Visual range obstacle weather identification method and device
Patent Information
- Application Number
- CN202511651860.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-02-10
Smart Images
Figure CN121502418A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of meteorological data processing technology, and in particular to a method, system, computer-readable storage medium, and computer device for identifying visibility-impaired weather. Background Technology
[0002] With the increasing demand for meteorological, transportation, and environmental monitoring, data products that can provide detailed descriptions of atmospheric visibility are gaining more and more attention. Visibility obstruction phenomena (such as fog, haze, blowing dust, and sandstorms) not only affect people's travel safety but also have significant impacts on multiple fields, including aviation, shipping, road transport, and environmental quality assessment.
[0003] Traditional meteorological observations rely heavily on ground-based automatic weather stations to provide basic meteorological elements such as visibility, temperature, humidity, air pressure, wind speed, and wind direction, while environmental monitoring stations provide data on pollutants such as PM2.5 and PM10. Although these observation data points are reliable, their distribution is limited, making it difficult to comprehensively reflect changes in visibility over a large area.
[0004] In recent years, researchers have attempted to use gridded data products (such as national meteorological element grid data generated through interpolation algorithms) to more comprehensively and continuously reflect meteorological phenomena. Research on visibility and visibility impairment phenomena has gradually shifted from relying on point values from single stations to automatic identification methods that integrate multiple elements and data sources.
[0005] Furthermore, with the development of image processing technology and edge computing, all-weather image acquisition equipment such as fisheye cameras have been gradually deployed in meteorological stations, highways, and urban monitoring networks, providing a new data dimension for image recognition and analysis of visibility impairment phenomena. However, currently, image data is mainly used for manual analysis and has not yet been integrated with gridded data to form a systematic processing method.
[0006] Currently, technologies related to automatic visual obstacle recognition and grid data generation can be categorized as follows:
[0007] Visibility threshold-based methods rely on data from observation stations, either manually or automatically, to determine the presence of visibility obstructions based on predefined thresholds. For example, visibility less than 1000 meters and humidity greater than 90% is considered fog; visibility less than 5000 meters and PM2.5 concentration above a certain threshold is considered haze; a sudden drop in visibility accompanied by strong winds or dust storms is considered a sandstorm. The main problems with this method are its reliance on point-based data, poor spatial continuity, high false alarm rate, and inability to adapt to complex multi-factor conditions. It's like a "simple conditional judgment table," relying on data from each meteorological station to check if several conditions are met. However, due to uneven data distribution and diverse weather patterns, it's prone to misjudgment or omission.
[0008] Gridded methods based on multivariate interpolation: Some studies use interpolation algorithms (such as Inverse Distance Weighted Method (IDW), Kriging, spline interpolation, etc.) to grid meteorological elements, and then identify visibility obstacles on the gridded data. This approach can compensate for the uneven distribution of observation points and construct a spatially continuous "national visibility obstacle distribution map". The drawback is that the interpolation results are sensitive to the algorithm, and there is a lack of linkage logic between different meteorological elements, resulting in a still coarse judgment model. This is similar to the method of "completing a map"—"covering" the visibility and other data from 2460 observation points across the country to "estimate" visibility even in areas without stations. However, it's like "only drawing colors" without combining other information for comprehensive judgment, and therefore lacks intelligence.
[0009] Image recognition-based line-of-sight identification methods: In recent years, some studies have proposed using all-weather fisheye images to identify low-visibility weather based on indicators such as image sharpness, color scattering, and edge contours. However, most methods remain in the research or local application stage, and image analysis is mostly offline processing or manual assistance, lacking an automated system that integrates with meteorological data. This is essentially "reading the weather from a photograph." For example, a blurry image might indicate fog. However, currently, these images are rarely automatically combined with meteorological data.
[0010] In summary, existing visual obstacle recognition technologies have the following main shortcomings:
[0011] 1. Insufficient spatial continuity of location data
[0012] Currently, the assessment mainly relies on visibility, meteorological data, and air quality data from ground-based observation stations. However, these stations are unevenly distributed spatially, with sparse coverage in some areas, making it difficult to establish continuous spatial information on the distribution of visibility obstacles. This results in information gaps in national or regional analyses.
[0013] 2. Determine whether the model depends on a single factor or empirical rules.
[0014] Existing identification methods often employ threshold-based single-factor judgments (such as visibility, PM2.5, relative humidity, etc.), relying primarily on experience and failing to systematically consider the interrelationships between multiple meteorological and environmental factors. In situations with complex meteorological conditions and overlapping types of visibility obstacles, the accuracy of identification may be limited.
[0015] 3. Image data has not yet been fully integrated into the automatic recognition process.
[0016] Although some studies have attempted to use fisheye images or surveillance images to identify visibility impairments, in current systems, image data is mostly used for manual verification and rarely participates in the automatic identification process. A standardized processing procedure that can be integrated with meteorological data has not yet been formed.
[0017] 4. Limited national-scale product production capacity.
[0018] Because the processing flow relies heavily on observation point data and local processing rules, existing solutions still suffer from problems such as inconsistent calculation processes and limited model versatility in constructing nationwide, gridded visibility obstacle data products. Summary of the Invention
[0019] The present invention aims to solve at least one of the technical problems existing in the prior art or related art.
[0020] Therefore, the purpose of this invention is to provide a method and device for visual impairment weather identification, which constructs a spatial interpolation method for multi-source meteorological and environmental data, interpolates ground station data into grid data at a national scale, and realizes the continuous expression of visual impairment related elements.
[0021] To achieve the above objectives, the first aspect of the present invention provides a method for identifying visibility-impaired weather, comprising the following steps:
[0022] Acquire multi-source meteorological data related to visibility impairment;
[0023] Multi-source meteorological data includes at least one of the following: observation data from ground meteorological stations, pollutant data from environmental monitoring stations, image data from ground meteorological stations, satellite remote sensing data, and numerical model data; observation data from ground meteorological stations includes at least one of the following: visibility, relative humidity, precipitation, wind speed, wind direction, temperature, and air pressure; pollutant data from environmental monitoring stations includes at least one of the following: PM2.5 concentration and PM10 concentration.
[0024] Spatiotemporal standardization processing is performed on multi-source meteorological data to map the multi-source data to a preset latitude and longitude grid, thereby obtaining discrete station data;
[0025] Spatial interpolation algorithms are used to interpolate discrete station data to generate a nationwide, spatially continuous multi-element grid dataset.
[0026] Based on a set of multi-factor logical rules, point-by-point identification is performed on a multi-element grid dataset to output visual obstacle classification labels;
[0027] The station image data and ground meteorological station observation data are input into the multimodal fusion model to verify or correct the identification results of point-by-point identification;
[0028] Based on the identification, verification, or correction results of point-by-point identification, output standardized grid data containing visual obstacle types for each grid point nationwide.
[0029] In the above technical solution, preferably, the site image data includes at least one of fisheye images, all-sky camera images, and fixed-point high-definition camera images; the satellite remote sensing data includes at least one of multispectral visible light channel data and far-infrared channel data.
[0030] In the above technical solution, preferably, the spatiotemporal standardization process includes the following steps:
[0031] The nearest neighbor matching method is used to associate the observed data with the target grid cell for the site data. Specifically:
[0032] ;
[0033] in, Let o be the spatial coordinates of the observed data point. The spatial coordinates of the center points of all cells in the target grid G; the station data includes at least one of the following: observation data from ground meteorological stations, pollutant data from environmental monitoring stations, and image data from ground meteorological stations.
[0034] Spatial resampling is used for satellite remote sensing data and numerical model data. The resampled values of the target grid cells are calculated by weighting the resampled values according to the area overlap ratio. Specifically, the values V of the original data points that spatially overlap with N grid cells of 0.01° × 0.01° in the target grid G are calculated. O The resampled value V of the target mesh cell is obtained by weighting according to the area overlap ratio. G The calculation process can be represented by the following formula:
[0035] ;
[0036] in, The area of a target grid cell. For the first The overlap area between the original input data pixels and the target cell, For the N input data cells that overlap with the target grid cell, the first... One pixel.
[0037] In the above technical solution, preferably, in the spatial interpolation algorithm, the point x to be corrected is surrounded by n observation points ( ), the observed value The point to be corrected x and the i-th observation point The distance is Then the first Weight of each observation point Defined as:
[0038] ;
[0039] in, Represented as the first The distance correction coefficient for each observation point, p is the distance decay coefficient, which controls the rate at which the weights decay with distance:
[0040] The larger p is, the faster the weight decays with distance, and the more significant the influence of nearby observation points; the smaller p is, the greater the influence of distant observation points, and the smoother the result.
[0041] To make the sum of the weights equal to 1, the final weights are... for: ;
[0042] The interpolated lattice field value A(x) is: ;
[0043] in, Indicates the first The weight of each observation point This represents the i-th observation. Indicates the first Distance correction coefficient for each observation point.
[0044] In the above technical solution, preferably, a multi-factor logical rule set include:
[0045] R1, precipitation phenomenon: precipitation > 0.1 is classified as precipitation;
[0046] R2, Fog: Visibility < 1000m and relative humidity > 90%, wind speed ≤ 3m / s;
[0047] R3, Haze: PM 2.5 / PM 10 Concentration > 75 µg / m³, relative humidity 40% ≤ 80%, visibility < 5000m;
[0048] R4, Dust Storm: Wind speed ≥10m / s, PM 10 Concentration ≥300), humidity <40% and visibility <2000m;
[0049] R5, Dust storm: Wind speed ≥ 5m / s, PM 10 Concentration ≥200, humidity <50%, and visibility <3000m;
[0050] R6, Dust: Wind speed <5m / s, PM 10 ≥100 and visibility <5000m;
[0051] The visual impairment classification labels include: 0, No barrier; 1, Precipitation; 2, Fog; 3, Haze; 4, Sandstorm; 5, Dust; 6, Dust.
[0052] In the above technical solution, preferably, the multimodal fusion model expression is:
[0053] ;
[0054] in, For the normalization function, specifically for Each feature is normalized, and its definition is as follows:
[0055] ;
[0056] Where Q represents meteorological feature, K represents image feature, and V represents image feature representation. For all features, the total number is indivual, For the first feature The index normalization term is denoted as F, the integrated feature after fusion is denoted as M, the meteorological and environmental element data is denoted as I, and the image data is denoted as I.
[0057] In the above technical solution, preferably, it includes standardized grid data of various grid point visibility obstacle types across the country. Specifically:
[0058] ;
[0059] in, Meteorological and environmental data, Image data.
[0060] The second aspect of the present invention provides a visibility-impairing weather recognition system, comprising:
[0061] The acquisition module is configured to acquire multi-source meteorological data related to visibility impairment;
[0062] The spatiotemporal standardization processing module is configured to perform spatiotemporal standardization processing on multi-source meteorological data, so that the multi-source data is mapped to a preset latitude and longitude grid to obtain discrete station data;
[0063] The interpolation processing module is configured to use spatial interpolation algorithms to interpolate discrete station data to generate a nationwide and spatially continuous multi-element grid dataset.
[0064] The visual obstacle classification module is configured to perform point-by-point identification of multi-factor grid datasets based on a set of multi-factor logical rules and output visual obstacle classification labels.
[0065] The multimodal fusion module is configured to input station image data and ground meteorological station observation data into the multimodal fusion model to verify or correct the identification results of point-by-point identification.
[0066] The visibility obstacle grid output module is configured to output standardized grid data containing visibility obstacle types for each grid point nationwide, based on the identification, verification, or correction results of point-by-point identification.
[0067] The third aspect of the present invention provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the visibility obstacle weather recognition method provided by the first aspect of the present invention.
[0068] The fourth aspect of the present invention provides a computer device, including a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the steps of the visibility obstacle weather recognition method provided by the first aspect of the present invention.
[0069] Compared with existing technologies, the advantages of the visibility obstacle weather recognition method and device provided by this invention are as follows:
[0070] 1. A spatial interpolation method for multi-source meteorological and environmental data was constructed, which interpolates ground station data (visibility, air pressure, temperature, humidity, wind speed, wind direction, PM2.5, PM10, etc.) into gridded data at the national scale, thereby realizing the continuous expression of visibility-related elements.
[0071] 2. Introduce a fisheye image-assisted recognition mechanism. Use images and meteorological elements to perform cross-validation between the recognition results of the visibility obstacle phenomenon and the meteorological model results to improve recognition accuracy. This is especially useful for correcting results in uncertain areas or low-confidence areas of the model boundary.
[0072] 3. It has realized an automatic generation process for nationwide visual obstacle grid data, including acquisition, processing, identification, fusion and output, and has a unified automated processing framework that supports real-time or near real-time product generation.
[0073] 4. Logically couple the physical quantity interpolation data with the image information to make the system more adaptable and have greater spatial coverage. Attached Figure Description
[0074] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:
[0075] Figure 1 A flowchart of the visibility impairment weather recognition method according to an embodiment of the present invention is shown;
[0076] Figure 2 A structural block diagram of the visibility obstacle weather recognition system according to an embodiment of the present invention is shown. Detailed Implementation
[0077] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.
[0078] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.
[0079] like Figure 1 As shown, a visibility-impeded weather recognition method according to an embodiment of the present invention includes the following steps:
[0080] S1, acquire multi-source meteorological data related to visibility impairment;
[0081] Multi-source meteorological data includes at least one of the following: observation data from ground meteorological stations, pollutant data from environmental monitoring stations, image data from ground meteorological stations, satellite remote sensing data, and numerical model data; observation data from ground meteorological stations includes at least one of the following: visibility, relative humidity, precipitation, wind speed, wind direction, temperature, and air pressure; pollutant data from environmental monitoring stations includes at least one of the following: PM2.5 concentration and PM10 concentration.
[0082] S2 performs spatiotemporal standardization processing on multi-source meteorological data, mapping the multi-source data to a preset latitude and longitude grid to obtain discrete station data;
[0083] S3 uses a spatial interpolation algorithm to interpolate discrete station data, generating a nationwide, spatially continuous multi-element grid dataset.
[0084] S4, based on a set of multi-factor logical rules, performs point-by-point identification on a multi-element grid dataset and outputs visual obstacle classification labels;
[0085] S5 inputs the station image data and ground meteorological station observation data into the multimodal fusion model to verify or correct the identification results of point-by-point identification;
[0086] S6 outputs standardized grid data containing visual obstacle types for each grid point nationwide, based on the identification, verification, or correction results of the point-by-point identification.
[0087] This method for identifying visibility-impaired weather conditions has the following technical advantages:
[0088] Comprehensive data support: By acquiring multi-source meteorological data (covering ground meteorological station observations, environmental monitoring station pollutants, station images, satellite remote sensing, numerical models, etc., and each type of data contains a variety of subdivided elements, such as visibility and relative humidity in meteorological observations, and PM2.5 and PM10 in pollutants), rich and diverse basic data support is provided for visibility obstacle identification, ensuring the comprehensiveness of the identification basis.
[0089] Spatial processing standardization: After spatiotemporal standardization, multi-source data is mapped to a preset latitude and longitude grid, and then spatial interpolation is used to generate a nationwide spatially continuous multi-element grid dataset. This solves the problem of spatially discrete data and achieves continuous spatial coverage across the country, providing a unified and spatially continuous data source for subsequent point-by-point identification.
[0090] Accuracy of identification logic: Based on a set of multi-factor logical rules, the multi-factor grid dataset is identified point by point. By integrating the logical relationships of multiple factors such as meteorology and pollution, the accuracy and logic of visibility obstacle classification and identification are improved, and different types of visibility obstacles (such as fog and haze) can be effectively distinguished.
[0091] Multimodal verification of reliability: A multimodal fusion model is introduced, which combines station image data (fisheye images, all-sky camera images, etc.) with ground meteorological station observation data to verify or correct the point-by-point identification results from both visual and meteorological observation dimensions, thereby further reducing the misjudgment rate and enhancing the reliability of the identification results.
[0092] Application value and practicality: It outputs standardized grid data of various visibility obstacle types across the country, clearly presenting the spatial distribution of visibility obstacles. It can provide accurate spatial information for fields such as weather forecasting, traffic management, and environmental monitoring, and has high practical value and application prospects.
[0093] In the above embodiments, preferably, the site image data includes at least one of fisheye images, all-sky camera images, and fixed-point high-definition camera images; the satellite remote sensing data includes at least one of multispectral visible light channel data and far-infrared channel data.
[0094] In this embodiment, visual features of the sky and environment (such as the morphological manifestations of fog and haze) are intuitively captured from a visual dimension by using at least one type of station image data, such as fisheye images, all-sky camera images, and fixed-point high-definition camera images. This provides a visual basis for verification or correction of the point-by-point identification results of visibility obstacles, effectively reducing the misjudgment rate and improving the reliability of the identification results.
[0095] Meanwhile, by leveraging at least one type of satellite remote sensing data from the multispectral visible light channel and far-infrared channel, the optical properties (such as particulate matter distribution) and thermal information of the atmosphere can be analyzed at a macro scale, enabling large-scale monitoring across the country. This provides macro-level data support for generating spatially continuous multi-element gridded datasets, enhancing the spatial coverage and macro-analysis capabilities of visibility obstacle identification, and facilitating the accurate identification and spatial distribution of visibility obstacle types at the national scale.
[0096] In the above embodiments, preferably, the spatiotemporal standardization process includes the following steps:
[0097] The nearest neighbor matching method is used to associate the observed data with the target grid cell for the site data. Specifically:
[0098] ;
[0099] in, Let o be the spatial coordinates of the observed data point. The spatial coordinates of the center points of all cells in the target grid G; the station data includes at least one of the following: observation data from ground meteorological stations, pollutant data from environmental monitoring stations, and image data from ground meteorological stations.
[0100] Spatial resampling is used for satellite remote sensing data and numerical model data. The resampled values of the target grid cells are calculated by weighting the resampled values according to the area overlap ratio. Specifically, the values V of the original data points that spatially overlap with N grid cells of 0.01° × 0.01° in the target grid G are calculated. O The resampled value V of the target mesh cell is obtained by weighting according to the area overlap ratio. G The calculation process can be represented by the following formula:
[0101] ;
[0102] in, The area of a target grid cell. For the first The overlap area between the original input data pixels and the target cell, For the N input data cells that overlap with the target grid cell, the first... One pixel.
[0103] If the spatial resolution of the original input data is significantly smaller than that of the target grid, the above formula can be further simplified by directly taking the mean of all data points whose coordinates or center point coordinates fall within the target grid.
[0104] In this embodiment, there are many meteorological factors that affect the occurrence of visibility impairment, and the types of observation data are also very diverse, with different spatiotemporal resolutions and coverages. Therefore, spatiotemporal standardization is performed on the multi-source data.
[0105] The data primarily includes station observation data, such as meteorological variables like visibility, air pressure, temperature, relative humidity, wind speed, wind direction, and precipitation; environmental data on PM2.5 and PM10 particulate matter concentrations observed by environmental monitoring stations; and, in addition to meteorological element data, fisheye images deployed at each station, which can also be used to identify visibility-impairing weather phenomena. The spatial location of meteorological station observation data is at fixed stations, exhibiting local representativeness and high accuracy, but overall it is characterized by dispersion and randomness.
[0106] Besides station observation data, multispectral data from satellite remote sensing can also be used to identify visibility-impeded weather phenomena. For example, fog areas show a high correlation with differences in visible light (VIS) and far-infrared (IR) window channels; dust scattering / absorption characteristics differ significantly from those of cloud droplets, ice crystals, and the ground surface, requiring joint monitoring of dust using multiple channels, including near-infrared, deep blue, and mid-to-far-infrared. Satellite remote sensing data is characterized by high spatial coverage and relatively uniform spatial grid, but the resolution and spatial extent of different remote sensing data vary.
[0107] To generate visibility obstacle data on a spatially continuous standard grid, it is first necessary to perform time synchronization, spatial unification, and standardization processing on various types of input observation data, transferring the data to a standardized grid space. For example, if the target grid is set to have a latitude and longitude of 0.01° and the target grid is G, different data need to be matched to this grid using specific methods.
[0108] In the above embodiments, preferably, in the spatial interpolation algorithm, the point x to be corrected is assumed to have n observation points around it ( ), the observed value The point to be corrected x and the i-th observation point The distance is Then the first Weight of each observation point Defined as:
[0109] ;
[0110] in, Represented as the first The distance correction coefficient for each observation point, p is the distance decay coefficient, which controls the rate at which the weights decay with distance:
[0111] The larger p is, the faster the weight decays with distance, and the more significant the influence of nearby observation points; the smaller p is, the greater the influence of distant observation points, and the smoother the result.
[0112] To make the sum of the weights equal to 1, the final weights are... for: ;
[0113] The interpolated lattice field value A(x) is: ;
[0114] in, Indicates the first The weight of each observation point This represents the i-th observation. Indicates the first Distance correction coefficient for each observation point.
[0115] In this embodiment, visibility obstacle weather phenomenon data at each grid point nationwide is obtained using the above method. This data is two-dimensional grid data, which is usually saved in NetCDF format. It can effectively store information such as spatial coordinates, timestamps, algorithm version, and creator of the data.
[0116] In the above embodiments, preferably, the multi-factor logical rule set include:
[0117] R1, precipitation phenomenon: precipitation > 0.1 is classified as precipitation;
[0118] R2, Fog: Visibility < 1000m and relative humidity > 90%, wind speed ≤ 3m / s;
[0119] R3, Haze: PM 2.5 / PM 10 Concentration > 75 µg / m³, relative humidity 40% ≤ 80%, visibility < 5000m;
[0120] R4, Dust Storm: Wind speed ≥10m / s, PM 10 Concentration ≥300), humidity <40% and visibility <2000m;
[0121] R5, Dust storm: Wind speed ≥ 5m / s, PM 10 Concentration ≥200, humidity <50%, and visibility <3000m;
[0122] R6, Dust: Wind speed <5m / s, PM 10 ≥100 and visibility <5000m;
[0123] The visual impairment classification labels include: 0, No barrier; 1, Precipitation; 2, Fog; 3, Haze; 4, Sandstorm; 5, Dust; 6, Dust.
[0124] In this embodiment, the visibility obstacle identification model is to establish statistical laws and empirical models for the meteorological environmental elements M on each target unit, and construct logical rules R for multiple factors. The automatic judgment of each visibility obstacle weather phenomenon is achieved through these rules. The set of logical rules has the following requirements: (1) The union of the rules corresponds to all weather phenomena, that is, it can include all weather phenomena defined by visibility obstacles; (2) The intersection between the rules that determine different visibility obstacle weather phenomena must be empty, that is, given a set of meteorological environmental elements, only a unique visibility obstacle weather phenomenon can be output.
[0125] In the above embodiments, preferably, the multimodal fusion model expression is:
[0126] ;
[0127] in, For the normalization function, specifically for Each feature is normalized, and its definition is as follows:
[0128] ;
[0129] Where Q represents meteorological feature, K represents image feature, and V represents image feature representation. For all features, the total number is indivual, For the first feature The index normalization term is denoted as F, the integrated feature after fusion is denoted as M, the meteorological and environmental element data is denoted as I, and the image data is denoted as I.
[0130] In this embodiment, Q, K, and V are implemented through a meteorological element feature extraction submodule, an image feature extraction submodule, and an image feature representation submodule, respectively:
[0131] The meteorological element feature extraction submodule standardizes and embeds the observation data from meteorological stations, converts continuous elements (such as temperature, humidity, air pressure, visibility, etc.) and discrete features (such as weather phenomenon codes, wind direction categories, etc.) into vectorized features, and extracts high-dimensional representations of meteorological elements through multi-layer neural networks. These representations are used as query information in the cross-attention mechanism to guide the model to focus on image regions related to meteorological conditions.
[0132] Image feature extraction submodule: Performs visual Transformer feature extraction on images captured by fisheye cameras to obtain key features that reflect spatial structure and visual texture, which serve as "key" features in the attention mechanism for feature matching with meteorological element information.
[0133] Image Feature Representation Submodule: Shares the image feature extraction network with module K, but is used to generate a high-dimensional representation that reflects the overall semantics of the image and local meteorological features. Based on the correlation between meteorological features and image features, the model performs a weighted summation of the V features of different image regions to obtain the fused multimodal comprehensive features.
[0134] The multimodal cross-attention model F was used to identify fisheye images and meteorological elements. The identification categories included six weather phenomena: accessibility, precipitation, fog, sandstorm, blowing sand, and dust.
[0135] In the above embodiments, preferably, standardized grid data includes visual impairment types for each grid point nationwide. Specifically:
[0136] ;
[0137] in, Meteorological and environmental data, Image data.
[0138] like Figure 2 As shown, a visibility obstacle weather recognition system 100 according to another embodiment of the present invention includes:
[0139] Acquisition module 10 is configured to acquire multi-source meteorological data related to visibility impairment;
[0140] The spatiotemporal standardization processing module 20 is configured to perform spatiotemporal standardization processing on multi-source meteorological data, so that the multi-source data is mapped to a preset latitude and longitude grid to obtain discrete station data.
[0141] Interpolation processing module 30 is configured to use a spatial interpolation algorithm to interpolate discrete station data to generate a nationwide and spatially continuous multi-element grid dataset.
[0142] The visual obstacle classification module 40 is configured to perform point-by-point identification of a multi-factor grid dataset based on a set of multi-factor logical rules and output visual obstacle classification labels.
[0143] The multimodal fusion module 50 is configured to input station image data and ground meteorological station observation data into the multimodal fusion model to verify or correct the identification results of point-by-point identification.
[0144] The visibility obstacle grid output module 60 is configured to output standardized grid data containing visibility obstacle types for each grid point nationwide, based on the identification results, verification or correction results of point-by-point identification.
[0145] Based on the above, Figure 1Accordingly, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the visibility obstacle weather recognition method of any of the above embodiments.
[0146] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.
[0147] Based on the above, Figure 1 The method shown, and Figure 2 To achieve the above objectives, the present application also provides a computer device, including a storage medium and a processor, as shown in the virtual device embodiment. The storage medium is used to store a computer program; the processor is used to execute the computer program to implement the steps of the visibility obstacle weather recognition method of any of the above embodiments.
[0148] Optionally, the computer device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB ports, card reader ports, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Bluetooth interfaces, Wi-Fi interfaces), etc.
[0149] Those skilled in the art will understand that the computer device structure provided in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components, or combine certain components, or have different component arrangements.
[0150] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages and stores the hardware and software resources of a computer device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software within the physical device.
[0151] In the description of this specification, the terms "one embodiment," "some embodiments," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0152] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for identifying visibility-impaired weather, characterized in that, Includes the following steps: Acquire multi-source meteorological data related to visibility impairment; The multi-source meteorological data includes at least one of the following: ground meteorological station observation data, environmental monitoring station pollutant data, ground meteorological station image data, satellite remote sensing data, and numerical model data; the ground meteorological station observation data includes at least one of the following: visibility, relative humidity, precipitation, wind speed, wind direction, temperature, and air pressure; the environmental monitoring station pollutant data includes at least one of the following: PM2.5 concentration and PM10 concentration. The multi-source meteorological data is subjected to spatiotemporal standardization processing, so that the multi-source data is mapped to a preset latitude and longitude grid to obtain discrete station data; Spatial interpolation algorithms are used to interpolate the discrete station data to generate a nationwide, spatially continuous multi-element grid dataset. Based on a set of multi-factor logical rules, the multi-factor grid dataset is identified point by point, and visual obstacle classification labels are output. The image data of the stations and the observation data of the ground meteorological stations are input into the multimodal fusion model to verify or correct the identification results of the point-by-point identification. Based on the identification results, verification results, or correction results of the point-by-point identification, standardized grid data containing visual obstacle types for each grid point nationwide is output.
2. The method for identifying visibility-impaired weather according to claim 1, characterized in that: The site image data includes at least one of fisheye images, all-sky camera images, and fixed-point high-definition camera images; the satellite remote sensing data includes at least one of multispectral visible light channel data and far-infrared channel data.
3. The method for identifying visibility-impaired weather according to claim 2, characterized in that, The spatiotemporal standardization process includes the following steps: The nearest neighbor matching method is used to associate the observed data with the target grid cell for the site data. Specifically: ； in, Let o be the spatial coordinates of the observed data point. The spatial coordinates of the center points of all cells in the target grid G are defined; the station data includes at least one of ground meteorological station observation data, environmental monitoring station pollutant data, and ground meteorological station image data. Spatial resampling is used for satellite remote sensing data and numerical model data. The resampled values of the target grid cells are calculated by weighting the resampled values according to the area overlap ratio. Specifically, the values V of the original data points that spatially overlap with N grid cells of 0.01° × 0.01° in the target grid G are calculated. O The resampled value V of the target mesh cell is obtained by weighting according to the area overlap ratio. G The calculation process can be represented by the following formula: ； in, The area of a target grid cell. For the first The overlap area between the original input data pixels and the target cell, For the N input data cells that overlap with the target grid cell, the first... One pixel.
4. The method for identifying visibility-impaired weather according to claim 3, characterized in that, In the spatial interpolation algorithm, for the point x to be corrected, there are n observation points around it. ), the observed value The point to be corrected x and the i-th observation point The distance is Then the first Weight of each observation point Defined as: ； in, Represented as the first The distance correction coefficient for each observation point, p is the distance decay coefficient, which controls the rate at which the weights decay with distance: The larger p is, the faster the weight decays with distance, and the more significant the influence of nearby observation points; the smaller p is, the greater the influence of distant observation points, and the smoother the result. To make the sum of the weights equal to 1, the final weights are... for: ; The interpolated lattice field value A(x) is: ; in, Indicates the first The weight of each observation point This represents the i-th observation. Indicates the first Distance correction coefficient for each observation point.
5. The method for identifying visibility-impaired weather according to claim 4, characterized in that, Multi-factor logical rule set include: R1, precipitation phenomenon: precipitation > 0.1 is classified as precipitation; R2, Fog: Visibility < 1000m and relative humidity > 90%, wind speed ≤ 3m / s; R3, Haze: PM 2.5 / PM 10 Concentration > 75 µg / m³, relative humidity 40% ≤ 80%, visibility < 5000m; R4, Dust Storm: Wind speed ≥10m / s, PM 10 Concentration ≥300), humidity <40% and visibility <2000m; R5, Dust storm: Wind speed ≥ 5m / s, PM 10 Concentration ≥200, humidity <50%, and visibility <3000m; R6, Dust: Wind speed <5m / s, PM 10 ≥100 and visibility <5000m; The visual impairment classification labels include: 0, No barrier; 1, Precipitation; 2, Fog; 3, Haze; 4, Sandstorm; 5, Dust; 6, Dust.
6. The method for identifying visibility-impaired weather according to claim 5, characterized in that, The expression for the multimodal fusion model is: ； in, For the normalization function, specifically for Each feature is normalized, and its definition is as follows: ； Where Q represents meteorological feature, K represents image feature, and V represents image feature representation. For all features, the total number is indivual, For the first feature The index normalization term is F, the integrated feature after fusion is M, the meteorological and environmental element data is I, and the image data is I.
7. The method for identifying visibility-impaired weather according to claim 6, characterized in that, Standardized grid data including visual impairment types at various grid points nationwide Specifically: ； in, Meteorological and environmental data, Image data.
8. A visibility-impairment weather recognition system, characterized in that, include: The acquisition module is configured to acquire multi-source meteorological data related to visibility impairment; The spatiotemporal standardization processing module is configured to perform spatiotemporal standardization processing on the multi-source meteorological data, so that the multi-source data is mapped to a preset latitude and longitude grid to obtain discrete station data. The interpolation processing module is configured to perform interpolation processing on the discrete station data using a spatial interpolation algorithm to generate a nationwide and spatially continuous multi-element grid dataset. The visual obstacle classification module is configured to perform point-by-point identification of the multi-factor grid dataset based on a set of multi-factor logical rules and output visual obstacle classification labels. The multimodal fusion module is configured to input the station image data and the ground meteorological station observation data into the multimodal fusion model to verify or correct the identification results of the point-by-point identification. The visibility obstacle grid output module is configured to output standardized grid data containing visibility obstacle types for each grid point nationwide, based on the identification results, verification or correction results of the point-by-point identification.
9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the visibility obstacle weather recognition method as described in any one of claims 1 to 7.
10. A computer device, characterized in that, It includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the steps of the visibility obstacle weather recognition method as described in any one of claims 1 to 7.