Multi-source polar orbit satellite joint polar day sea fog / low cloud detection method, device and medium

Abstract

This invention discloses a method, device, and medium for detecting polar daytime sea fog / low clouds using multi-source polar-orbiting satellites. The method includes: acquiring and preprocessing remote sensing image data collected by multi-source polar-orbiting satellites; calculating a sea surface identification index using two preset band data from the preprocessed remote sensing image data; then, based on the sea surface identification index and near-infrared band reflectance data from the remote sensing image data, jointly removing areas related to seawater and sea ice / snow from the preprocessed remote sensing image data; calculating the gray-level co-occurrence matrix for each pixel in a third preset band of the remote sensing image data after removing seawater, sea ice, and snow, and statistically analyzing several texture features based on the gray-level co-occurrence matrix; inputting the statistically analyzed texture feature values ​​corresponding to all remote sensing images from polar-orbiting satellites into a trained random forest model to identify and remove mid-to-high clouds, leaving the remaining clouds as sea fog / low clouds. This invention is less affected by satellite differences and variations in solar altitude angle.

Description

Technical Field

[0001] This invention belongs to the field of image processing and meteorological services, specifically relating to a method, equipment, and medium for detecting polar daytime sea fog / low clouds using multi-source polar-orbiting satellites. Background Technology

[0002] Crossing the Arctic is much faster and less strenuous than previous routes. However, on the other hand, the melting of Arctic sea ice greatly promotes the formation of sea fog / low clouds. The positive relationship between sea fog / low cloud formation and sea ice melting has been widely proven. For example, steam fog forms when warm seawater comes into direct contact with cold air after sea ice melts, posing a serious potential threat to navigation safety. Therefore, conducting Arctic fog detection is crucial for navigation safety.

[0003] Polar-orbiting meteorological satellites possess high temporal and spatial resolution. A single satellite has a reentry period of approximately 100 minutes and a spatial resolution of ~1 km. When multiple medium-resolution multispectral polar-orbiting satellites are combined, the temporal resolution can reach up to 5 minutes, providing a good platform and data foundation for large-scale and accurate monitoring of Arctic sea fog / low clouds. However, there are certain differences in spectral bandwidth, Earth response sensitivity, radiometric calibration, and solar zenith angle among the multi-source sensors, resulting in significant errors in multi-source satellite sea fog / low cloud monitoring.

[0004] Traditional methods for detecting Arctic sea fog primarily rely on ground-based observation stations, ships, and buoys. However, the harsh geographical environment of the Arctic region, the sparse distribution of ground-based observation stations, and the limited availability of ships and buoys result in a severe shortage of monitoring data, making it difficult to effectively monitor large-scale sea fog along Arctic shipping routes.

[0005] Remote sensing algorithms based on visible-near-infrared reflectance characteristics and infrared radiation properties, widely used in mid- and low-latitude regions, face significant challenges in the Arctic. The Arctic surface is complex, with sea ice and seawater intermingled. Different land cover types exhibit relatively small differences in reflectance and are easily affected by the solar zenith angle, making the reflectance of sea ice / snow very similar to that of sea fog, thus increasing the difficulty of sea fog detection. Furthermore, the high latitude and weak solar radiation in the Arctic region make it difficult for traditional remote sensing algorithms to accurately distinguish sea fog from other land features. Summary of the Invention

[0006] This invention provides a method, equipment, and medium for detecting polar daytime sea fog / low clouds using a combination of multi-source polar-orbiting satellites, which is less affected by satellite differences and changes in solar altitude angle.

[0007] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0008] A method for detecting polar daytime sea fog / low clouds using multi-source polar-orbiting satellites, comprising:

[0009] Acquire remote sensing image data collected by multi-source polar-orbiting satellites and perform preprocessing;

[0010] The sea surface identification index is calculated using the first and second preset band data in the preprocessed remote sensing image data. Then, based on the sea surface identification index and the near-infrared band reflectance data in the remote sensing image data, the areas related to seawater and sea ice / snow in the preprocessed remote sensing image data are jointly removed.

[0011] For each pixel in the third preset band of remote sensing image data after removing seawater and sea ice / snow, calculate its gray-level co-occurrence matrix, and statistically analyze several texture features based on the gray-level co-occurrence matrix.

[0012] The texture feature values ​​corresponding to the remote sensing images of all polar-orbiting satellites are input into the trained random forest model to identify mid-to-high clouds. After removing mid-to-high clouds, the remaining clouds are sea fog / low clouds.

[0013] Furthermore, the remote sensing image data acquired by the multi-source polar-orbiting satellites includes: MODIS data, AVHRR data, VIRR data, and MERSI-II data.

[0014] Furthermore, the preprocessing includes: first, calibrating each band to reflectivity or brightness temperature; then, converting the calibrated remote sensing image data from different polar-orbiting satellites to the same coordinate system; and then performing masking processing on the Arctic land area in each remote sensing image data under the same coordinate system.

[0015] Furthermore, the sea surface identification index was calculated using reflectance data from the 0.8μm and 0.6μm bands, and is expressed as:

[0016]

[0017] In the formula, R represents the sea surface identification index. ~0.8μm and R ~0.6μm These are the reflectivities of each polar-orbiting satellite in the ~0.6μm and ~0.8μm bands, respectively. The ~0.6μm and ~0.8μm bands refer to the spectral bands covered by channels with center wavelengths of 0.6μm and 0.8μm, respectively.

[0018] Furthermore, the step of jointly removing areas related to seawater, sea ice, and snow from the preprocessed remote sensing image data based on the sea surface identification index and near-infrared reflectance data in the remote sensing image data specifically involves:

[0019] If the remote sensing image data is MODIS data, VIRR data, or MERSI-II data, the sea surface identification index is used to distinguish whether it belongs to seawater or sea ice / snow at dawn and dusk, and the near-infrared reflectance is used to distinguish whether it belongs to seawater or sea ice / snow at daytime.

[0020] If the remote sensing image data is AVHRR data, the sea surface identification index is used at any time during the day to distinguish whether it belongs to seawater, sea ice / snow.

[0021] Furthermore, for MODIS data at dawn and dusk, if the sea surface identification index of a pixel is less than 0.9, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For MODIS data during the day, if the near-infrared reflectance of a pixel is less than 0.02, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow.

[0022] For VIRR data at dawn and dusk, if the sea surface identification index of a pixel is less than 0.8, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For VIRR data during the day, if the near-infrared reflectance of a pixel is less than 0.04, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow.

[0023] For MERSI-II data at dawn and dusk, if the sea surface identification index of a pixel is less than 0.9, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For MERSI-II data during the day, if the near-infrared reflectance of a pixel is less than 0.04, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow.

[0024] For AVHRR data at any time during the day, if the sea surface identification index of a pixel is less than 0.75, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow.

[0025] Furthermore, the third preset band uses ~12μm band data.

[0026] Furthermore, when calculating the gray-level co-occurrence matrix, a gray-level co-occurrence matrix is ​​calculated for each of the four directions: 0°, 45°, 90°, and 135°. The average of the four gray-level co-occurrence matrices is then taken as the final gray-level co-occurrence matrix of the pixel. Several texture features are statistically derived from the gray-level co-occurrence matrix, including: mean, variance, and contrast.

[0027] An electronic device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to implement the polar daytime sea fog / low cloud detection method described above.

[0028] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described polar daytime sea fog / low cloud detection method.

[0029] Compared with existing technologies, the beneficial effects of this invention are as follows: Based on the analysis of visible light spectral characteristics, this invention eliminates the influence of spectral bandwidth, Earth response sensitivity, radiometric calibration differences, and solar altitude variations among multi-source polar satellite sensors by combining the solar zenith angle with a specific threshold, thus achieving the removal of elements such as seawater and sea ice / snow. Furthermore, based on the analysis of the texture differences between sea fog / low clouds and mid-to-high clouds in the thermal infrared water vapor absorption band, a method fusing gray-level co-occurrence matrix and random forest is used to extract sea fog / low clouds. Experimental results show that the method of this invention has high consistency and stability on different satellite platforms, and the detection effect is less affected by satellite differences and changes in solar altitude angle. The average detection accuracy (POD), false alarm rate (FAR), and stability (CSI) of the algorithm are 75.39%, 12.35%, and 68.17%, respectively, and the differences in accuracy, false alarm rate, and stability among different sensors are all less than 6%. Attached Figure Description

[0030] Figure 1 This is a flowchart of the polar daytime sea fog / low cloud detection method described in this invention.

[0031] Figure 2 This is a flowchart of an embodiment of the present invention.

[0032] Figure 3 The images show the land-sea distribution, remote sensing images, and CALIOP transit map of the study area in this embodiment of the invention.

[0033] Figure 4 This is a flowchart illustrating the fusion process of solar zenith angle data and multi-source polar-orbiting satellite imagery in an embodiment of the present invention.

[0034] Figure 5 This is a graph showing the time-series statistical results of AVHRR reflectance, brightness temperature, and SSRI of the observed object in an embodiment of the present invention.

[0035] Figure 6 This is a comparison image of the ~11μm and ~12μm channels of the observed object in an embodiment of the present invention, along with their textures.

[0036] Figure 7 This is a schematic diagram of the four directional angles of the gray-level co-occurrence matrix.

[0037] Figure 8 The results of detection and CALIOP verification on August 5, 2021, using the method described in this embodiment of the invention are shown. (a) and (b) are the false-color image and detection results at 12:10 on August 5, 2021, respectively. The blue line in (a) is the CALIOP motion trajectory, the red line in (b) represents the sea fog in the CALIOP VFM results, and (c) represents the observation results of CALIOP VFM.

[0038] Figure 9 The image shows the detection results of the method described in this embodiment of the invention for the time sequence from morning to noon on July 15, 2016; where (a) and (b) are the false color image and detection results at 03:30, respectively; (c) and (d) are the false color image and detection results at 05:35, respectively; (e) and (f) are the false color image and detection results at 07:31, respectively; (g) and (h) are the false color image and detection results at 10:05, respectively; and (i) and (j) are the false color image and detection results at 11:26, respectively. The blue line in (i) is the CALIOP motion trajectory, and the red line in (j) represents the sea fog in the CALIOP VFM results.

[0039] Figure 10 The image shows the time-series detection results of the method described in this embodiment of the invention from noon to dusk on July 15, 2016. Among them, (a) and (b) are the false color image and detection result at 13:07, respectively; (c) and (d) are the false color image and detection result at 15:05, respectively; (e) and (f) are the false color image and detection result at 16:40, respectively; (g) and (h) are the false color image and detection result at 17:42, respectively; and (i) and (j) are the false color image and detection result at 21:30, respectively.

[0040] Figures 11 to 15 The accuracy (UTC-11) of sea fog detection results for three different dates in 2016, 2018, 2019, 2020, and 2021 is shown in this embodiment of the invention. Figure 16 This represents the overall average test results for all satellites over the past five years. Detailed Implementation

[0041] The embodiments of the present invention will be described in detail below. These embodiments are based on the technical solutions of the present invention and provide detailed implementation methods and specific operation processes to further explain the technical solutions of the present invention.

[0042] This embodiment provides a method for detecting polar daytime sea fog / low clouds using a multi-source polar-orbiting satellite fusion approach, referencing... Figure 1 , Figure 2 As shown, it includes the following steps:

[0043] Step 1: Acquire remote sensing image data from multi-source polar-orbiting satellites and perform preprocessing.

[0044] The remote sensing image data acquired by the multi-source polar-orbiting satellites described in this embodiment includes: MODIS data, AVHRR data, VIRR data, and MERSI-II data. MODIS data is remote sensing image data acquired by the Moderate-resolution Imaging Spectraradiometer aboard the polar-orbiting meteorological satellites Terra and Aqua, abbreviated as MODIS data; AVHRR data (Advanced Very High Resolution Radiometer) is remote sensing image data acquired by the Very High Resolution Scanning Radiometer aboard the polar-orbiting meteorological satellites NOAA-18, NOAA-19, METOP-A, and METOP-B, abbreviated as AVHRR data; VIRR is remote sensing image data acquired by the Visible and Infra-Red Radiometer aboard the polar-orbiting meteorological satellites FY-3C and FY-3B, abbreviated as VIRR data; MERSI-II is remote sensing image data acquired by the Medium Resolution Spectral Imager-II aboard the polar-orbiting meteorological satellite FY-3D, abbreviated as MERSI-II data.

[0045] Step A1: Band calibration.

[0046] Step A1.1: The original MODIS data contains 36 bands, of which bands 1-19 and 26 are calibrated as reflectivity, and bands 20-25 and 27-36 are calibrated as brightness temperature.

[0047] Step A1.2: The original AVHRRR data has 5 bands, of which bands 1-2 are calibrated as reflectivity and bands 3-5 are calibrated as brightness temperature.

[0048] Step A1.3: The original VIRR data contains 10 bands, of which bands 1-8 are calibrated as reflectivity and bands 9-10 are calibrated as brightness temperature.

[0049] Step A1.4: The original MERSI-II data contains 25 bands, of which bands 1-19 are calibrated as reflectivity and bands 20-25 are calibrated as brightness temperature.

[0050] Step A2: Geometric positioning. Convert the calibrated MODIS, AVHRR, VIRR, and MERSI-II data to the WGS-84 coordinate system, selecting UTM projection as the projection method, and resample the data spatial resolution to 1km.

[0051] Step A3: Land Masking. Using a standard Arctic land boundary SHP file, mask the MODIS, AVHRR, VIRR, and MERSI-II data, setting the pixel values ​​of the land portion to 0.

[0052] Step A4: CALIOP data is laser sounding radar data jointly developed by CNES and NASA. The Vertical Feature Layer Distribution (VFM) data from the CALIOP Level 2 product is selected to verify the sea fog / low cloud detection results in this embodiment. The CALIOP VFM product has eight feature layer classification labels; sea fog, low clouds, and mid-to-high clouds are uniformly labeled as "Cloud." Clouds with an altitude below 400m are defined as sea fog. Furthermore, if the cloud's base height is 0m, meaning it is close to the ocean surface, even if its top height is greater than 400m, it is still defined as sea fog.

[0053] Figure 3 This embodiment demonstrates the land-sea distribution, remote sensing imagery, and CALIOP radar transit map of the study area of ​​interest.

[0054] Since this embodiment subsequently uses solar zenith angle data, the solar zenith angle data is further fused with the preprocessed remote sensing image data in the data preprocessing operation, referring to... Figure 4 As shown, the specific process is as follows:

[0055] Step B1: Resampling of solar zenith angle data. Solar zenith angle data is provided in the MODIS dataset, but the size of this dataset (271×406) differs from the size of the preprocessed MODIS data (data size 1354×2030). Therefore, it is necessary to resample the solar zenith angle data. The sampling method selected is triple convolution resampling, which uniformly samples the solar zenith angle data to 1354×2030.

[0056] Step B2: The solar zenith angle data obtained from Step B1 is fused with the MODIS, AVHRR, VIRR, and MERSI-II data obtained from Step A1, respectively. The specific steps are as follows:

[0057] Step B2.1: Georegistration. The GCP ground point control method is selected for registration, and the pixels in the solar zenith angle data are matched and corresponded pixel by pixel with the pixels in the MODIS, AVHRR, VIRR, and MERSI-II images.

[0058] Step B2.2: Land Masking of Solar Zenith Data. The solar zenith data is masked using a standard Arctic land boundary SHP file, setting the pixel values ​​of the land portion to 0.

[0059] Step B2.3: Data Fusion. The solar zenith angle data obtained in Step B2.2 is fused with the MODIS, AVHRR, VIRR, and MERSI-II images obtained in Step A3 to obtain fused solar zenith angle MODIS, AVHRR, VIRR, and MERSI-II images.

[0060] Step 2: Removal of seawater and sea ice / snow.

[0061] The reflectivity of seawater and sea ice / snow decreases with increasing wavelength. The multi-source polar satellite sensors described in this embodiment are all equipped with ~0.6μm and ~0.8μm channels, for example... Figure 5 (a) and Figure 5 (b) shows the reflectance of the observed objects on MODIS sensors B1 (0.630 μm) and B2 (0.850 μm) over time. The reflectance of seawater and sea ice / snow is lower than that of ~0.6 μm in the ~0.8 μm channel, while sea fog / low clouds do not have this feature. Figure 5 (c) The time-series statistical results of the Sea Surface Recognition Index (SSRI) for the observed AVHRR data show that the SSRI values ​​for seawater, sea ice / snow, sea fog / low clouds, and mid-to-high clouds remain stable and have low correlation with the solar zenith angle. Specifically, the SSRI values ​​for sea fog / low clouds and mid-to-high clouds are all higher than or close to 0.8, while the SSRI values ​​for seawater and sea ice / snow are all lower than 0.8. Therefore, the existing technology based on near-infrared reflectance R... NIR To address the issue that the removal of seawater and sea ice / snow based on thresholds is significantly affected by the solar altitude angle, this invention constructs a seawater and sea ice / snow removal method based on SSRI thresholds: the sea surface identification index is calculated using preset band data in the preprocessed remote sensing image data, and then regions containing seawater, sea ice, and snow in the preprocessed remote sensing image data are jointly removed based on the sea surface identification index and the near-infrared band reflectance data in the remote sensing image data.

[0062] Step C1: Calculate the sea surface identification index using reflectance data from the 0.8μm and 0.6μm wavelength bands. The specific calculation formula is as follows:

[0063] (1)

[0064] In the formula, R ~0.8μm and R ~0.6μm These represent the reflectivity of each polar-orbiting satellite in the 0.6 μm and 0.8 μm bands, specifically B1 (0.645 μm) and B2 (0.859 μm) in MODIS, B1 (0.650 μm) and B2 (0.865 μm) in MERSI-II, B1 (0.630 μm) and B2 (0.865 μm) in VIRR, and B1 (0.630 μm) and B2 (0.862 μm) in AVHRR.

[0065] Step C2: Remove seawater and sea ice / snow.

[0066] During periods of high solar altitude (0º≤SZA<75º), seawater and sea ice / snow exhibit low reflectivity in the near-infrared band (almost zero), while sea fog / low clouds show no such characteristic (RNIR between 7% and 15%). However, during twilight hours when solar altitude is low (75º≤SZA≤90º), the reflectivity of the three is relatively similar, leading to significant errors in sea fog detection algorithms based on near-infrared reflectivity RNIR thresholds. Therefore, an RNIR threshold is constructed for different sensors at different solar zenith angles. NIR The segmentation threshold lookup table for SSRI is shown in Table 1:

[0067]

[0068] Based on the segmentation threshold lookup table constructed above, seawater and sea ice / snow are identified and removed:

[0069] For MODIS data at dawn / dusk (75º≤solar altitude angle SZA≤90º), if the sea surface identification index of a pixel is less than 0.9, then the pixel belongs to seawater, sea ice, or snow; otherwise, it does not belong to seawater, sea ice, or snow. For MODIS data at daytime (0º≤solar altitude angle<75º), if the near-infrared reflectance of a pixel is less than 0.02, then the pixel belongs to seawater, sea ice, or snow; otherwise, it does not belong to seawater, sea ice, or snow.

[0070] For VIRR data at dawn and dusk, if the sea surface identification index of a pixel is less than 0.8, then the pixel belongs to seawater, sea ice, or snow; otherwise, it does not belong to seawater, sea ice, or snow. For VIRR data during the day, if the near-infrared reflectance of a pixel is less than 0.04, then the pixel belongs to seawater, sea ice, or snow; otherwise, it does not belong to seawater, sea ice, or snow.

[0071] For MERSI-II data at dawn and dusk, if the sea surface identification index of a pixel is less than 0.9, then the pixel belongs to seawater, sea ice, or snow; otherwise, it does not belong to seawater, sea ice, or snow. For MERSI-II data during the day, if the near-infrared reflectance of a pixel is less than 0.04, then the pixel belongs to seawater, sea ice, or snow; otherwise, it does not belong to seawater, sea ice, or snow.

[0072] For AVHRR data at any time during the day, if the sea surface identification index of a pixel is less than 0.75, then the pixel belongs to seawater, sea ice, or snow; otherwise, it does not belong to seawater, sea ice, or snow.

[0073] Step 3: For each pixel in the third preset band of the remote sensing image data after removing seawater, sea ice and snow, calculate its gray-level co-occurrence matrix, and statistically analyze several texture features based on the gray-level co-occurrence matrix.

[0074] Building upon the removal of surface elements such as seawater and sea ice / snow, further work is needed to differentiate and extract sea fog / low clouds from mid- to high-level clouds. Sea fog / low clouds have smooth tops, while mid- to high-level clouds have undulating tops, with the former being lower than the latter. These characteristics give sea fog / low clouds a uniform texture and consistent brightness in the thermal infrared channel, while mid- to high-level clouds exhibit fragmented texture, strong brightness variations, and are accompanied by shadows. Figure 6 As shown, this forms the physical basis for remote sensing extraction of sea fog / low clouds and mid-to-high clouds.

[0075] Due to its sensitivity to water vapor, the ~12μm channel exhibits greater textural differences compared to the ~11μm channel, particularly in mid-to-high clouds, where texture undulations are more pronounced. Figure 6 (Blue circle box) Further statistical analysis was conducted on the temporal characteristics of brightness and temperature of objects observed by the AVHRR sensor Band5 (12.00μm) during the day. Figure 5 (d) Except for seawater, the brightness temperature of the observed object is basically unaffected by changes in solar altitude. Therefore, this embodiment of the invention selects the brightness temperature of the 12μm band and uses the gray-level co-occurrence matrix to statistically analyze texture features.

[0076] Given the significant textural differences between sea fog / low clouds and mid-to-high clouds, this embodiment reflects the frequency and variation characteristics of image grayscale combinations by calculating the joint distribution of pixel pairs at different directions and distances. That is, from grayscale (coordinates)... Starting from a pixel, when the interval distance is d and the direction is θ, the pixel value of that pixel is... normalized probability The formula is defined as follows:

[0077] (2)

[0078] In the formula For two pixels in Distance difference in direction, This represents the grayscale level. To improve computational efficiency, this embodiment compresses the image grayscale to 3 bits and uses a 5×5 window with a distance of 1 to traverse the image; the direction θ is as follows: Figure 7 As shown, four directions—0°, 45°, 90°, and 135°—are selected to fully reflect the joint distribution characteristics of gray levels. The mean of the gray-level co-occurrence matrix in these four directions is chosen as the feature value of the central pixel.

[0079] Eight texture statistics are defined on the gray-level co-occurrence matrix to describe the gray-level correlation and spatial distribution of the image. Considering the information redundancy caused by the commonality between statistics, this embodiment selects three statistics—mean, variance, and contrast—to distinguish between sea fog / low clouds and mid-to-high clouds. Their calculation formulas are as follows:

[0080] (3)

[0081] The mean describes the regularity of an image's texture; a higher mean value indicates greater uniformity in the image. Indicates the number of gray levels (0~7). It is the probability value in the gray-level co-occurrence matrix corresponding to the sum of pixel pairs being i.

[0082] (4)

[0083] Variance describes the periodicity of an object's texture; a higher variance value indicates that the texture features in the image have more obvious repetition and regularity. These are two index variables, used to iterate through the rows and columns of the gray-level co-occurrence matrix, corresponding to different gray levels. This represents the mean, which is the average grayscale value of the pixels within the selected area. The element value at position (i,j) in the gray-level co-occurrence matrix represents the joint probability that a pixel with gray level i and a pixel with gray level j appear simultaneously.

[0084] (5)

[0085] Contrast describes the difference in gray levels between image pixels. A higher Contrast value indicates a coarser texture in the image.

[0086] Step 4: Input the statistical texture feature values ​​of the ~12μm band corresponding to all polar-orbiting satellite remote sensing images into the trained random forest model to identify the mid-to-high clouds. After removing the mid-to-high clouds, the remaining clouds are sea fog and low clouds.

[0087] The texture feature values ​​corresponding to the ~12μm band of remote sensing images from polar-orbiting satellites are a set of texture feature statistics under multiple scales, including directionality, uniformity, and complexity. The combination of these texture feature statistics gives the image high-dimensional features. Random Forest (RF) is not easily affected by the "curse of dimensionality" and can effectively avoid the computation and storage problems caused by high-dimensional data. At the same time, it can obtain good classification results without excessive parameter tuning.

[0088] In this embodiment of the invention, the texture statistics (mean, variance, contrast) obtained by extracting the remote sensing data corresponding to multi-source polar satellites are input into a trained random forest model to identify mid-to-high clouds. After removing mid-to-high clouds, the remaining data are sea fog and low clouds.

[0089] The parameter settings for the Gray-Level Co-occurrence Matrix (GLCM) and Random Forest (RF) in this embodiment are shown in Table 2:

[0090]

[0091] The accuracy of the onion model detection results of this invention will be verified next:

[0092] The algorithm's universality was quantitatively validated by selecting 15 fog cases from different dates, including 2016 and the summer months (June-August) of 2018-2021. Figures 11 to 16 As shown in the figure. The results show that the method of the present invention has an average detection accuracy (POD) of 75.39%, an average false alarm rate (FAR) of 12.35%, and a stability (CSI) of 68.17%. The differences of the three indicators among different sensors are all less than 6%, indicating that the detection performance of the algorithm is basically unaffected by changes in the sensor.

[0093] Figure 8 The results of MODIS / Aqua detection during this period are as follows: Figure 8 (a) Figure 8 As shown in (b), a large number of sea fog detection results were missed, compared with the detection results of CALIOP. Figure 8 (c) the white dashed box) and Figure 8 (a) (white solid line circle) The main areas of missed sea fog are thin sea fog with obvious texture shadows, thin thickness, and complex underlying surface distribution. The thinner sea fog has limited shielding of the radiation energy of the underlying surface, which means that the brightness temperature value of this part of the sea fog is mixed with the energy of the underlying surface at ~12um, that is, spectral mixing occurs. Since the brightness temperature difference between sea ice / snow and seawater in the underlying surface is obvious and they are interspersed, the temperature of this part of the sea fog fluctuates greatly. At the same time, due to the movement of the sea airflow, compared with normal sea fog, the texture of this part of the thin sea fog is coarser and has a "shadow" phenomenon, that is, the fog top has high and low undulations in a short distance. Pixel mixing, underlying surface temperature fluctuations, and thin fog shadows make the texture characteristics of this part of the thin sea fog coarse and change more drastically, so it is misclassified as clouds.

[0094] like Figure 9 and Figure 10 As shown, two fog areas with different thematic morphologies appeared in the time-series detection results during the day. Figure 9A green circle marks an independent leaf-shaped fog area. The algorithm detected the changes in the shape and movement of this leaf-shaped fog area within the study area relatively completely. It was generated in the floating ice area in the northwestern Chukchi Sea outside the study area and entered the study area at 03:30 am. It then gradually moved southward under the influence of wind and showed signs of dissipation by breaking up at 17:42 pm. By 21:30 pm, the leaf-shaped fog area had disappeared from the study area.

[0095] Meanwhile, satellite time-series images show that the density and morphology of this blade-shaped fog area are correlated with the intensity of solar radiation, indicating that it is affected by diurnal variations. From 03:30 to 05:35, the fog area was denser and larger, almost completely obscuring the underlying surface radiation. As solar radiation intensified, the fog area began to shrink inward and gradually became elongated, while its density gradually decreased. By 13:07, the fog area reached its lowest density, exposing the sea surface below. From 13:07 to 15:05, as the sun's altitude decreased and radiation weakened, the fog area gradually expanded, and its density increased again. Subsequently, influenced by wind, the blade-shaped fog area began to break apart from the main vein, and its density decreased accordingly. By 15:05, the blade-shaped fog area had broken into a circle, and the underlying sea surface radiation was visible. By 21:30, the fog area had disappeared from the study area.

[0096] Beyond the foliated sea fog / low cloud area, a large-scale planar advection fog also existed within the study area. This advection fog occurred in the periglacial zone in the southern part of the study area. Figure 9 The time-series detection results shown indicate that the warm and humid air mass ( Figure 9 The fog (within the white box in the middle) entered the study area around 3:30 AM and began to form advection fog after encountering cold air. Subsequently, the warm and moist air mass gradually moved southeast under the influence of the wind, continuously transforming into sea fog / low clouds during this process. Around 3:05 PM, the warm and moist air mass completely transformed into sea fog / low clouds. Afterward, the main body of the sea fog / low cloud area remained relatively stable and gradually moved eastward under the influence of the wind. By around 9:30 PM, the entire sea fog / low cloud area had drifted to the coast of the North American continent.

[0097] The CALIOP dataset was selected to quantitatively validate the detection results of the method of this invention. The evaluation metrics included accuracy (POD), false alarm rate (FAR), and reliability factor (CSI). POD represents detection accuracy; a higher POD value indicates higher algorithm accuracy. CSI represents reliability; a higher CSI value indicates a more reliable algorithm. The calculation method is as follows:

[0098]

[0099]

[0100]

[0101] Where: N H The number of pixels representing both CALIOP and MODIS detection results is the number of pixels in sea fog / low clouds; N F N represents the number of false positives (i.e., CALIOP represents the number of non-sea fog / low cloud cells, while MODIS represents the number of sea fog / low cloud cells), and vice versa. M This represents the number of missed pixels, i.e., the number of pixels that CALIOP detects as sea fog / low clouds, while MODIS detects as not being sea fog / low clouds.

[0102] This invention utilizes nine polar-orbiting meteorological satellites to detect Arctic sea fog / low clouds. The method demonstrates high consistency and stability across different satellite platforms, with detection results minimally affected by satellite variations and changes in solar altitude angle. The fog detection results show that the method comprehensively reflects the formation and dissipation processes of sea fog, and its detection effectiveness is unaffected by changes in solar zenith angle or differences in satellite sensors, further confirming the reliability of the invention.

[0103] The above embodiments are preferred embodiments of this application. Those skilled in the art can make various changes or improvements based on them. Without departing from the overall concept of this application, these changes or improvements should fall within the scope of protection claimed in this application.

Claims

1. A multi-source track satellite joint polar diurnal sea fog / low cloud detection method, characterized in that, include: Acquire remote sensing image data collected by multi-source polar-orbiting satellites and perform preprocessing; The sea surface identification index is calculated using the first and second preset band data in the preprocessed remote sensing image data. Then, based on the sea surface identification index and the near-infrared band reflectance data in the remote sensing image data, the areas related to seawater and sea ice / snow in the preprocessed remote sensing image data are jointly removed. The sea surface identification index was calculated using reflectance data from the 0.8μm and 0.6μm bands, and is expressed as: ； In the formula, R represents the sea surface identification index. ~0.8μm and R ~0.6μm These are the reflectivities of each polar-orbiting satellite in the ~0.6μm and ~0.8μm bands, respectively. The ~0.6μm and ~0.8μm bands refer to the spectral bands covered by channels with center wavelengths of 0.6μm and 0.8μm, respectively. The step of jointly removing areas related to seawater, sea ice, and snow from the preprocessed remote sensing image data based on the sea surface identification index and near-infrared reflectance data in the remote sensing image data is as follows: If the remote sensing image data is MODIS data, VIRR data, or MERSI-II data, the sea surface identification index is used to distinguish whether it belongs to seawater or sea ice / snow at dawn and dusk, and the near-infrared reflectance is used to distinguish whether it belongs to seawater or sea ice / snow at daytime. If the remote sensing image data is AVHRR data, the sea surface identification index is used at any time during the day to distinguish whether it belongs to seawater, sea ice / snow; For each pixel in the third preset band of remote sensing image data after removing seawater and sea ice / snow, calculate its gray-level co-occurrence matrix, and statistically analyze several texture features based on the gray-level co-occurrence matrix. The third preset band uses ~12μm band data; The texture feature values ​​corresponding to the remote sensing images of all polar-orbiting satellites are input into the trained random forest model to identify mid-to-high clouds. After removing mid-to-high clouds, the remaining clouds are sea fog / low clouds.

2. The polar day sea fog / low cloud detection method according to claim 1, characterized in that, The remote sensing image data acquired by the multi-source polar-orbiting satellites includes: MODIS data, AVHRR data, VIRR data, and MERSI-II data.

3. The polar daytime sea fog / low cloud detection method according to claim 1, characterized in that, The preprocessing includes: first, calibrating each band to reflectivity or brightness temperature; then, converting the calibrated remote sensing image data from different polar-orbiting satellites to the same coordinate system; and then performing masking processing on the Arctic land area in each remote sensing image data under the same coordinate system.

4. The polar daytime sea fog / low cloud detection method according to claim 1, characterized in that, For MODIS data at dawn and dusk, if the sea surface identification index of a pixel is less than 0.9, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For MODIS data during the day, if the near-infrared reflectance of a pixel is less than 0.02, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For VIRR data at dawn and dusk, if the sea surface identification index of a pixel is less than 0.8, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For VIRR data during the day, if the near-infrared reflectance of a pixel is less than 0.04, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For MERSI-II data at dawn and dusk, if the sea surface identification index of a pixel is less than 0.9, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For MERSI-II data during the day, if the near-infrared reflectance of a pixel is less than 0.04, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow. For AVHRR data at any time during the day, if the sea surface identification index of a pixel is less than 0.75, then the pixel belongs to seawater or sea ice / snow; otherwise, it does not belong to seawater or sea ice / snow.

5. The polar day sea fog / low cloud detection method according to claim 1, characterized in that, When calculating the gray-level co-occurrence matrix, one gray-level co-occurrence matrix is ​​calculated for each of the four directions: 0°, 45°, 90° and 135°. The average of the four gray-level co-occurrence matrices is then taken as the final gray-level co-occurrence matrix of the pixel. Several texture features based on the statistics of the gray-level co-occurrence matrix include: mean, variance, and contrast.

6. An electronic device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, the processor causes the processor to implement the method as described in any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

Images