High-precision sea fog image recognition and prediction method based on ensemble learning

Abstract

The invention relates to the field of sea fog image recognition, discloses a high-precision sea fog image recognition and prediction method based on ensemble learning, and aims to improve multi-source data fusion efficiency and reasonable utilization and improve the stability of a model so as to realize accurate recognition of sea fog and more accurate time-space evolution prediction. According to the method, a three-dimensional feature space is constructed through visible light wavelet texture analysis, infrared dynamic temperature correction and microwave liquid water path inversion, an image branch adopts improved DeepLabV3 + to carry out fog region semantic segmentation, and fog region identification and trajectory prediction are realized in combination with a thermodynamic phase change criterion and a fluid mechanics diffusion model. And finally, a high-precision sea fog image identification and prediction system is formed. A black box model is replaced with a physical equation, and the defects that a traditional method is poor in interpretability and insufficient in real-time performance are overcome.

Description

Technical Field

[0001] The present invention relates to the field of sea fog image recognition, and particularly to a high-precision sea fog image recognition and prediction method based on ensemble learning. Background Art

[0002] Sea fog is a common weather phenomenon at sea, posing a great threat to the safety of maritime operations. Therefore, accurately recognizing and predicting sea fog images is of great significance for ensuring the safety of maritime activities.

[0003] Currently, the methods for sea fog image recognition mainly include threshold segmentation, feature extraction, and machine learning methods. Among them, threshold segmentation is simple and direct, but is greatly affected by external factors such as illumination and weather. Feature extraction relies on manual extraction, which is time-consuming and laborious, and key information is easily missed. The methods for sea fog forecasting mainly include synoptic methods, numerical forecasting methods, and machine learning methods. Synoptic methods mainly rely on the subjective experience and professional level of forecasters, and it is difficult to guarantee the accuracy under complex weather conditions. Numerical forecasting methods have a high accuracy in large-scale sea fog forecasting, but have a low forecasting accuracy for micro-scale sea fog affected by local complex factors.

[0004] In recent years, machine learning methods have been gradually applied in the fields of sea fog recognition and forecasting, and the classification accuracy has been improved by automatically learning features. However, these methods have the following limitations: Data dependence: A large number of labeled samples are required to train the model, but it is difficult to obtain sea fog samples, and overfitting or "curse of dimensionality" is likely to occur in small-sample scenarios; Poor interpretability: The black-box model is difficult to associate with physical mechanisms and cannot reveal the key driving factors for the formation of sea fog; Insufficient real-time performance: Complex models are time-consuming to calculate, difficult to meet the needs of minute-level forecasting, and rely on high computing power resources.

[0005] Therefore, we propose a high-precision sea fog image recognition and prediction method based on ensemble learning to solve the above problems. Summary of the Invention

[0006] The present invention provides a high-precision sea fog image recognition and prediction method based on ensemble learning, which provides a high-precision and low-dependence path for sea fog prediction through a three-element architecture of physical mechanism modeling - multi-source hard fusion - dynamic feedback calibration.

[0007] The first aspect of the present invention provides a high-precision sea fog image recognition and prediction method based on ensemble learning. The high-precision sea fog image recognition and prediction method based on ensemble learning includes: obtaining visible light satellite remote sensing images and infrared band data, performing multi-scale texture analysis on the visible light satellite remote sensing images, extracting texture features of the fog area edge, generating a visible light texture feature map, calculating the brightness temperature difference for the infrared band data, and generating an infrared fog area probability map in combination with the sea surface temperature; obtaining the sea surface temperature and the air temperature above the sea surface measured by ocean buoys, obtaining a critical humidity threshold, and generating a thermodynamic phase change criterion matrix in combination with the edge sharpness quantization value in the visible light texture feature map; obtaining microwave radiometer data, retrieving the liquid water path based on the microwave brightness temperature data, and performing logical fusion based on the infrared fog area probability map, the thermodynamic phase change criterion matrix, and the liquid water path threshold to generate a multi-source fusion fog area recognition map; based on the initial fog area distribution provided by the multi-source fusion fog area recognition map, combining wind field data and boundary layer height data to obtain the horizontal diffusion rate and direction of the fog area, simulating the future movement trajectory, and generating a fog band diffusion vector field; superimposing the multi-source fusion fog area recognition map and the fog band diffusion vector field to generate a time series prediction result. When it is predicted that the fog area covers a port or a waterway and the coverage probability is ≥ a preset value, a hierarchical alarm signal is triggered, and a comprehensive sea fog prediction alarm map is generated. <

[0008] Optionally, in the first implementation manner of the first aspect of the present invention, it includes: performing two-dimensional discrete wavelet transform on the visible light satellite remote sensing image, decomposing it to the 3rd layer, extracting the energy values of the high-frequency components, and generating a wavelet high-frequency energy map; obtaining the total high-frequency energy of each pixel point based on the wavelet high-frequency energy map, mapping the energy value to the [0,1] interval, and generating a visible light texture feature map; obtaining the brightness temperature difference BTD based on the channel brightness temperature data and the sea surface temperature SST, and performing piecewise linear correction on BTD according to SST to generate a corrected brightness temperature difference map; mapping the brightness temperature difference BTD to the fog existence probability through an S-shaped function to generate an infrared fog area probability map.

[0009] Optionally, in the second implementation manner of the first aspect of the present invention, it includes: based on the sea surface temperature SST measured by the ocean buoy in real time and the air temperature 2 meters above the sea surface , obtaining the sea-air temperature difference: ; dividing the thermodynamic state according to the positive and negative of ΔT to obtain a sea-air temperature difference state table; based on the sea-air temperature difference state table, obtaining the critical relative humidity threshold under different ΔT: when ΔT > 0, ; when ΔT ≤ 0, ; obtaining a dynamic critical humidity threshold table; based on the visible light texture feature map and the dynamic critical humidity threshold table, performing weighted superposition of the edge sharpness quantization value and the critical relative humidity threshold: if the edge sharpness ≥ 0.7, then is reduced by 5%; if the edge sharpness < 0.7, then Maintain the original value; obtain the thermodynamic criterion matrix integrating edge features; based on the thermodynamic criterion matrix of the integrated edge features and the real-time relative humidity field, compare the real-time relative humidity RH with the corrected critical relative humidity threshold pixel by pixel : When RH ≥ , it is determined as a fog area; when RH < , it is determined as a non-fog area; obtain the thermodynamic phase change criterion matrix.

[0010] Optionally, in the third implementation manner of the first aspect of the present invention, it includes: performing radiometric calibration and geolocation correction on the brightness temperature data, removing the outliers affected by abnormal sea surface roughness, and obtaining the corrected microwave brightness temperature map; based on the corrected microwave brightness temperature map, inversely calculating the liquid water path based on the microwave radiative transfer model to obtain the liquid water path distribution map; resampling the liquid water path distribution map to a resolution of 1 km by bilinear interpolation, and processing it based on the infrared fog area probability map, the thermodynamic criterion matrix, and the liquid water path distribution map to obtain the preliminary screened integrated fog area distribution map; in the preliminary screened fog area, if the edge sharpness in the visible light texture feature map is ≥ 0.7, it is determined as a reliable fog area, and the misdetected areas with edge sharpness < 0.7 are removed to generate a multi-source integrated fog area identification map.

[0011] Optionally, in the fourth implementation manner of the first aspect of the present invention, it includes: converting the multi-source integrated fog area identification map into a rasterized concentration field to generate an initial fog area concentration field, and adjusting the wind field intensity according to the boundary layer height: for every 100 m decrease in the boundary layer height, the wind field speed decays by 5% to generate a corrected wind field vector map; based on the initial fog area concentration field and the corrected wind field vector map, with a time step of 10 minutes, iteratively calculate the movement trajectory of the fog area with the wind field in the next 3 hours, record the fog area front position at each time step to generate a fog area trajectory sequence; obtain the fog area displacement amount between adjacent time steps based on the fog area trajectory sequence, extract the diffusion speed and direction, and remove the abnormal trajectory points caused by terrain occlusion to generate a fog band diffusion vector field.

[0012] Optionally, in the fifth implementation manner of the first aspect of the present invention, it includes: performing spatio-temporal alignment on the multi-source fusion fog area recognition map and the fog band diffusion vector field, interpolating the fog area front position at 10-minute intervals to generate a time series fog area prediction grid set; according to the time series fog area prediction grid set and the preset port and waterway geographic information database, for each port / waterway area, counting the proportion of the predicted grid number covered by the fog area to obtain the area coverage probability. When the coverage probability of the same area for three consecutive time steps ≥ 70%, it is marked as a high-risk area to obtain a key area coverage probability table; according to the key area coverage probability table, setting an alarm threshold according to the risk level: yellow alarm: 70% ≤ probability < 85%; orange alarm: 85% ≤ probability < 95%; red alarm: probability ≥ 95%; combining the area priority weight to generate a hierarchical alarm instruction set; superimposing the time series fog area prediction grid set and the hierarchical alarm instruction set on the electronic chart base map, marking the time series of the fog area front and the alarm area to generate a sea fog comprehensive prediction alarm map.

[0013] Optionally, in the sixth implementation manner of the first aspect of the present invention, it further includes the following steps: comparing the fog area range in the sea fog comprehensive prediction alarm map with the visibility observation data real-time transmitted by the ocean buoy to obtain the regional prediction error, where the regional prediction error = predicted coverage rate - measured coverage rate; if the absolute value of the prediction error for three consecutive time steps > 15%, then trigger calibration: adjusting the critical humidity threshold of the thermodynamic phase change criterion; correcting the wind field attenuation coefficient; generating an optimized model parameter set.

[0014] The mechanism of the present invention is as follows: The present invention is based on a dynamic closed-loop prediction system that couples multi-source physical characteristics, and through physical model-driven instead of data-driven, it realizes a breakthrough innovation in sea fog recognition and prediction.

[0015] Beneficial effects: Visible light wavelet decomposition quantifies the fog area edge texture, breaking through the dependence on color features in traditional image segmentation. The piecewise correction of the infrared bright temperature difference and the sea surface temperature solves the problem of misjudgment of the fog area in low sea temperature regions. The microwave liquid water path inversion formula enhances the perception of the vertical information of the fog area through the physical radiation model; Dynamically adjusting the critical humidity threshold based on the sea-air temperature difference to realize the real-time optimization of the thermodynamic phase change criterion. The boundary layer height corrects the wind field attenuation rate and incorporates the vertical stratification effect into the horizontal diffusion model; Using physical equation-driven instead of data-driven to solve the inherent problems of traditional methods that rely on labeled data and are black-box and unexplainable; Forming a closed loop from feature extraction, diffusion prediction to real-time calibration, breaking through the limitations of the fixed parameters of numerical models. Description of the Drawings

[0016] Figure 1Schematic diagram of an embodiment of the high-precision sea fog image recognition and prediction method based on ensemble learning in the embodiments of the present invention; Figure 2 Schematic diagram of another embodiment of the high-precision sea fog image recognition and prediction method based on ensemble learning in the embodiments of the present invention; Figure 3 Schematic diagram of an embodiment of the high-precision sea fog image recognition and prediction device based on ensemble learning in the embodiments of the present invention; Figure 4 Schematic diagram of an embodiment of the high-precision sea fog image recognition and prediction device based on ensemble learning in the embodiments of the present invention; Figure 5 It is the improved DeepLabV3+ network structure diagram in the present invention. Detailed implementation manners

[0017] The embodiments of the present invention provide a high-precision sea fog image recognition and prediction method based on ensemble learning, which provides a high-precision and low-dependency path for sea fog prediction through a ternary architecture of physical mechanism modeling - multi-source hard fusion - dynamic feedback calibration. In addition, the term "including" or "having" and any of its variants are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to Figure 1 An embodiment 1 of the high-precision sea fog image recognition and prediction method based on ensemble learning in the embodiments of the present invention includes: 101. Feature extraction of multi-spectral satellite remote sensing images. Input data: Visible light satellite remote sensing images: from the VISSR sensor of Fengyun-4A satellite, with a spatial resolution of 500 meters; Infrared band data: Brightness temperature data of the 10.8μm and 12.0μm channels from the AHI sensor of Himawari-8 satellite; Perform multi-scale texture analysis on the visible light image, extract the texture features of the fog area edge, and generate a visible light texture feature map; Calculate the brightness temperature difference (the brightness temperature difference value between 10.8μm and 12.0μm) for the infrared band data, and generate an infrared fog area probability map in combination with the sea surface temperature; Output products: Visible light texture feature map (including the edge sharpness quantization value); Infrared fog area probability map (with a resolution of 2 kilometers, and the pixel value represents the fog presence probability, ranging from 0 to 1); It can be understood that the execution entity of the present invention can be a high-precision sea fog image recognition and prediction device based on integrated learning, or it can also be a terminal or a server. Specifically, it is not limited here. In the embodiments of the present invention, the server is taken as the execution entity for illustration.

[0019] It should be noted that the input data is the visible light satellite remote sensing image (spatial resolution 500 meters) obtained by the VISSR sensor of Fengyun-4A satellite.

[0020] Operation process: Multi-scale texture analysis: Sliding window setting: Three window sizes of 3×3, 5×5, and 7×7 are adopted to extract texture features of different scales respectively. Gray-level co-occurrence matrix (GLCM) calculation: For each window, texture parameters such as contrast (CON), entropy (ENT), and correlation (COR) are calculated. The fog area edge usually shows low contrast (CON≤5) and high entropy value (ENT≥2.5), reflecting the smoothness and randomness of the fog area texture. Edge sharpness quantization: The gradient magnitude is calculated through the Sobel operator, and the edge sharpness index (value range 0~1, the larger the value, the clearer the edge) is generated in combination with the GLCM features.

[0021] Feature fusion and dimensionality reduction: The multi-scale texture parameters are input into the principal component analysis (PCA), and the first principal component (contribution rate ≥85%) is extracted as the core index of the visible light texture feature map. The output resolution is maintained at 500 meters and unified to a 1-kilometer grid through bilinear interpolation for subsequent fusion.

[0022] Output product: Visible light texture feature map (1-kilometer resolution, pixel value is the edge sharpness quantization value). If the edge sharpness of a certain area is 0.8, it indicates that there is a significant fog area boundary in this area.

[0023] Generation of infrared fog area probability map, input data: Brightness temperature data of the 10.8μm and 12.0μm channels of the AHI sensor of Himawari-8 satellite (original resolution 2 kilometers).

[0024] Operation process: Brightness temperature difference calculation: Calculate the brightness temperature difference ( , ,

[0024] , ,

[0023] , ,

[0022] , , 10.8 , , 12.0 ,

[0025] ) between the 10.8μm (BT 12.0 ) and 12.0μm (BT ) channels. Sea surface temperature (SST) constraint: Combine the remotely sensed retrieved SST data (MODIS SST product) to screen the areas where the difference between SST and dew point temperature ≤2°C (thermodynamic conditions for fog formation). Construct a logistic regression model: , where σ is the Sigmoid function, and the output probability range is 0~1. Spatial interpolation and resolution adjustment: Use the Kriging interpolation method to unify the probability values from a 2-kilometer resolution to a 1-kilometer grid and spatially align them with the visible light feature map.

[0025] Output product: Infrared fog area probability map (2-kilometer resolution, example of probability value: if the pixel value in a certain sea area is 0.75, it means there is a 75% probability of sea fog at that location).

[0026] 102. Modeling of thermodynamic phase transition criterion, input data: Sea surface temperature and air temperature 2 meters above the sea surface measured by ocean buoys; Visible light texture feature map generated in step 101; Calculate the critical humidity threshold based on the thermodynamic equation, and combine with the edge sharpness quantization value in the visible light texture feature map to generate a thermodynamic phase transition criterion matrix; Output product: Thermodynamic phase transition criterion matrix (resolution is 1 kilometer, 0 represents non-fog, 1 represents fog); It should be noted that the measured data: Sea surface temperature measured by ocean buoy (SST = 20°C), air temperature 2 meters above the sea surface (Ta = 18°C), need to be converted from discrete point data to 1-kilometer resolution grid data through Kriging interpolation method. Visible light texture feature map (from step 101): Resolution is 1 kilometer, and the edge sharpness value ranges from 0.3 to 0.9 (the edge sharpness of the fog area is usually ≥0.6).

[0027] Calculation of critical humidity threshold, based on the thermodynamic phase transition criterion, when the air humidity reaches saturation (i.e., the air temperature is close to the dew point temperature), sea fog may form. The specific steps are as follows: Calculation of dew point temperature (Td): Use the measured air temperature (Ta) and relative humidity (RH), and calculate the dew point temperature through the Magnus formula: ; Among them, the unit of 237.3 is °C, and the measured RH of the buoy = 90%, then the calculated Td≈16.5°C.

[0028] Criterion condition: If , then the air is close to the saturation state, meeting the humidity condition for fog formation. In this example, , which meets the condition.

[0029] Constraint of sea surface temperature (SST): When SST≥15°C and the difference between SST and Ta≤3°C, sufficient water vapor evaporation is conducive to fog generation. In this example, SST = 20°C, , which meets the condition.

[0030] Fusion of visible light texture features, extract the edge sharpness value from the visible light texture feature map (set the sharpness of a certain area to 0.7), when the sharpness≥0.6, it indicates that there is a clear fog area boundary in this area.

[0031] Logical fusion rule: If a certain pixel meets all of the following conditions simultaneously: ; ; Edge sharpness≥0.6; Then it is marked as a fog area (set to 1 in the criterion matrix), otherwise it is 0.

[0032] Output product example Table 1, Thermodynamic phase change criterion matrix (1 km resolution): Table 1 Position coordinates Criterion value Key parameter (120°E, 30°N) 1 Ta = 18℃, Td = 16.5℃, Sharpness = 0.7 (120.1°E, 30°N) 0 Ta = 19℃, Td = 14℃ (humidity condition not satisfied)

[0033] If the measured RH fluctuates to 85% (Td ≈ 15°C), then °C, the criterion fails and the model needs to be recalibrated; through ROC curve verification, when the sharpness threshold is increased from 0.6 to 0.65, the false positive rate decreases by 5%, but the missed detection rate increases by 3%, and a trade-off needs to be made for selection.

[0034] 103. Multi-source data fusion and fog area identification, input data: infrared fog area probability map generated in step 101; thermodynamic phase change criterion matrix generated in step 102; microwave radiometer data: 36.5 GHz vertically polarized brightness temperature data from the AMSR2 sensor of the GCOM-W1 satellite; Invert the liquid water path based on the microwave brightness temperature data, and logically fuse the infrared fog area probability map, the thermodynamic phase change criterion matrix, and the liquid water path threshold to generate a multi-source fusion fog area identification map; Output product: Multi-source fusion fog area identification map (resolution 1 km, rasterized binary distribution, 0 represents non-fog area, 1 represents fog area); It should be noted that the infrared fog area probability map (from step 101): resolution 2 km, pixel value range in a certain sea area is 0.2 (clear sky) to 0.9 (high fog probability), and the infrared probability value in a certain area (coordinate 120.5°E, 32°N) is 0.8, indicating an 80% probability of sea fog at that location.

[0035] Thermodynamic phase change criterion matrix (from step 102): resolution 1 km, binary data (0 or 1), and the thermodynamic criterion value of 1 in the same area indicates that the humidity and temperature conditions for fog formation are met.

[0036] Microwave radiometer data (GCOM-W1 satellite AMSR2 sensor): 36.5 GHz vertically polarized brightness temperature data (original resolution 10 km), which needs to be downscaled to a 1 km grid through bilinear interpolation and invert the liquid water path (LWP). The inverted LWP is 200 g / m 2 (threshold set to ≥ 150 g / m 2 when judged as fog).

[0037] Multi-source data logical fusion, fusion rule: infrared probability threshold: when the pixel value ≥ 0.6 (probability ≥ 60%), it is regarded as a candidate fog area. Thermodynamic criterion: matrix value is 1 (meeting the thermodynamic conditions). Liquid water path threshold: LWP ≥ 150 g / m 2 (lower limit of liquid water content in fog area).

[0038] Example scenario: Parameters of a 1-kilometer grid (coordinates 120.5°E, 32°N): Infrared probability: 0.8; Thermodynamic criterion: 1; LWP: 200 g / m 2 ; Fusion result: Meeting all three conditions simultaneously, marked as 1 indicating a fog area.

[0039] Parameters of another grid (coordinates 120.6°E, 32°N): Infrared probability: 0.7; Thermodynamic criterion: 0 (insufficient humidity); LWP: 180 g / m 2 ; Fusion result: Thermodynamic criterion not met, marked as 0 indicating a non-fog area.

[0040] Spatial alignment and resolution unification, Infrared probability map: Interpolated from 2 kilometers to a 1-kilometer grid, using bicubic interpolation to reduce edge blurring; Microwave data: Downscaled from 10 kilometers to 1 kilometer, combined with weighted averaging of the brightness temperature values of neighboring pixels to ensure spatial consistency with the thermodynamic matrix.

[0041] Example of output product Table 2: Multi-source fusion fog area identification map: 1-kilometer resolution, binary distribution.

[0042] Table 2 Position coordinates Criterion value Key parameter (infrared probability / thermodynamics / LWP) (120.5°E, 32°N) 1 0.8 / 1 / 200 g / m2 (120.6°E, 32°N) 0 0.7 / 0 / 180 g / m2

[0043] Parameter sensitivity analysis: Adjustment of LWP threshold: If the threshold is lowered to 100 g / m 2 , the fog area coverage rate may increase by 15%, but the misjudgment rate rises (e.g., misjudging thin clouds as fog).

[0044] Optimization of infrared probability threshold: Verified by the ROC curve, a threshold of 0.6 balances the missed detection rate (8%) and the false detection rate (12%), suitable for most ocean scenarios.

[0045] 104. Prediction of hydrodynamic diffusion trajectory, Input data: Multi-source fusion fog area identification map generated in step 103; 850 hPa wind field data and boundary layer height data provided by numerical weather prediction; Calculate the horizontal diffusion rate and direction of the fog area based on the hydrodynamic equations, simulate the movement trajectory in the next 3 hours, and generate a fog band diffusion vector field; Output product: Fog band diffusion vector field (including the spatio-temporal distribution of diffusion speed and direction, with the speed unit of m / s); It should be noted that the multi-source fusion fog area identification map (from step 103): Resolution is 1 kilometer, binary grid data (0 for non-fog area, 1 for fog area), and a grid in a certain sea area (coordinates 120.5°E, 32°N) is marked as

[0046] Numerical weather prediction data: 850 hPa wind field data: The horizontal wind field speed is 12 m / s, and the wind direction is northeast (45°). Boundary layer height: The current boundary layer height is 800 meters, which affects the vertical diffusion limit of the fog area.

[0047] Construction of the hydrodynamic diffusion model, based on the Navier-Stokes equation and the advection-diffusion equation, to calculate the horizontal diffusion rate and direction of the fog area: Calculation of the advection term: The wind field dominates the advection direction. According to the 850 hPa wind field speed (12 m / s) and direction (45°), calculate the hourly advection displacement of the fog area: ; ; The advection direction is consistent with the wind field, and the cumulative displacement is about 91.2 kilometers every 3 hours.

[0048] Calculation of the turbulent diffusion term: According to the boundary layer height (800 meters) and the atmospheric stability (neutral condition), use the eddy diffusion coefficient K = 50 m 2 / s to calculate the horizontal diffusion rate: ; Comprehensive movement trajectory simulation: Superimpose advection and diffusion to generate a diffusion vector field of the fog belt. The predicted coordinates of the center point (120.5°E, 32°N) of a fog area after 3 hours are (121.3°E, 32.5°N), the diffusion speed is 12 m / s, the direction is 45°, and the diffusion range radius is 1.8 kilometers.

[0049] Boundary layer height constraint: The boundary layer height restricts vertical turbulent mixing, and the fog area only diffuses below 800 meters, and the vertical speed is negligible.

[0050] Terrain correction: If the predicted path encounters the coastline (Yangtze River Estuary), the diffusion direction is adjusted to be along the terrain shear direction (deflected by 20°).

[0051] Output product example is shown in Table 3: Diffusion vector field of the fog belt: Spatiotemporal distribution with a resolution of 1 kilometer, and each grid contains speed and direction.

[0052] Table 3 Position coordinates Speed (m / s) Direction (°) Key parameter (advection / diffusion) (120.5°E, 32°N) 12 45 Wind field dominant, diffusion radius 1.8 km (121.0°E, 32.3°N) 10 50 Topographic shear correction, diffusion radius 1.5 km

[0053] Verification method: Compare the actually measured fog area position with the predicted trajectory after 3 hours. If the error ≤ 5 kilometers, it is considered valid.

[0054] Parameter Sensitivity: If the boundary layer height drops to 500 meters, the diffusion radius shrinks to 1.2 kilometers while the velocity direction remains unchanged. If the wind field velocity increases to 15 m / s, the advection displacement increases to 113 kilometers / 3 hours and the diffusion range expands.

[0055] 105. Comprehensive Prediction and Alarm Generation, Input Data: The multi-source fusion fog area identification map generated in Step 103; the fog band diffusion vector field generated in Step 104. Overlay the multi-source fusion fog area identification map and the fog band diffusion vector field to generate a time series prediction result; when the predicted fog area covers the port or waterway and the coverage probability ≥ 70%, trigger a graded alarm signal and generate a comprehensive sea fog prediction alarm map. Output Product: Comprehensive Sea Fog Prediction Alarm Map (including geographical coordinates, impact time, and alarm level). It should be noted that the multi-source fusion fog area identification map (from Step 103): Binary raster data with a resolution of 1 kilometer (0 for non-fog areas, 1 for fog areas). There is currently a fog area (pixel value 1) in a certain sea area (coordinates 120.3°E, 36.1°N) with a coverage area of approximately 50 square kilometers.

[0056] The fog band diffusion vector field (from Step 104): Includes the diffusion velocity (12 m / s) and direction (northeast wind at 45°). It is predicted that the fog area will move approximately 91.2 kilometers in the southeast direction in the next 3 hours and the diffusion radius will expand to 1.8 kilometers.

[0057] Time Series Prediction: Overlay the fog area identification map and the diffusion vector field, and simulate the position of the fog area every 30 minutes in the next 3 hours based on the advection-diffusion model: Current Moment (T0): The center coordinates of the fog area are 120.3°E, 36.1°N, and the coverage area is 50 km 2 .

[0058] T0 + 1 hour: The center moves to 120.8°E, 36.5°N, and the coverage area expands to 58 km 2 (the diffusion radius increases by 0.6 km).

[0059] T0 + 3 hours: The center reaches 121.6°E, 37.1°N, and the coverage area reaches 75 km 2 , covering part of the waterway of Qingdao Port.

[0060] Coverage Probability Calculation: Based on the uncertainty of the diffusion vector field (wind speed error ±2 m / s), use Monte Carlo simulation to generate 100 sets of trajectories, and statistically calculate the fog area coverage probability in the port area (120.5°E - 121.5°E, 36.5°N - 37.5°N). The coverage probability of the waterway of Qingdao Port at T0 + 3 hours is 75% (exceeding the threshold of 70%).

[0061] Alarm trigger conditions: Level 1 alarm (red): Coverage probability ≥ 90% and impact time ≥ 6 hours.

[0062] Level 2 alarm (orange): Coverage probability ≥ 70% and impact time 3 - 6 hours.

[0063] Level 3 alarm (yellow): Coverage probability ≥ 50% and impact time ≤ 3 hours.

[0064] Example scenario: The predicted coverage probability of the waterway in Qingdao Port is 75%, and the impact time is 08:00 - 11:00 (3 hours), triggering an orange alarm. The alarm information includes: Geographical coordinates: 120.8°E - 121.6°E, 36.5°N - 37.1°N.

[0065] Impact time: From 08:00 to 11:00 on May 9, 2025.

[0066] Alarm level: Level 2 (orange). It is recommended that the port suspend the entry and exit of ships and activate fog navigation control.

[0067] Spatial consistency check: Compare with the measured visibility data of buoys in the same period (such as the visibility of a certain buoy drops to 300 meters), and verify that the position error of the fog area ≤ 3 kilometers.

[0068] Parameter sensitivity: If the error of the diffusion speed exceeds 15%, the alarm level may be downgraded to yellow (coverage probability corrected to 65%).

[0069] The output example is shown in Table 4: Table 4 Geographical coordinate range Influence time Coverage probability Alarm level Suggested measures 120.8°E - 121.6°E, 36.5°N - 37.1°N 08:00-11:00 75% Orange Restrict passage of waterways, activate fog light navigation

[0070] In the embodiments of the present invention, for the first time, visible light texture features at the 500-meter level, infrared brightness temperature probability maps at the 2-kilometer level, and microwave liquid water path data at the 10-kilometer level are integrated. Through bilinear interpolation and Kriging interpolation, 1-kilometer grid alignment is achieved, breaking through the resolution limit of a single sensor. The thermodynamic phase change criterion is logically fused with the visible light texture features to construct a criterion matrix combining physical driving and data driving, reducing the misjudgment rate. 3×3, 5×5, and 7×7 sliding windows are used to extract GLCM parameters, and combined with the Sobel operator to generate an edge sharpness index to quantify the blurriness of the fog area boundary. Based on the Navier-Stokes equation, the advection term and the turbulent diffusion term are simulated to achieve fog band trajectory prediction, and terrain shear correction is introduced. The LWP is retrieved through the 36.5 GHz vertical polarization brightness temperature data, and a threshold dynamic adjustment strategy is established. A secondary alarm system is constructed by combining the coverage probability and the influence time, and the Monte Carlo simulation is used to quantify the uncertainty of the coverage probability caused by the wind speed error. Through double verification of edge sharpness quantification and infrared probability maps, the confusion between light fog and low clouds is reduced. By combining the SST-Ta difference and the dew point temperature difference constraints, interference from non-fog areas is excluded. Based on the 850 hPa wind field and the boundary layer height, the advection displacement and the diffusion radius are calculated, and the verification error ≤ 5 kilometers, which is better than the prediction error of the traditional numerical model. When the coverage probability ≥ 70%, an orange alarm is triggered, supporting hierarchical control of port channels and reducing ship detention time. It is adapted to the data of Fengyun-4A satellite (FY-4A), Himawari-8 satellite, and GCOM-W1 satellite, compatible with MODIS SST products and numerical weather prediction (NWP) wind fields, reducing the deployment cost.

[0071] Please refer to Figure 2 , Another embodiment two of the high-precision sea fog image recognition and prediction method based on ensemble learning in the embodiments of the present invention includes: 201. Feature extraction of multi-spectral satellite remote sensing images, input data: Visible light satellite remote sensing images: from the VISSR sensor of Fengyun-4A satellite, with a spatial resolution of 500 meters; Infrared band data: Brightness temperature data of the 10.8μm and 12.0μm channels from the AHI sensor of Himawari-8 satellite; Perform multi-scale texture analysis on the visible light image, extract the texture features of the fog area edge, and generate a visible light texture feature map; Calculate the brightness temperature difference (the brightness temperature difference between 10.8μm and 12.0μm) for the infrared band data, and combine it with the sea surface temperature to generate an infrared fog area probability map; Output products: Visible light texture feature map (including edge sharpness quantification value); Infrared fog area probability map (with a resolution of 2 kilometers, and the pixel value represents the fog existence probability, ranging from 0 to 1).

[0072] Specifically, sub-step S1-1: Input data: Satellite remote sensing image in the visible light band (0.45~0.90μm) obtained by the VISSR sensor of Fengyun-4A satellite, with an original resolution of 500 meters; perform two-dimensional discrete wavelet transform on the visible light image using Daubechies-4 wavelet basis function and decompose it to the 3rd layer; extract the energy values of the high-frequency components (horizontal, vertical, and diagonal directions) to generate a wavelet high-frequency energy map; Output product: Wavelet high-frequency energy map (resolution 500 meters, quantifying the sharpness of the fog area edge); Sub-step S1-2: Edge sharpness quantization and normalization, Input data: Wavelet high-frequency energy map generated in sub-step S1-1; calculate the total high-frequency energy of each pixel; map the energy value to the interval [0,1] through linear normalization to generate a visible light texture feature map (resolution 500 meters, with pixel values representing edge sharpness); Product use: As the input of step 202 for thermodynamic criterion modeling.

[0073] Sub-step S1-3: Infrared brightness temperature difference calculation and sea surface temperature correction, Input data: Brightness temperature data of the 10.8μm and 12.0μm channels of the AHI sensor of Himawari-8 satellite (resolution 2 km); Sea surface temperature (SST) provided by ERA5 reanalysis data (resolution 5 km); calculate the brightness temperature difference ( ); perform piecewise linear correction on BTD according to SST: when SST < 285K, correction coefficient α = 0.8; when SST ≥ 285K, correction coefficient α = 1.2; generate a corrected brightness temperature difference map (resolution 2 km); Product use: As the input of sub-step S1-4.

[0074] Sub-step S1-4: Infrared fog area probability mapping, Input data: Corrected brightness temperature difference map generated in sub-step S1-3; set the threshold range: when the corrected , it is determined as a potential fog area; map the brightness temperature difference to the fog existence probability through an S-shaped function to generate an infrared fog area probability map (resolution 2 km, pixel values ∈[0,1]); Product use: As the input of step 203 to participate in multi-source data fusion.

[0075] It should be noted that taking the visible light image (500-meter resolution) of the East China Sea region obtained by the VISSR sensor of Fengyun-4A satellite at 08:00 UTC on May 10, 2025 as an example: Perform 3-layer decomposition using Daubechies-4 wavelet basis function, and calculate the energy of each high-frequency component (horizontal, vertical, and diagonal directions) of each layer: The high-frequency energy of the 3rd layer of a pixel (coordinates E123°45', N30°15') at the edge of a certain fog area is 12.5, 14.3, and 11.7 W / m 2 ·sr·μm respectively, and the total is 38.5 W / m 2 ·sr·μm; The energy values are mapped to the interval [0, 1] through linear normalization. Assume that the maximum high-frequency energy of the whole image is 200 W / mW / m 2 ·sr·μm, and the normalized value of this pixel is 38.5 / 200 = 0.192, reflecting the edge sharpness feature; Infrared bright temperature difference correction and fog area probability mapping, using the data of the AHI sensor of the Himawari-8 satellite in the same period: The original bright temperature values in a certain sea area (E125°00', N31°20'): BT10.8μm = 285K, BT12.0μm = 283K, initial BTD = 2K; Combined with the ERA5 sea surface temperature data (SST = 282K < 285K), applying the correction coefficient α = 0.8, after correction ; Mapping the probability through the S-shaped function: when , the probability , indicating that there is a 78.5% probability of a fog area in this pixel; The 5km SST data of ERA5 is downscaled to a resolution of 2km through bilinear interpolation and aligned with the AHI infrared data. The original SST at an interpolation point (E124°30', N30°45') is 286K (5km grid point), and the interpolated 2km grid point SST = 285.3K, triggering the correction coefficient α = 1.2; The corrected bright temperature difference map is resampled to a resolution of 500 meters through the nearest neighbor method and spatially registered with the visible light feature map to ensure the subsequent fusion accuracy; In the visible light texture feature map, a continuous high-sharpness area (>0.7) is detected in the sea area around the Zhoushan Archipelago, and the coincidence rate with the visibility <1km area reported by the actual Automatic Identification System (AIS) of ships reaches 87%; The infrared fog area probability map shows a strip-shaped high-probability area (P > 0.8) outside the Yangtze River Estuary, and the spatial overlap rate with the area where the microwave retrieved liquid water path of the GCOM-W1 satellite in the same period > 0.2 kg / m 2 reaches 92%.

[0076] 202. Modeling of the thermodynamic phase change criterion, input data: the sea surface temperature measured by the ocean buoy and the air temperature 2 meters above the sea surface; the visible light texture feature map generated in step 201; Calculating the critical humidity threshold based on the thermodynamic equation, and combining with the edge sharpness quantization value in the visible light texture feature map to generate a thermodynamic phase change criterion matrix (resolution is 1 kilometer, 0 represents non-fog, 1 represents fog); Specifically, sub-step S2-1: Dynamically calculating the sea-air interface temperature difference, input data: the sea surface temperature (SST) measured by the ocean buoy in real time; the air temperature (Tair) 2 meters above the sea surface; calculating the sea-air temperature difference: ; Divide the thermodynamic state according to the positive or negative of ΔT (ΔT>0 indicates evaporation-dominated, ΔT≤0 indicates cooling-dominated); Output product: Sea-air temperature difference state table (including ΔT values and thermodynamic state marks at each buoy location); Product use: As the input of sub-step S2-2.

[0077] Sub-step S2-2: Critical humidity threshold calculation, Input data: The sea-air temperature difference state table generated in sub-step S2-1; Calculate the critical relative humidity threshold at different ΔT values: When ΔT>0, ; When ΔT≤0, ; Output product: Dynamic critical humidity threshold table (spatial resolution 1 km, generated by spatial interpolation of buoy data); Product use: As the input of sub-step S2-3.

[0078] Sub-step S2-3: Visible light edge sharpness fusion, Input data: The visible light texture feature map (including edge sharpness quantization values) generated in step 201; The above-generated dynamic critical humidity threshold table; Perform weighted superposition of the edge sharpness quantization value and the critical humidity threshold: If the edge sharpness ≥ 0.7, then the threshold is reduced by 5%; If the edge sharpness < 0.7, then the threshold remains unchanged; Output product: Thermodynamic criterion matrix with fused edge features (resolution 1 km, including the corrected critical humidity threshold); Product use: As the input of sub-step S2-4.

[0079] Sub-step S2-4: Binary fog area determination, Input data: The thermodynamic criterion matrix with fused edge features generated in the sub-step; The real-time relative humidity field provided by ERA5 reanalysis data (resolution 1 km); Compare the relative humidity with the corrected critical threshold pixel by pixel: When RH≥ , it is determined as a fog area (marked as 1); When RH< , it is determined as a non-fog area (marked as 0); Output product: Thermodynamic phase change criterion matrix (resolution 1 km, binary fog area distribution); Product use: As the input of step 203, participating in multi-source data fusion; It should be noted that the following is a specific implementation case of the thermodynamic phase change criterion modeling in the East China Sea region on May 10, 2025 (focusing on step 202): Input data: The measured SST of buoy A (E122.5°, N30.2°) in the Zhoushan sea area is 290K, and the sea surface air temperature T_air at 2 meters is 288K; Calculate (evaporation-dominated state), mark the thermodynamic state as "evaporation-dominated"; Output product: Generate a sea-air temperature difference status table containing 20 floating points, where the ΔT range is [-1K, +3K], and after spatial interpolation, a 1km resolution grid is formed; For the calculation of the dynamic critical humidity threshold, when ΔT = +2K at the grid point where buoy A is located, the formula is applied; When ΔT = -0.5K for the adjacent buoy B (E123.0°, N30.5°), ; Generate a 1km resolution threshold map through Kriging interpolation, and the threshold distribution in the Zhoushan Port area is 83.2 - 85.6%; For the fusion of visible light edge features, extract the pixel edge sharpness value of 0.75 at (E122.8°, N30.3°) in the visible light texture feature map; Since the sharpness ≥ 0.7, this point is adjusted downward by 5% from the interpolation reference value of 84.3% to 79.8%; The adjacent area (E122.9°, N30.4°) with an edge sharpness < 0.7 maintains the original threshold of 85.1%; For the binary fog area determination, call the ERA5 real-time relative humidity field, and the RH at the open sea outside Zhoushan Port (E122.7°, N30.25°) is 82%; After comparison and correction = 79.8%, satisfying RH ≥ condition, marked as a fog area (value 1); At a certain point in the open sea outside the Yangtze River Estuary (E123.2°, N31.0°), RH = 78% is lower than the threshold of 82.1%, marked as a non-fog area (value 0); Verified by AIS ship visibility reports, the accuracy rate of the criterion matrix in the Zhoushan Port area reaches 89%; the dynamic adjustment of the critical threshold increases the fog area recognition rate in the northern part of the East China Sea by 12% (compared with the static threshold method); the edge sharpness fusion mechanism successfully eliminates 3 misjudged areas caused by cloud shadows.

[0080] 203. Multi-source data fusion and fog area recognition, input data: The infrared fog area probability map generated in step 201; The thermodynamic phase change criterion matrix generated in step 202; Microwave radiometer data: The 36.5GHz vertically polarized brightness temperature data from the AMSR2 sensor of the GCOM-W1 satellite; Invert the liquid water path based on the microwave brightness temperature data, and perform logical fusion of the infrared fog area probability map, the thermodynamic phase change criterion matrix, and the liquid water path threshold to generate a multi-source fusion fog area recognition map (with a resolution of 1 kilometer, rasterized binary distribution, 0 represents non-fog, and 1 represents fog); Specifically, sub-step S3-1: preprocessing of microwave brightness temperature data, input data: 36.5 GHz vertically polarized brightness temperature data collected by the AMSR2 sensor of the GCOM-W1 satellite (original resolution 25 km); radiometric calibration and geolocation correction of the brightness temperature data, and removal of outliers affected by abnormal sea surface roughness; output product: corrected microwave brightness temperature map (resolution 25 km, valid data area after removal of outliers); product use: as input for sub-step S3-2.

[0081] Sub-step S3-2: Liquid water path physical inversion, input data: the corrected microwave brightness temperature map generated in sub-step S3-1; based on the microwave radiation transmission model, invert the liquid water path (LWP) and obtain the liquid water path distribution map (resolution 25 km, mark LWP ≥ 0.15 kg / m 2 product use: as input to sub-step S3-3.

[0082] Sub-step S3-3: Multi-source logical fusion operation, input data: infrared fog area probability map generated in step 201 (resolution 2 km); thermodynamic phase change criterion matrix generated in step 202 (resolution 1 km); liquid water path distribution map generated in sub-step S3-2 (resolution 25 km); resample the liquid water path distribution map to 1 km resolution through bilinear interpolation; perform the infrared fog area probability map (P ≥ 0.6), thermodynamic criterion matrix (value 1) and LWP ≥ 0.15 kg / m 2 Perform logical AND operation on the area; output product: preliminary screening and fusion fog area distribution map (resolution 1 km, binary marking of potential fog areas); product use: as input of sub-step S3-4.

[0083] Sub-step S3-4: Spatial consistency check, input data: the initial screening fusion fog area distribution map generated by sub-step S3-3; the visible light texture feature map (resolution 500 meters) generated in step 201; in the initial screening fog area, if the edge sharpness in the visible light texture feature map is ≥0.7, it is determined to be a reliable fog area; eliminate the false detection areas with edge sharpness <0.7, and generate a multi-source fusion fog area identification map (resolution 1 kilometer, rasterized binary distribution, 0 / 1 represents non-fog / fog); product use: as input to step 204, driving fluid mechanics diffusion prediction.

[0084] It should be noted that the following is a specific implementation case of multi-source data fusion and fog area identification in the East China Sea region on May 10, 2025 (focusing on step 203): Input data: 36.5 GHz vertically polarized brightness temperature data obtained from the AMSR2 sensor of the GCOM-W1 satellite (in the area of E125° - 128°, N30° - 32°), with an original resolution of 25 km. The original brightness temperature value of a certain pixel (E126.5°, N31.2°) is abnormal (TB = 280K, and the average value in the adjacent area is 160K ± 10K), which is determined to be an interference signal caused by abnormal sea surface roughness; After radiometric calibration, the brightness temperature data is spatially aligned with the ERA5 wind field data through geocorrection, and the outliers deviating more than 3σ from the standard deviation are removed. The effective data coverage rate after processing reaches 98.7%; Based on the microwave radiative transfer model. In a certain sea area (E127.3°, N30.8°), LWP = 0.75 kg / m 2 , exceeding the threshold of 0.15 kg / m 2 ; The generated liquid water path distribution map shows that there is a high-value area with LWP > 0.5 kg / m in the southeast of the Zhoushan Archipelago, and the spatial coincidence degree with the area where the measured visibility of the buoy is < 500 m reaches 85%; 2 The probability value P = 0.8 of a certain grid (E126° - 128°, N30.5° - 31.5°) in the infrared fog area probability map (with a resolution of 2 km) is increased to a resolution of 1 km through bicubic interpolation; The thermodynamic criterion matrix shows that 80% of the pixels in this area are marked as fog areas (value 1). After bilinear interpolation of the liquid water path distribution, 85% of the area has LWP ≥ 0.15 kg / m ; 2 ; After performing a logical AND operation, a preliminary screened fog area is generated: the area that meets the three conditions of P ≥ 0.6, thermodynamic criterion = 1, and LWP ≥ 0.15 accounts for 72% of the total area of this area; In the preliminary screened fog area (E127.1°, N31.0°), the edge sharpness value of the visible light texture feature map is extracted: the sharpness of the pixel points outside the Zhoushan Port is 0.75 (≥ 0.7), which is determined to be a reliable fog area; the sharpness of a certain area in the Yangtze Estuary is 0.6 (< 0.7), and combined with the AIS ship report, it is confirmed that the visibility in this area is > 1 km, so it is excluded; The final generated multi-source fusion fog area identification map has a resolution of 1 km. Compared with the vertical profile observation of the CALIPSO satellite, the fog area identification accuracy rate reaches 89%, and the false detection rate is reduced to 7%.

[0085] 204. Prediction of hydrodynamic diffusion trajectory, input data: the multi-source fusion fog area identification map generated in step 203; 850 hPa wind field data and boundary layer height data provided by numerical weather prediction; Calculate the horizontal diffusion rate and direction of the fog area based on the hydrodynamic equations, simulate the movement trajectory for the next 3 hours, and generate a fog belt diffusion vector field (including the spatio-temporal distribution of diffusion velocity and direction, with the velocity unit of m / s). Specifically, sub-step S4-1: Construction of the initial fog field. Input data: The multi-source fusion fog area identification map (resolution 1 km, binary fog area distribution) generated in step 203; Convert the fog area identification map into a rasterized concentration field (fog areas are marked as '1', non-fog areas are marked as '0') to generate the initial fog concentration field (resolution 1 km, rasterized concentration value 0 / 1); Product use: As the input for sub-step S4-2.

[0086] Sub-step S4-2: Dynamic correction of the wind field. Input data: 850 hPa wind field data (u / v components, time resolution 1 hour) provided by ECMWF numerical weather forecast; Boundary layer height data (derived from the vertical profile inversion of CALIPSO satellite); Adjust the wind field intensity according to the boundary layer height: For every 100 m decrease in the boundary layer height, the wind field speed decays by 5%; Generate the corrected wind field vector map (including horizontal wind speed and direction, resolution 1 km, wind speed unit m / s); Product use: As the input for sub-step S4-3.

[0087] Sub-step S4-3: Advection-diffusion trajectory simulation. Input data: The initial fog concentration field generated in sub-step S4-1; The corrected wind field vector map generated in sub-step S4-2; Based on the advection-diffusion equation, with a time step of 10 minutes, iteratively calculate the movement trajectory of the fog area with the wind field for the next 3 hours; Record the position of the fog area front at each time step to generate a fog area trajectory sequence (including the fog area boundary coordinates every 10 minutes); Product use: As the input for sub-step S4-4.

[0088] Sub-step S4-4: Vector field generation and verification. Input data: The fog area trajectory sequence generated in sub-step S4-3; Calculate the displacement of the fog area between adjacent time steps, extract the diffusion velocity and direction; Eliminate abnormal trajectory points caused by terrain occlusion to generate a fog belt diffusion vector field (resolution 1 km, including the magnitude of velocity and direction angle); Product use: As the input for step 205 to generate a comprehensive prediction and warning map.

[0089] It should be noted that the following is a specific implementation case of the fog area diffusion trajectory prediction in the East China Sea area on May 10, 2025 (focusing on step 204): Input data: The multi-source fusion fog area identification map generated in step 203 shows that there is a strip-shaped fog area with an area of about 12,000 km 2 in the central part of the East China Sea (E123.5°-126.8°, N29.8°-31.5°), with a resolution of 1 km. Convert this binary raster into a concentration field, mark the fog area as '1' (concentration value 100%), and non-fog areas as '0' to form the initial concentration field grid.

[0090] Key features: The concentration gradient at the edge of the fog area near the Zhoushan Archipelago (E122.8°, N30.3°) reaches 0.8 / km, reflecting a clear diffusion interface.

[0091] The initial wind speed at 850 hPa provided by ECMWF is 8 m / s (southeast wind), with a time resolution of 1 hour. Combining with the boundary layer height data retrieved by the CALIPSO satellite, the boundary layer height in the central East China Sea at 08:00 is 800 meters, which has decreased by 120 meters compared to the previous time period.

[0092] Corrected according to the rule of "attenuating 5% for every 100-meter decrease": , and the wind direction is adjusted to 15° east of south. A corrected wind field vector map with a resolution of 1 km is generated, and the wind speed gradient in the open sea outside Zhoushan is 7.2 - 7.8 m / s.

[0093] Based on the advection-diffusion equation, iterative calculations are performed with a time step of 10 minutes. At the initial moment, the front of the fog area is located at E125.2°, N31.0°. Driven by a wind speed of 7.6 m / s in the first 10 minutes, it displaces 4.56 km in the northwest direction.

[0094] After 18 iterations (3 hours), the boundary coordinate sequence of the fog area is recorded. At 08:50, the fog front reaches 30 km outside Zhoushan Port (E122.5°, N30.2°), and the horizontal diffusion rate reaches 0.85 m / s, with a coincidence degree of 91% with the moving trend of the liquid water path measured by the microwave radiometer.

[0095] Extract the displacement amounts at adjacent time steps: The average diffusion speed from 08:00 to 09:00 is 0.78 m / s (in the northwest direction of 22°). After terrain correction, 5 abnormal trajectory points around the Zhoushan Archipelago are removed (due to abnormal speeds > 2σ caused by island occlusion).

[0096] The output 1-km resolution vector field shows that the fog belt in the central East China Sea spreads fan-shaped. The area with the maximum diffusion speed is located at E124.5°, N30.8° (0.92 m / s), and the area with the minimum speed is at the mouth of Hangzhou Bay (0.65 m / s). Verified by the actual measurement of the Zhoushan VTS system, the positioning error of the fog area front at 09:00 is < 1.5 km.

[0097] 205. Comprehensive prediction and warning generation, input data: The multi-source fusion fog area identification map generated in step 203; the fog belt diffusion vector field generated in step 204; Overlay the multi-source fusion fog area identification map and the fog belt diffusion vector field to generate a time series prediction result; when the predicted fog area covers a port or waterway and the coverage probability ≥ 70%, trigger a graded warning signal and generate a comprehensive sea fog prediction and warning map (including geographical coordinates, impact time, and warning level); Specifically, sub-step S5-1: Spatiotemporal superposition and trajectory interpolation. Input data: The multi-source fusion fog area recognition map generated in step 203 (resolution 1 km, binary fog area distribution); the fog band diffusion vector field generated in step 204 (resolution 1 km, including velocity and direction). Align the fog area recognition map and the diffusion vector field spatiotemporally, and interpolate the fog area front positions at 10-minute intervals. Generate a time series fog area prediction raster set (including the fog area distribution every 10 minutes in the next 3 hours, time resolution 10 minutes, space resolution 1 km). Product use: As the input for sub-step S5-2.

[0098] Sub-step S5-2: Calculation of the coverage probability of key areas. Input data: The time series fog area prediction raster set generated in sub-step S5-1; the pre-set port and waterway geographic information database (including coordinate range and priority weight). For each port / waterway area, count the proportion of the predicted raster number covered by the fog area, and calculate the area coverage probability. When the coverage probability of the same area for three consecutive time steps (30 minutes) ≥ 70%, mark it as a high-risk area. Output product: The key area coverage probability table (including area name, coordinates, probability value, and risk level). Product use: As the input for sub-step S5-3.

[0099] Sub-step S5-3: Triggering of dynamic hierarchical alarms. Input data: The key area coverage probability table generated in sub-step S5-2. Set alarm thresholds according to the risk level: Yellow alarm: 70% ≤ probability < 85%; Orange alarm: 85% ≤ probability < 95%; Red alarm: probability ≥ 95%. Combine the area priority weight (international waterway weight × 1.2, local port × 1.0) to generate a hierarchical alarm instruction set (including alarm area, trigger time, level, and recommended measures). Product use: As the input for sub-step S5-4.

[0100] Sub-step S5-4: Synthesis of multi-modal alarm maps. Input data: The time series fog area prediction raster set generated in sub-step S5-1; the hierarchical alarm instruction set generated in sub-step S5-3. Overlay the prediction raster set and the alarm instructions on the electronic chart base map, and mark the time series of the fog area front and the alarm area. Generate a standardized format sea fog comprehensive prediction and alarm map (supporting GeoTIFF / Shapefile format, including geographic coordinates, impact time, alarm level, and visualization layer). Product use: As the final output result, transmitted to the shipping management terminal.

[0101] It should be noted that the following is a specific implementation case of the comprehensive prediction and alarm generation in the East China Sea region on May 10, 2025 (focusing on step 205): Input data: The multi-source fusion fog area recognition map shows that there is an area of about 1,200 km off the outer sea of Zhoushan Port 2The banded fog area (E122.5° - 123.8°, N29.8° - 30.5°), the diffusion vector field shows a southeast wind of 7.8 m / s and a diffusion rate of 0.85 m / s; After spatio-temporal alignment, the fog area boundary coordinates every 10 minutes for the next 3 hours are generated by bicubic interpolation. At 08:50, it is predicted that the fog front will reach E122.3°, N30.2°, and the distance to the main channel of Zhoushan Port is shortened to 18 km; The preset database contains the coordinate ranges of 6 key ports such as Zhoushan Port (E122.1°, N30.0°) and Yangshan Port (E122.1°, N30.6°); Statistics for the Zhoushan Port area (5×5 km grid): The coverage probabilities for 3 consecutive time steps during 08:50 - 09:20 are 72%, 78%, and 83% respectively; Apply priority weights: Zhoushan Port, as an international shipping hub (weight × 1.2), the calculated weighted coverage probability reaches 89.6%; According to the threshold setting: An orange alert is triggered at 09:10 (85% ≤ probability < 95%). Suggested measures include starting secondary navigation control and issuing hourly visibility bulletins; For the Yangshan Port area, the consecutive time step probabilities are 69%, 75%, and 82%. Since it does not meet the standard of ≥70% for 3 consecutive time steps, only a single-time yellow alert is triggered; Overlaid elements: 1 km resolution fog area prediction grid (GeoTIFF format), warning area polygon (Shapefile), port heat map layer; Visualization effect: A red warning circle (radius 15 km) is marked in the Zhoushan Port area, and an animation of the fog area expansion from 09:00 to 12:00 is shown on the timeline.

[0102] 206. Based on real-time observation for dynamic parameter calibration and feedback optimization, input data: Visibility observation data (updated at minute intervals) transmitted in real-time by ocean buoys; The sea fog comprehensive prediction and warning map generated in step 205; The multi-source fusion fog area identification map generated in step 203; Dynamic error detection: Compare the fog area range in the prediction and warning map with the visibility data measured by the buoy, and calculate the regional prediction error (error = predicted coverage rate - measured coverage rate); If the absolute value of the prediction error for 3 consecutive time steps (30 minutes) > 15%, then calibration is triggered: Adjust the critical humidity threshold of the thermodynamic phase change criterion in step S2 ( ±2%); Correct the wind field attenuation coefficient in step S4 (the attenuation rate per 100 meters of the boundary layer height is adjusted from 5% to 4% or 6%); Generate an optimized model parameter set (including the corrected critical humidity threshold and wind field attenuation coefficient); Product Use: Provide real-time feedback to Steps 202 and 204, replace the original fixed parameters, and drive the model operation in the next prediction cycle.

[0103] It should be noted that the following is a specific implementation case of dynamic parameter calibration and feedback optimization in the East China Sea region on May 10, 2025 (focusing on Step 206): Input Data: The coverage rate of the measured visibility less than 1 km from 08:30 to 09:00 at the outer buoy A of Zhoushan Port (E122.5°, N30.2°) is 65%, while the predicted coverage rate of this area shown in the predicted warning map generated in Step 205 is 85%; The absolute errors for three consecutive time steps (08:40 / 08:50 / 09:00) are 18%, 22%, and 19% respectively, all exceeding the 15% threshold; Thermodynamic Model Calibration: Critical humidity threshold It is adjusted down by 2% from the reference value of 84% to 82%. The humidity determination conditions in the surrounding area of Zhoushan Port (E122° - 123.5°, N29.8° - 30.5°) are relaxed, and the low cloud area that was originally misjudged as a fog area is corrected; Fluid Mechanics Correction: The CALIPSO satellite monitors that the boundary layer height drops from 800 meters to 650 meters. According to the rule of "attenuating 5% for every 100 - meter drop", the wind field attenuation coefficient is adjusted from 5% to 6%. The initial wind speed of 8 m / s in a certain sea area (E125.3°, N31.1°) is corrected to 8×(1 - 150 / 100×6%) = 7.28 m / s; Prediction from 09:10 to 09:40 after parameter update: The predicted coverage rate outside Zhoushan Port drops from 85% to 68%, and the error with the measured value of the buoy of 63% is reduced to 5%; After the wind field is corrected in the Yangshan Port area (E122.1°, N30.6°), the prediction error of the fog area diffusion speed drops from 0.25 m / s to 0.12 m / s, and the matching degree with the AIS ship's measured track is increased to 91%; After parameter adjustment, 3 originally misjudged low cloud areas (such as E123.8°, N30.7°) are excluded, and the excluded area reaches 120 km 2 . The positioning accuracy of the real fog area edge in the Yangtze Estuary is improved, and the proportion of reliable fog areas with a sharpness > 0.7 in the visible light texture map (500 meters) increases from 78% to 85%.

[0104] In the embodiment of the present invention, by integrating three - dimensional data of visible light texture features, infrared brightness temperature difference probability, and microwave liquid water path, the limitation of a single sensor is broken through. Through the Daubechies - 4 wavelet basis, 3 - layer decomposition of visible light images is realized, and the high - frequency energy is extracted to quantify the sharpness of the fog area edge, forming a spatial scale complementarity with the infrared fog area probability map and the microwave - inverted liquid water path; when the visible light texture sharpness ≥ 0.7, the thermodynamic threshold It is reduced by 5%, and 3 misjudged cloud shadow areas are successfully eliminated, increasing the recognition rate of the fog area in the northern East China Sea by 12%; for every 100-meter decrease in the boundary layer, the wind speed decays by 5%. After correction with the measured data of the CALIPSO satellite, the prediction error of the fog area diffusion speed is reduced from 0.25 m / s to 0.12 m / s; when the error between the measured visibility of the buoy and the predicted coverage rate is continuously greater than 15% for 30 minutes, the threshold value and the wind field attenuation coefficient are automatically adjusted, reducing the prediction error of Zhoushan Port from 22% to 5%; through multi-source data resampling and interpolation, a prediction result with a resolution of 1 km is finally output, with the accuracy improved by 5 times compared to traditional satellite products; the dynamic parameter calibration mechanism enables minute-level updates, supports the release of visibility reports per hour, and meets the real-time decision-making requirements of international waterways; standardized GeoTIFF / Shapefile format products are output, which can be directly overlaid on the electronic chart system, reducing the manual interpretation time by about 40%; after the orange / red alarm is triggered, it is recommended that the ship adjust its speed (when the fog area diffusion rate is 0.85 m / s, the recommended speed ≤ 10 knots), reducing fuel consumption by about 15%; the thermodynamic phase change criterion and the hydrodynamic diffusion equation are deeply coupled to achieve cross-scale simulation from the formation of microscopic fog droplets to the movement of macroscopic fog areas; integrating the data of Fengyun-4A satellite, Himawari-8 (infrared), and GCOM-W1 (microwave) satellites, combined with the vertical detection of the CALIPSO lidar, to construct a three-dimensional observation network, representing the current forefront of marine meteorological monitoring technology.

[0105] Under the condition that other conditions are the same, an embodiment three of the high-precision sea fog image recognition and prediction method based on ensemble learning includes, which is basically the same as Embodiment 1, except that: Based on the visible light satellite remote sensing image, improved DeepLabV3+ is used for fog area semantic segmentation to obtain a fog area probability map, and the final fog area report is obtained based on the comprehensive sea fog prediction and warning map and the fog area probability map.

[0106] Refer to Figure 5 , the image backbone architecture needs to construct an improved DeepLabV3+ model, whose backbone network adopts the EfficientNet-B5 architecture, which has efficient feature extraction ability and can improve the feature expression ability while reducing the computational amount. An adaptive atrous convolution module is added to the decoder layer, and the dynamic range of the atrous rate is set to (2, 6). By dynamically adjusting the atrous rate, the model can capture feature information of different scales. The weighted combination of DiceLoss and FocalLoss is used as the loss function, and the weight coefficient α = 0.7.

[0107] In the training process of the DeepLabV3+ model of the present invention, the parameters of the model are continuously adjusted to minimize the loss function until the Coefficient (Mean Intersection over Union) ≥ 0.85. Among them (MeanIntersectionoverUnion) is an index for predicting the accuracy of fog area pixels, and the calculation method is: ; Among them is the number of categories (fog / non-fog), is the number of correctly predicted fog area pixels.

[0108] It should be noted that in this embodiment, the visible light remote sensing image (resolution 500 meters) obtained by the FY-4A VISSR sensor is used as the input, and the improved DeepLabV3+ model is used for fog area semantic segmentation, and the final fog area report is generated by combining the thermodynamic criterion matrix, the infrared fog area probability map and the diffusion prediction result.

[0109] Improvement of the DeepLabV3+ model architecture and training, backbone network optimization: EfficientNet-B5 is used as the backbone network, and its compound scaling factor (depth 1.2, width 2.2, resolution 456) significantly improves the feature extraction ability.

[0110] Adaptive atrous convolution module: An ASPP module (atrous rate range 2-6) for dynamically adjusting the atrous rate is embedded in the decoder. The optimal atrous rate is automatically selected according to the spatial distribution of the input feature map through learnable parameters. A small atrous rate (r = 2) is used to capture details at the fog area edge (such as the coastline), and a large atrous rate (r = 6) is used to extract global features in the large area fog region.

[0111] Loss function design: The combination of DiceLoss (weight 0.3) + FocalLoss (weight 0.7) is adopted to solve the problem of unbalanced number of fog area and non-fog area pixels. γ of FocalLoss = 2, focusing on difficult-to-classify pixels (the area where thin fog is confused with low clouds).

[0112] Training data: 10,000 labeled fog area images (resolution 500 meters) are used, including 8000 in the training set and 2000 in the validation set. Data augmentation includes random rotation (±30°), horizontal flipping and Gaussian noise injection (σ = 0.05).

[0113] Training result: After the model converges, the test set reaches 0.87, among which the fog area is 0.89, the non-fog area is 0.85, and the predicted fog area probability of a certain sample (coordinate 120.5°E, 32°N) is 0.92, and the verification accuracy of the measured visibility reaches 91%.

[0114] Generation of fog area probability map and multi-source fusion, input and processing flow: Visible light image preprocessing: The input image is normalized to [0, 1], and the contrast is enhanced by histogram equalization.

[0115] The model outputs a fog area probability map (resolution 500 meters, pixel values 0 - 1). For example, the fog area probability in a certain sea area (coordinates 121°E, 35°N) is 0.88, indicating an 88% probability of sea fog in this area.

[0116] Multi - source data fusion rules: Spatial alignment: The fog area probability map, thermodynamic criterion matrix (1 km), and infrared fog area probability map (1 km interpolation) are unified to a 1 - km grid, and bilinear interpolation is used to ensure spatial consistency.

[0117] Logical fusion conditions: Fog area probability map threshold: ≥0.7 (if lower than this value, it is judged as non - fog even if other conditions are met).

[0118] Thermodynamic criterion matrix: Must be 1 (meeting the humidity and temperature conditions).

[0119] Infrared fog area probability map: ≥0.6 (probability ≥60%).

[0120] Example: The parameters of a certain grid (121.2°E, 35.5°N) are: Fog area probability = 0.88, thermodynamic criterion = 1, infrared probability = 0.75 → The fusion result is a fog area.

[0121] If the infrared probability = 0.55 (not meeting the threshold), it is judged as a non - fog area.

[0122] Final fog area report generation, spatio - temporal weighting and alarm triggering: Time - series prediction superposition: The predicted fog area position in 3 hours by the diffusion vector field (the predicted coverage probability of the Qingdao Port waterway is 75% at T + 3 hours) is superposed with the current fusion result to generate a spatio - temporal probability distribution map.

[0123] Alarm rules: Red alarm: Coverage probability ≥90% and impact time ≥6 hours (such as persistent thick fog in a certain waterway in the Yellow Sea).

[0124] Orange alarm: Coverage probability ≥70% and impact time 3 - 6 hours (Example: Qingdao Port waterway, probability 75%, triggering an orange alarm).

[0125] Output format: Geographical range: 120.8°E - 121.6°E, 36.5°N - 37.1°N; Impact time: From 08:00 to 11:00 on May 20, 2025; Alarm level: Orange (Level 2); Suggested measures: Restrict the entry and exit of ships and activate the fog navigation system.

[0126] The above describes the high-precision sea fog image recognition and prediction method based on ensemble learning in the embodiments of the present invention. Next, the high-precision sea fog image recognition and prediction device based on ensemble learning in the embodiments of the present invention will be described. Please refer to Figure 3 One embodiment of the high-precision sea fog image recognition and prediction device based on ensemble learning in the embodiments of the present invention includes: an acquisition module 301, configured to acquire visible light satellite remote sensing images and infrared band data, perform multi-scale texture analysis on the visible light satellite remote sensing images, extract the texture features of the fog area edge, generate a visible light texture feature map, calculate the brightness temperature difference for the infrared band data, and generate an infrared fog area probability map in combination with the sea surface temperature; a processing module 302, configured to acquire the sea surface temperature and the air temperature data above the sea surface measured by an ocean buoy, obtain a critical humidity threshold, and generate a thermodynamic phase change criterion matrix in combination with the edge sharpness quantization value in the visible light texture feature map; an identification module 303, configured to acquire microwave radiometer data, invert the liquid water path based on the microwave brightness temperature data, and perform logical fusion based on the infrared fog area probability map, the thermodynamic phase change criterion matrix, and the liquid water path threshold to generate a multi-source fusion fog area identification map; a simulation module 304, configured to obtain the horizontal diffusion rate and direction of the fog area based on the initial fog area distribution provided by the multi-source fusion fog area identification map, combine the wind field data and the boundary layer height data, simulate the future movement trajectory, and generate a fog band diffusion vector field; an allocation module 305, configured to superimpose the multi-source fusion fog area identification map and the fog band diffusion vector field to generate a time series prediction result. When it is predicted that the fog area covers a port or a waterway and the coverage probability ≥ a preset value, a hierarchical alarm signal is triggered, and a comprehensive sea fog prediction alarm map is generated.

[0127] In the embodiments of the present invention, the thermodynamic phase change conditions are converted into matrix operations, avoiding the discretization error of traditional threshold comparison and improving the real-time processing efficiency; the inversion of the microwave liquid water path can capture the tiny water droplets that are difficult to identify by visible light / infrared, improving the recognition accuracy of low visibility fog areas; the simulation based on the diffusion vector field can realize the prediction of the fog area trajectory in the next 6-12 hours, winning valuable time for waterway control; through the superposition of the vector fields, the refined management requirements at the port scale are met.

[0128] Above Figure 3 The high-precision sea fog image recognition and prediction device based on ensemble learning in the embodiments of the present invention is described in detail from the perspective of modular functional entities. Next, the high-precision sea fog image recognition and prediction device based on ensemble learning in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0129] Figure 4It is a schematic structural diagram of a high-precision sea fog image recognition and prediction device based on ensemble learning provided by an embodiment of the present invention. The high-precision sea fog image recognition and prediction device 400 based on ensemble learning may vary greatly due to different configurations or performances. The device 400 includes a transmitter 401, a receiver 402, and a processor 403. Among them, the processor 403 can also be a controller. Figure 4 It is represented as "controller / processor 403" in the figure. Optionally, the device 400 may further include a modulation and demodulation processor 405. Among them, the modulation and demodulation processor 405 may include an encoder 406, a modulator 407, a decoder 408, and a demodulator 409.

[0130] In one example, the transmitter 401 adjusts (for example, analog conversion, filtering, amplification, and upconversion, etc.) the output sample and generates an uplink signal, and the uplink signal is transmitted to the access network device via the antenna. On the downlink, the antenna receives the downlink signal transmitted by the access network device. The receiver 402 adjusts (for example, filtering, amplification, downconversion, and digitization, etc.) the signal received from the antenna and provides an input sample. In the modulation and demodulation processor 405, the encoder 406 receives the service data and signaling messages to be transmitted on the uplink, and processes (for example, formats, encodes, and interleaves) the service data and signaling messages. The modulator 407 further processes (for example, symbol mapping and modulation) the encoded service data and signaling messages and provides an output sample. The demodulator 409 processes (for example, demodulates) the input sample and provides symbol estimation. The decoder 408 processes (for example, deinterleaves and decodes) the symbol estimation and provides the decoded data and signaling messages sent to the device 400. The encoder 406, the modulator 407, the demodulator 409, and the decoder 408 can be implemented by the integrated modulation and demodulation processor 405. These units process according to the radio access technology adopted by the radio access network (for example, the access technology of LTE and other evolved systems). It should be noted that when the device 400 does not include the modulation and demodulation processor 405, the above functions of the modulation and demodulation processor 405 can also be completed by the processor 403.

[0131] The processor 403 controls and manages the actions of the device 400, and is used to execute the processing procedures performed by the device 400 in the above embodiments of the present disclosure. For example, the processor 403 is further used to execute each step of the transmitting device or the receiving device in the above method embodiments, and / or other steps of the technical solutions described in the embodiments of the present disclosure.

[0132] Furthermore, the device 400 may further include a memory 404, and the memory 404 is used to store program codes and data for the device 400.

[0133] It can be understood that Figure 4Only a simplified design of the device 400 is shown. In practical applications, the device 400 may include any number of transmitters, receivers, processors, modem processors, memories, etc., and all devices that can implement the embodiments of the present disclosure are within the protection scope of the embodiments of the present disclosure.

[0134] The present invention also provides a high-precision sea fog image recognition and prediction device based on integrated learning. The high-precision sea fog image recognition and prediction device based on integrated learning includes a memory and a processor. Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, the processor is caused to execute the steps of the high-precision sea fog image recognition and prediction method in the above respective embodiments.

[0135] The present invention also provides a computer-readable storage medium. The computer-readable storage medium may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are run on a computer, the computer is caused to execute the steps of the high-precision sea fog image recognition and prediction method.

[0136] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0137] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0138] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-precision sea fog image recognition and prediction method based on ensemble learning, characterized in that, The high-precision sea fog image recognition and prediction method based on ensemble learning includes: Obtain visible light satellite remote sensing images and infrared band data, perform multi-scale texture analysis based on the visible light satellite remote sensing images, extract texture features of the fog area edge, generate a visible light texture feature map, calculate the brightness temperature difference based on the infrared band data, and generate an infrared fog area probability map in combination with the sea surface temperature; Obtaining sea surface temperature and air temperature data measured by ocean buoys to obtain a critical humidity threshold, and combining this with the edge sharpness quantification value in the visible light texture feature map to generate a thermodynamic phase change criterion matrix; Acquire microwave radiometer data, invert the liquid water path based on the microwave brightness temperature data, perform logical fusion based on the infrared fog area probability map, the thermodynamic phase change criterion matrix and the liquid water path threshold, and generate a multi-source fusion fog area identification map; Based on the initial distribution of the fog area provided by the multi-source fusion fog area identification map, combined with wind field data and boundary layer height data, the horizontal diffusion rate and direction of the fog area are obtained, the future movement trajectory is simulated, and the fog belt diffusion vector field is generated; The multi-source fusion fog area identification map and the fog band diffusion vector field are superimposed to generate a time series prediction result. When the fog area is predicted to cover the port or waterway and the coverage probability is ≥ the preset value, a graded alarm signal is triggered to generate a comprehensive sea fog prediction alarm map.

2. The high-precision sea fog image recognition and prediction method based on ensemble learning according to claim 1, wherein include: Perform two-dimensional discrete wavelet transform on visible light satellite remote sensing images, decompose them to the third layer, extract the energy value of high-frequency components, and generate wavelet high-frequency energy maps; The sum of high-frequency energy of each pixel is obtained based on the wavelet high-frequency energy map to generate a visible light texture feature map; The brightness temperature difference (BTD) is obtained based on the channel brightness temperature data and the sea surface temperature (SST). The BTD is piecewise linearly corrected according to the SST to generate a corrected brightness temperature difference map. The brightness temperature difference (BTD) is mapped to the probability of fog presence through the S-type function to generate an infrared fog area probability map.

3. The high-precision sea fog image recognition and prediction method based on ensemble learning according to claim 2, characterized in that, include: Based on the sea surface temperature SST measured in real time by ocean buoys and the air temperature 2 meters above the sea surface , the sea-air temperature difference is obtained: ； The thermodynamic state is divided according to the positive and negative nature of ΔT, and the sea-temperature difference state table is obtained; According to the sea-temperature difference state table, the critical relative humidity threshold under different ΔT is obtained: When ΔT > 0, ; When ΔT ≤ 0, ; Get the dynamic critical humidity threshold table; According to the visible light texture feature map and the dynamic critical humidity threshold table, the edge sharpness quantization value and the critical relative humidity threshold are weighted superimposed: If the edge sharpness ≥ 0.7, then Reduce by 5%; If the edge sharpness < 0.7, then Maintain the original value; Obtain the thermodynamic criterion matrix of fused edge features; Based on the thermodynamic criterion matrix integrating edge features and the real-time relative humidity field, the real-time relative humidity RH is compared pixel by pixel with the corrected critical relative humidity threshold : When RH ≥ it is determined as a fog area; When RH < , it is determined as a non-fog area; The thermodynamic phase transition criterion matrix is obtained.

4. The high-precision sea fog image recognition and prediction method based on ensemble learning according to claim 3, wherein include: The brightness temperature data were radiometrically calibrated and geo-positioned to remove outliers affected by sea surface roughness anomalies, thus obtaining a corrected microwave brightness temperature map. According to the corrected microwave brightness temperature image and the microwave radiation transmission model, the liquid water path is inverted to obtain the liquid water path distribution map; The liquid water path distribution map was resampled to 1 km resolution through bilinear interpolation and processed according to the infrared fog area probability map, thermodynamic criterion matrix and liquid water path distribution map to obtain the initial screening fusion fog area distribution map; In the initial screening fog area, if the edge sharpness in the visible light texture feature map is ≥0.7, it is determined to be a reliable fog area. The false detection areas with edge sharpness <0.7 are eliminated to generate a multi-source fusion fog area identification map.

5. The high-precision sea fog image recognition and prediction method based on ensemble learning according to claim 4, characterized in that include: The multi-source fusion fog area identification map is converted into a rasterized concentration field to generate the initial concentration field of the fog area. The wind field intensity is adjusted according to the boundary layer height to generate a corrected wind field vector map. Based on the initial concentration field of the fog area and the corrected wind field vector diagram, the movement trajectory of the fog area along the wind field in the next 3 hours is iteratively calculated with a time step of 10 minutes. The front position of the fog area at each time step is recorded to generate a fog area trajectory sequence. The fog area displacement at adjacent time steps is obtained based on the fog area trajectory sequence, the diffusion speed and direction are extracted, the abnormal trajectory points caused by terrain occlusion are eliminated, and the fog belt diffusion vector field is generated.

6. The high-precision sea fog image recognition and prediction method based on ensemble learning according to claim 5, characterized in that include: The multi-source fusion fog area identification map is temporally and spatially aligned with the fog band diffusion vector field, and the fog area front position is interpolated at 10-minute intervals to generate a time series fog area prediction grid set. Based on the time series fog area prediction grid set and the preset port and waterway geographic information database, for each port / waterway area, the proportion of the predicted grids covered by fog is counted to obtain the regional coverage probability. When the coverage probability of the same area for three consecutive time steps is ≥70%, it is marked as a high-risk area, and a key area coverage probability table is obtained; Based on the key area coverage probability table, set the alarm threshold according to the risk level, and generate a hierarchical alarm instruction set based on the regional priority weight; The time series fog area prediction grid set and the graded alarm instruction set are superimposed on the electronic nautical chart base map, the fog area front time series and the alarm area are marked, and a comprehensive sea fog prediction and alarm map is generated.

7. The high-precision sea fog image recognition and prediction method based on ensemble learning according to claim 6, characterized in that The following steps are also included: Compare the fog area range in the comprehensive sea fog forecast warning map with the visibility observation data transmitted back by ocean buoys in real time to obtain the regional forecast error, which is calculated as follows: regional forecast error = predicted coverage rate - measured coverage rate; If the absolute value of the prediction error for three consecutive time steps is greater than 15%, calibration is triggered: Adjust the critical humidity threshold of the thermodynamic phase transition criterion; Corrected wind field attenuation coefficient; Generate optimized model parameter set.

8. The high-precision sea fog image recognition and prediction method based on ensemble learning according to claim 1, characterized in that, The improved DeepLabV3+ is used to perform semantic segmentation of fog areas based on visible light satellite remote sensing images to obtain a fog area probability map, and the final fog area report is obtained based on the sea fog comprehensive forecast warning map and the fog area probability map.

9. The high-precision sea fog image recognition and prediction method based on ensemble learning according to claim 8, characterized in that The improved DeepLabV3+ model includes: the backbone network adopts the EfficientNet-B5 architecture; the decoder layer adds an adaptive hole convolution module, and the dynamic range of the hole rate is set to (2, 6); the loss function is a weighted combination of Dice Loss and Focal Loss, with a weight coefficient α=0.7.

Images