Patent Application
20210318253
This application claims priority of the Chinese Patent Application No. 201910454155.5, filed on May 29, 2019, which is incorporated herein by reference in its entirety.
The present invention relates to the field of information remote sensing technologies, and more particularly to a method for retrieving atmospheric aerosol based on statistical segmentation.
The atmospheric aerosol is a multi-phase system formed by atmosphere as well as solid and liquid particles suspending therein. It is a mixture consisting of molecular groups and liquid or solid particles suspending in the air and having certain stability, a small sedimentation rate, and a size ranging from one micrometer to dozens of micrometers. Aerosol particles mainly originate from artificial sources such as industrial activities, biological burning and the like, as well as natural sources such as dusts, offshore marine particles and the like.
Research shows that the atmospheric aerosol not only affects radiation balance of an earth surface layer system, but also causes the earth-atmosphere system to cool by adjusting a reflectivity of an earth atmosphere system through scattering solar shortwave radiation. It also causes a heating process through absorbing the solar radiation, and changes microphysical properties of clouds by affecting the microphysical process of formation of the clouds. Aerosol also significantly influences atmospheric chemical processes and biogeochemical cycle. At the same time, the aerosol particles contain some particles harmful to human body, which may cause serious problem to the health of human beings, especially in heavily populated urban areas with many industries. Therefore, retrieval of the atmospheric aerosol has a significant meaning to research of global climate change as well as monitoring and management of atmospheric pollution.
Currently, researchers home and abroad have made quite some achievements in terms of aerosol remote sensing retrieval. They also pay much attention to monitoring of the aerosol in designing novel sensors. However, most of the conventional retrieval algorithms are directed to large-size aerosol having average properties, and quite limited in terms of applications. For example, such algorithms are only adapted to dark ground surface. At present, there is no effective aerosol retrieval algorithm adapted to bright ground surface areas such as cities.
An object of the present invention is to provide a method for retrieving atmospheric aerosol based on statistical segmentation. Two parts are partitioned by way of statistical segmentation, and aerosol thickness values thereof are retrieved using different methods, thereby improving accuracy and resolution of the retrieval result of the bright ground surface area.
In order to meet the above-mentioned objective of the present invention, the method for retrieving atmospheric aerosol based on statistical segmentation in the present invention comprises steps of:
S1: acquiring a multi-band high-resolution remote sensing image of an analyzed area using satellites, and performing radiometric calibration to obtain a multi-band remote sensing image comprising an apparent reflectance, wherein the multi-band remote sensing image comprises a mid-infrared 1.6 micrometer band and a mid-infrared 2.1 micrometer band, and selecting a blue/costal band from the multi-band remote sensing image as an retrieval band;
S2: inputting the multi-band remote sensing image into an atmospheric radiative transfer model, obtaining a corresponding aerosol optical thickness look-up table according to the retrieval band, wherein the look-up table comprises parameters corresponding to each aerosol optical thickness value AOT for calculating the apparent reflectance;
S3: for all pixels in the multi-band remote sensing image, performing categorization using a statistical segmentation method, wherein the segmentation method comprises steps of:
S3.1: partitioning an apparent reflectance range of the mid-infrared 2.1 micrometer band into N segments φn, n=1, 2, . . . , N, according to a predetermined interval λ1;
S3.2: based on apparent reflectances of respective pixels in the multi-band remote sensing image in the mid-infrared 2.1 micrometer band, allocating the respective pixels into corresponding segments, to obtain pixel sets xn corresponding to respective segments g;
S3.3: for each pixel set xn obtained in step S3.2, in condition that a number of pixels in the pixel set xn|xn≥T1, T1 being a predetermined threshold value, retaining the pixel set, otherwise deleting the pixel set; marking retained pixel sets as xn′, n′=1, 2, . . . , N′, N′ being a number of retained pixel sets;
S3.4: partitioning an apparent reflectance range of the mid-infrared 1.6 micrometer band into M segments γm, m=1, 2, . . . , M according to a predetermined interval λ2;
S3.5: based on apparent reflectances of pixels in each pixel set xn′ in the mid-infrared 1.6 micrometer band, allocating respective pixels to a corresponding segment, to obtain pixel sets γn′,m corresponding to respective segments γm in the pixel sets;
S3.6: in condition that a number of pixels in the pixel set yn′,m|yn′,m|≥T2, T2 being a predetermined threshold value, retaining the pixel set yn′,m, otherwise deleting it; marking retained pixel sets as yk, k=1, 2, . . . , K, K being a number of the retained pixel sets;
S3.7: when an apparent reflectance of a pixel in the pixel set yk in the blue/coast band ρy
S4: for the target set Ys, acquiring the aerosol optical thickness value AOT of each pixel through the steps of:
S4.1: marking a p-th pixel set in the set Ys as ys,p, p=1, 2, . . . , |Ys|, |Ys| being a number of pixel sets in the set Ys, for each pixel set ys,p, acquiring a pixel ys,p,min with the minimum apparent reflectance in the blue/coast band, the pixel ys,p,min being a clean pixel;
S4.2: setting an aerosol optical thickness value AOT of the clean pixel as ε, ε being a preset minimum value, retrieving a ground surface reflectivity of the clean pixel Ys,p,min from the aerosol optical thickness look-up table generated in S2; setting ground surface reflectivities of all pixels in the pixel set ys,p as being equal to the ground surface reflectivity of the clean pixel ys,p,min;
S4.3: retrieving aerosol optical thickness values AOTs of all pixels in the target set Ys;
S5: performing retrieval on the error categorization set Yt through the steps of:
S5.1: traversing all pixels in the set Ys, to search for pixels with no values within a predetermined radius, designating an aerosol thickness value of a pixel with no value as the aerosol thickness value of the pixel, and marking a set comprising all designated pixels as Ps;
S5.2: for each pixel in the set Ps, obtaining parameters corresponding to the designated aerosol optical thickness value from the aerosol optical thickness look-up table, and then calculating and obtaining a ground surface reflectivity of the pixel according to the apparent reflectance;
S5.3: gridding the multi-band remote sensing image according to a predetermined side length;
S5.4: marking a q-th pixel set in the set Yt as yt,q=1, 2, . . . , |Yt|, |Yt| being a number of the pixel sets in the set Yt; for each pixel set yt,q, finding an intersection with Ps from each grid partitioned in step S5.3, and setting the ground surface reflectivity of all pixels of the pixel set yt,q in a current grid as being equal to an average ground surface reflectivity value of all pixels in the intersection;
S5.5: performing aerosol thickness value retrieval on each pixel assigned with the ground surface reflectivity in the step S5.4, to obtain the aerosol thickness value of the pixel.
In the method for retrieving atmospheric aerosol based on statistical segmentation in the present invention, the aerosol optical thickness look-up table of the retrieval band are obtained first, then the pixels are partitioned and screened according to the apparent reflectance segments of the mid-infrared 2.1 micrometer band. After that the retained pixel sets are further partitioned and screened according to the apparent reflectance segments of the mid-infrared 1.6 micrometer band. Finally the obtained pixel sets are partitioned into two categories according to the respective OTSU threshold for apparent reflectance in the coast band, with pixels smaller than the threshold in one category, and the remaining pixels in another category. The category with pixels smaller than the threshold is taken as the target set for retrieval. Specifically, each of the pixel sets are first searched for the clean pixel, then the ground surface reflectivity of the clean pixel is taken as the ground surface reflectivity of the whole pixel set, thereby obtaining the aerosol optical thickness value through retrieval. After that these pixels are taken as references to perform retrieval on the other category.
The present invention adopts a novel method when calculating the ground surface reflectivity. According to this method, the ground surface reflectivity is determined according to the clean pixel, instead of depending on a dark pixel. Therefore, as long as completion of the statistical segmentation is satisfied, the retrieval can be performed. Consequently, the present invention also has good retrieval effect for bright ground surface area, and has wider applicability compared with the conventional dark pixel algorithm.
In the following, specific embodiments of the present invention will be described in detail with reference to the accompanying drawings such that a person skilled in the art can better understand the invention. In the following description, it is noted that well-known functions and configurations are not described in detail to avoid obscuring the present invention.
S101: acquiring a multi-band remote sensing image.
A multi-band high-resolution remote sensing image of an analyzed area is acquired through satellites, and radiometric calibration is performed to obtain a multi-band remote sensing image comprising an apparent reflectance. The multi-band remote sensing image comprises a mid-infrared 1.6 micrometer band and a mid-infrared 2.1 micrometer band, and a blue/coast band is selected from the multi-band remote sensing image as a retrieval band.
Currently the LANDSAT satellite is used to acquire the multi-band high-resolution remote sensing image. In a LANDSAT 8 OLI image there are 11 bands, namely, Band1 costal band, Band2 blue band, Band3 green band, Band4 red band, Band5 near-infrared band, Band6 mid-infrared 1.6 micrometer band, Band1 mid-infrared 2.1 micrometer band, Band8 full-color band, Band9 cirrus band, Band10 thermal infrared band 1, and Band11 thermal infrared band 2. Band6 mid-infrared band and Band1 mid-infrared band are used in the present invention. The retrieval band selected in the present example is Band1, this is because Band1 in the LANDSAT 8 OLI image is more sensitive to the aerosol, and can therefore improve the retrieval accuracy.
An image taken on the 42nd day of the year 2015 in Beijing is used in the present example.
S102: acquiring an aerosol optical thickness (AOT) look-up table.
The multi-band remote sensing image obtained in step S101 is input into an atmospheric radiative transfer model, thereby arriving at a corresponding aerosol optical thickness look-up table according to the retrieval band. The look-up table includes parameters corresponding to each aerosol thickness value AOT for calculating the apparent reflectance. In the present example, assuming that an underlying surface is a Lambertian surface, its apparent reflectance ρTOA is calculated as:
where ρ0 is an atmospheric path radiation reflectivity, ρs a ground surface reflectivity, S is a downward hemispherical emissivity of a lower atmospheric layer, Ts is a total transmittance of incident light from top of the atmosphere to the ground surface, Tv is a total transmittance of light upwardly entering a satellite sensor field. Therefore it can be seen that in the present example the parameters which need to be included in the look-up table include the atmospheric path radiation reflectivity, the atmosphere hemispherical emissivity, and a product of the total transmittance of the incident light from the top of the atmosphere to the ground surface and the total transmittance upwardly entering a satellite sensor field.
Currently, there are many atmospheric radiative transfer models. In the embodiment, the most common 6S radiative transfer model is used. In addition to the multi-band remote sensing image, other parameters that need to be input into the 6S radiative transfer model include: date, zenith angle, and azimuth angle corresponding to the multi-band remote sensing image, an aerosol type and atmosphere mode, a step size of the aerosol optical thickness value and so on. The atmosphere mode selected in the present example is mid-latitude winter, the aerosol type is a continental type, and the selected retrieval band is Band1 costal band. Table 1 is a look-up table acquired in the present example.
S103: performing statistical segmentation on the pixels.
For all pixels in the multi-band remote sensing image, segmentation is performed using a statistical segmentation method. As the mid-infrared 1.6 micrometer band and mid-infrared 2.1 micrometer band have relatively long wavelengths, and they are not susceptible to aerosol interference. Therefore, the apparent reflectances of the mid-infrared 1.6 micrometer band and the mid-infrared 2.1 micrometer band are used to segment the pixels in the present invention, and aerosol optical thickness value retrieval is performed with the costal band. Since the LANDSAT 8 OLI image is used in the present example, the mid-infrared 1.6 micrometer band is Band6, the mid-infrared 2.1 micrometer band is Band7, and the costal band is Band1.
S301: partitioning the apparent reflectance of Band 7 into segments.
The apparent reflectance range of Band7 is partitioned into N segments φn, n=1, 2, . . . , N, according to a predetermined interval λ1. The partitioning interval λ1 can be determined as needed. Generally, the interval λ1 is within a range of 0.002≤λ1≤0.01.
S302: allocating the pixels.
Based on apparent reflectances in Band7 of respective pixels of the multi-band remote sensing image, the pixels are allocated to corresponding segments, to obtain pixel sets xn corresponding to respective segments φn. Assuming that an apparent reflectance of a pixel at Band7 is marked as p, then a serial number of its corresponding segment will be ┌ρ/λ┐, ┌ ┐ is rounding up to an integer.
S303: screening pixel sets.
N pixel sets xn obtained in step S302 are evaluated one by one. If a number of pixels in the pixel set xn|xn|≥T1, T1 being a predetermined threshold value, this pixel set is retained, otherwise it is deleted. The retained pixel sets are marked as xn′, n′=1, 2, . . . , N′, N′ is the number of the pixel sets retained after the screening. The value of the threshold T1 may also be set as needed. Apparently, the higher the resolution of the multi-band remote sensing image and the larger the interval λ1 is, and the more pixels are included in each segment. Correspondingly, the threshold value T1 may also be set to be slightly higher, otherwise the threshold value T1 may be set to be slightly lower. In the present example, T1=100000.
S304: partitioning the apparent reflectance of Band6 into segments.
The apparent reflectance range of Band6 is partitioned into M segments γm, m=1, 2, . . . , M, according to a predetermined interval λ2. Likewise, the partitioning interval λ2 can be determined as needed, and may be within the range of 0.002≤λ2≤0.01.
S305: further partitioning the pixel sets.
Each pixel set xn′ retained in step S303 is further partitioned. The further partitioning may be performed in the following way: each pixel within each pixel set xn′ is allocated into a corresponding segment γm according to an apparent reflectance of the pixel at Band6, to obtain a pixel set yn′,m corresponding to respective segment γm in this pixel set. Apparently, after further partitioning the pixel sets, N′×M pixel sets in total are obtained.
S306: re-screening the pixel sets.
The N′×M pixel sets obtained in step S305 are screened again. If a number of pixels in the pixel set yn′,m|yn′,m|≥T2, T2 being a predetermined threshold value, this pixel set is retained, otherwise it is deleted. The pixel set retained after the screening is marked as yk, k=1, 2, . . . , K, K is the number of the pixel sets retained after the screening. Like the threshold value T1, the threshold value T2 may also be set as needed. Since the pixel set yn′,m is a pixel set obtained after further partitioning, the number of pixels in the pixel set yn′,m will definitely be less, therefore T2 is definitely smaller than T1. In the present example, T2=5000.
S307: categorizing the pixel sets in Band1.
The K pixel sets yk obtained after screening in step S306 are categorized according to an OTSU threshold T and an apparent reflectance of each set yk in Band1. If a pixel in the pixel set yk has an apparent reflectance ρy
The pixels in the multi-band remote sensing image are segmented according to the apparent reflectances of Band6 and Band1 through the above steps, and get the final categorization result in Band1.
S104: performing retrieval on the sets Ys.
In the present invention, the sets Ys, taken as a reference set, are processed using a method of retrieving polluted pixels based on clean pixels.
In the retrieval of aerosol, it is very important to obtain the ground surface reflectivity. It is shown by experiments that each pixel segment in the target sets Ys has few errors, this is mainly because each pixel set in the target sets Ys is the same type of regular object, for example, vegetation and bare earth and so on. In the meantime, some errors can be further eliminated by the OTSU threshold method. Moreover, the probability that each pixel set in Ys containing clean pixels is relatively high. The p-th pixel set in Ys is denoted as ys,p, p=1, 2, . . . , |Ys|, |Ys| being a number of the pixel sets in the sets Ys. Thus, as long as the ground surface reflectivities of the clean pixels in each pixel set are obtained, the ground surface reflectivities of all pixels in the pixel set can be obtained. Therefore, it is necessary to obtain the ground surface reflectivities of the clean pixels in each pixel set first.
S701: searching for a clean pixel.
For each pixel set ys,p in the set Ys, a pixel with the minimum apparent reflectance in Band1 is searched out. The minimum value cannot be less than or equal to 0, and it is recorded as a clean pixel ys,p,min.
S702: calculating the ground surface reflectivity.
Assuming that the aerosol thickness value AOT of the clean pixel Ys,p,min is ε, where ε is a preset minimum value, various parameters corresponding to the aerosol optical thickness value ε are searched out from the aerosol optical thickness look-up table generated in S102. The ground surface reflectivity is calculated based on the apparent reflectance of the clean pixel ys,p,min. In the present embodiment, ε=0.01.
S703: performing retrieval on the set ys,p.
It can be seen from the previous analysis that the ground surface reflectivity corresponding to the clean pixie ys,p,min is actually the ground surface reflectivity of the whole pixel set ys,p. Therefore, it can be used to retrieve the aerosol optical thickness value of each pixel in the pixel set ys,p. The retrieval method may be selected as needed. A retrieval method used in the present example is as follows:
obtaining parameters corresponding to each aerosol thickness value AOT from the look-up table, and calculating an apparent reflectance ρi corresponding to each aerosol thickness value AOT, where i=1, 2, . . . , L, L is the number of the aerosol thickness value in the look-up table.
For each pixel, an apparent reflectance closest to its apparent reflectance is searched from the L apparent reflectances ρi, and an aerosol thickness value AOT corresponding thereto is just the aerosol thickness value of this pixel.
S105: performing retrieval on the set Yt.
Since retrieval is performed on the pixels in the set Ys in step S104 to obtain the aerosol optical thickness value of each pixel, retrieval for the pixels in the set Yt is performed by taking the pixels in the set Ys as references.
S801: expanding reference pixels.
Each pixel in the set Ys is traversed, to search for pixels with no value within a predetermined radius. The aerosol thickness value of this pixel is designated as aerosol thickness values of the pixels with no value. A pixel set of all filled and assigned pixels is marked as Ps. The searching radius r is generally set as needed, and the smaller the radius is, the more precise the filled aerosol thickness value will be, though the number of pixels remained with no value will be increased. Generally, the searching radius r is within the range of 1≤r≤10.
S802: calculating the ground surface reflectivities of the pixels in the set Ps.
For each pixel in the set Ps, parameters corresponding to the filled aerosol thickness value are obtained from the aerosol optical thickness look-up table, and then the ground surface reflectivity of this pixel is calculated and obtained according to the apparent reflectance.
S803: gridding the multi-band remote sensing image.
Considering that the objects have certain correlation in geographic space, the multi-band remote sensing image is partitioned into grids according to a predetermined side length. The predetermined side length can be set according to a resolution of the multi-band remote sensing image.
S804: acquiring the ground surface reflectivity.
A q-th pixel set in the set Yt is marked as yt,q, q=1, 2, . . . , |Yt|, |Yt| representing the number of the pixel sets in the set Yt. For each pixel set yt,q, an intersection with Ps is obtained for each grid partitioned in step S803. Then the ground surface reflectivity of all pixels of the pixel set yt,q in the current grid is set to be equal to an average ground surface reflectivity value of all pixels in this intersection. Thus, the ground surface reflectivities of the pixels in the set Yt are obtained.
S805: retrieving the aerosol thickness value.
Aerosol thickness value retrieval is performed for each pixel assigned with the ground surface reflectivity in step S804, to obtain the aerosol thickness value of this pixel.
Since the pixel sets are screened in step S103 in the present invention, the sets Ys and Yt do not contain all pixels of the multi-band remote sensing image, neither can retrieval be performed on all of the pixels in step S105. A number of the remaining pixels with no value can be reduced by adjusting the threshold of each segment or adjusting the grid size, so the remaining pixels with no value have little effect on the overall retrieval result.
In order to prove the accuracy of the present invention, data of AERONET (Aerosol Robotic Network) ground observation network is downloaded from the website http://aeronet.gsfc.nasa.gov/. The retrieval results of the present invention are compared with the ground observation data and the retrieval results of the dark pixel algorithm, and two selected sites are both located in the city, i.e. belonging to bright ground surface areas. Table 2 shows comparison of the retrieval results of the present invention, the dark pixel algorithm retrieval results and the ground surface observation data of the graph as shown in
As shown in Table 2, the retrieval results of the present invention are closer to the observation data of the actual sites, and their errors are smaller. It can be seen that for the bright ground surface area, the present invention has good retrieval results. Therefore compared with the dark pixel method, the present invention has a wider application scope.
The above is only the preferred embodiments of the invention and is not intended to limit the invention. For a person skilled in the art, the invention may have a variety of changes and modifications. Any change, equivalent replacement, improvement made within the spirit and principle of the present invention should be included in the protection scope of the invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 201910454155.5 | May 2019 | CN | national |
