Adaboost-based on-ice sea fog / low cloud detection method and equipment in north polar day and medium

Abstract

The invention discloses a north polar day ice sea fog / low cloud detection method and device based on Adaboost and a medium. The method comprises the following steps: acquiring remote sensing image data acquired by an MODIS (Moderate Resolution Imaging Spectroradiometer) and preprocessing the remote sensing image data; calculating a plurality of spectral features and a plurality of texture features by using preset waveband data in the preprocessed remote sensing image data; inputting the calculated spectral features and texture features into an Adaboost-based and trained on-ice sea fog / low cloud classification model in the north polar day to obtain a sea fog / low cloud classification result in the remote sensing image data; and removing residual medium and high clouds in the sea fog / low clouds obtained by classification by using 11.030 [mu] m waveband brightness temperature data in the preprocessed remote sensing image data to obtain the finally detected sea fog / low clouds. According to the invention, the on-ice sea fog / low cloud in the north polar day can be accurately identified.

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 Arctic daytime ice fog / low clouds based on Adaboost. Background Technology

[0002] Under the trend of global warming, the Arctic region is rapidly warming, and the area and thickness of sea ice are continuously decreasing. Before the middle of this century, the ice sheets of the Northwest Passage (NWP), Central Passage (TSR), and Northeast Passage (NSR) will further transform into floating ice or flake ice areas, potentially allowing for long-term navigation during the summer. However, on the other hand, the rapid melting of Arctic sea ice leads to an increasing frequency of heat and kinetic energy exchange between the Arctic Ocean and the atmosphere, resulting in larger open water surfaces, stronger water vapor evaporation, and lower underlying surface temperatures. This makes sea fog and low clouds highly likely to form in the mid-to-high latitudes of the Arctic during the summer, affecting the safety of navigation. Therefore, conducting research on sea fog / low cloud detection on Arctic sea ice is of great significance for ensuring safe navigation in the Arctic.

[0003] Traditional sea fog / low cloud observations primarily rely on ship-based stations, offering high accuracy. However, the harsh Arctic environment and the high manpower and material costs of ship-based observations hinder large-scale real-time monitoring. Satellite remote sensing, with its unique on-orbit motion and high spatiotemporal resolution in the Arctic, offers a potentially effective and accurate technology for near-real-time detection of large-scale sea fog in the Arctic. Currently, sea fog / low cloud remote sensing data detection methods are mainly applied to mid- and low-latitude sea areas, primarily based on the radiation characteristics of different objects to extract sea fog / low clouds. However, due to weak sunlight in the Arctic, the spectral information of sea fog / low clouds, sea ice, and mid- to high-latitude clouds in the mid- to high-latitude Arctic is quite similar in the visible and near-infrared bands, making differentiation difficult. Furthermore, the complex underlying surface conditions and the influence of diurnal variations in solar altitude angle in the Arctic result in a lack of reliable and stable separation features for sea fog / low cloud detection in the mid- to high-latitude Arctic. Therefore, there is an urgent need to construct effective detection features and establish suitable detection models for sea fog / low cloud detection in the mid- to high-latitude Arctic.

[0004] In recent years, with the rapid development of artificial intelligence, combining multi-scale features for land cover classification holds promise for providing a powerful tool for accurate sea fog detection. However, AI models typically require a large amount of sample data and computational resources, and the harsh weather and significant climate variations in the Arctic make it difficult to obtain a large amount of labeled data. Summary of the Invention

[0005] This invention provides a method, device, and medium for detecting Arctic daytime ice fog / low clouds based on Adaboost, which can accurately identify Arctic daytime ice fog / low clouds.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0007] A method for detecting Arctic daytime ice fog / low clouds based on Adaboost includes:

[0008] Acquire MODIS remote sensing image data collected by polar-orbiting satellites and perform preprocessing;

[0009] Using preset band data from preprocessed remote sensing image data, multiple spectral features and multiple texture features are calculated.

[0010] The calculated spectral and texture features are input into the Arctic daytime ice fog / low cloud classification model based on Adaboost and trained to obtain the classification results of fog / low cloud in remote sensing image data.

[0011] Using the 11.030μm band brightness temperature data from the preprocessed remote sensing image data, residual mid-to-high clouds in the classified sea fog / low clouds were removed to obtain the final detected sea fog / low clouds.

[0012] Furthermore, the preprocessing includes: first, calibrating each band to reflectance or brightness temperature; then, converting the calibrated MODIS remote sensing image data to the WGS-84 coordinate system; and then performing masking processing on the Arctic land area in the MODIS remote sensing image data in the WGS-84 coordinate system.

[0013] Furthermore, the calculated spectral features include: Normalized Difference Water Index (NDSWI), Sea Ice Identification Index (SIRI), and Medium-High Cloud Identification Index (MHCRI).

[0014] Furthermore, NDSWI was calculated using reflectance data from the 0.645μm and 1.240μm bands, and is expressed as:

[0015]

[0016] In the formula, and These represent the reflectance of the 0.645μm and 1.240μm bands in MODIS remote sensing image data, respectively.

[0017] Furthermore, SIRI was calculated using reflectance data from the 1.240 μm and 0.4125 μm bands, and expressed as:

[0018]

[0019] In the formula, and These represent the reflectance of the 1.240μm and 0.4125μm bands in the MODIS remote sensing image data, respectively.

[0020] Furthermore, the MHCRI was calculated using reflectance data from the 0.905 μm and 0.936 μm bands, and is expressed as:

[0021]

[0022] In the formula, and These represent the reflectance of the 0.905μm and 0.936μm bands in the MODIS remote sensing image data, respectively.

[0023] Furthermore, using brightness temperature data in the 11.030μm band, the gray-level co-occurrence matrix of each pixel was calculated, and then texture features were calculated based on the gray-level co-occurrence matrix; among which, the calculated texture features include: homogeneity. Contrast and / or entropy .

[0024] Furthermore, homogeneity Contrast ,entropy The formula for calculation is:

[0025]

[0026]

[0027]

[0028] In the formula, This indicates that the gray-level co-occurrence matrix represents the gray-level value pairs. The value at this location represents the gray level. The pixels and gray levels are The joint probability of pixels appearing simultaneously.

[0029] 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 causes the processor to implement the Adaboost-based Arctic daytime ice fog / low cloud detection method described above.

[0030] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the Adaboost-based method for detecting Arctic daytime ice fog / low clouds as described above.

[0031] Beneficial effects

[0032] This invention analyzes the diurnal variation characteristics of the reflectance spectra of sea fog, sea ice, seawater, and mid-to-high clouds to construct three indices: Normalized Diurnal Seawater Index (NDSWI), Sea Ice Recognition Index (SIRI), and Mid-to-High Cloud Recognition Index (MHCRI). Simultaneously, it calculates the 11.030 μm band brightness temperature (BT) based on the gray-level co-occurrence matrix. 11.030μm The model extracts three texture features: homogeneity, contrast, and entropy. Then, using the constructed spectral index and texture features as samples, and a decision tree as a weak learner, iteratively trains a strong learner classification model based on the Adaboost ensemble learning algorithm. This model is used for ice fog / low cloud extraction and employs Bitwise Transform (BT) algorithm. 11.030μm Thresholding removes missed mid-to-high cloud formations, ultimately extracting sea fog / low clouds. The average detection accuracy (POD), false alarm rate (FAR), and stability (CSI) of this invention are 81.46%, 11.40%, and 73.41%, respectively. The algorithm exhibits high accuracy and effectively avoids the influence of diurnal variations in solar altitude angle, demonstrating strong stability and reliability. Attached Figure Description

[0033] Figure 1 This is a flowchart of the polar daytime ice fog / low cloud detection method described in this invention.

[0034] Figure 2 The images show the land-sea distribution and remote sensing images of the study area in this embodiment of the invention.

[0035] Figure 3 This is a time-series statistical result diagram of the reflectance of the B1, B5, B8, B17, and B18 bands and the brightness temperature of the B31 band for the observed objects in this embodiment of the invention, corresponding sequentially to... Figure 3 (a) to Figure 3 (f).

[0036] Figure 4 This image shows the brightness temperature image of the observed object in the 11.030μm (B31) channel and the calculated entropy of the texture index on July 4, 2019, in an embodiment of the present invention. Figure 4 (a) and Figure 4 (b) Calculation results corresponding to brightness temperature image and texture index "entropy", respectively. Figure 4 (c) and Figure 4 (d) are respectively Figure 4 (a) Enlarged view within the blue and red boxes Figure 4 (e) and Figure 4 (f) are respectively Figure 4 (b) Enlarged view within the blue and red boxes.

[0037] Figure 5The above refers to the time-series statistical results of the spectral index of the observed object in this embodiment of the invention; wherein... Figure 5 (a) to Figure 5 (c) Corresponding to the spectral indices NDSWI, SIRI, and MHCRI in sequence.

[0038] Figure 6 This is a schematic diagram illustrating the Adaboost ensemble learning principle in an embodiment of the present invention.

[0039] Figure 7 This is a graph showing the detection results of the method described in this embodiment of the invention for the time sequence from morning to noon on July 4, 2019; wherein, Figure 7 (a) and Figure 7 (b) shows the false-color image and detection results at 03:25, respectively. Figure 7 (c) and Figure 7 (d) shows the false-color image and detection results at 06:40, respectively. Figure 7 (e) and Figure 7 (f) shows the false-color image and detection results at 09:55, respectively. Figure 7 (g) and Figure 7 (h) shows the false-color image and detection results at 12:50, respectively; Figure 7 (a) and Figure 7 The blue line in (e) represents the CALIOP motion trajectory. Figure 7 (b) and Figure 7 The red line in (f) represents sea fog / low clouds in the CALIOP VFM results.

[0040] Figure 8 The image shows the time-series detection results from the afternoon to dusk of July 4, 2019, using the method described in this embodiment of the invention; wherein, Figure 8 (a) and Figure 8 (b) shows the false-color image and detection results at 16:05, respectively. Figure 8 (c) and Figure 8 (d) shows the false-color image and detection results at a 19:20 scale, respectively. Figure 8 (e) and Figure 8 (f) shows the false-color image and detection results at 21:00, respectively. Figure 8 (g) and Figure 8 (h) shows the false color image and detection results at a 23:15 scale. 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 Arctic daytime ice fog / low clouds based on Adaboost, referencing... Figure 1 As shown, it includes the following steps:

[0043] Step 1: Acquire MODIS remote sensing image data collected by polar-orbiting satellites and perform preprocessing.

[0044] The MODIS remote sensing image data acquired by the polar-orbiting satellites in this embodiment refers to the remote sensing image data obtained by the Moderate-resolution Imaging Spectraradiometer carried by the polar-orbiting meteorological satellites Terra and Aqua, abbreviated as MODIS 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 A2: Geometric positioning. Convert the calibrated MODIS data to the WGS-84 coordinate system, selecting UPS_North projection as the projection method, and resample the data spatial resolution to 1km.

[0048] Step A3: Land Masking. The MODIS data is masked using a standard Arctic land boundary SHP file, setting the pixel values ​​of the land portion to 0.

[0049] 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 2000m are defined as sea fog / low clouds.

[0050] Figure 2 This embodiment demonstrates the land-sea distribution and remote sensing images of the study area of ​​interest.

[0051] Step 2: Calculate spectral and texture features.

[0052] The main objects in the Arctic research area during the daytime include four categories: mid-to-high clouds, sea fog / low clouds, sea ice, and seawater. To conduct sea fog / low cloud detection, it is necessary to construct appropriate spectral and texture indicators, which can then be used to classify and identify each pixel in the MODIS data using a model, thereby obtaining the sea fog / low cloud information.

[0053] Based on the object classification results obtained from CALIOP data on July 3, 2019, the diurnal variation curves of visible-near-infrared band reflectance radiation characteristics of different objects in the study area were obtained by analyzing MODIS time-series sample data. Figure 3 ), and the results of calculating the luminance temperature image and texture index "entropy" for the 11.030μm channel on July 4, 2019 ( Figure 4 Therefore, this embodiment constructs a model to identify input data for sea fog / low clouds by calculating the following spectral and texture features.

[0054] Step B1: Calculate spectral characteristics.

[0055] Step B1.1: Based on the reflectance differences between sea fog / low clouds and seawater in the B1 (0.645μm) and B5 (1.240μm) bands, calculate the Normalized Difference Seawater Index (NDSWI) as a feature for seawater removal.

[0056] (1)

[0057] In the formula, and The values ​​represent the spectral reflectance in the B1 (0.645 μm) and B5 (1.240 μm) bands of MODIS, respectively. The NDSWI time-series calculation results for different objects are shown below. Figure 5 (a) serves as a feature separation index to separate the two.

[0058] Step B1.2: Based on the differences in reflectance characteristics between sea ice and other objects in the B5 (1.240 μm) and B8 (0.4125 μm) bands, calculate the Sea Ice Recognition Index (SIRI) as a sea ice removal feature:

[0059] (2)

[0060] In the formula, and The reflectance values ​​are for the B5 (1.240 μm) and B8 (0.4125 μm) bands of MODIS data, respectively. SIRI time-series calculation results for different objects are shown below. Figure 5 (b) serves as a feature separation index to separate the two.

[0061] Step B1.3: Based on the reflectance variation characteristics of high clouds and sea fog / low clouds in the shortwave infrared (B17 (0.905μm) and B18 (0.936μm) spectra, calculate the Mid-High Cloud Recognition Index (MHCRI) as a feature for removing mid-high clouds:

[0062] (3)

[0063] In the formula, and The values ​​represent the reflectance of the MODIS B17 (0.905 μm) and B18 (0.936 μm) shortwave infrared bands, respectively. MHCRI time-series calculation results for different objects are shown below. Figure 5 (c) can be used as a feature separation index to separate the two.

[0064] Step B2: Calculate texture features.

[0065] Given the texture differences between sea fog / low clouds and mid-to-high clouds in thermal infrared brightness temperature images, this invention uses three texture features calculated by the gray-level co-occurrence matrix—homogeneity, contrast, and entropy—as the separation and detection features between the two.

[0066] Step B2.1: Calculate the gray-level co-occurrence matrix. Based on... Figure 4 The thermal infrared texture feature analysis results of mid-to-high clouds and sea fog / low clouds were used to select the 11.030μm band (B). 11.030μm Brightness temperature image calculation for each gray level With gray level In direction and distance Frequency of occurrence :

[0067] (4)

[0068] In the formula For two pixels in Distance difference in direction, This represents the number of gray levels. To improve computational efficiency, this embodiment compresses the image gray level to 3 bits and uses a 7×7 window with a distance of 1 to traverse the image. The directions θ are selected as 0°, 45°, 90° and 135°. To fully reflect the joint distribution characteristics of gray levels, the mean of the gray-level co-occurrence matrices in the four directions is selected as the feature value of the center pixel to obtain the final gray-level co-occurrence matrix of the center pixel.

[0069] Step B2.2: Calculate homogeneity. Homogeneity reflects the local uniformity of image texture. Sea fog / low clouds typically exhibit relatively uniform texture, while mid-to-high cloud textures are more complex, such as cloud clusters and gaps, containing more structural features. The calculation formula is as follows:

[0070] (5)

[0071] Step B2.3: Calculate contrast. Contrast reflects the degree of grayscale difference in an image. Sea fog and low clouds show relatively small grayscale changes, with indistinct edges and structures, while mid- to high-altitude clouds typically exhibit significant edge, cloud cluster, and texture variations, resulting in larger grayscale differences. The calculation formula is as follows:

[0072] (6)

[0073] Step B2.4: Calculate entropy. Entropy reflects the complexity of an image. A higher entropy value indicates a greater degree of disorder, more local information, and higher image complexity, and vice versa. Sea fog and low clouds have relatively simple textures, concentrated gray-level distributions, and low randomness, while mid-to-high clouds have complex textures, discrete gray-level distributions, and higher randomness. The calculation formula is as follows:

[0074] (7)

[0075] Among the three formulas above These are two index variables used to traverse the rows and columns of the gray-level co-occurrence matrix, corresponding to different gray levels, and refer to the mean of the pixel gray values ​​within the selected area. This indicates that the gray-level co-occurrence matrix represents the gray-level value pairs. The value at this location represents the gray level. The pixels and gray levels are The joint probability of pixels appearing simultaneously.

[0076] Step B3: Construct and train the Adaboost model. Based on different time-series MODIS image data, extract the type labels corresponding to each pixel: sea fog / low clouds (147,355 pixels), seawater (55,477 pixels), sea ice (297,008 pixels), and mid-to-high clouds (194,633 pixels). Simultaneously, use all calculated spectral and texture features as input features to construct a training sample set, with 80% of the sample set used as the training set and 20% as the test set. Based on the Adaboost ensemble learning principle (… Figure 6 Using a decision tree as a weak learner, the decision tree parameters were set to train and obtain a classification model of Arctic summer daytime ice fog / low cloud. The specific parameter settings of the Adaboost ensemble learner are shown in Table 1.

[0077]

[0078] Step B4: Input the calculated spectral and texture features into the trained Adaboost Arctic daytime ice fog / low cloud classification model to obtain the fog / low cloud classification results.

[0079] In this invention, the input and output formats of the Adaboost Arctic daytime ice fog / low cloud classification model are both to input and output all pixels of the entire MODIS image in Excel or CSV format, thereby achieving the classification of pixels within the image.

[0080] Step 3: Remove residual mid-to-high clouds.

[0081] Step C1: Some mid-to-high clouds are distributed above 2km, much higher than sea fog / low clouds. However, the spectral characteristics and GLCM texture features of these clouds are quite similar to those of sea fog / low clouds, resulting in a small number of such mid-to-high clouds appearing in the sea fog / low cloud detection results of the Adaboost model. 11.030μm band (B 11.030μm Brightness temperature data can effectively reflect the radiation temperature of cloud tops and sea surfaces. The brightness temperature of these mid-to-high cloud types is below 265K. Figure 3 (f)), while sea fog / low clouds show the opposite. Based on the 11.030μm band (B 11.030μm The brightness temperature index is used to further separate mid-to-high clouds and sea fog / low clouds based on the classification results in step 2. If the brightness temperature value at the corresponding location in the image is less than 265K, it is classified as a mid-to-high cloud and removed; otherwise, it is the final sea fog / low cloud detection result.

[0082] Spectral and texture indices were calculated based on MODIS data. Detection results were obtained using an Adaboost-based Arctic daytime ice fog / low cloud classification model, and then analyzed using B... 11.030μm The brightness temperature band was further used to remove mid-to-high level clouds, resulting in sea fog / low cloud classification results.

[0083] The accuracy of this invention will be verified below:

[0084] This invention takes a sea fog event that occurred on July 4, 2019 as an example. Figure 7 and Figure 8 The study will monitor daytime sea fog / low clouds in the Arctic region.

[0085] According to MODIS false-color imagery and CALIOP data, at 03:25 (UTC-7) on that day Figure 7 (a) A large area of ​​sea fog / low clouds has formed over Baffin Bay, covering 95% of the sea area together with mid-to-high clouds; the fog area is relatively thin, and the underlying surface exhibits typical ice-water cross-contamination characteristics, making it a typical advection fog event in the Arctic ice break zone. 06:40 (UTC-7) Figure 7 (c) until 16:05 (UTC-7) Figure 8 During (a) period, driven by the northwesterly airflow, the fog area continued to advance towards the Canadian Arctic Islands, successively reaching 12:50 (UTC-7). Figure 7 (g) At 16:05 (UTC-7), the Lancaster Strait will be covered by 60% of the waterway. Figure 8 (a) At time 19:20 (UTC-7), a band of fog appeared in the northern ice sheet of Devon Island. Figure 8 (c) until 21:00 (UTC-7) Figure 8(e) During this period, it gradually covered the Lancaster Channel and spread towards the Viscount Melville Channel. 23:15 (UTC-7) Figure 8 At time (g), a large area of ​​sea fog / low clouds remained in the Baffin Bay area and continued to spread toward the Arctic Islands under the continuous influence of airflow.

[0086] This sea fog event has CALIOP transit data at two times, one at 03:25 (UTC-7) Figure 7 (b) At that time, VFM data showed that the fog height in Baffin Bay remained stable at 400m ± 100m, and the sea fog / low cloud detected by this invention was consistent with the VFM results. 09:55 (UTC-7) Figure 7 (f) The VFM data at the corresponding time point shows that the fog height in the study area is within the range of 450m±100m. Except for the detection results of some thin fog edges which are inconsistent with the VFM data, the detection results of the remaining fog areas are consistent with the VFM data. Therefore, this invention reflects the formation and dissipation process of the sea fog / low clouds in a good way.

[0087] This invention utilizes CALIOP VFM product data to verify the detection accuracy of the invention. First, the MODIS data detection results and CALIOP data matrix are obtained, as shown in Table 2. Second, verification indicators including accuracy (POD), false alarm rate (FAR), and reliability factor (CSI) are selected to evaluate the detection accuracy of the invention. The specific calculations are shown in formulas (8)-(10).

[0088]

[0089] (8) (9) (10)

[0090] In the formula, X represents the number of pixels that are both detected by CALIOP and MODIS as sea fog / low clouds; Y represents the number of falsely detected pixels (i.e., the number of pixels that are not sea fog / low clouds in CALIOP, but sea fog / low clouds in MODIS); Z represents the number of missed pixels (i.e., the number of pixels that are not sea fog / low clouds in CALIOP, but not sea fog / low clouds in MODIS).

[0091] This invention randomly selected 12 remote sensing images from MODIS in the summer of 2020 and 2021, and evaluated the accuracy of the sea fog / low cloud algorithm based on CALIOPVFM data. The evaluation results are shown in Table 3.

[0092]

[0093] In Table 3, the average accuracy indicators of POD, FAR, and CSI are 81.46%, 11.40%, and 73.41%, respectively, indicating that the algorithm has high accuracy and reliability. During certain periods (July 28, 2020 and August 11, 2021), FAR was abnormal (>30%), mainly due to the area where CALIOP passed through being at the edge of fog. For example, there was a large time difference (≥30 min) between the validation data (July 28, 2020, 10:01 (UTC-7)) and the detection data (July 28, 2020, 10:35 (UTC-7)). The rapidly moving fog caused deviations in the validation. Meanwhile, the algorithm's detection performance varied significantly across different months. The POD accuracy in July and August (average 90.30%) was much higher than in September (average 69.10%). This was mainly due to the heterogeneity in the texture of sea fog / low clouds, leading to missed detections. Although GLCM texture features (contrast, entropy, etc.) effectively characterize most sea fog / low cloud textures, September saw a large number of patchy sea fog / low clouds. While CALIOP data showed that the cloud top heights in the sea fog / low cloud areas were generally consistent and uniform, MODIS imagery and GLCM texture feature calculations revealed a significant increase in texture heterogeneity, with GLCM texture values ​​closer to those of mid-to-high clouds. This unique texture phenomenon resulted in some sea fog / low cloud being missed in September, leading to lower POD values. Apart from these special cases, the detection accuracy remained high and stable in other time periods. Therefore, this invention can clearly detect sea fog / low clouds, and it has a certain degree of stability in the face of underlying surfaces with inconsistent ice-water mixing ratios and images with different solar altitude angles. It effectively solves the problems of missed detection and inconsistent detection caused by spatiotemporal heterogeneity in traditional sea fog detection algorithms, and better reflects the formation and dissipation process of sea fog / low clouds, further proving the reliability of the method of this invention.

[0094] 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. An Adaboost-based polar mesopause fog / low cloud detection method, characterized in that, The method comprises the following steps: acquiring MODIS remote sensing image data collected by a polar orbit satellite and preprocessing the data; calculating a plurality of spectral features and a plurality of texture features by using preset band data in the preprocessed remote sensing image data; inputting the calculated spectral features and texture features into an Arctic polar day sea ice fog / low cloud classification model based on Adaboost and trained, to obtain a sea ice fog / low cloud classification result in the remote sensing image data; removing residual medium and high clouds in the classified sea ice fog / low cloud by using 11.030 μm band brightness temperature data in the preprocessed remote sensing image data, to obtain finally detected sea ice fog / low cloud.

2. The Adaboost-based polar mesospheric ice fog / low cloud detection method according to claim 1, characterized in that, The preprocessing comprises the following steps: firstly, calibrating each band as reflectivity or brightness temperature; secondly, converting the calibrated MODIS remote sensing image data to a WGS-84 coordinate system; and thirdly, performing mask processing on an Arctic land area in the MODIS remote sensing image data in the WGS-84 coordinate system.

3. The Adaboost-based polar mesospheric ice fog / low cloud detection method according to claim 1, wherein, The calculated spectral features comprise a normalized sea water index NDSWI, a sea ice recognition index SIRI, and a medium and high cloud recognition index MHCRI.

4. The Adaboost-based polar mesospheric ice fog / low cloud detection method according to claim 1, wherein, The normalized sea water index is calculated by using reflectivity data of 0.645 μm band and 1.240 μm band, and is denoted as NDSWI, with a calculation formula as follows: ； wherein, and represent the reflectance of the 0.645 μm and 1.240 μm bands, respectively, in the MODIS remote sensing imagery data.

5. The Adaboost-based polar mesospheric ice fog / low cloud detection method according to claim 1, wherein, The sea ice recognition index is calculated by using reflectivity data of 1.240 μm band and 0.4125 μm band, and is denoted as SIRI, with a calculation formula as follows: ； wherein, and represent the reflectance of the 1.240 μm and 0.4125 μm bands, respectively, of the MODIS remote sensing imagery data.

6. The Adaboost-based polar mesospheric ice fog / low cloud detection method according to claim 1, wherein, The medium and high cloud recognition index is calculated by using reflectivity data of 0.905 μm band and 0.936 μm band, and is denoted as MHCRI, with a calculation formula as follows: ； wherein and represent the reflectance of the 0.905 μm and 0.936 μm bands, respectively, in MODIS remote sensing imagery data.

7. The Adaboost-based polar mesospheric ice fog / low cloud detection method according to claim 1, wherein, Using the brightness temperature data of 11.030 μm band, a gray level co-occurrence matrix of each pixel is calculated, and then texture features are calculated based on the gray level co-occurrence matrix; wherein the calculated texture features include: homogeneity , contrast , and / or entropy .

8. The Adaboost-based polar mesospheric ice fog / low cloud detection method according to claim 7, characterized in that, homogeneity , contrast , entropy The calculation formula is: ； ； ； wherein denotes the value of the gray level co-occurrence matrix at the gray value pair represents the joint probability of a pixel with gray level and a pixel with gray level appearing simultaneously.

9. An electronic device comprising a memory and a processor, said memory having stored therein a computer program, characterized in that, The computer program is executed by the processor, so that the processor implements the method according to any one of claims 1-8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor, so that the processor implements the method according to any one of claims 1-8.

Images