. 6 is a conceptual diagram showing the processing steps in Example 1. [Figure 7] Figure 7 is a photograph in place of a drawing showing an example of a sample image taken. [Figure 8] FIG. 8 is a graph that replaces a drawing when deriving reflectance. [Figure 9] FIG. 9 is a conceptual diagram showing the processing steps in Example 2. DETAILED DESCRIPTION OF THE INVENTION

[0013] The following describes embodiments of the present invention with reference to the drawings. The present invention is not limited to the embodiments described below, and also includes appropriate modifications of the embodiments below within the scope obvious to those skilled in the art.

[0014] 1 is a flowchart showing an example of a method for estimating aerosol concentration in a target area. S indicates a step. The invention described in this specification relates to a method for estimating aerosol concentration in a target area using a computer.

[0015] FIG. 2 is a block diagram showing an example of a system for estimating aerosol concentration. This system 1 has an imaging unit 3, a physical quantity acquisition unit 5, an estimation value acquisition unit 7, and an aerosol concentration estimation unit 9. This system is computer-based, and each process is performed by the computer. The computer has an input unit, an output unit, a control unit, a calculation unit, and a memory unit, and each element is connected by a bus or the like to enable information exchange. For example, the memory unit may store a control program or various information. When predetermined information is input from the input unit, the control unit reads out the control program stored in the memory unit. The control unit then reads out the information stored in the memory unit as appropriate and transmits it to the calculation unit. The control unit also transmits the input information as appropriate to the calculation unit. The calculation unit performs calculation processing using the various received information and stores it in the memory unit. The control unit reads out the calculation results stored in the memory unit and outputs them from the output unit. In this way, various processes and steps are performed. These various processes are performed by the various units and means.

[0016] FIG. 3 is a conceptual diagram showing an example of the configuration of an aerosol concentration estimation system. This system may include a terminal connected to a network such as the Internet or an intranet, and a server connected to the network. Of course, a single computer or mobile terminal may function as the device of the present invention, or multiple servers may exist. In the example of FIG. 3, input terminals 31, 33, 35, and 37 are connected to server 41 via network 39. Examples of input terminals are camera 31, smartphone 33, portable game console 35, and laptop computer 37. Any device with a photographing unit, such as these, can function as an input terminal for this system. The photographed sample images may be processed within these terminals, or may be sent to server 41 and processed there.

[0017] Next, we will explain an example of a method for estimating aerosol concentration in a target area using a computer. As shown in Figure 1, this method includes a physical quantity acquisition process (S101), a numerical value acquisition process for estimation (S102), and an aerosol concentration estimation process (S103). This series of algorithms for estimating aerosol concentration is called SNAP-CII.

[0018] The physical quantity acquisition process (S101) is a process for obtaining physical quantities from a sample image of a target area. The target area may be any area where aerial photography can be performed. The target area may be a certain region or an area photographed from a certain location. The physical quantities are values ​​used to obtain the estimation numerical values ​​used to estimate the aerosol concentration in the aerosol concentration estimation process. To obtain a sample image, a sample image photographed with a known photographing device may be input into a computer. The sample image may be a digital image of space obtained by photographing the target area with a photographing device. The physical quantities may be those of the entire sample image or those of each pixel. The physical quantities may also be those of a specific region of the sample image. Examples of physical quantities are the luminance I per pixel, the luminance I per color per pixel, the value per color per pixel, the luminance I of a region of the sample image, the luminance per color of a region, the value per color of a region, and brightness. The computer reads the sample image (or pixels of the sample image) from the memory unit, analyzes the read sample image, and obtains the specified physical quantities. As described below, this invention uses machine learning to estimate aerosol concentration, so any physical quantity obtained from a sample image can be used to estimate aerosol concentration. In particular, physical quantities that change depending on aerosol concentration are preferred. The calculated physical quantities are stored in a storage unit as appropriate.

[0019] The estimation value acquisition step (S102) is a step for obtaining an estimation value for estimating the aerosol concentration using the obtained physical quantity. The computer reads the physical quantity obtained in the physical quantity acquisition step from the memory unit. Then, based on instructions from the program, the control unit reads out the necessary numerical values ​​from the memory unit as appropriate, and uses these numerical values ​​together with the physical quantity to perform a calculation in the calculation unit to obtain the estimation value. This estimation value may be any value as long as it can be confirmed using a technique such as machine learning that there is a correlation between the actual aerosol concentration and the estimated value. The computer stores the obtained estimation value in the memory unit as appropriate.

[0020] The aerosol concentration estimation step (S103) is a step for estimating the aerosol concentration using the estimation value. The computer reads out the estimation value from the memory unit. The control unit of the computer estimates the aerosol concentration using the estimation value based on instructions from the program. The computer may be equipped with, for example, a machine learning engine, and may estimate the aerosol concentration using the estimation value through machine learning. The calculated aerosol concentration may be stored in the memory unit. The computer may also output the aerosol concentration. The computer may also use the estimated aerosol concentration to determine other values ​​such as visibility, view, or air pollution level. The meteorological condition determination step may use the methods described below as appropriate.

[0021] FIG. 4 is a flowchart showing an example of a process for determining the aerosol concentration from the reflectance.

[0022] Advance preparation process The advance preparation step is a step for determining a reference illuminance L0 used to estimate the aerosol concentration in the target area. The advance preparation step is not particularly limited as long as it can achieve the above. Examples of the advance preparation step include a reference image data acquisition step (S111), a reference illuminance L0 derivation step (S112), and a reference illuminance L0 storage step (S113).

[0023] Reference image data acquisition step (S111) The reference image data acquisition process is a process for acquiring reference image data in order to derive the reference illuminance L0 of the target area. An example of this image data is photographed image data. For example, the target area from which the air pollution situation is to be estimated is photographed in advance using a photographing device such as a camera (photographing unit 3) in a predetermined direction from a location where the target area is photographed. In this way, the reference image data can be obtained. The photographing device may be something like a CCD or a smartphone camera. Of course, a photographing direction control device may be used to control the photographing direction so that the target area can be photographed in a fixed direction. Another example of a photographing device is a fixed camera. The computer may appropriately store the obtained reference image data in the memory unit.

[0024] Standard illuminance L0 derivation process (S112) The reference illuminance L0 derivation step is a step for deriving the reference illuminance L0. The reference image data usually includes multiple pixels. The reference illuminance L0 is preferably the illuminance L0 for each pixel of the reference image data. The computer's control unit reads the reference image data from the storage unit and, based on program instructions, causes the calculation unit to perform a calculation to determine the luminance (I0) for each pixel. Based on program instructions, the control unit performs a calculation to determine the illuminance L0 for each pixel from the luminance (I0) for each pixel of the reference image data. An example of this calculation processing is calculation processing using a bilateral filter. Examples of calculation processing using a bilateral filter are publicly known, as described in JP 2019-180002 A and JP 2021-057866 A. A preferred example of the reference illuminance L0 is the illuminance obtained using a spatial image of the same target area as the sample image, photographed at the same photographing location as the sample image described below. The sample image preferably includes a partial area image of one or more of the upper sky area, the lower sky area, the horizon and boundary area, and the building area. However, the reference illuminance L0 may be a value estimated from the reference illuminance L0 in a nearby target area.

[0025] Standard illuminance L0 storage process (S113) The reference illuminance L0 storage step is a step for storing the reference illuminance L0 in the storage unit. The control unit of the computer stores the reference illuminance L0 calculated by the calculation process in the storage unit. In this way, by performing bilateral filter processing on the image data, the reference illuminance L0 for each pixel can be calculated and stored in the storage unit. It is not necessary to calculate the reference illuminance L0 for all pixels in the image data. The reference illuminance L0 may be calculated for pixels included in a portion of the image data that is necessary for calculating the reflectance.

[0026] Weather condition determination process The meteorological condition determination process is a process for determining whether or not the meteorological conditions are suitable for estimating the aerosol concentration (whether or not the shooting location matches the clear weather conditions). Another example of the meteorological condition determination process is a process for determining the meteorological conditions to be used in determining when the method for estimating the aerosol concentration differs depending on the meteorological conditions, or when the parameters for estimating the aerosol concentration differ depending on the meteorological conditions. The meteorological condition determination process is not particularly limited as long as it can achieve the above. This process may also be performed in the advance preparation process.

[0027] An example of the weather condition determination process includes a weather condition input process (S121) and a condition determination process (S122). The weather condition input process (S121) is a process for inputting information required for determination into a computer, and the condition determination process (S122) is a process for determining weather conditions based on the input information. A specific example of the weather condition determination process is determining weather conditions using global solar radiation (GSI) and diffuse solar radiation (DSI). International Publication WO2018 / 100957 describes a daylighting system. The weather detection unit of this daylighting system includes a direct solar radiation meter that detects direct solar radiation and a pyranometer that detects global solar radiation. This system then calculates diffuse solar radiation by subtracting direct solar radiation from global solar radiation, and calculates the ratio of diffuse solar radiation to direct solar radiation. In this way, a system for determining global solar radiation and diffuse solar radiation using a direct pyranometer and a pyranometer is known. The computer may be connected to the direct pyranometer and the pyranometer, or may be configured to input observation data from the direct pyranometer and the pyranometer.

[0028] An example of a weather condition determination process is GSI > 550 W / m 2 , DSI / GSI < 0.15 is the criterion for determining whether the weather is fine. In this example, the global solar radiation (GSI) and the diffuse solar radiation (DSI) are input to the system. The system stores the input global solar radiation (GSI) and diffuse solar radiation (DSI) in the memory unit as appropriate. Based on the program's instructions, the control unit reads the global solar radiation (GSI) and the threshold value (for example, 550 W / m 2) The control unit causes the calculation unit to perform a calculation to compare the global solar radiation (GSI) with a threshold value, and if the global solar radiation (GSI) is greater than the threshold value, proceed to the next step. On the other hand, the control unit causes the calculation unit to perform a calculation to compare the global solar radiation (GSI) with a threshold value, and if the global solar radiation (GSI) is less than the threshold value, determines that the weather conditions are not sunny. In the next step, the control unit reads the global solar radiation (GSI) and the diffuse solar radiation (DSI) from the memory unit, and causes the calculation unit to perform a calculation process to calculate DSI / GSI based on the program instructions. When this process is performed by hardware, a reverse circuit and a multiplication circuit may be used to calculate 1 / global solar radiation (GSI) from the global solar radiation (GSI), and the calculated value may be multiplied by the diffuse solar radiation (DSI) to calculate DSI / GSI. The control unit stores the calculated DSI / GSI in the memory unit. The control unit reads the DSI / GSI and a threshold value (e.g., 0.15) from the storage unit, and causes the calculation unit to perform a calculation to compare the DSI / GSI with the threshold value based on instructions from the program. If the DSI / GSI is smaller than the threshold value, it is determined that the weather conditions are sunny. Information indicating that the weather conditions are sunny may be stored in the storage unit as appropriate, or may be used for subsequent processing.

[0029] The computer's calculation unit may read sample images from the storage unit, perform pattern matching processing based on program instructions, and classify the sample images into sunny areas, cloudy areas, rain, buildings, trees, and others, and determine weather conditions based on the results of the calculation. In this case, for example, the weather conditions may be determined by reading a threshold value that serves as a judgment criterion from the storage unit and comparing it with the threshold value.

[0030] The weather conditions of the target area may be input from the input unit of the computer and stored in the memory unit. Then, the control unit of the computer may cause the calculation unit to perform appropriate calculation processing based on instructions from the program, and determine whether the input weather conditions (e.g., one or more of weather, temperature, humidity, and precipitation probability) are suitable for estimating the aerosol concentration. Furthermore, a coefficient for calculating the aerosol concentration may be used appropriately based on the input weather conditions, and may be used in the process of estimating the aerosol concentration described below.

[0031] For example, if it is determined in the meteorological condition determination step that the meteorological conditions at the time of measurement are suitable for estimating the aerosol concentration, the step of estimating the aerosol concentration described below may be carried out.

[0032] Estimation process of aerosol concentration in the target area The step of estimating the aerosol concentration in the target area is a step of estimating the aerosol concentration in the target area using a computer. This step includes a brightness acquisition step (S101'), a reflectance acquisition step (S102'), and an aerosol concentration estimation step (S103'). These steps may be followed by an aerosol concentration output step (S104').

[0033] Luminance acquisition step (S101') The brightness acquisition process (S101') is a process for determining the brightness I of a sample image of a target area. The target area refers to the spatial region of the area where it is desired to estimate the aerosol concentration. The meaning of "photographing" is as explained above. The photographing unit 3 is an element for photographing a sample image. The photographing unit 3 photographs the sample image. The sample image of the target area is input to a computer and appropriately stored in a memory unit. At this time, one or more pieces of information from the photographing location (identification information), the photographing time, and the weather conditions at the time of photographing may also be input to the computer and appropriately stored in the memory unit in association with the input sample image.

[0034] Luminance calculation process The brightness calculation unit 5 calculates the brightness I of the sample image. For example, the control unit, calculation unit, and storage unit of a computer function as the brightness calculation unit 5. For example, the control unit of the computer reads out the sample image from the storage unit based on a command from a program, and calculates the brightness (I tgt In this case, the computer calculates the brightness (I tgt ) may be calculated. At this time, the luminance may be calculated for each color such as RGB. The luminance value for each color such as RGB may be corrected by gamma correction or the like. The three axes of the color components are not limited to RGB, and may be any three axes such as CMY, CIE XYZ, CIE xyY, CIE u'v'Y, CIE LUV, CIE LAB, CIE LCh, etc. The calculated luminance (I tgt ) may be stored in a storage unit as appropriate.

[0035] The brightness calculation unit 5 may divide the sample image into multiple regions and extract regions for estimating the aerosol concentration. For example, the sample image may be divided into partial region images that are divided into one or more of the upper sky region, the lower sky region, the horizon and boundary region, and the building region. The sample image may also be analyzed to exclude regions with obstacles such as buildings and trees from the region for calculating the aerosol concentration. Furthermore, in the sky region, the amount of clouds may be analyzed and excluded from the region for calculating the aerosol concentration, or a coefficient or constant (correction value) for calculating the aerosol concentration may be calculated.

[0036] Figure 5 is a conceptual diagram showing an example of pattern matching. In the example of Figure 5, areas including parts that are estimated to be buildings or trees (areas with obstacles) based on the RGB values ​​of each pixel contained in the sample image are removed from the area for estimating aerosol concentration, and the area of ​​the sample image that exists above the area with obstacles is set as the area for estimating aerosol concentration. In the example of Figure 5, sample image 51 is classified into area 53 that is not used for estimating aerosol concentration because it contains obstacles such as buildings, area 55 above that and where there are no clouds, and area 57 further above that and where there are clouds.

[0037] Reflectance acquisition process (S102') The reflectance acquisition step (S102') is a step for obtaining the reflectance R of the target area using the reference illuminance L0 and the brightness I of the sample image obtained by photographing the target area. The reflectance acquisition unit 7 obtains the reflectance R of the area where the sample image was photographed using the brightness I of the sample image and the reference illuminance L0. For example, the control unit, calculation unit, and memory unit of the computer function as the reflectance acquisition unit 7. For example, the control unit of the computer obtains the reference illuminance L0 and the brightness (I tgt ) and calculate the reflectance (R tgt ) is calculated. An example of a calculation to calculate reflectance is R tgt =I tgt In the calculation process to obtain the reflectance, coefficients and constants according to the time and weather conditions may be used as appropriate. tgt Also, for each color such as RGB, tgt The values ​​for each color such as RGB may be corrected by gamma correction or the like. The three axes of the color components are not limited to RGB, but may be any three axes such as CMY, CIE XYZ, CIE xyY, CIE u'v'Y, CIE LUV, CIE LAB, CIE LCh, etc. The following explanation will be given using RGB as an example. A preferred example is when R is calculated for multiple pixels and for each color. tgt What is required is multiple R tgt is,R tgt [1],R tgt [2],R tgt [3],···R tgt It is written as [n]. tgt are stored in the storage unit as appropriate.

[0038] Aerosol concentration estimation process (S102) The aerosol concentration estimation step (S102) is a step for estimating the aerosol concentration using the reflectance R. The aerosol concentration estimation unit 9 estimates the aerosol concentration using the reflectance R. For example, a control unit, an arithmetic unit, and a memory unit of a computer function as the aerosol concentration estimation unit 9. A preferred example of the aerosol concentration estimation step (S102) is a step for estimating the aerosol concentration using the reflectance R by using machine learning. The reflectance R preferably includes the reflectance for each of a plurality of color components. The computer preferably has a machine learning engine. This machine learning engine includes a program and a memory unit. The machine learning engine then estimates the aerosol concentration by using R tgt The machine learning engine then stores the input reflectance R (e.g., multiple R tgt ) to estimate the aerosol concentration. The machine learning engine uses the input reflectance R (e.g., multiple R tgt ) as well as information such as time and weather conditions, the aerosol concentration may be estimated. In this way, the computer can estimate the aerosol concentration using the reflectance R. The estimated aerosol concentration is stored in the storage unit as appropriate.

[0039] Another example of the aerosol concentration estimation process includes a process of estimating the aerosol concentration using the ratio of reflectance for each of a plurality of color components. An example of the reflectance for each color component is the reflectance for each RGB value. The values ​​for each color, such as RGB, may be corrected by gamma correction or the like. The three axes of the color components are not limited to RGB, and any color component can be used. The following explanation uses RGB values ​​as an example. An example of the ratio of reflectance for each color is R B / R G , R G / R R , and R R / R B The control unit of the computer is the R R R G R B The value is used and R is used for the calculation. B / R G , R G / R R , and RR / R B Then, the control unit of the computer performs the calculation to find the R B / R G , R G / R R , and R R / R B The aerosol concentration estimation unit has, for example, a machine learning engine. B / R G , R G / R R , and R R / R B The aerosol concentration is estimated using the value of . In this way, the computer can estimate the aerosol concentration using the ratio of the reflectance for each of the multiple color components. The estimated aerosol concentration is stored in the storage unit as appropriate.

[0040] Aerosol concentration output process (S103) The estimated aerosol concentration may be output as appropriate. For example, estimated values ​​of aerosol concentration from multiple locations are output to a server. This makes it possible to collect information on aerosol concentration over a wide area. As a result, for example, if the aerosol concentration is high near a certain factory and low away from the factory, it is inferred that the factory is the source of contamination. Also, for example, if the computer is a user's mobile terminal (including smartphones, mobile phones, and wristwatches), the aerosol concentration may be displayed on the display of the mobile terminal. The aerosol concentration may also be output to a server that can exchange information with the computer. In this way, information on aerosol concentration is output from multiple users' mobile terminals and aggregated on the server.

[0041] The computer may also determine the level of air pollution based on the estimated aerosol concentration. In this case, the control unit may use the estimated aerosol concentration to read a threshold value from the memory unit, and determine the level of air pollution by comparing the estimated aerosol concentration with the threshold value. The determined level of air pollution may be stored in the memory unit or output as appropriate. The computer may also output the aerosol concentration and the level of air pollution together with weather forecast information.

[0042] This specification also provides a program for causing a computer (or processor) to function as the above-mentioned aerosol concentration estimation system, and an information recording medium that can be read by a computer and that stores such a program. Such a program causes a computer to function as an aerosol concentration estimation system having an imaging unit that captures a sample image, a brightness calculation unit that calculates the brightness I of the sample image, a reflectance acquisition unit that calculates the reflectance R of the area where the sample image was captured using the brightness I of the sample image and the reference illuminance L0, and an aerosol concentration estimation unit that estimates the aerosol concentration using the reflectance R.

[0043] Next, we will explain an invention that estimates aerosol concentration using the values ​​of each color for each pixel of a sample image. This invention can estimate aerosol concentration without using a reference illuminance L0. Therefore, it does not require control such as a fixed camera. For example, this can promote the use of applications that can be installed on mobile devices such as smartphones. This makes it possible to collect multiple pieces of information on aerosol concentration, enabling the status of environmental pollution to be grasped quickly and over a wide area.

[0044] This invention also relates to a computer-based aerosol concentration estimation system. A photographing unit for photographing a sample image; an RGB value acquisition unit that calculates a value (e.g., an RGB value) for each color of each pixel of the sample image; an aerosol concentration estimation unit that estimates the aerosol concentration using the RGB value or the ratio of the RGB values ​​for each pixel of the sample image; It has. This system takes a sample image and acquires the sample image of the photographing part. Calculate the color values ​​(e.g., RGB values) for each pixel of the sample image, The RGB values ​​or ratios of RGB values ​​for each pixel of the sample image are used to estimate the aerosol concentration.

[0045] This system may also perform a weather condition determination process. The photographing unit that photographs the sample images is the same as that described above.

[0046] The RGB value acquisition unit obtains a value for each color (e.g., RGB value) for each pixel of the sample image. The values ​​for each color, such as RGB, may be corrected by gamma correction or the like. The three axes of the color components are not limited to RGB, and any color component can be used. For simplicity, the present invention will be explained below using RGB values ​​as an example. In this case, the RGB values ​​may be obtained for all pixels, or for some pixels. As explained above, RGB is an example for each color, and values ​​after correction may be used, or other color classifications may be used. The following explanation will be given using an example that is classified into RGB. For example, the control unit, calculation unit, and memory unit of a computer function as the RGB value acquisition unit. For example, the control unit of a computer reads a sample image from the memory unit based on instructions from a program, and obtains the RGB values ​​for each pixel for a specific area of ​​the sample image. The obtained RGB values ​​(R[1], R[2], R[3], ···R[n]; G[1], G[2], G[3], ···G[n]; B[1], B[2], B[3], ···B[n]) may be stored in a memory unit as appropriate.

[0047] The aerosol concentration estimation unit may estimate the aerosol concentration using the RGB values ​​for each pixel of the sample image. For example, the control unit, calculation unit, and memory unit of a computer function as the aerosol concentration estimation unit. A preferred example of the aerosol concentration estimation step is a step of estimating the aerosol concentration using the RGB values ​​for each pixel using machine learning. The computer preferably has a machine learning engine. This machine learning engine includes a program and a memory unit. The machine learning engine stores the RGB values ​​for each pixel and a large amount of information related to the aerosol concentration. The machine learning engine estimates the aerosol concentration using the input RGB values ​​for each pixel. Note that the machine learning engine may estimate the aerosol concentration using not only the input RGB values ​​for each pixel but also information such as the time and weather conditions. In this way, the computer can estimate the aerosol concentration using the RGB values ​​for each pixel. The estimated aerosol concentration is appropriately stored in the memory unit.

[0048] The aerosol concentration estimation unit may use the ratio of each pixel color. Examples of the ratio of each color are the B / G, G / R, and R / B values. The computer's control unit uses the RGB values ​​of each pixel of the sample image and causes the calculation unit to perform calculation processing to determine B / G, G / R, and R / B. The computer's control unit then stores the determined B / G, G / R, and R / B values ​​in the storage unit as appropriate. The aerosol concentration estimation unit then estimates the aerosol concentration using the B / G, G / R, and R / B values. As above, the computer has a machine learning engine and can estimate the aerosol concentration using the B / G, G / R, and R / B values. [Example]

[0049] Figure 6 is a conceptual diagram showing the processing steps in Example 1. In order to extract the wavelength dependency of light scattering by aerosols, reflectance (R) was derived from the image data. Reflectance R is described as R = I / L, which is the ratio of the intensity (I) of the image data to the illuminance (L) from the light source. Reflectance R was derived from the target image data, and a machine learning model was created that estimates aerosol concentration from reflectance R, using this reflectance R as an input variable and the ground-based aerosol measurement values ​​as an output variable.

[0050] The algorithm flow in the example is shown in Figure 6. First, in order to accurately extract changes in images due to aerosol concentration, image data that is significantly affected by clouds is removed using solar radiation data. The solar radiation data is a combination of two variables, global solar radiation (GSI) and diffuse solar radiation (DSI), and the data is filtered out if GSI > 550 W / m 2 A DSI / GSI < 0.15 was used as the criterion for determining a clear sky. The reference illuminance L0 was derived by applying a bilateral filter to reference image data from a certain angle of view, and the reflectance R was derived by dividing the brightness I of the sample image by L0. By narrowing the image range to be analyzed to a narrow range such as the upper sky, lower sky, the boundary with the horizon, and building areas, the wavelength dependence of aerosol light scattering was extracted as reflectance R. Furthermore, by analyzing each RGB color component, we succeeded in deriving multiple reflectances R from a single image data, making it possible to build a machine learning model with little training data.

[0051] The verification was performed using two months of image data taken in Fukuoka City from January to February 2021. Figure 7 is a photograph in place of a drawing showing an example of a sample image taken. Here, two-class classification was performed using the K-nearest neighbor method on the concentration of suspended particulate matter (SPM) with a particle size of 10 μm or less in aerosols. Figure 8 is a graph in place of a drawing used when deriving reflectance. The vertical axis of the graph is reflectance and the horizontal axis is SPM particle size. The upper left graph shows the correlation between blue (B) reflectance and SPM particle size in the sky (2) area in Figure 7. The upper right graph shows the correlation between green (G) reflectance and SPM particle size in the sky (1) area in Figure 7. The lower graph shows the correlation between red (R) reflectance and SPM particle size across the entire area. The classification threshold was set at 25 μg / m, the average SPM observed concentration in January. 3 In Test 1, the January data was randomly divided into training and testing data in a 6:4 ratio, and a model constructed using only the training data was applied to the testing data, calculating the accuracy rate. In Test 2, the January data was used for training and the February data for testing. For each test, a comparison was made between two cases: one using 28 reflectance values ​​(7 image range options x 4 color component options) ("Input Variable 1") and the other using three image range reflectance values ​​with the highest correlation between SPM concentration and each RGB component ("Input Variable 2"). Accuracy rates were calculated for a total of four cases: two data division methods x two input variables. The results are shown in the table below. Test 2 and Input Variable 2, which most closely resemble actual operation, achieved an accuracy rate of 86%. Note that verification was performed using only sunny-weather data extracted from solar radiation data obtained from Himawari-8 satellite observations.

[0052] Input variable 1 Input variable 2 Test 1 100% 92% Test 2 86% 86%

[0053] As demonstrated by this example, it is possible to estimate the amount of aerosols from image data from fixed-point cameras used in weather forecasts and other applications. Cameras are easier to obtain than specialized measuring instruments used in methods that exploit the optical properties of aerosols, dramatically expanding the observation range and increasing the number of observation points. This is of great academic significance, enabling big data analysis using spatiotemporally dense observation data on aerosol concentrations, and is expected to be applied to forecasting air pollution such as aerosols. Furthermore, since this method can be applied to images taken not only with fixed-point cameras but also with smartphones, all citizens will be able to observe aerosols. It is expected to increase the general public's understanding and awareness of global atmospheric observation, not just air pollution, and has great potential for social applications in addition to its academic significance. [Example]

[0054] Example 2 Figure 9 is a conceptual diagram showing the processing steps in Example 2. In this example, the aerosol concentration can be determined simply by taking a sample image. As in Example 1, in order to extract changes in the image due to aerosol concentration with high precision, solar radiation data was used to remove image data that was significantly affected by clouds. The solar radiation data was a combination of two variables, global solar radiation (GSI) and diffuse solar radiation (DSI), and was used to remove images with GSI > 550 W / m 2A DSI / GSI < 0.15 was used as the criterion for determining a clear sky. The image range to be analyzed was narrowed to include the upper sky, lower sky, the boundary with the horizon, and building areas. Sample images were taken, and information about the time and location of the image was entered into the system. This information was stored in association with the sample image. RGB values ​​were calculated for each pixel in the area to be analyzed in the sample image. These RGB values ​​were used to calculate their ratio (pixel value ratio). The pixel value ratios were calculated as B / G, G / R, and R / B. The relationship between pixel value ratios and aerosol concentration values ​​was stored in advance in association with the time and location of the image. This information made it possible to build a machine learning model with limited training data. The pixel value ratios obtained by analyzing the target area of ​​the sample image were used to calculate the aerosol concentration in the target area using a machine learning engine. [Industrial Applicability]

[0055] This invention can be used in fields such as weather forecasting. [Explanation of symbols]

[0056] 1 System 3. Filming Department 5 Physical quantity acquisition unit (brightness calculation unit) 7. Estimation value acquisition unit (reflectance acquisition unit) 9. Aerosol concentration estimation section

Claims

1. 1. A method for estimating aerosol concentration in a target area using a computer, comprising: a physical quantity acquisition step of acquiring physical quantities from a sample image obtained by photographing the target area; an estimation value acquisition step of obtaining an estimation value for estimating an aerosol concentration using the physical quantity; an aerosol concentration estimation step of estimating an aerosol concentration using the estimation value; A method for estimating aerosol concentration, comprising: the physical quantity is the luminance of a sample image obtained by photographing the target area; A method for estimating aerosol concentration, in which the estimation value is the reflectance of the target area, calculated using a reference illuminance and the brightness of a sample image of the target area.

2. 2. The method of claim 1 , A method in which the sample image includes a partial area image that is aread into one or more of a sky upper portion, a sky lower portion, a horizon and boundary portion, and a building portion.

3. A method according to claim 1, the reference illuminance is an illuminance obtained using a spatial image of the target area photographed at the same photographing location as the sample image; The aerosol concentration estimation step is a step of estimating the aerosol concentration using the estimation value by using machine learning. method.

4. 2. The method of claim 1 , The method, wherein the reflectance includes a reflectance for each of a plurality of color components.

5. 2. The method of claim 1 , the physical quantity is a value for each color for each pixel, the estimation value is a ratio of the values ​​for each color, The aerosol concentration estimation step is a step of estimating the aerosol concentration using the estimation value by using machine learning. Using the reflectance ratio to estimate aerosol concentration.

6. A computer-based system for estimating aerosol concentrations in a target area, comprising: a photographing unit for photographing a sample image; a physical quantity acquisition unit for acquiring a predetermined physical quantity from the sample image; an estimation value acquisition unit that acquires an estimation value for estimating an aerosol concentration using the physical quantity; an aerosol concentration estimation unit that estimates an aerosol concentration using the estimation value; and the physical quantity is the luminance of a sample image obtained by photographing the target area; An aerosol concentration estimation system, wherein the estimation value is the reflectance of the target area, calculated using a reference illuminance and the brightness of a sample image taken of the target area.

Images