A method for visibility analysis based on satellite observations and sequential images

Abstract

This invention proposes a visibility analysis method based on satellite observation and continuous imagery. It acquires continuous image data from fixed cameras, hourly visibility observation data from ground monitoring stations, and visibility data retrieved from satellites. Based on these data, it performs multi-source data dynamic spatial fusion, using appropriate interpolation algorithms according to different data sources and dynamically adjusting the contribution weight of each data source based on data distribution density to ensure the model has accurate spatial prediction capabilities. The model is then built, trained, and fine-tuned to ultimately achieve visibility recognition from continuous images. In the model application stage, it no longer relies on ground monitoring station data and satellite data, but instead utilizes real-time camera images for visibility recognition. This method, by introducing a time-weighted mechanism, can fully utilize camera images from consecutive moments, improving the stability and accuracy of visibility prediction and avoiding the significant impact of noise or outliers at a single moment on the results.

Description

Technical Field

[0001] This invention relates to the field of atmospheric sounding technology, specifically a visibility analysis method based on satellite observation and continuous imagery. Background Technology

[0002] Visibility refers to the maximum distance at which a person with normal vision can identify a target in a specific background environment. Among many influencing factors, various weather phenomena such as fog, haze, sand, smoke, dust, snow, and rain have a significant impact on visibility. In particular, localized weather phenomena, such as sandstorms, dense fog, and storms, can easily cause a sharp decrease in visibility, leading to low visibility conditions.

[0003] Visibility plays a crucial role in urban air pollution control. In cities, good visibility helps people promptly detect the level of air pollution and take appropriate preventative measures. It is also closely related to public transportation safety. Whether it's vehicles on roads or urban rail transit, sufficient visibility is essential for safe operation. In maritime ports and waterways, traffic safety is even more critically linked to visibility. In low visibility conditions, ship navigation faces significant risks, potentially leading to serious collisions. Low visibility can easily trigger serious safety accidents, posing a significant threat to people's lives and property. Once an accident occurs, it not only causes casualties and heavy losses to families but also inflicts substantial damage on the socio-economic system. Therefore, accurate classification and refined monitoring of atmospheric visibility are of paramount importance.

[0004] Currently, in my country's meteorological departments, the mainstream methods for visibility observation and identification are mainly divided into manual observation and instrumental measurement. Manual observation involves observers determining visibility values ​​by visually assessing the clarity of objects at different distances at several specific times of day. The visibility observed manually here usually refers to effective horizontal visibility, which is the maximum horizontal distance at which objects can be seen within more than half of the surrounding field of vision. However, manual observation has several drawbacks. First, the number of observations is limited, making it impossible to monitor changes in visibility in real time. Second, the observation results are greatly influenced by the observer's subjectivity; different observers may arrive at different results. Finally, the accuracy is relatively low, making it difficult to meet the needs of modern meteorological observation.

[0005] Instrumental methods of visibility measurement utilize optical equipment to measure specific optical properties of the atmosphere, such as extinction coefficient and transmittance, and then calculate the visibility value. Common instruments include forward scattering visibility meters and single-end projection portable measurement systems. Visibility measurements obtained using instrumental methods are relatively accurate. These instruments can accurately measure the optical properties of the atmosphere, thereby calculating a relatively accurate visibility value. However, such instruments are expensive to design and manufacture, which can pose a significant financial burden for some meteorological departments. Furthermore, these instruments require specialized installation and calibration, demanding not only professional technicians but also considerable time and effort. In addition to their high cost, they also lack flexibility; once installed in a fixed location, they are difficult to move or adjust.

[0006] In addition, recent studies have attempted to apply machine learning methods to image-based visibility inversion. This method uses computer technology to analyze and process a large number of images to derive visibility values. However, this method still has some problems. For example, the quality of image acquisition is unstable and may be affected by factors such as lighting and weather. The image data is not synchronized with environmental monitoring data, which affects the accuracy of the inversion results. Under low visibility conditions, the recognition effect is also unsatisfactory, making it difficult to accurately determine the visibility value. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention proposes a visibility analysis method based on satellite observation and continuous imagery. By constructing a visibility analysis system with strong generalization capabilities based on visual and satellite observation technologies, high-precision visibility observation can be achieved.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A visibility analysis method based on satellite observation and continuous imagery, characterized by the following steps:

[0010] S1. Data Preparation:

[0011] A system database was established to collect continuous image data from fixed cameras, hourly visibility observation data from ground monitoring stations, and visibility data retrieved from satellites.

[0012] S11, Continuous image data from a fixed camera:

[0013] Continuous image data from fixed cameras is provided by fixed cameras, which in turn come from traffic cameras, urban security cameras, or other dedicated cameras deployed by meteorological departments. The cameras continuously acquire images at fixed intervals to form a time-series image dataset.

[0014] Data preprocessing of time-series image datasets:

[0015] S111, Uniform image size:

[0016] All images are scaled to a uniform size, and the image size is uniformly calculated using either bilinear interpolation or cubic spline interpolation.

[0017] I resize =resize(I original ,800,600)

[0018] Among them I resize For the processed image, I original Original image;

[0019] S112, Color Channel Standardization:

[0020] To accommodate the input requirements of the neural network, the three color channels of the image (RGB) need to be normalized separately. The normalization formula is as follows:

[0021]

[0022] Where I(x, y, c) is the value of the color channel c of pixel (x, y), μ c and μ c These are the mean and standard deviation of color channel c, respectively.

[0023] S113, Noise Reduction and Enhancement:

[0024] Median filtering or bilateral filtering can be used to denoise the image.

[0025] I denoise =medianFilter(I norm k)

[0026] Where k is the size of the filtering window.

[0027] Image enhancement is achieved through contrast stretching and brightness adjustment.

[0028] S114, Data Augmentation:

[0029] Image data augmentation is performed using random rotation, cropping, and flipping techniques:

[0030] I aug =randomAugment(I denoise )

[0031] The data-augmented images will be used together with the original images for training the visibility regression model.

[0032] S115, Time Alignment:

[0033] Time alignment is performed using an interpolation algorithm to ensure that the timestamps of the images are consistent with the hourly visibility observation data from ground monitoring stations and the visibility data retrieved from satellites. If a camera image is missing at a certain moment, it can be filled by interpolating images from adjacent moments.

[0034]

[0035] This ensures that continuous image data is available for subsequent analysis at all points in time;

[0036] S12, Hourly visibility observation data from ground monitoring stations:

[0037] Hourly visibility data from ground monitoring stations is provided by the meteorological department. The meteorological department monitors visibility through ground monitoring stations in various cities. The data from these stations includes hourly visibility data and meteorological data, with the meteorological data used to assist in the analysis of visibility changes.

[0038] Data preprocessing was performed on the hourly visibility observation data:

[0039] S121, Time Alignment:

[0040] The hourly visibility observation data was interpolated over time to ensure consistency with the camera image data collected by the fixed camera.

[0041]

[0042] Where V(t) is the visibility value at time t, and t0 and t1 are two adjacent integer times.

[0043] S122, Missing Value Handling:

[0044] If ground monitoring station data is missing at a certain moment, it can be supplemented by interpolation or by using interpolation methods based on historical data;

[0045] S13, Satellite inversion visibility data:

[0046] Visibility data retrieved via satellite is acquired through polar-orbiting or geostationary satellites, with resolution depending on the respective satellite system.

[0047] S131. Remove outliers:

[0048] In the preprocessing stage, outliers in the satellite inversion visibility data are removed, and error data is detected through upper and lower limit constraints and statistical methods.

[0049] S132, Time Alignment:

[0050] Linear interpolation is used to fill the temporal gaps in the satellite-retrieved visibility data, ensuring that camera image data, hourly visibility observation data from ground monitoring stations, and satellite-retrieved visibility data correspond to each other at the same time. S14. All processed data will be stored in the system database, categorized and managed by region and time.

[0051] Data management includes: the storage of raw data and the storage of processed data.

[0052] Each processed camera image, hourly visibility observation, and satellite-retrieved visibility data is timestamped and spatially labeled to ensure consistency with the input of the visibility regression model.

[0053] S2. Dynamic spatial fusion of multi-source data:

[0054] By introducing a mesh model, spatial fusion and weight adjustment are performed on three types of data: camera image data, hourly visibility observation data, and satellite-retrieved visibility data, as detailed below:

[0055] Appropriate interpolation algorithms are used based on different data sources, and the contribution weight of each data source is dynamically adjusted based on the data distribution density to ensure that the visibility regression model has accurate predictive capabilities in space.

[0056] Based on the highest resolution grid of satellite visibility data, the target area is geospatially gridded. The grid division is based on satellite resolution, with the grid where the fixed camera is located as the core grid, and then eight surrounding grids are extended outward to form a 3x3 nine-square grid structure.

[0057] Of these nine grids, the core grid integrates data from fixed cameras, ground monitoring stations, and satellite systems, while the surrounding grids are used for interpolation of satellite and ground station data.

[0058] The spatial data fusion method is adjusted according to the distribution of ground monitoring stations within the grid, and is divided into four cases:

[0059] Scenario 1: No site

[0060] There are no ground monitoring stations within the core grid and the eight surrounding grids. Due to the lack of ground monitoring station data, the visibility of the core grid relies entirely on the satellite system. Its calculation process does not require interpolation algorithms; the visibility value is determined solely based on the observation data from the satellite system. The data weight allocation is as follows:

[0061] Satellite data weights within the core grid: ω sat =1

[0062] Ground station data weight: ω site =0

[0063] Core grid visibility algorithm formula:

[0064] V core =V sat (core)

[0065] Among them, V core V represents the visibility of the core grid after data fusion. sat (core) represents satellite observation visibility data within the core grid.

[0066] Scenario 2: Clustered Sites

[0067] If there is at least one ground monitoring station within the core grid, but no stations within the surrounding eight grids, then it is necessary to combine station data from the core grid with satellite data.

[0068] If there is only one ground monitoring station in the core grid, then the data from that ground monitoring station is used directly as the station visibility V for the core grid. site (core),

[0069] If the core grid has more than one ground monitoring station, interpolation between the ground monitoring stations is required:

[0070]

[0071] Where V(x) i ) represents the visibility observation value at station i, d(x0, x) i ) is the distance between camera position x0 and station i, and p is the distance attenuation factor.

[0072] Data weight allocation:

[0073] Satellite data weights within the core grid: ω sat =0.5

[0074] Weight of ground monitoring station data within the core grid: ω site =0.5

[0075] Taking into account both satellite and station data for the core grid, the visibility calculation formula for the core grid is as follows:

[0076] V core =ω site ×V site (core)+ω sat ×V sat (core)

[0077] Scenario 3: Sparse sites

[0078] There are no ground monitoring stations within the core grid, but at least one ground monitoring station exists in one of the eight surrounding grids. In this case, it is necessary to interpolate the visibility of the core grid using data from the surrounding ground monitoring stations.

[0079] If there is only one ground monitoring station within the surrounding grid, then the data from that ground monitoring station is directly used as the visibility V for the surrounding grid's ground monitoring stations. site (out),

[0080] If there are multiple ground monitoring stations within the surrounding grid, use Kriging interpolation to interpolate the data from the ground monitoring stations in the surrounding grid into the core grid:

[0081]

[0082] Where V(x) i ) represents the visibility observation value of the surrounding ground monitoring station i, and λ represents the visibility value. i These are interpolation weights, determined based on geospatial correlation, satisfying...

[0083] Data weight allocation:

[0084] Satellite data weights for the core grid: ω sat =0.8

[0085] Ground monitoring station data weight: ω site =0.2

[0086] The interpolated ground monitoring station data is combined with the satellite data of the core grid. The visibility calculation formula for the core grid is as follows:

[0087] V core =ω site ×V site (out)+ω sat ×V sat (core)

[0088] Scenario 4: Multiple sites

[0089] Ground monitoring stations are located in the core grid and the eight surrounding grids. Therefore, the ground monitoring station data in the core grid and the surrounding grids need to be processed separately. The ground monitoring station data in the core grid is interpolated using the inverse distance weighting method, while the ground monitoring station data in the surrounding grids is interpolated to the core grid using Kriging interpolation. The interpolation algorithm for the ground monitoring station data in the core grid is as follows:

[0090]

[0091] The interpolation algorithm for the ground monitoring stations in the surrounding grid is as follows:

[0092]

[0093] Data weight allocation:

[0094] Weight of ground monitoring station data in the core grid: ω site (core) = 0.6

[0095] Satellite data weights for the core grid: ω sat =0.3

[0096] Weight of ground monitoring station data in the surrounding grid: ω site (out) = 0.1

[0097] Combining interpolated data from ground monitoring stations in the core and surrounding grids, as well as satellite data from the core grid, the visibility calculation formula for the core grid is as follows:

[0098] V core =ω site (core)×V site (core)+ω sat ×V sat (core)+ω site (out)×V site (out);

[0099] S3. Visibility Regression Model Training:

[0100] S31. Visual image feature extraction:

[0101] The backbone network is used to extract features from visual images captured by fixed cameras to obtain high-dimensional image features. The backbone network is also used to extract features from satellite data acquired by the satellite system to obtain high-dimensional satellite data features. The satellite data is then standardized to obtain satellite values.

[0102] S32. Establish a fully connected regression model:

[0103] A fully connected neural network regression model is established by combining the high-dimensional features of visual images and the high-dimensional features or satellite values ​​extracted from satellite data with visibility label data.

[0104] S33. Model Training: Use multi-region training samples over a long time scale to train the model;

[0105] S34. Model Fine-tuning: In model training, the model trained using samples from multiple locations and over long time scales needs to be fine-tuned for real-world use. In model application, a model fine-tuning function is set up to enable fine-tuning of the model.

[0106] S4. Visibility Recognition:

[0107] In the application phase of the visibility regression model, instead of relying on ground monitoring station data and satellite-derived visibility data, visibility identification is performed using real-time images from fixed cameras.

[0108] By introducing a time-weighted mechanism, real-time images from fixed cameras at continuous intervals can be fully utilized.

[0109] For a target time t, image data from 2n+1 preceding and following times are extracted from the image sequence captured by a fixed camera, and visibility is identified at each time.

[0110] Then, based on the time difference between the image capture time and the target time, different weights are assigned to each time point;

[0111] Finally, the visibility value at time t is obtained through a weighted calculation method.

[0112]

[0113] Where V(t+i) is the visibility estimate at time t+i; ω i The weights are inversely proportional to the time difference |i|; the smaller the time difference, the larger the weight.

[0114] Weight Explanation: The weight ω0 is maximized at the current time t, ensuring that the visibility regression model is dominated by the observations at the current time; data from previous and subsequent times are used as auxiliary data, with weights ω0 and ω0 respectively. -2 ω -1 ω1 and ω2 gradually decrease as the time distance increases.

[0115] As a preferred technical solution of the present invention: in step S11, the data acquisition interval of the fixed camera is every 5 minutes or 10 minutes.

[0116] As a preferred technical solution of the present invention: in step S12, the monitoring station data also includes meteorological data, which is used to assist in the analysis of visibility changes.

[0117] As a preferred technical solution of the present invention: in step S13, the resolution range of the satellite inversion visibility data is from 100 meters to 9,000 meters.

[0118] As a preferred technical solution of the present invention: in step S33, data enhancement is achieved by scaling and transposing the visual image and satellite data.

[0119] As a preferred technical solution of the present invention: In step S34, the model fine-tuning function uses manual visual inspection or a mobile visibility monitoring device to measure the visibility value of the model application location, accumulates a certain amount of data to fine-tune the basic model, and optimizes the model effect.

[0120] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0121] 1. This invention provides higher precision visibility observation: By constructing a visibility recognition system based on satellite and visual images, and combining different observation methods with artificial intelligence, the accuracy of visibility observation can be improved.

[0122] 2. This invention offers broader visibility observation capabilities: Automated visibility observation is no longer limited to expensive observation equipment with specific construction requirements. Satellite data provides global coverage, and combined with numerous camera resources, visibility observation can be conducted more extensively.

[0123] 3. This invention enhances forecasting capabilities: For weather forecasting technology, better reproducing the current state of the atmosphere is a key factor in improving the accuracy of predicting future weather patterns. High-precision, high-density visibility observations can better help reproduce the current state of the atmosphere, thereby improving weather forecasting capabilities. Attached Figure Description

[0124] Figure 1 This is a schematic diagram of the structure of the present invention;

[0125] Figure 2 This is a schematic diagram of a neural network regression model;

[0126] Figure 3 These are real-time weather images;

[0127] Figure 4 The images were obtained by satellite. Detailed Implementation

[0128] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0129] like Figure 1-2 As shown, this invention proposes a visibility analysis method based on satellite observation and continuous imagery, comprising the following steps:

[0130] S1. Data Preparation:

[0131] A system database was established to collect continuous image data from fixed cameras, hourly visibility observation data from ground monitoring stations, and visibility data retrieved from satellites.

[0132] S11, Continuous image data from a fixed camera:

[0133] like Figure 3As shown, continuous image data from fixed cameras is provided by fixed cameras, which in turn come from traffic cameras, urban security cameras, or other dedicated cameras deployed by meteorological departments. The cameras continuously acquire images at fixed intervals to form a time-series image dataset.

[0134] The data characteristics of these images include:

[0135] Capture at consecutive time points: Images from adjacent time points have a high correlation, which provides a basis for subsequent time-weighted processing;

[0136] Resolution differences: Different cameras may have different resolutions, with common image resolutions ranging from 640x480 to 1920x1080 pixels;

[0137] Different cameras have different shooting angles and field of view, which need to be standardized during preprocessing.

[0138] Before model training, various preprocessing operations must be performed on the acquired continuous images to ensure the consistency of the image data and its suitability for deep learning model training. The specific processing methods are as follows:

[0139] Data preprocessing of time-series image datasets:

[0140] S111, Uniform image size:

[0141] All images are scaled to a uniform size using either bilinear or cubic spline interpolation to ensure relatively clear details are maintained after scaling.

[0142] I resize =resize(I original ,800,600)

[0143] Among them I resize For the processed image, I original Original image;

[0144] S112, Color Channel Standardization:

[0145] To accommodate the input requirements of the neural network, the three color channels of the image (RGB) need to be normalized separately. The normalization formula is as follows:

[0146]

[0147] Where I(x, y, c) is the value of the color channel c of pixel (x, y), μ c and σ c These are the mean and standard deviation of color channel c, respectively.

[0148] S113, Noise Reduction and Enhancement:

[0149] Image denoising aims to remove the effects of sensor noise and environmental interference on images. Median filtering or bilateral filtering is commonly used for image denoising.

[0150] I denoise =medianFilter(I norm k)

[0151] Where k is the size of the filtering window.

[0152] Meanwhile, considering that some images may be affected by factors such as haze and insufficient light, methods such as contrast stretching and brightness adjustment can be used to enhance the images and improve their clarity.

[0153] S114, Data Augmentation:

[0154] Data augmentation helps improve the generalization ability of a model and avoid overfitting. Data augmentation can be performed on images using random rotation, cropping, and flipping techniques.

[0155] I aug =randomAugment(I denoise )

[0156] The data-augmented images will be used together with the original images to train the visibility regression model, ensuring that the model can adapt to different camera angles, lighting conditions, and weather conditions.

[0157] S115, Time Alignment:

[0158] Since data from fixed cameras is collected at fixed time intervals, it is necessary to ensure that the timestamps of the images are consistent with the timestamps of the ground monitoring station and satellite data. Interpolation algorithms are used for time alignment to ensure that the timestamps of the images are consistent with the hourly visibility observation data from the ground monitoring station and the visibility data retrieved from the satellite. If a camera image is missing at a certain moment, it can be filled by interpolating images from adjacent moments.

[0159]

[0160] This ensures that continuous image data is available for subsequent analysis at all points in time;

[0161] S12. Hourly visibility observation data from ground monitoring stations are shown in the table below:

[0162] Current status of station 54416 (Miyun)

[0163] meteorological elements value Update time instantaneous temperature 19.0 2024-11-01 12:00+0800 24-hour temperature regulation 1.9 2024-11-01 12:00+0800 Ground air pressure 1015.3 2024-11-01 12:00+0800 relative humidity 53 2024-11-01 12:00+0800 2-minute average wind direction (NE) 2024-11-01 12:00+0800 2-minute average wind speed (2) 2024-11-01 12:00+0800 1 hour of precipitation 0.0 2024-11-01 12:00+0800 24-hour precipitation 0.0 2024-11-01 12:00+0800 10-minute average visibility 6.348 2024-11-01 12:00+0800

[0164] Hourly visibility data from ground monitoring stations is provided by the meteorological department. The meteorological department monitors visibility through ground monitoring stations in various cities. The data from these stations includes hourly visibility data and meteorological data, with the meteorological data used to assist in the analysis of visibility changes.

[0165] Data preprocessing was performed on the hourly visibility observation data:

[0166] S121, Time Alignment:

[0167] Time synchronization with fixed camera images and satellite data is crucial. Ground monitoring station data is typically collected hourly, but fixed camera data may be collected minute by minute. Therefore, hourly visibility observation data is interpolated to ensure consistency with the camera image data collected by fixed cameras.

[0168]

[0169] Where V(t) is the visibility value at time t, and t0 and t1 are two adjacent integer times.

[0170] S122, Missing Value Handling:

[0171] If ground monitoring station data is missing at a certain moment, it can be supplemented by interpolation or by using interpolation methods based on historical data;

[0172] S13, Satellite inversion visibility data:

[0173] Satellite inversion visibility data is obtained through polar-orbiting or geostationary satellites, such as... Figure 4 As shown, its resolution depends on the corresponding satellite system. Satellite data can provide wide-area visibility information, compensating for the shortcomings of sparse distribution of ground stations.

[0174] S131. Remove outliers:

[0175] Since satellite observations may be affected by cloud cover, sensor malfunctions, etc., and there may be some errors, outliers in the satellite inversion visibility data are removed in the preprocessing stage, and error data are detected by upper and lower limit constraints and statistical methods.

[0176] S132, Time Alignment:

[0177] Satellite data is typically acquired at long intervals, so linear interpolation or more complex interpolation methods are needed to fill these time gaps and ensure that camera images, monitoring station data, and satellite data correspond at the same point in time.

[0178] Linear interpolation is used to fill the temporal gaps in the satellite-retrieved visibility data, ensuring that camera image data, hourly visibility observation data from ground monitoring stations, and satellite-retrieved visibility data correspond to each other at the same time. S14. All processed data will be stored in the system database, categorized and managed by region and time.

[0179] Data management includes: the storage of raw data and the storage of processed data.

[0180] Each processed camera image, hourly visibility observation, and satellite-retrieved visibility data is timestamped and spatially labeled to ensure consistency with the input of the visibility regression model.

[0181] S2. Dynamic spatial fusion of multi-source data:

[0182] Multi-source data spatial fusion is one of the key steps in visibility analysis systems.

[0183] By introducing a mesh model, spatial fusion and weight adjustment are performed on three types of data: camera image data, hourly visibility observation data, and satellite-retrieved visibility data, as detailed below:

[0184] Appropriate interpolation algorithms are used based on different data sources, and the contribution weight of each data source is dynamically adjusted based on the data distribution density to ensure that the visibility regression model has accurate predictive capabilities in space.

[0185] Based on the highest resolution grid of satellite visibility data, the target area is geospatially gridded. The grid division is based on satellite resolution, with the grid where the fixed camera is located as the core grid, and then eight surrounding grids are extended outward to form a 3x3 nine-square grid structure.

[0186] Of these nine grids, the core grid integrates data from fixed cameras, ground monitoring stations, and satellite systems, while the surrounding grids are used for interpolation of satellite and ground station data.

[0187] The spatial data fusion method is adjusted according to the distribution of ground monitoring stations within the grid, and is divided into four cases:

[0188] Scenario 1: No site

[0189] There are no ground monitoring stations within the core grid and the eight surrounding grids. Due to the lack of ground monitoring station data, the visibility of the core grid relies entirely on the satellite system. Its calculation process does not require interpolation algorithms; the visibility value is determined solely based on the observation data from the satellite system. The data weight allocation is as follows:

[0190] Satellite data weights within the core grid: ω sat =1

[0191] Ground station data weight: ω site =0

[0192] Core grid visibility algorithm formula:

[0193] V core =V sat (core)

[0194] Among them, V core V represents the visibility of the core grid after data fusion. sat (core) represents satellite observation visibility data within the core grid.

[0195] Scenario 2: Clustered Sites

[0196] If there is at least one ground monitoring station within the core grid, but no stations within the surrounding eight grids, then it is necessary to combine station data from the core grid with satellite data.

[0197] If there is only one ground monitoring station in the core grid, then the data from that ground monitoring station is used directly as the station visibility V for the core grid. site (core),

[0198] If the core grid has more than one ground monitoring station, interpolation between the ground monitoring stations is required:

[0199]

[0200] Where V(x) i ) represents the visibility observation value at station i, d(x0, x) i ) is the distance between camera position x0 and station i, and p is the distance attenuation factor.

[0201] Data weight allocation:

[0202] Satellite data weights within the core grid: ω sat =0.5

[0203] Weight of ground monitoring station data within the core grid: ω site =0.5

[0204] Taking into account both satellite and station data for the core grid, the visibility calculation formula for the core grid is as follows:

[0205] V core =ω site ×V site (core)+ω sat ×V sat (core)

[0206] Scenario 3: Sparse sites

[0207] There are no ground monitoring stations within the core grid, but at least one ground monitoring station exists in one of the eight surrounding grids. In this case, it is necessary to interpolate the visibility of the core grid using data from the surrounding ground monitoring stations.

[0208] If there is only one ground monitoring station within the surrounding grid, then the data from that ground monitoring station is directly used as the visibility V for the surrounding grid's ground monitoring stations. site (out),

[0209] If there are multiple ground monitoring stations within the surrounding grid, use Kriging interpolation to interpolate the data from the ground monitoring stations in the surrounding grid into the core grid:

[0210]

[0211] Where V(x) i ) represents the visibility observation value of the surrounding ground monitoring station i, and λ represents the visibility value. i These are interpolation weights, determined based on geospatial correlation, satisfying...

[0212] Data weight allocation:

[0213] Satellite data weights for the core grid: ω sat =0.8

[0214] Ground monitoring station data weight: ω site =0.2

[0215] The interpolated ground monitoring station data is combined with the satellite data of the core grid. The visibility calculation formula for the core grid is as follows:

[0216] V core =ω site ×V site (out)+ω sat ×V sat (core)

[0217] Scenario 4: Multiple sites

[0218] Ground monitoring stations are located in the core grid and the eight surrounding grids. Therefore, the ground monitoring station data in the core grid and the surrounding grids need to be processed separately. The ground monitoring station data in the core grid is interpolated using the inverse distance weighting method, while the ground monitoring station data in the surrounding grids is interpolated to the core grid using Kriging interpolation. The interpolation algorithm for the ground monitoring station data in the core grid is as follows:

[0219]

[0220] The interpolation algorithm for the ground monitoring stations in the surrounding grid is as follows:

[0221]

[0222] Data weight allocation:

[0223] Weight of ground monitoring station data in the core grid: ω site (core) = 0.6

[0224] Satellite data weights for the core grid: ω sat =0.3

[0225] Weight of ground monitoring station data in the surrounding grid: ω site (out) = 0.1

[0226] Combining interpolated data from ground monitoring stations in the core and surrounding grids, as well as satellite data from the core grid, the visibility calculation formula for the core grid is as follows:

[0227] V cor =ω site (core)×V site (core)+ω sat ×V sat (core)+ω site (out)×V site (out);

[0228] S3. Visibility Regression Model Training:

[0229] S31. Visual image feature extraction:

[0230] The backbone network is used to extract features from visual images captured by fixed cameras to obtain high-dimensional image features. The backbone network is also used to extract features from satellite data acquired by the satellite system to obtain high-dimensional satellite data features. The satellite data is then standardized to obtain satellite values.

[0231] Various backbone networks such as CNN, Swin-Transformer, ViT, and DeiT can be used for feature extraction from visual images. These backbone networks are pluggable; as neural network technology continues to evolve, the neural network used for extracting visual image features in the system can be replaced at any time.

[0232] S32. Establish a fully connected regression model:

[0233] A fully connected neural network regression model is established by combining the high-dimensional features of visual images and the high-dimensional features or satellite values ​​extracted from satellite data with visibility label data.

[0234] S33. Model Training: Use multi-region training samples over a long time scale to train the model;

[0235] S34. Model Fine-tuning: In model training, the model trained using samples from multiple locations and over long time scales needs to be fine-tuned for real-world use. In model application, a model fine-tuning function is set up to enable fine-tuning of the model.

[0236] S4. Visibility Recognition:

[0237] In the application phase of the visibility regression model, instead of relying on ground monitoring station data and satellite-derived visibility data, visibility identification is performed using real-time images from fixed cameras.

[0238] By introducing a time-weighted mechanism, real-time images from fixed cameras at continuous intervals can be fully utilized.

[0239] Unlike traditional single-image recognition, this method introduces a time-weighted mechanism, which fully utilizes camera images from consecutive moments to improve the stability and accuracy of visibility prediction, and avoids the significant impact of noise or outliers at a single moment on the results.

[0240] For a target time t, image data from 2n+1 preceding and following times are extracted from the image sequence captured by a fixed camera, and visibility is identified at each time.

[0241] For example, when n=2, images are extracted at five time points: t-2, t-1, t, t+1, and t+2, and visibility is identified at each time point.

[0242] Then, based on the time difference between the image capture time and the target time, different weights are assigned to each time point;

[0243] Finally, the visibility value at time t is obtained through a weighted calculation method.

[0244]

[0245] Where V(t+i) is the visibility estimate at time t+i; ω i The weights are inversely proportional to the time difference |i|; the smaller the time difference, the larger the weight.

[0246] Weight Explanation: The weight ω0 is maximized at the current time t, ensuring that the visibility regression model is dominated by the observations at the current time; data from previous and subsequent times are used as auxiliary data, with weights ω0 and ω0 respectively. -2 ω -1 ω1 and ω2 gradually decrease as the time distance increases.

[0247] For example, if the weights are calculated according to ω0 = 0.4, ω -1=ω1=0.2, ω -2 The proportional distribution with ω² = 0.1 is calculated using the following formula:

[0248] V(t)=0.4×V(t)+0.2×(V(t-1)+V(t+1))+0.1×(V(t-2)+V(t+2))

[0249] This time-weighted processing can fully reference image data from adjacent times based on the image at the current time, thereby reducing the impact of noise or instantaneous fluctuations in a single frame on the results and obtaining more stable visibility estimation results.

[0250] In step S11, the data acquisition interval of the fixed camera is every 5 minutes or 10 minutes.

[0251] In step S12, the monitoring station data also includes meteorological data, which is used to assist in the analysis of visibility changes.

[0252] In step S13, the resolution range of the satellite inversion visibility data is from 100 meters to 9,000 meters.

[0253] In step S33, data augmentation is achieved by scaling and transposing the visual images and satellite data.

[0254] In step S34, the model fine-tuning function uses manual visual inspection or mobile visibility monitoring equipment to determine the visibility value at the model application location, accumulates a certain amount of data to fine-tune the basic model, and optimizes the model effect.

[0255] This invention proposes a visibility analysis method based on satellite observation and continuous imagery. It utilizes satellite observation and continuous shooting at fixed intervals by a fixed camera. The method acquires continuous image data from the fixed camera, hourly visibility observation data from ground monitoring stations, and visibility data retrieved from the satellite. After preprocessing, the data is stored in a system database and categorized by region and time. Not only is the original data stored and preserved for subsequent algorithm optimization, but each processed image data from the fixed camera, visibility observation data from the ground monitoring station, and visibility data retrieved from the satellite is also tagged with timestamps and spatial labels to ensure consistency with the model's input.

[0256] After processing and storing the data, multi-source data spatial dynamic fusion is performed. Multi-source data spatial fusion is one of the key steps in visibility analysis methods. By introducing a mesh-like 3x3 grid model, spatial fusion and weight adjustment are performed on three types of data from fixed cameras, ground monitoring stations and satellite observations. This process requires the use of appropriate interpolation algorithms according to different data sources and dynamic adjustment of the contribution weight of each data source based on data distribution density to ensure that the visibility model has accurate predictive capabilities in space.

[0257] After achieving dynamic fusion of multi-dimensional data space, the model is built, trained, and fine-tuned to make the method more adaptable. In actual use, a certain degree of fine-tuning is required. This can be done by visual inspection or by using mobile visibility monitoring equipment to measure the visibility values ​​at the model application location. Accumulate a certain amount of data to fine-tune the basic model in order to optimize the model effect.

[0258] Once the model is complete, it can be used immediately. During the application phase, it no longer relies on ground monitoring station data and satellite data, but instead utilizes real-time images from fixed cameras for visibility identification. Unlike traditional single-image recognition, this method introduces a time-weighted mechanism, which fully utilizes camera images from consecutive moments, improving the stability and accuracy of visibility prediction and avoiding the significant impact of noise or outliers at a single moment on the results. For the target time t, the system extracts image data from the fixed camera image sequence for 2n+1 moments before and after it. For example, when n=2, it extracts images from five moments: t-2, t-1, t, t+1, and t+2, and performs visibility identification for each moment. Then, based on the time difference between the fixed camera image capture time and the target time, different weights are assigned to each moment. Finally, the visibility value at time t is obtained through a weighted calculation method.

[0259]

[0260] Where V(t+i) is the visibility estimate at time t+i; ω i The weights are inversely proportional to the time difference |i|, with larger weights for smaller time differences.

[0261] Weight Explanation: The weight ω0 is maximum at the current time t, ensuring that the model is dominated by the observations at the current time; data from previous and subsequent time steps are used as auxiliary data, with weights ω... -2 ω -1 ω1 and ω2 gradually decrease as the time distance increases.

[0262] For example, if the weights are calculated according to ω0 = 0.4, ω -1 =ω1=0.2, ω -2 The proportional distribution with ω² = 0.1 is calculated using the following formula:

[0263] V(t)=0.4×V(t)+0.2×(V(t-1)+V(t+1))+0.1×(V(t-2)+V(t+2))

[0264] This time-weighted processing can fully reference image data from adjacent times based on the image at the current time, thereby reducing the impact of noise or instantaneous fluctuations in a single frame on the results and obtaining more stable visibility estimation results.

[0265] 1. This invention provides higher precision visibility observation: By constructing a visibility recognition system based on satellite and visual images, and combining different observation methods with artificial intelligence, the accuracy of visibility observation can be improved.

[0266] 2. This invention offers broader visibility observation capabilities: Automated visibility observation is no longer limited to expensive observation equipment with specific construction requirements. Satellite data provides global coverage, and combined with numerous camera resources, visibility observation can be conducted more extensively.

[0267] 3. This invention enhances forecasting capabilities: For weather forecasting technology, better reproducing the current state of the atmosphere is a key factor in improving the accuracy of predicting future weather patterns. High-precision, high-density visibility observations can better help reproduce the current state of the atmosphere, thereby improving weather forecasting capabilities.

[0268] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A visibility analysis method based on satellite observation and continuous imagery, characterized in that: Includes the following steps: S1. Data Preparation: A system database was established to collect continuous image data from fixed cameras, hourly visibility observation data from ground monitoring stations, and visibility data retrieved from satellites. S11, Continuous image data from a fixed camera: Continuous image data from fixed cameras is provided by fixed cameras, which in turn come from traffic cameras, urban security cameras, or other dedicated cameras deployed by meteorological departments. The cameras continuously acquire images at fixed intervals to form a time-series image dataset. Data preprocessing of time-series image datasets: S111, Uniform image size: All images are scaled to a uniform size, and the image size is uniformly calculated using either bilinear interpolation or cubic spline interpolation. I resize =resize(I original ,800,600) Where I resize For the processed image, I original Original image; S112, Color Channel Standardization: To accommodate the input requirements of the neural network, the three color channels of the image (RGB) need to be normalized separately. The normalization formula is as follows: Where I(x, y, c) is the value of the color channel c of pixel (x, y), μ c and σ c These are the mean and standard deviation of color channel c, respectively. S113, Noise Reduction and Enhancement: Median filtering or bilateral filtering can be used to denoise the image. I denoise =medianFilter(I norm ,k) Where k is the size of the filtering window. Image enhancement is achieved through contrast stretching and brightness adjustment. S114, Data Augmentation: Image data augmentation is performed using random rotation, cropping, and flipping techniques: I aug =randomAugment(I denoise ) The data-augmented images will be used together with the original images for training the visibility regression model. S115, Time Alignment: Time alignment is performed using an interpolation algorithm to ensure that the timestamps of the images are consistent with the hourly visibility observation data from ground monitoring stations and the visibility data retrieved from satellites. If a camera image is missing at a certain moment, it can be filled by interpolating images from adjacent moments. This ensures that continuous image data is available for subsequent analysis at all points in time; S12, Hourly visibility observation data from ground monitoring stations: Hourly visibility data from ground monitoring stations is provided by the meteorological department. The meteorological department monitors visibility through ground monitoring stations in various cities. The data from these stations includes hourly visibility data and meteorological data, with the meteorological data used to assist in the analysis of visibility changes. Data preprocessing was performed on the hourly visibility observation data: S121, Time Alignment: The hourly visibility observation data was interpolated over time to ensure consistency with the camera image data collected by the fixed camera. Where V(t) is the visibility value at time t, and t0 and t1 are two adjacent integer times. S122, Missing Value Handling: If ground monitoring station data is missing at a certain moment, it can be supplemented by interpolation or by using interpolation methods based on historical data; S13, Satellite inversion visibility data: Visibility data retrieved via satellite is acquired through polar-orbiting or geostationary satellites, with resolution depending on the respective satellite system. S131. Remove outliers: In the preprocessing stage, outliers in the satellite inversion visibility data are removed, and error data is detected through upper and lower limit constraints and statistical methods. S132, Time Alignment: Linear interpolation is used to fill the temporal gaps in the visibility data retrieved from satellites, ensuring that camera image data, hourly visibility observation data from ground monitoring stations, and satellite-retrieved visibility data correspond to each other at the same time. S14. All processed data will be stored in the system database, categorized and managed by region and time. Data management includes: the storage of raw data and the storage of processed data. Each processed camera image, hourly visibility observation, and satellite-retrieved visibility data is timestamped and spatially labeled to ensure consistency with the input of the visibility regression model. S2. Dynamic spatial fusion of multi-source data: By introducing a mesh model, spatial fusion and weight adjustment are performed on three types of data: camera image data, hourly visibility observation data, and satellite-retrieved visibility data, as detailed below: Appropriate interpolation algorithms are used based on different data sources, and the contribution weight of each data source is dynamically adjusted based on the data distribution density to ensure that the visibility regression model has accurate predictive capabilities in space. Based on the highest resolution grid of satellite visibility data, the target area is geospatially gridded. The grid division is based on satellite resolution, with the grid where the fixed camera is located as the core grid, and then eight surrounding grids are extended outward to form a 3x3 nine-square grid structure. Of these nine grids, the core grid integrates data from fixed cameras, ground monitoring stations, and satellite systems, while the surrounding grids are used for interpolation of satellite and ground station data. The spatial data fusion method is adjusted according to the distribution of ground monitoring stations within the grid, and is divided into four cases: Scenario 1: No site There are no ground monitoring stations within the core grid and the eight surrounding grids. Due to the lack of ground monitoring station data, the visibility of the core grid relies entirely on the satellite system. Its calculation process does not require interpolation algorithms; the visibility value is determined solely based on the observation data from the satellite system. The data weight allocation is as follows: Satellite data weights within the core grid: ω sat =1 Ground station data weight: ω site =0 Core grid visibility algorithm formula: V core =V sat (core) Among them, V core V represents the visibility of the core grid after data fusion. sat (core) represents satellite observation visibility data within the core grid. Scenario 2: Clustered Sites If there is at least one ground monitoring station within the core grid, but no stations within the surrounding eight grids, then it is necessary to combine station data from the core grid with satellite data. If there is only one ground monitoring station in the core grid, then the data from that ground monitoring station is used directly as the station visibility V for the core grid. site (core), If the core grid has more than one ground monitoring station, interpolation between the ground monitoring stations is required: Where V(x) i ) represents the visibility observation value at station i, d(x0, x) i ) is the distance between camera position x0 and station i, and p is the distance attenuation factor. Data weight allocation: Satellite data weights within the core grid: ω sat =0.5 Weight of ground monitoring station data within the core grid: ω site =0.5 Taking into account both satellite and station data for the core grid, the visibility calculation formula for the core grid is as follows: V core =ω site ×V site (core)+ω sat ×V sat (core) Scenario 3: Sparse sites There are no ground monitoring stations within the core grid, but at least one ground monitoring station exists in one of the eight surrounding grids. In this case, it is necessary to interpolate the visibility of the core grid using data from the surrounding ground monitoring stations. If there is only one ground monitoring station within the surrounding grid, then the data from that ground monitoring station is directly used as the visibility V for the surrounding grid's ground monitoring stations. site (out), If there are multiple ground monitoring stations within the surrounding grid, use Kriging interpolation to interpolate the data from the ground monitoring stations in the surrounding grid into the core grid: Where V(x) i ) represents the visibility observation value of the surrounding ground monitoring station i, and λ represents the visibility value. i These are interpolation weights, determined based on geospatial correlation, satisfying... Data weight allocation: Satellite data weights for the core grid: ω sat =0.8 Ground monitoring station data weight: ω site =0.2 The interpolated ground monitoring station data is combined with the satellite data of the core grid. The visibility calculation formula for the core grid is as follows: V core =ω site ×V site (out)+ω sat ×V sat (core) Scenario 4: Multiple sites Ground monitoring stations are located in the core grid and the eight surrounding grids. Therefore, the ground monitoring station data from the core grid and the surrounding grids need to be processed separately. The ground monitoring station data within the core grid is interpolated using the inverse distance weighting method, while the ground monitoring station data from the surrounding grids is interpolated to the core grid using Kriging interpolation. The interpolation algorithm for ground monitoring stations within the core grid is as follows: The interpolation algorithm for the ground monitoring stations in the surrounding grid is as follows: Data weight allocation: Weight of ground monitoring station data in the core grid: ω site (core) = 0.6 Satellite data weights for the core grid: ω sat =0.3 Weight of ground monitoring station data in the surrounding grid: ω site (out) = 0.1 Combining interpolated data from ground monitoring stations in the core and surrounding grids, as well as satellite data from the core grid, the visibility calculation formula for the core grid is as follows: V core =ω site (core)×V site (core)+ω sat ×V sat (core)+ω site (oμt)×V site (out): S3. Visibility Regression Model Training: S31. Visual image feature extraction: The backbone network is used to extract features from visual images captured by fixed cameras to obtain high-dimensional image features. The backbone network is also used to extract features from satellite data acquired by the satellite system to obtain high-dimensional satellite data features. The satellite data is then standardized to obtain satellite values. S32. Establish a fully connected regression model: A fully connected neural network regression model is established by combining the high-dimensional features of the visual images and the high-dimensional features or satellite values ​​extracted from the satellite data with the visibility label data. S33. Model Training: Use multi-region training samples over a long time scale to train the model; S34. Model Fine-tuning: In model training, the model trained using samples from multiple locations and over long time scales needs to be fine-tuned for real-world use. In model application, a model fine-tuning function is set up to enable fine-tuning of the model. S4. Visibility Recognition: In the application phase of the visibility regression model, instead of relying on ground monitoring station data and satellite-derived visibility data, visibility identification is performed using real-time images from fixed cameras. By introducing a time-weighted mechanism, real-time images from fixed cameras at continuous intervals can be fully utilized. For a target time t, image data from 2n+1 preceding and following times are extracted from the image sequence captured by a fixed camera, and visibility is identified at each time. Then, based on the time difference between the image capture time and the target time, different weights are assigned to each time point; Finally, the visibility value at time t is obtained through a weighted calculation method. Where V(t+i) is the visibility estimate at time t+i; ω i The weights are inversely proportional to the time difference i; the smaller the time difference, the larger the weight. Weight Explanation: The weight ω0 is maximized at the current time t, ensuring that the visibility regression model is dominated by the observations at the current time; data from previous and subsequent times are used as auxiliary data, with weights ω0 and ω0 respectively. -2 ω -1 ω1 and ω2 gradually decrease as the time distance increases.

2. The visibility analysis method based on satellite observation and continuous imagery according to claim 1, characterized in that: In step S11, the data acquisition interval of the fixed camera is every 5 minutes or 10 minutes.

3. The visibility analysis method based on satellite observation and continuous imagery according to claim 1, characterized in that: In step S12, the monitoring station data also includes meteorological data, which is used to assist in the analysis of visibility changes.

4. The visibility analysis method based on satellite observation and continuous imagery according to claim 1, characterized in that: In step S13, the resolution range of the satellite inversion visibility data is from 100 meters to 9,000 meters.

5. The visibility analysis method based on satellite observation and continuous imagery according to claim 1, characterized in that: In step S33, data augmentation is achieved by scaling and transposing the visual images and satellite data.

6. The visibility analysis method based on satellite observation and continuous imagery according to claim 1, characterized in that: In step S34, the model fine-tuning function uses manual visual inspection or mobile visibility monitoring equipment to determine the visibility value at the model application location, accumulates a certain amount of data to fine-tune the basic model, and optimizes the model effect.

Images