Subtropical morning terrain fog detection method, equipment and medium

Abstract

The invention discloses a subtropical morning terrain fog detection method, device and medium based on Himawari-8 data, wherein the method comprises: obtaining Himawari-8 data in the morning period of a subtropical study area; preprocessing the Himawari-8 data, including band clipping, useless band removal, and calculation of a surface suppression index SSI; using a subtropical morning terrain fog detection model MTFDM, removing the surface, low clouds and medium-high clouds from the preprocessed Himawari-8 data, and obtaining terrain fog in the subtropical study area; wherein the MTFDM model uses the Farneback optical flow method to perform preliminary surface removal on the time series image of the calculated surface suppression index SSI, and uses a given threshold of the brightness temperature difference time series increment to perform surface re-removal. The invention can accurately, efficiently and quickly realize subtropical morning terrain fog detection.

Description

Technical Field

[0001] The invention relates to the field of environmental monitoring and weather forecasting, and in particular to a subtropical morning terrain fog detection method, equipment and medium based on Himawari-8 data. Background Art

[0002] Subtropical terrain fog is mostly formed in the early morning in mountainous and hilly areas with high air humidity and low wind, which has a serious impact on mountainous traffic. Near-real-time terrain fog detection in the morning is extremely important to reduce the loss of life and property caused by fog. Traditional fog detection mostly relies on observations from ground meteorological stations, but ground stations in mountainous and hilly areas are sparsely distributed, making it difficult to achieve large-scale real-time fog detection. Remote sensing has the advantages of large observation range, low cost, and high temporal resolution. It provides an important means to achieve large-scale real-time dynamic fog detection and study the formation and disappearance process of terrain fog.

[0003] Traditional daytime fog detection mainly uses the differences in spectral texture characteristics between clouds, fog and the ground surface, and uses fixed thresholds or dynamic thresholds to carry out fog detection; nighttime fog uses brightness temperature difference images and combines spectral characteristics to construct an index to separate low clouds. Compared with polar-orbiting satellites, geostationary satellites have a higher observation frequency and can capture the movement characteristics of fog in near real time. The third-generation geosynchronous meteorological satellite Himawari-8 has high temporal resolution, spectral resolution and spatial resolution, and shows great potential in near-real-time fog detection. Summary of the invention

[0004] The present invention provides a subtropical morning terrain fog detection method, equipment and medium based on Himawari-8 data, which can accurately, efficiently and quickly realize subtropical morning terrain fog detection.

[0005] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0006] A subtropical morning terrain fog detection method based on Himawari-8 data, comprising:

[0007] Step 1, obtain Himawari-8 data in the morning period in the subtropical study area;

[0008] Step 2, preprocessing the acquired Himawari-8 data, including: band clipping, useless band removal, and calculation of the surface suppression index SSI of the Hiamwari-8 data;

[0009] Step 3, using the subtropical morning terrain fog detection model MTFDM, remove the surface, low clouds and medium and high clouds from the preprocessed Himawari-8 data to obtain the terrain fog in the subtropical study area;

[0010] The MTFDM model uses the Farneback optical flow method to perform preliminary surface removal on the time series images of the calculated surface suppression index SSI, and then uses a given threshold of the brightness temperature difference time series increment to perform surface removal.

[0011] Furthermore, the preprocessing of Himawari-8 data in step 2 includes:

[0012] Step B1.1, radiation calibration: calibrate the 1st to 6th bands of the acquired original Himawari-8 data to reflectivity, and calibrate the 7th to 16th bands to brightness temperature;

[0013] Step B1.2, projection conversion: convert the calibrated Himawari-8 data to the WGS-84 coordinate system, select UTM projection as the projection mode, and resample the data spatial resolution to the preset value;

[0014] Step B1.3, land mask: mask the Himawari-8 data with the help of the Chinese boundary SHP file, and assign 0 to the pixel value of the ocean part;

[0015] Step B1.4, useless band elimination and band cropping: After the 16 bands after mask processing, the useless bands are eliminated and 8 bands numbered 2, 7, 11, 12, 13, 14, 15, and 16 are retained, and the corresponding wavelengths are: 0.51μm, 3.9μm, 8.6μm, 9.6μm, 10.4μm, 11.2μm, 12.3μm, and 13.3μm.

[0016] Furthermore, the surface suppression index SSI is calculated based on the bands numbered 2 and 14 after band clipping:

[0017]

[0018] In the formula, R B2(0.51μm) and BT B14(11.2μm) They are the reflectivity of Himawari-8B2 band and the brightness temperature of B14 band respectively.

[0019] Furthermore, the Farneback optical flow method is used to perform preliminary surface removal on the time series images of the calculated surface suppression index SSI, specifically including:

[0020] The Farneback optical flow method is used to compare the grayscale value changes of pixels between each adjacent frame in the SSI time series image, and the magnitude and direction of the movement speed of each pixel from the previous frame to the current frame are obtained, that is, the optical flow vector, where the magnitude of the movement speed is the length of the optical flow vector;

[0021] The optical flow vector length of each pixel is compared with the corresponding given threshold, and the pixels smaller than the corresponding given threshold are classified as the ground surface and removed.

[0022] Furthermore, the given threshold of the brightness temperature difference time series increment is used to remove the surface, specifically:

[0023] First, the brightness temperature difference BTD is calculated based on the bands numbered 7 and 14 after band clipping:

[0024] BTD=BT B7(3.9μm) -BT B14(11.2μm)

[0025] Where, BT B7(3.9μm) and BT B14(11.2μm) They are the brightness temperature of Himawari-8B7 band and the brightness temperature of Himawari-8B14 band;

[0026] Then calculate the brightness temperature difference BTD of the two adjacent frames of Himawari-8 data and calculate the brightness temperature difference time series increment BTDTI:

[0027] BTDTI=BTD current -BTD former

[0028] Where, BTD current and BTD former are the brightness temperature difference between the current frame and the previous frame of Himawari-8 image;

[0029] Finally, the pixels whose brightness temperature difference time series increment BTDTI is lower than the preset threshold are classified as surface and removed.

[0030] Furthermore, three different low cloud identification indices LCRI1, LCRI2, and LCRI3 are calculated to identify and remove low clouds:

[0031] LCRI1=(BT B11(8.6μm) -BT B12(9.6μm) ) / (BT B11(8.6μm) +BT B12(9.6μm) )

[0032] LCRI2=(BT B15(12.3μm) -BT B16(13.3μm) ) / (BT B15(12.3μm) +BT B16(13.3μm) )

[0033] LCRI3=(BT B15(12.3μm) -BT B13(10.4μm) ) / (BT B15(12.3μm) +BT B13(10.4μm) )

[0034] Where, BTB11(8.6μm) , BT B12(9.6μm) , BT B13(10.4μm) , BT B15(12.3μm) , BT B16(13.3μm) They are the brightness temperatures of Himawari-8B11, B12, B13, B15 and B16 bands;

[0035] Pixels that satisfy one of the following conditions: LCRI1≤H, LCRI2≤M, LCRI3≤N are identified as low clouds and removed; H, M, and N are the low cloud separation thresholds of the three low cloud identification indices LCRI1, LCRI2, and LCRI3, respectively.

[0036] Furthermore, the low cloud separation thresholds H, M, and N are determined by looking up a table according to the current solar altitude angle; wherein the low cloud separation thresholds in the lookup table are determined according to statistical experience.

[0037] Furthermore, the brightness temperature of the B12 band of the adjacent frames of the Himawari-8 data is used in turn. B12(9.6μm) The ratio of Himawari-8B14 band brightness temperature BT B14(11.2μm) The fixed thresholds for removing medium and high clouds include:

[0038] Step C3.1, calculate the brightness temperature BT of the B12 band after removing the surface and low clouds in the current frame and the previous frame of Himawari-8 data B12(9.6μm) The pixels with a ratio not equal to 1 are classified as medium to high clouds and removed;

[0039] Step C3.2: for the detection results obtained in step C3.1, pixels whose brightness temperature in the B14 band is lower than the corresponding preset medium-high cloud segmentation threshold P are classified as medium-high clouds and removed.

[0040] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor implements any of the above-mentioned subtropical morning terrain fog detection methods.

[0041] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements any of the above-mentioned subtropical morning terrain fog detection methods.

[0042] Beneficial Effects

[0043] Compared with the prior art, the invention has the following beneficial effects: Aiming at the gap in the research on subtropical terrain fog detection, the invention proposes a terrain fog detection algorithm suitable for the morning in the subtropics, and solves the influence of the change of the solar zenith angle on the detection effect of terrain fog. The optical flow method and BTDTI based on the atmospheric windows of B2 (0.51μm) and B14 (11.2μm) to construct the surface suppression index SSI solve the problem of the difficulty in separating terrain fog from the surface in the morning in the previous algorithm. At the same time, the coupling of LCRI1, LCRI2, and LCRI3 low cloud removal index effectively improves the accuracy of separation of terrain fog and low clouds, and has high reliability. In terms of medium and high cloud removal, the invention proposes to use the band ratio of B12 (9.6μm) images at adjacent moments to remove medium and high clouds, and by coupling the fixed threshold of the single band of B14, the removal accuracy of medium and high clouds is improved. The verification results of ground station data show that the overall accuracy of the invention is high, and the accuracy, false alarm rate and reliability of terrain fog detection are 79.3%, 20.7% and 73.1% respectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The flowchart of the fast detection algorithm of subtropical morning terrain fog based on Himawari-8 of the present invention;

[0045] Figure 2 The topography, remote sensing images and observation site distribution map of the study area of ​​the embodiment of the present invention;

[0046] Figure 3 This is a flow chart of an embodiment of the present invention using the MTFDM model to detect morning terrain fog;

[0047] Figure 4 It is a curve diagram of the reflectivity of the observed object changing with time;

[0048] Figure 5 It is the motion characteristic map of fog and low clouds;

[0049] Figure 6 The curves of BTD and BTDTI values ​​of fog and ground surface changing with time;

[0050] Figure 7 The curves of radiation characteristic values ​​of fog, low cloud interior and low cloud edge changing with time;

[0051] Figure 8 The curves of LCRI1, LCRI2 and LCRI3 changing with time;

[0052] Fig. 9 These are the false color images and terrain fog detection results at 7:00, 7:30, 8:00 and 8:30 on January 27, 2017. DETAILED DESCRIPTION

[0053] The following is a detailed description of an embodiment of the present invention. This embodiment is based on the technical solution of the present invention, and provides a detailed implementation method and a specific operation process to further explain the technical solution of the present invention.

[0054] This embodiment provides a method for rapid detection of subtropical morning terrain fog. Figure 1 As shown, the following steps are included:

[0055] Step 1: Obtain Himawari-8 data during the morning hours over the subtropical study area.

[0056] Himawari-8 data is remote sensing image data acquired by the AHI sensor (Advanced Himawari Imager) carried by the geostationary meteorological satellite H8 (Himawari-8).

[0057] Definition of dawn and dusk: It includes both morning and dusk. The solar altitude angle in the morning gradually increases from 0, while the altitude angle in the dusk gradually decreases to 0. According to the solar altitude angle (the angle between the incident direction of sunlight and the horizon, which is complementary to the solar zenith angle, the higher the solar altitude angle, the stronger the downward radiation of the sun, and the solar altitude angle is the highest at noon), the remote sensing images containing dawn and dusk in the study area are screened. The solar zenith angle can be directly obtained in the AHI data, and the solar altitude angle at dawn and dusk is defined as [0°, 30°], that is, the solar zenith angle [60°, 90°]. Considering that subtropical terrain fog often exists in the morning, for the convenience of processing, the time series image data of [7:00, 8:30] Beijing time is obtained.

[0058] Step 2: Preprocess the Himawari-8 data and ground station data.

[0059] Step B1: The preprocessing steps of Himawari-8 data include:

[0060] Step B1.1, radiation calibration: calibrate the 1st to 6th bands of the acquired original Himawari-8 data to reflectivity, and calibrate the 7th to 16th bands to brightness temperature;

[0061] Step B1.2, projection conversion: convert the calibrated Himawari-8 data to the WGS-84 coordinate system, select UTM projection as the projection mode, and resample the data spatial resolution to 2km;

[0062] Step B1.3, land mask: mask the Himawari-8 data with the help of the Chinese boundary SHP file, and assign 0 to the pixel value of the ocean part;

[0063] Step B1.4, useless band removal and band cropping: After removing the useless bands from the 16 bands after mask processing, 8 bands numbered 2, 7, 11, 12, 13, 14, 15, and 16 are retained. The corresponding wavelengths are: 0.51μm, 3.9μm, 8.6μm, 9.6μm, 10.4μm, 11.2μm, 12.3μm, and 13.3μm, respectively. The wavelengths of the 8 band data bands are represented as R B2(0.51μm) , BT B7(3.9μm) , BT B11(8.6μm) , BT B12(9.6μm) , BT B13(10.4μm) , BT B14(11.2μm) , BT B15(12.3μm) , BT B16(13.3μm) .

[0064] Step B1.5, band calculation: perform band calculation on the bands numbered 2 and 14 after band clipping to construct the SSI index:

[0065]

[0066] In the formula, R B2(0.51μm) and BT B14(11.2μm) They are the reflectivity of Himawari-8B2 band and the brightness temperature of B14 band respectively.

[0067] Step B2: Ground observation data refers to the full-element Micaps-diamond1 mapping data recorded by ground meteorological stations, which is used to verify the morning fog detection results of this embodiment. Here, the preprocessing marks the ground stations with "visibility" less than or equal to 1000m as foggy stations, and those greater than 1000m as fog-free stations;

[0068] Step 3, using the subtropical morning terrain fog detection model MTFDM, remove the surface, remove low clouds, and remove medium and high clouds from the preprocessed Himawari-8 data to obtain terrain fog.

[0069] The subtropical region is composed of four types of objects: surface, fog, low clouds, and medium and high clouds. To carry out terrain fog detection, it is necessary to build appropriate spectral indicators and models to achieve the sequential separation and distinction of each object. The specific steps for morning fog detection are as follows:

[0070] Step C1, ground surface removal: the Farneback optical flow method and the BTDTI fixed threshold are used to remove the ground surface;

[0071] Step C1.1: Preliminary surface removal: Figure 4As shown in (a) and (d), the change trends of the surface at B2 (0.56 μm) and B14 (11.2 μm) are similar. Therefore, the surface suppression index (SSI) is constructed to reduce the changes of the surface caused by the change of the solar altitude angle. The formula for constructing SSI is as follows:

[0072]

[0073] In the formula, R B2(0.51μm) and BT B14(11.2μm) They are the Himawari-8 B2 band reflectivity and B14 band brightness temperature respectively.

[0074] At the same time, if Figure 5 As shown, the characteristic that the ground surface is stationary in the time-series image while fog and low clouds are moving, and this characteristic is more prominent in the SSI time-series image, this embodiment inputs the SSI time-series image into the Farneback optical flow method, that is, the Farneback optical flow method is used to calculate the optical flow vector field of the SSI time-series image, specifically including: the optical flow method compares the gray value changes of the pixel points between each adjacent frame in the SSI time-series image, calculates the movement speed and direction of each pixel point from the previous frame to the current frame, that is, the optical flow vector, the magnitude of the movement speed is the optical flow vector length (Vector Length), calculates the optical flow vector of each pixel point to form an optical flow vector field, and can obtain the optical flow vector field representing the movement of clouds, fog and ground surface. Since the ground surface is stationary, the optical flow vector length threshold is set to 1, and the pixels with Vector Length<1 are classified as the ground surface and removed.

[0075] Step C1.2: Surface removal: Figure 6 As shown in (a), in the brightness temperature difference (BTD) image, the BTD value of the surface changes little with the change of the solar altitude angle. When constructing the brightness temperature difference timing increment (BTDTI), the BTDTI value of the surface is small, so BTDTI=1 is selected to optimize the surface removal result of step C1.1, and the pixels with BTDTI≤1 are classified as the surface and then removed. The calculation formulas of BTD and BTDTI are as follows:

[0076] BTD=BT B7(3.9μm) -BT B14(11.2μm)

[0077] BTDTI=BTD current -BTD former

[0078] Where, BT B7(3.9μm) and BT B14(11.2μm) The brightness temperatures of Himawari-8B7 (3.9 μm) and B14 (11.2 μm) bands are respectively; BTD current and BTD former They are the brightness temperature difference between the current frame and the previous frame respectively.

[0079] Step C2, low cloud removal: Use appropriate fixed thresholds of LCRI1, LCRI2, and LCRI3 to remove low clouds;

[0080] like Figure 7 As shown in Figure 2, fog and low cloud interior and edge show different trends in different infrared bands. Based on the radiation differences between fog and low cloud in B11 and B12, B15 and B16, and B15 and B13, different low cloud identification indices (LCRI) are constructed. Figure 8 , remove low clouds. The calculation formulas of the three different low cloud identification indexes LCRI1, LCRI2, and LCRI3 are as follows:

[0081] LCRI1=(BT B11(8.6μm) -BT B12(9.6μm) ) / (BT B11(8.6μm) +BT B12(9.6μm) )

[0082] LCRI2=(BT B15(12.3μm) -BT B16(13.3μm) ) / (BT B15(12.3μm) +BT B16(13.3μm) )

[0083] LCRI3=(BT B15(12.3μm) -BT B13(10.4μm) ) / (BT B15(12.3μm) +BT B13(10.4μm) )

[0084] Where, BT B11(8.6μm) , BT B12(9.6μm) , BT B13(10.4μm) , BT B15(12.3μm) , BT B16(13.3μm) They are the brightness temperatures of Himawari-8B11, B12, B13, B15 and B16 bands respectively.

[0085] The appropriate fixed thresholds for the low cloud recognition indices LCRI1, LCRI2, and LCRI3 are set as H, M, and N respectively. As long as a pixel meets one of the following conditions: LCRI1 ≤ H, LCRI2 ≤ M, LCRI3 ≤ N, the pixel can be determined as a low cloud and removed. Since the reflectivity and radiation characteristics of fog and low clouds in the infrared band are greatly affected by the solar altitude angle, according to statistical experience, a lookup table for the segmentation thresholds of the fog detection index at different solar altitude angles (Table 1) is established. According to the index and solar altitude angle corresponding to this lookup table, the values of H, M, and N are obtained to achieve the removal of low clouds.

[0086] Table 1 Lookup table for the segmentation thresholds of the fog detection index at different solar zenith angles

[0087]

[0088] Step C3, removal of medium and high clouds: Successively use the ratio of the B12 (9.6μm) band of adjacent images and BT B14(11.2μm) Remove medium and high clouds with a fixed threshold.

[0089] Step C3.1, Medium and high clouds often exist in areas with unstable atmosphere and are more active under the influence of upper-air currents. Use the brightness temperature BT of the B12 (9.6μm) band of the current frame and the previous frame image B12(9.6μm) Ratio to remove medium and high clouds from the cloud and fog targets after step C2. Pixels with a ratio of the B12 band of the front and rear images not equal to 1 are classified as medium and high clouds and removed again;

[0090] Step C3.2, Medium and high clouds are at a higher altitude and lower temperature, with a large difference from other observed objects and are not affected by the solar zenith angle. As shown in Figure 4 (d), use the fixed threshold of the B14 (11.2μm) single band to perform secondary optimization on the detection result of terrain fog after step C3.1. Pixels with BT B14(11.2μm) < P in the terrain fog detection result are classified as medium and high clouds and removed. Among them, the value of P is also affected by the season. According to statistical experience, a lookup table for the segmentation threshold P in different months (Table 2) is established.

[0091] Table 2 Lookup table for the BT TIR11.2μm Segmentation threshold

[0092]

[0093] Next, the accuracy verification of the detection result of the MTFDM model of the present invention is carried out:

[0094] Select the ground station data at 8:00 Beijing time to carry out the detection of terrain fog in subtropical China at different time periods on January 27, 2017, and qualitatively verify the accuracy of the detection algorithm at other times.

[0095] Fig. 9 The Himawari-8 false color images and terrain fog detection results at 7:00, 7:30, 8:00 and 8:30 Beijing time on January 27, 2017 are shown. The terrain fog remote sensing detection results show that the terrain fog occurred in seven provinces including Zhejiang and Fujian. Fig. 9 It can be seen that: (1) The detection results at the four moments cover more than 90% of the fog area, with few false detections. The Farneback optical flow method shows good detection results at different moments, especially in the terrain fog with weak signals in the morning line crossing area where the traditional fog detection algorithm fails. This algorithm can also effectively separate low clouds from fog, which is suitable for terrain fog detection in the morning. (2) However, since some low clouds are relatively low in height and thin, and their properties are very similar to those of fog (green elliptical area), the low cloud recognition index cannot completely separate such low clouds from fog, and some low cloud edge pixels are also misdetected. (3) Due to the limitations of the passive remote sensing sensor used in this paper, this algorithm fails to detect fog under clouds, such as missing some fog under clouds near Chongqing (orange box area).

[0096] The ground station data is selected to quantitatively verify the algorithm detection results. The detection evaluation indicators include the accuracy rate POD, the false alarm rate FAR and the reliability factor CSI. Among them, POD represents the detection accuracy. The higher the POD value, the higher the algorithm accuracy. CSI represents the reliability. The higher the CSI value, the more reliable the algorithm. The calculation method is as follows:

[0097]

[0098]

[0099] Where N H N represents the number of stations where the ground station observation results are consistent with the Himawari-8 detection results, i.e., correct detection; F represents the number of stations where there is no fog in ground observation but there is fog in satellite detection, i.e., false detection; N M It represents the number of stations where ground observations show fog but satellite detections show no fog, i.e. missed detections.

[0100] Table 4 below shows the accuracy of terrain fog detection in five periods in January, November and December 2017. As can be seen from Table 4, the terrain fog detection method has high accuracy, and the average values ​​of POD, FAR and CSI are 72.2%, 27.8% and 67.1%, respectively, indicating that the method of the present invention has high stability and reliability for subtropical morning terrain fog detection. Further analysis shows that among the five days, the FAR was the highest and the CSI was the lowest on December 7, 2017. The main reasons are: (1) there was fog under the clouds near the Yunnan-Guizhou Plateau and the Sichuan Basin, and the passive sensor could not penetrate the clouds to obtain the information of the fog under the clouds, resulting in missed detection of the fog area; (2) the fog area in the middle and lower reaches of the Yangtze River Plain and some areas of the southeastern hills was small, and the station showed fog, but there was no fog reflection radiation feature in the image, resulting in missed detection by the algorithm. In addition, except for December 7, the POD value remained stable (average 72.3%) for the rest of the time, further confirming the reliability of the method of the present invention.

[0101] Table 4. Accuracy of terrain fog detection results at 8:00 am

[0102]

[0103] In order to test the seasonal stability of the algorithm, 10 days of data were selected from February to October for evaluation. Two foggy weather cases in June and July were used to verify the stability of the algorithm. The algorithm detection results are shown in Table 5. In general, the average values ​​of POD, FAR, and CSI for 8 days were 0.793, 0.207, and 0.731, respectively. The accuracy of the algorithm detection results was high and the false alarm rate was low, indicating that the algorithm had strong seasonal adaptability and good stability. The NH and NF of the two control experiments in June and July were both 0, indicating that the algorithm had good stability for foggy weather in foggy areas. The accuracy verification results of 10 days in different seasons showed that the algorithm can be applied to fog detection in different seasons, but the accuracy of the algorithm was affected by fog under clouds, which easily led to low CSI. Among the 10 days, May 5, 2017 had the highest FAR value and the lowest CSI value. Further analysis showed that the main reasons were: (1) some foggy areas in Fujian and Hubei provinces were blocked by medium and high clouds, resulting in missed detection of foggy areas; (2) some low clouds in Anhui and Jiangxi provinces were relatively low in height and thickness, and their reflected radiation characteristics were too similar to those of foggy areas, resulting in the failure of the low cloud identification index and misdetection of foggy areas.

[0104] Table 5 Accuracy of terrain fog detection results at 8:00 in the morning in different seasons

[0105]

[0106] The above embodiments are preferred embodiments of the present application. Ordinary technicians in this field can also make various changes or improvements on this basis. Without departing from the overall concept of the present application, these changes or improvements should fall within the scope of protection required by the present application.

Claims

1. A subtropical morning terrain fog detection method based on Himawari-8 data, characterized in that: include: Step 1, obtain Himawari-8 data in the morning period in the subtropical study area; Step 2, preprocessing the acquired Himawari-8 data, including: band clipping, useless band removal, and calculation of the surface suppression index SSI of the Hiamwari-8 data; Among them, the surface suppression index SSI is calculated based on the bands numbered 2 and 14 after band clipping: In the formula, R B2(0.51μm) and BT B14(11.2μm) They are Himawari-8 B2 band reflectance and B14 band brightness temperature; Step 3, using the subtropical morning terrain fog detection model MTFDM, remove the surface, low clouds and medium and high clouds from the preprocessed Himawari-8 data to obtain the terrain fog in the subtropical study area; Among them, the MTFDM model uses the Farneback optical flow method to perform preliminary surface removal on the time series image of the calculated surface suppression index SSI, and uses a given threshold of the brightness temperature difference time series increment to perform surface re-removal; Low clouds are identified and removed by calculating three different low cloud identification indices: LCRI1, LCRI2, and LCRI3: LCRI1=(BT B11(8.6μm) -BT B12(9.6μm) ) / (BT B11(8.6μm) +BT B12(9.6μm) ) LCRI2=(BT B15(12.3μm) -BT B16(13.3μm) ) / (BT B15(12.3μm) +BT B16(13.3μm) ) <h2 style=";text-align:left;direction:ltr">LCRI3=(BT<h2 style=";text-align:left;direction:ltr"> B15(12.3μm) <h2 style=";text-align:left;direction:ltr"> -BT<h2 style=";text-align:left;direction:ltr"> B13(10.4μm) <h2 style=";text-align:left;direction:ltr"> ) / (BT<h2 style=";text-align:left;direction:ltr"> B15(12.3μm) <h2 style=";text-align:left;direction:ltr"> +BT<h2 style=";text-align:left;direction:ltr"> B13(10.4μm) <h2 style=";text-align:left;direction:ltr"> ) Where, BT B11(8.6μm) , BT B12(9.6μm) , BT B13(10.4μm) , BT B15(12.3μm) , BT B16(13.3μm) They are the brightness temperatures of Himawari-8B11, B12, B13, B15 and B16 bands; Pixels that satisfy one of the following conditions: LCRI1≤H, LCRI2≤M, LCRI3≤N are identified as low clouds and removed; H, M, and N are the low cloud separation thresholds of the three low cloud identification indices LCRI1, LCRI2, and LCRI3, respectively; The brightness temperature of the B12 band of the adjacent frames of the Himawari-8 data is used in turn. B12(9.6μm) The ratio of Himawari-8B14 band brightness temperature BT B14(11.2μm) A fixed threshold is used to remove medium and high clouds.

2. The subtropical morning terrain fog detection method according to claim 1, characterized in that: Step 2: Preprocessing of Himawari-8 data includes: Step B1.1, radiation calibration: calibrate the 1st to 6th bands of the acquired original Himawari-8 data to reflectivity, and calibrate the 7th to 16th bands to brightness temperature; Step B1.2, projection conversion: convert the calibrated Himawari-8 data to the WGS-84 coordinate system, select UTM projection as the projection mode, and resample the data spatial resolution to the preset value; Step B1.3, land mask: mask the Himawari-8 data with the help of the Chinese boundary SHP file, and assign 0 to the pixel value of the ocean part; Step B1.4, useless band elimination and band cropping: After the 16 bands after mask processing, the useless bands are eliminated and 8 bands numbered 2, 7, 11, 12, 13, 14, 15, and 16 are retained, and the corresponding wavelengths are: 0.51μm, 3.9μm, 8.6μm, 9.6μm, 10.4μm, 11.2μm, 12.3μm, and 13.3μm.

3. The subtropical morning terrain fog detection method according to claim 1, characterized in that: The Farneback optical flow method is used to perform preliminary surface removal on the time series images of the calculated surface suppression index SSI, specifically including: The Farneback optical flow method is used to compare the grayscale value changes of pixels between each adjacent frame in the SSI time series image, and the magnitude and direction of the movement speed of each pixel from the previous frame to the current frame are obtained, that is, the optical flow vector, where the magnitude of the movement speed is the length of the optical flow vector; The optical flow vector length of each pixel is compared with the corresponding given threshold, and the pixels smaller than the corresponding given threshold are classified as the ground surface and removed.

4. The subtropical morning terrain fog detection method according to claim 1, characterized in that: The given threshold of the brightness temperature difference time series increment is used to remove the surface, specifically: First, the brightness temperature difference BTD is calculated based on the bands numbered 7 and 14 after band clipping: <h2 style=";text-align:left;direction:ltr">BTD=BT<h2 style=";text-align:left;direction:ltr"> B7(3.9μm) <h2 style=";text-align:left;direction:ltr"> -BT<h2 style=";text-align:left;direction:ltr"> B14(11.2μm) Where, BT B7(3.9μm) and BT B14(11.2μm) They are the brightness temperature of Himawari-8B7 band and the brightness temperature of Himawari-8B14 band; Then calculate the brightness temperature difference BTD of the two adjacent frames of Himawari-8 data and calculate the brightness temperature difference time series increment BTDTI: BTDTI=BTD current -BTD former Where, BTD current and BTD former are the brightness temperature difference between the current frame and the previous frame of Himawari-8 image; Finally, the pixels whose brightness temperature difference time series increment BTDTI is lower than the preset threshold are classified as surface and removed.

5. The subtropical morning terrain fog detection method according to claim 1, characterized in that: The low cloud separation thresholds H, M, and N are determined by looking up a table based on the current solar altitude angle; wherein the low cloud separation thresholds in the lookup table are determined based on statistical experience.

6. The subtropical morning terrain fog detection method according to claim 1, characterized in that: The brightness temperature of the B12 band of the adjacent frames of the Himawari-8 data is used in turn. B12(9.6μm) The ratio of Himawari-8B14 band brightness temperature BT B14(11.2μm) The fixed thresholds for removing medium and high clouds include: Step C3.1, calculate the brightness temperature BT of the B12 band after removing the surface and low clouds in the current frame and the previous frame of Himawari-8 data B12(9.6μm) The pixels with a ratio not equal to 1 are classified as medium to high clouds and removed; Step C3.2: for the detection results obtained in step C3.1, pixels whose brightness temperature in the B14 band is lower than the corresponding preset medium-high cloud segmentation threshold P are classified as medium-high clouds and removed.

7. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the computer program is executed by the processor, the processor is caused to implement the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images