Satellite cloud classification method and device suitable for multiple scenes, server and medium

Abstract

The invention provides a satellite cloud classification method and device suitable for multiple scenes, a server and a medium, and relates to the technical field of data processing, and the method comprises the steps: receiving a cloud classification request sent by a user for a specified cloud classification scene and a specified star source, the cloud classification request carrying an identifier of the specified cloud classification scene and an identifier of the specified star source; based on the identifier of the specified cloud classification scene and the identifier of the specified star source, determining target cloud basic product data corresponding to the cloud classification request from a cloud basic product data set corresponding to various star sources obtained by pre-inversion; and calling a cloud classification algorithm corresponding to the specified cloud classification scene based on the identifier of the specified cloud classification scene, so as to process the target cloud basic product data by using the cloud classification algorithm corresponding to the specified cloud classification scene to obtain a cloud classification response result. According to the method, the problem of limitation of application scene simplification of a current cloud classification product can be relieved, so that the requirements of accurate recognition and quantitative analysis of cloud features in different industries or fields are met.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a satellite cloud classification method, device, server and medium applicable to multiple scenarios. Background Art

[0002] Current studies on cloud classification based on satellite data usually use the International Satellite Cloud Climatology Project (ISCCP) standard, which divides clouds into nine categories: Cirrus, Cirrostratus, Deep Convection, Altocumulus, Altostratus, Nimbostratus, Cumulus, Stratocumulus, and Stratus, based on the optical thickness and cloud top height of clouds. However, existing cloud classification results often cannot meet the needs of diverse application scenarios.

[0003] In practical applications, different industries or fields have different concerns about cloud characteristics. For example, in the fields of aviation and weapon launch, the focus is usually on the height of the cloud in order to accurately assess flight safety and target hits. Therefore, the classification of cloud height becomes the primary task, which is divided into high, medium, and low clouds. In weather forecasting, atmospheric chemistry, atmospheric component research, and artificial weather modification operations, the composition of clouds is a more concerned aspect. For example, clouds of different components such as ice clouds and water clouds have a huge impact on atmospheric characteristics. In addition, the aviation and energy industries may pay more attention to the shape characteristics of clouds, because different forms of clouds have different effects on the generation and utilization of energy such as solar radiation and wind energy. Therefore, clouds need to be divided into different forms such as layered clouds, pile clouds, stratocumulus clouds, thin clouds, and rain clouds.

[0004] Therefore, the current cloud classification results have the limitation of single application scenarios and cannot meet the diverse needs of cloud features in different fields. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a satellite cloud classification method, device, server and medium suitable for multiple scenarios, so as to alleviate the limitation of current cloud classification products in terms of single application scenarios, thereby meeting the needs of different industries or fields for accurate identification and quantitative analysis of cloud features.

[0006] In a first aspect, the present invention provides a satellite cloud classification method applicable to multiple scenarios. The method is applied to a satellite cloud classification system. The satellite cloud classification system is configured with cloud classification algorithms corresponding to multiple cloud classification scenarios. The method includes:

[0007] receiving a cloud classification request sent by a user for a specified cloud classification scenario and a specified star source, where the cloud classification request carries an identifier of the specified cloud classification scenario and an identifier of the specified star source;

[0008] Based on the identifier of the specified cloud classification scene and the identifier of the specified star source, the target cloud basic product data corresponding to the cloud classification request is determined from the cloud basic product data sets corresponding to the various star sources obtained in advance by inversion;

[0009] Based on the identifier of the specified cloud classification scenario, a cloud classification algorithm corresponding to the specified cloud classification scenario is called to process the target cloud basic product data using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result.

[0010] In one embodiment, before receiving a cloud classification request sent by a user for a specified cloud classification scene and a specified star source, the method further includes:

[0011] Obtain satellite observation data and satellite cloud detection products from multiple star sources;

[0012] For any star source, the numerical model data and static data corresponding to the satellite observation data of the star source are obtained, and the satellite observation data, satellite cloud detection products, numerical model data and static data of the star source are preprocessed, quality controlled, time matched and normalized to obtain the target satellite observation data, target satellite cloud detection products, target numerical model data and target static data corresponding to the star source;

[0013] Through the inversion model corresponding to the star source, based on the target satellite observation data, target satellite cloud detection products, target numerical model data and target static data, the cloud basic product data set corresponding to the star source is inverted;

[0014] Among them, the cloud basic product data set includes target satellite observation data, target satellite cloud detection products, cloud optical thickness inversion products, cloud top pressure inversion products and cloud top temperature inversion products.

[0015] In one embodiment, the cloud classification scenario includes a cloud classification scenario based on the ISCCP classification standard, and the cloud classifications in the cloud classification scenario include cirrus, cirrostratus, deep convection, altocumulus, altostratus, nimbostratus, cumulus, stratocumulus, and stratus;

[0016] Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, target cloud basic product data corresponding to the cloud classification request is determined from the cloud basic product data sets corresponding to the various star sources obtained by pre-inversion, including: extracting cloud optical thickness inversion products and cloud top pressure inversion products from the cloud basic product data sets corresponding to the designated star source as target cloud basic product data corresponding to the cloud classification request;

[0017] The cloud classification algorithm corresponding to the specified cloud classification scenario is used to process the target cloud basic product data to obtain the cloud classification response result, including: comparing the cloud optical thickness inversion product and the cloud top pressure inversion product in the target cloud basic product data with the threshold set in the pre-set ISCCP classification standard to obtain the cloud classification response result under the cloud classification scenario based on the ISCCP classification standard.

[0018] In one embodiment, the cloud classification scenario includes a cloud classification scenario based on cloud top height, and the cloud classification in the cloud classification scenario includes high clouds, middle clouds, low clouds, and straight clouds;

[0019] Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, target cloud basic product data corresponding to the cloud classification request is determined from the cloud basic product data sets corresponding to the various star sources obtained by pre-inversion, including: extracting target satellite observation data, cloud optical thickness inversion products, cloud top pressure inversion products, and cloud top temperature inversion products from the cloud basic product data sets corresponding to the designated star source as target cloud basic product data corresponding to the cloud classification request;

[0020] The target cloud basic product data are processed using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain the cloud classification response result, including: comparing the cloud top pressure inversion product in the target cloud basic product data with the thresholds corresponding to high clouds, middle clouds and low clouds respectively, and comparing the brightness temperature data, cloud optical thickness inversion product and cloud top temperature inversion product of the target satellite observation data in the target cloud basic product data with the thresholds corresponding to straight-spread clouds, so as to obtain the cloud classification response result in the cloud classification scenario based on cloud top height.

[0021] In one embodiment, the cloud classification scenario includes a cloud classification scenario based on cloud components, and the cloud classifications in the cloud classification scenario include water clouds, ice clouds, mixed phase clouds, supercooled water clouds, deep convective clouds, precipitation clouds, aerosol and pollution clouds, and dust rolls;

[0022] Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, target cloud basic product data corresponding to the cloud classification request is determined from the cloud basic product data sets corresponding to the various star sources obtained by pre-inversion, including: extracting target satellite observation data, cloud optical thickness inversion products and cloud top temperature inversion products from the cloud basic product data sets corresponding to the designated star source as target cloud basic product data corresponding to the cloud classification request;

[0023] The target cloud basic product data are processed by using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain the cloud classification response result, including: performing radiation calibration and cloud mask processing on the target satellite observation data in the target cloud basic product data to obtain the brightness temperature of multiple bands, and obtaining the first coarse classification result based on the difference between the brightness temperatures of different bands, the cloud optical thickness inversion product and the cloud top temperature inversion product; and obtaining the second coarse classification result based on the target satellite observation data, the cloud optical thickness inversion product and the cloud top temperature inversion product in the target cloud basic product data through the first neural network model; the first coarse classification result and the second coarse classification result are corrected based on the classification rules of the fusion spectral characteristics and microphysical parameters to obtain the cloud classification response result in the cloud classification scenario based on cloud components.

[0024] In one embodiment, the cloud classification scenario includes a cloud classification scenario based on cloud shape, and the cloud classifications in the cloud classification scenario include layered clouds, pile clouds, stratocumulus clouds, thin clouds and rain clouds;

[0025] Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, target cloud basic product data corresponding to the cloud classification request is determined from cloud basic product data sets corresponding to multiple star sources obtained by pre-inversion, including: extracting target satellite observation data from the cloud basic product data set corresponding to the designated star source as the target cloud basic product data corresponding to the cloud classification request;

[0026] The target cloud basic product data are processed using a cloud classification algorithm corresponding to a specified cloud classification scenario to obtain a cloud classification response result, including: extracting two-dimensional satellite images from target satellite observation data within the target cloud basic product data, and stacking the two-dimensional satellite images of multiple consecutive time frames into a three-dimensional data block; through a second neural network, generating a cloud classification response result in a cloud classification scenario based on cloud shape based on the three-dimensional data block.

[0027] In one embodiment, the cloud classification scenario includes a custom cloud classification scenario, and the method includes:

[0028] Receive a cloud classification definition request sent by a user to determine a custom cloud classification, and receive a cloud classification label annotation request sent by a user to determine a custom cloud classification label corresponding to the custom cloud classification;

[0029] Receive a cloud basic product data selection request sent by a user to determine the custom cloud basic product data from the cloud basic product data set corresponding to the star source;

[0030] The third neural network is trained using the custom cloud basic product data, the custom cloud classification and its corresponding custom cloud classification labels, so as to generate a cloud classification response result in the custom cloud classification scenario through the trained third neural network.

[0031] In a second aspect, the present invention further provides a satellite cloud classification device applicable to multiple scenarios, the device is applied to a satellite cloud classification system, and the satellite cloud classification system is configured with cloud classification algorithms corresponding to multiple cloud classification scenarios, and the device includes:

[0032] A request receiving module, used to receive a cloud classification request sent by a user for a specified cloud classification scenario and a specified star source, wherein the cloud classification request carries an identifier of the specified cloud classification scenario and an identifier of the specified star source;

[0033] A data determination module is used to determine target cloud basic product data corresponding to the cloud classification request from cloud basic product data sets corresponding to multiple star sources obtained by pre-inversion based on the identifier of the specified cloud classification scene and the identifier of the specified star source;

[0034] The cloud classification module is used to call the cloud classification algorithm corresponding to the specified cloud classification scenario based on the identifier of the specified cloud classification scenario, so as to use the cloud classification algorithm corresponding to the specified cloud classification scenario to process the target cloud basic product data to obtain a cloud classification response result.

[0035] In a third aspect, the present invention further provides a server, comprising a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement any one of the methods provided in the first aspect.

[0036] In a fourth aspect, the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement any one of the methods provided in the first aspect.

[0037] The present invention provides a satellite cloud classification method, device, server and medium suitable for multiple scenarios. If a cloud classification request sent by a user for a specified cloud classification scenario and a specified star source is received, based on the identifier of the specified cloud classification scenario and the identifier of the specified star source carried in the cloud classification request, the target cloud basic product data corresponding to the cloud classification request is determined from the cloud basic product data sets corresponding to the multiple star sources obtained by pre-inversion. Finally, based on the identifier of the specified cloud classification scenario, the cloud classification algorithm corresponding to the specified cloud classification scenario is called to process the target cloud basic product data using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result. The above method uses cloud classification algorithms corresponding to multiple cloud classification scenarios and generates cloud classification response results under multiple cloud classification scenarios based on cloud basic product data sets corresponding to multiple star sources, so as to alleviate the problem of the limitation of the current cloud classification products in the single application scenario, thereby meeting the needs of different industries or fields for accurate identification and quantitative analysis of cloud features.

[0038] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A schematic diagram of a flow chart of a satellite cloud classification method applicable to multiple scenarios provided by an embodiment of the present invention;

[0042] Figure 2 An overall flow chart of a satellite cloud classification method applicable to multiple scenarios provided by an embodiment of the present invention;

[0043] Figure 3 A DNN model structure diagram provided by an embodiment of the present invention;

[0044] Figure 4 A schematic diagram of ISCCP cloud classification provided by an embodiment of the present invention;

[0045] Figure 5 A schematic diagram of the structure of a satellite cloud classification device applicable to multiple scenarios provided by an embodiment of the present invention;

[0046] Figure 6 A schematic diagram of the structure of a server provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described in combination with the embodiments below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] At present, the current cloud classification results have the limitation of single application scenarios, which makes it difficult to meet the needs of different fields for diversified cloud features. Based on this, the present invention implements a satellite cloud classification method, device, server and medium suitable for multiple scenarios to alleviate the problem of the limitation of single application scenarios of current cloud classification products, thereby meeting the needs of different industries or fields for accurate identification and quantitative analysis of cloud features.

[0049] To facilitate understanding of the present embodiment, a satellite cloud classification method applicable to multiple scenarios disclosed in an embodiment of the present invention is first introduced in detail. The method is applied to a satellite cloud classification system. The satellite cloud classification system is configured with cloud classification algorithms corresponding to multiple cloud classification scenarios. The multiple cloud classification scenarios include one or more of a cloud classification scenario based on the ISCCP classification standard, a cloud classification scenario based on cloud top height, a cloud classification scenario based on cloud components, a cloud classification scenario based on cloud shape, and a custom cloud classification scenario. See Figure 1 The flowchart of a satellite cloud classification method applicable to multiple scenarios is shown in FIG. 1 , and the method mainly includes the following steps S102 to S106:

[0050] Step S102: receiving a cloud classification request sent by a user for a specified cloud classification scene and a specified star source, wherein the cloud classification request carries an identifier of the specified cloud classification scene and an identifier of the specified star source.

[0051] In one example, the front end of a satellite cloud classification system displays a variety of cloud classification scenarios and a variety of star sources through a graphical user interface, responds to a user's selection operation on a cloud classification scenario to determine a specified cloud classification scene, and responds to a user's selection operation on a star source to determine a specified star source, and then generates a cloud classification request based on the identifier of the specified cloud classification scene and the identifier of the specified star source.

[0052] Step S104, based on the identifier of the designated cloud classification scene and the identifier of the designated star source, the target cloud basic product data corresponding to the cloud classification request is determined from the cloud basic product data sets corresponding to the various star sources obtained by pre-inversion.

[0053] In one example, the satellite cloud classification system monitors the satellite platform and timely obtains data from multiple star sources provided by the satellite platform, and then inverts multiple cloud basic product data based on the data to obtain cloud basic product data sets corresponding to multiple star sources. The cloud basic product data sets include target satellite observation data, target satellite cloud detection products, cloud optical thickness inversion products, cloud top pressure inversion products, and cloud top temperature inversion products.

[0054] In one example, different cloud classification scenarios require different cloud basic product data, and the mapping relationship between the identifier of each cloud classification scenario and the type of cloud basic product data can be pre-configured. On this basis, first, the identifier of the specified star source is used as the search condition to filter out the cloud basic product data set corresponding to the specified star source from the cloud basic product data sets corresponding to multiple star sources; then, the identifier of the specified cloud classification scenario is used as the search condition to filter out the target cloud basic product data from the cloud basic product data set corresponding to the specified star source.

[0055] Step S106: Based on the identifier of the specified cloud classification scenario, a cloud classification algorithm corresponding to the specified cloud classification scenario is called to process the target cloud basic product data using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result.

[0056] In one example, the data processing flow corresponding to each cloud classification scenario can be compiled into a cloud classification algorithm in advance, and the target cloud basic product data can be processed using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain the cloud classification response result under the specified cloud classification scenario. The cloud classification response result is used to characterize the cloud classification (also called cloud type) to which the pixel belongs.

[0057] Exemplarily, cloud classifications in cloud classification scenarios based on the ISCCP classification standard include cirrus clouds, cirrostratus clouds, deep convection, altocumulus clouds, high-level clouds, nimbostratus clouds, cumulus clouds, stratocumulus clouds, and stratus clouds; cloud classifications in cloud classification scenarios based on cloud top height include high clouds, middle clouds, low clouds, and straight-spread clouds; cloud classifications in cloud classification scenarios based on cloud components include water clouds, ice clouds, mixed phase clouds, supercooled water clouds, deep convective clouds, precipitation clouds, aerosol and pollution clouds, and dust rolls; cloud classifications in cloud classification scenarios based on cloud shape include layered clouds, heap clouds, stratocumulus clouds, thin clouds, and rain clouds.

[0058] The satellite cloud classification method applicable to multiple scenarios provided in the embodiment of the present invention utilizes cloud classification algorithms corresponding to multiple cloud classification scenarios and generates cloud classification response results under multiple cloud classification scenarios based on cloud basic product data sets corresponding to multiple satellite sources, so as to alleviate the limitation of the current cloud classification products in terms of single application scenarios, thereby meeting the needs of different industries or fields for accurate identification and quantitative analysis of cloud features.

[0059] For ease of understanding, the present invention provides a specific implementation of a satellite cloud classification method applicable to multiple scenarios, see Figure 2 The overall flow chart of a satellite cloud classification method applicable to multiple scenarios is shown, including:

[0060] (1) Satellite source selection:

[0061] In order to improve the autonomous controllable capability, the embodiment of the present invention uses the Fengyun series satellites to carry out the classification of cloud types, including the existing commercial Fengyun-4B satellite, Fengyun-3D satellite (afternoon satellite, passing around 13:30 pm) and Fengyun-3F satellite (morning satellite, passing around 10:30 am). Users can choose to use the following satellite sources according to the scope of the area of ​​concern and the requirements for time frequency.

[0062] Fengyun-4B (FY-4B) is a geostationary satellite located in a geosynchronous orbit about 36,000 kilometers above the equator. The subsatellite coordinates of FY-4B are approximately 104.7°E and 0°N, close to the sky above China. This enables it to continuously monitor meteorological conditions in East Asia and the Western Pacific. FY-4B has the advantage of high frequency, completing a full-disk image every 15 minutes; but its spatial resolution is low, and its cloud products are all 4 km, so cloud classification based on FY-4B is suitable for a large area.

[0063] Fengyun-3D (FY-3D) and Fengyun-3F (FY-3F) are two polar-orbiting meteorological satellites in the Fengyun-3 series. They cover global observations in the afternoon and morning respectively, complement each other, and provide all-weather, multi-level atmospheric and climate data, providing strong technical support for meteorological services, climate research and disaster warning systems. Through joint observations, the two can provide cloud products with a temporal resolution of about 6 hours and a spatial resolution of up to 1 km. Compared with FY-4B, the spatial scale is more refined, so cloud classification with higher spatial resolution can be carried out for key smaller areas.

[0064] The following cloud classification results of various standards can be provided for FY-3D, FY-3F and FY-4B.

[0065] (2) Preparation and inversion of cloud-based products:

[0066] In one implementation, an embodiment of the present invention provides a specific implementation of cloud basic product preparation and inversion, including:

[0067] (2.1) Obtain satellite observation data and satellite cloud detection products from multiple star sources.

[0068] The embodiment of the present invention considers classifying cloud types from different perspectives such as cloud temperature, height, composition, and shape. Prior to this, it is necessary to prepare cloud basic products in advance, including cloud detection, cloud top temperature, cloud top pressure, and cloud optical thickness, so that they can be added as characteristic factors into the cloud classification model.

[0069] The Fengyun Satellite Remote Sensing Data Service Network (https: / / satellite.nsmc.org.cn / ) provides the following data:

[0070] The L1-level observation data and L2-level cloud detection (CLM) products, cloud top temperature (CTT) products, and cloud top pressure (CTP) products of the AGRI sensor of the FY-4B satellite; the L1-level observation data and L2-level cloud detection (CLM) products of the MERSI sensor of the FY-3D satellite; and the L1-level observation data and L2-level cloud detection (CLM) products of the MERSI sensor of the FY-3F satellite.

[0071] The above data and products can be downloaded directly from the website for use. Figure 2 As shown, it is necessary to complete the inversion of cloud optical thickness (COT) products for the FY-4B satellite; and to complete the inversion of cloud top temperature (CTT), cloud top pressure (CTP) and cloud optical thickness (COT) products for the FY-3D and FY-3F satellites.

[0072] (2.2) For any satellite source, obtain the numerical model data and static data corresponding to the satellite observation data of the star source, and perform preprocessing, quality control, time matching and normalization on the satellite observation data, satellite cloud detection products, numerical model data and static data of the star source to obtain the target satellite observation data, target satellite cloud detection products, target numerical model data and target static data corresponding to the star source;

[0073] (2.3) The cloud basic product data set corresponding to the star source is inverted based on the target satellite observation data, the target satellite cloud detection product, the target numerical model data and the target static data through the inversion model corresponding to the star source. The inversion model may include a cloud optical thickness inversion model based on a deep neural network (DNN), a cloud top temperature inversion model based on a deep neural network (DNN), and a cloud top pressure inversion model based on a deep neural network (DNN).

[0074] The embodiments of the present invention provide specific implementation methods for inversion of cloud optical thickness (COT), cloud top temperature (CTT), and cloud top pressure (CTP):

[0075] In one example, the process of cloud optical thickness (COT) inversion is as follows:

[0076] Generally, traditional physical models consider that in the visible light band, the reflectivity of clouds is mainly related to the value of cloud optical thickness; while in the near-infrared or mid-infrared band, the reflectivity of clouds is mainly affected by the effective particle radius of cloud droplets. Therefore, based on satellite dual-channel observations, the radiation transfer model is used to jointly invert cloud optical thickness and cloud effective particle radius. However, in the simulation process using the radiation transfer model (such as RTTOV), water clouds and ice clouds are regarded as single-layer clouds, and the atmosphere is regarded as a parallel atmosphere; but in reality, most clouds are multi-layer clouds, the atmosphere is approximately spherical, and there is no unique solution to the radiation transfer equation, which will cause errors in simulation and calculation.

[0077] To address this problem, an embodiment of the present invention proposes to use a machine learning model to learn to invert cloud optical thickness directly from the input multi-band observation data. Machine learning models have advantages in processing complex inputs and nonlinear relationships, and can overcome the limitations of multiple assumptions of radiation transfer models. Satellite remote sensing data is large in volume and updated frequently. Deep neural networks have good scalability, can process large amounts of data through parallel computing, and are also highly efficient in the estimation process, making them suitable for real-time applications. Therefore, an embodiment of the present invention proposes to use a deep neural network (DNN) model to carry out cloud optical thickness inversion.

[0078] The technical methods are as follows:

[0079] 1) Build a training sample library:

[0080] Since FY-4B, FY-3D, and FY-3F all require cloud optical thickness inversion, when the input satellite data is FY-4B, the corresponding true value is the sunflower-9 cloud optical thickness product; when the input satellite data is FY-3D or FY-3F, the corresponding true value is the MODIS cloud optical thickness product.

[0081] First, we collect satellite observation data (including FY-4B, FY-3D, FY-3F observation channel data, GEO data), satellite cloud detection products (including FY-4B, FY-3D, FY-3F cloud detection products), numerical model data (including 2m atmospheric temperature T2m, relative humidity RH, surface air pressure SP, downlink shortwave radiation DSWRF, total atmospheric column water content TCW, 10-m U wind U10, 10-mV wind V10, total precipitation TP, atmospheric boundary layer height PBLH, vertical velocity VV and other elements), target cloud optical thickness products (H9COT, MODIS COT), static data (including digital elevation model, surface cover type) and other multi-source data.

[0082] Then the above data were preprocessed, quality controlled, time matched and normalized. Computer raster processing tools were used to perform data analysis, projection conversion, spatial scale conversion, time scale aggregation, etc. on multi-source data to obtain a unified resolution (data related to FY-4B COT inversion was unified to 4 km, data related to FY-3D / FCOT inversion was unified to 1 km) and a unified time scale (data related to FY-4B COT inversion was unified to 1 hour, data related to FY-3D / F COT inversion was unified to 1 day) raster dataset in longitude and latitude coordinate systems such as WGS84.

[0083] The main purpose of normalization is to unify the data into a certain range to achieve consistent iteration speed of gradient descent parameters, unify the range of each feature, make the iteration speed consistent, obtain the optimal solution at the same time, converge at the same time, reduce the number of iterations, and speed up the model.

[0084] The normalization method is mainly the maximum and minimum normalization, and its formula is:

[0085]

[0086] Among them, X nom is the normalized value, X is the original value, and X max and X min are the maximum and minimum values ​​of the original sequence respectively.

[0087] 2) DNN model training:

[0088] Deep neural network (DNN) is a technology in the field of machine learning. It is a neural network model for deep learning developed on the basis of perceptron. Figure 3 The structure diagram of a DNN model shown in Figure 1 shows that according to the position of different layers, the neural network layers inside the DNN can be divided into: input layer, hidden layer and output layer. Generally, the first layer is the input layer, the last layer is the output layer, and the layers in between are all hidden layers. The layers are fully connected, that is, any neuron in the i-th layer must be connected to any neuron in the i+1-th layer.

[0089] Like other machine learning methods, the core idea of ​​DNN is to establish a stable nonlinear function between input variables and outputs based on multi-source auxiliary data. In the DNN used in the embodiment of the present invention, the output is H9 COT or MODISCOT, and the input is the channel visible light, near infrared, and long-wave infrared data of the pixel position (including 0.65um reflectivity data, 1.6um reflectivity data, 11um brightness temperature data), numerical model data (T2m, RH, SP, DSWRF, TCW, U10, V10, TP, PBLH, VV and other elements) and auxiliary data describing pixel feature information (including cloud detection products CLM, digital elevation model DEM, surface cover type LC, longitude and latitude Lat / lon, and number of days in the year DOY). Therefore, COT-DNN aims to establish the following relationship:

[0090] COT=f DNN (VIS 0.65 ,VIS 1.6 ,TB 11,T2m,RH,SP,DSWRF,TCW,U10,V10,TP,PBLH,VV,DEM,LC,Lat,Lon,DOY);

[0091] Among them, DEM represents the influence of terrain, and Lat, Lon, and DOY represent the influence of three-dimensional data. In specific implementation, the inversion of FY-4B, FY-3D, and FY-3F cloud optical thickness can be realized based on the above-constructed model.

[0092] In one example, the process of cloud top temperature (CTT) inversion and cloud top pressure (CTP) inversion is as follows:

[0093] The embodiments of the present invention need to carry out cloud top temperature and cloud top pressure inversion for FY-3D and FY-3F. The Satellite Meteorological Center has not publicly provided cloud top temperature and cloud top pressure products of FY-3D and FY-3F on its official website, but its public research provides cloud top temperature and cloud top pressure products based on FY-3D. The algorithm and accuracy test results of the MERSI-II cloud top product, taking the MODIS cloud product as the true value, found that the cloud top temperature accuracy of FY-3D / MERSI-II water clouds is -1.2±4.6K, the cloud top height accuracy is 1.4±1.8km, and the cloud top pressure accuracy is -140.9±114.5hPa; the cloud top temperature accuracy of thick ice clouds is 7.0±6.0K, the cloud top height accuracy is -1.0±0.9km, and the cloud top pressure accuracy is 37.1±36.0hPa; the cloud top temperature accuracy of mixed clouds is 1.5±8.5K, the cloud top height accuracy is 0.8±2.2km, and the cloud top pressure accuracy is -87.4±157.8hPa. The inversion deviations of single-layer cirrus clouds and multi-layer clouds are relatively large. Radiative transfer models play a critical role in the inversion of cloud top properties, but currently there is insufficient understanding of the properties of ice clouds, especially cirrus clouds. Therefore, how to accurately describe the radiation characteristics of ice crystals and improve the simulation accuracy of radiation transfer of ice clouds, especially cirrus clouds, will be the next focus of work.

[0094] Considering that the physical method of using the radiation transfer model to invert cloud top products has limitations that have not yet been overcome, and MODIS cloud products are currently more recognized cloud products, including cloud top temperature, cloud top pressure, etc., which can be downloaded for free on the official website. Therefore, the embodiment of the present invention proposes to use a machine learning method to construct a relationship model between the morning star TerraMODIS cloud product MOD06 and the FY-3F MERSI-III observation data and auxiliary data with the MODIS official cloud product as the true value; and to construct a relationship model between the afternoon star AquaMODIS cloud product MYD06 and the FY-3D MERSI-II observation data and auxiliary data, so as to realize the inversion of FY-3D and FY-3F cloud top temperature and cloud top pressure.

[0095] The machine learning model also chooses to use the deep neural network DNN model. The overall implementation technical idea refers to the cloud optical thickness inversion, which will not be repeated here.

[0096] (3) Cloud classification scenario selection:

[0097] The system has four preset cloud classification scenarios, including Scenario 1: Cloud classification based on ISCCP classification standards; Scenario 2: Cloud classification based on cloud top height; Scenario 3: Cloud classification based on cloud composition; Scenario 4: Cloud classification based on cloud shape. In addition, Scenario 5: Customized cloud classification scenario is also supported, which allows users to customize the cloud classification standards according to their needs.

[0098] Scenario 1: Cloud classification scenario based on the ISCCP classification standard:

[0099] From the cloud basic product data set corresponding to the specified star source, the cloud optical thickness inversion product and the cloud top pressure inversion product are extracted as the target cloud basic product data corresponding to the cloud classification request; the cloud optical thickness inversion product and the cloud top pressure inversion product in the target cloud basic product data are compared with the threshold set in the pre-set ISCCP classification standard to obtain the cloud classification response result in the cloud classification scenario based on the ISCCP classification standard.

[0100] For details, see Figure 4 The diagram shows an ISCCP cloud classification diagram. The ISCCP cloud classification classifies clouds into 9 categories based on cloud top height and cloud optical thickness. It is the most widely used classification standard for cloud classification using satellite remote sensing. Using the cloud top pressure and cloud optical thickness products of the Fengyun series satellites (FY-4B / FY-3D / FY-3F), clouds are divided into 9 categories according to the above classification standards: cirrus, cirrostratus, deep convection, altocumulus, altostratus, nimbostratus, cumulus, stratocumulus and stratus.

[0101] Scenario 2: Cloud classification based on cloud top height:

[0102] From the cloud basic product data set corresponding to the specified star source, the target satellite observation data, cloud optical thickness inversion product, cloud top pressure inversion product and cloud top temperature inversion product are extracted as the target cloud basic product data corresponding to the cloud classification request; the cloud top pressure inversion product in the target cloud basic product data is compared with the thresholds corresponding to high clouds, middle clouds and low clouds respectively, and the brightness temperature data, cloud optical thickness inversion product and cloud top temperature inversion product of the target satellite observation data in the target cloud basic product data are compared with the thresholds corresponding to straight-spread clouds, so as to obtain the cloud classification response results in the cloud classification scenario based on cloud top height.

[0103] Specifically, considering that air pressure and altitude are closely related in the atmosphere, cloud top pressure can reflect the cloud top altitude. According to the ISCCP standard, based on the cloud top pressure product, using the threshold method, clouds can be divided into three types: high clouds, middle clouds, and low clouds. The specific classification standard is as follows:

[0104] High cloud: CTP≤440hpa;

[0105] Medium cloud: 440hpa<CTP≤680hpa;

[0106] Low cloud: CTP>680hpa;

[0107] High clouds mean clouds at an altitude of about 6 kilometers or more, medium clouds have an altitude between 2 and 6 kilometers, and low clouds are usually below 2 kilometers.

[0108] In addition, there are some clouds with a large vertical range, such as Cumulonimbus (Cb), whose cloud tops can extend from the lower layer to the upper layer and even into the stratosphere, and may cover the range of high clouds, middle clouds and low clouds. The cloud base is low but the cloud body is thick, often accompanied by severe convective weather. This type of cloud is defined as a straight-spread cloud. This type of cloud with a large vertical development is not easy to distinguish by cloud top pressure alone, so it is combined with multiple other cloud parameters for comprehensive identification. The identification method is as follows:

[0109] Zhizhan Cloud: Among them, BT 11um This is the brightness temperature data for the 11um band.

[0110] Therefore, in this scenario, clouds are divided into four categories: high clouds, medium clouds, low clouds and straight clouds.

[0111] Scenario 3: Cloud classification based on cloud components:

[0112] Clouds are mainly composed of water droplets, ice crystals and aerosols. The classification of clouds not only helps to understand their physical properties and formation mechanisms, but also has important significance for weather forecasting and climate research. According to the different components of clouds, clouds are divided into eight categories: water clouds, ice clouds, mixed phase clouds, supercooled water clouds, deep convective clouds, precipitation clouds, aerosols and pollution clouds, and dust clouds.

[0113] Among them, water clouds are composed of liquid water droplets, usually appear at lower altitudes, with higher cloud top temperatures and higher reflectivity; ice clouds are composed of ice crystals, usually appear in the upper atmosphere, with lower cloud top temperatures, and have a strong scattering effect on the infrared band; the temperature of mixed phase clouds is usually below 0℃, but they are not completely frozen. The clouds contain supercooled water droplets and ice crystals at the same time, and usually appear in mesosphere clouds or in cumulonimbus clouds and other convective clouds; supercooled water clouds are composed of water droplets with a temperature below 0℃ but still in liquid form, and usually appear in mountainous areas or in the middle and upper atmosphere. Supercooled water clouds pose a greater risk of icing to aircraft; deep convective clouds usually contain a large number of liquid water droplets, supercooled water droplets and ice crystals, with strong vertical development. The cloud tops can reach the stratosphere, and the clouds The top temperature is extremely low and the cloud optical thickness is extremely large. Deep convective clouds are accompanied by weather phenomena such as heavy precipitation, thunderstorms, and hail. Precipitation clouds are mainly composed of liquid water droplets, supercooled water droplets, and ice crystals, and are usually accompanied by continuous precipitation. Their cloud top temperature is relatively high, but their optical thickness is large and their reflectivity is also high. Aerosol and pollution clouds are composed of aerosol particles, water droplets and other pollutants suspended in the atmosphere. They are usually formed over urban areas or industrial areas. They have strong absorption of visible light and short-wave radiation, which often leads to a decrease in atmospheric transparency. Dust clouds contain a large number of sand and dust particles and are commonly seen in desert areas or accompanied by sandstorms. Due to the presence of sand and dust, the cloud optical thickness increases, and the reflection of short-wave radiation is strong, and the scattering and absorption of infrared radiation are more significant.

[0114] Considering the performance characteristics of various types of clouds in visible light, infrared, cloud optical thickness, cloud top temperature, etc., this topic proposes a comprehensive cloud type identification method based on cloud composition. Specifically, the algorithm combines satellite multi-band data and machine learning methods, and uses the band difference (BTD) method, physical model inversion and deep learning technology to accurately classify different types of clouds. By using the multi-band observation data of Fengyun satellites, the algorithm is not only based on simple threshold judgment, but also combines comprehensive characteristics such as cloud optical thickness, cloud top temperature, albedo, etc., which improves the ability to distinguish cloud types and the advancement of the algorithm.

[0115] The implementation methods are as follows:

[0116] a. Data input and processing:

[0117] From the cloud basic product data set corresponding to the specified star source, the target satellite observation data, cloud optical thickness inversion products and cloud top temperature inversion products are extracted as the target cloud basic product data corresponding to the cloud classification request. The target satellite observation data in the target cloud basic product data are subjected to radiation calibration and cloud mask processing to obtain multi-band brightness temperature.

[0118] Specifically, the multi-band data of FY-3D, FY-3F and FY-4A satellites, including visible light and near infrared, short-wave infrared, long-wave infrared and microwave, are used to perform radiation calibration and cloud mask processing on the data:

[0119] Visible light and near infrared: 0.55μm, 0.67μm, 0.87μm;

[0120] Short-wave infrared: 1.6μm, 2.2μm;

[0121] Long-wave infrared: 10.8μm, 11.2μm, 12.0μm;

[0122] Microwave: 6.9GHz, 10.7GHz.

[0123] b. Preliminary classification based on the inter-band difference method: Based on the difference between the brightness temperatures in different bands, the cloud optical thickness inversion product and the cloud top temperature inversion product, the first rough classification result is obtained.

[0124] Specifically, the band difference (BTD) is used in combination with the cloud optical thickness COT and the cloud top temperature CTT to preliminarily distinguish water clouds, ice clouds, mixed phase clouds and supercooled water clouds.

[0125] Water cloud: BTD (11.2-12.0)>0, COT>20, CTT>0℃;

[0126] Ice cloud: BTD (10.8-11.2) < 0, COT < 10, CTT < -20℃;

[0127] Mixed phase cloud: BTD (10.8-11.2) ≈ 0, COT medium, CTT ≈ 0℃;

[0128] Supercooled water cloud: BTD (11.2-12.0)>0, CTT<0℃, mainly liquid water droplets.

[0129] c. Deep learning model fine classification: The second coarse classification result is obtained through the first neural network model based on the target satellite observation data, cloud optical thickness inversion product and cloud top temperature inversion product in the target cloud basic product data. Among them, the first neural network model is obtained by pre-training the convolutional neural network (CNN).

[0130] Specifically, a convolutional neural network (CNN) is used to identify the spatial structure and texture features of clouds. The model input includes the multi-band brightness temperature difference of clouds, cloud optical thickness, cloud top temperature, albedo, etc. By training samples of different types of clouds in the dataset, the model automatically learns the classification rules of complex cloud structures, which is particularly suitable for complex cloud types such as deep convective clouds, dust clouds and aerosol pollution clouds.

[0131] d. Classification rules integrating spectral features and microphysical parameters: Based on the classification rules integrating spectral features and microphysical parameters, the first coarse classification results and the second coarse classification results are corrected to obtain the cloud classification response results in the cloud classification scenario based on cloud components.

[0132] Specifically, for the detailed classification of various types of clouds, the satellite observation spectral characteristics, physical inversion parameters and machine learning model output are integrated, and the final cloud classification judgment is made by using a multi-dimensional feature combination method:

[0133] Deep convective clouds: COT>50, CTT<-50℃, strong vertical development, using the spatial structure information extracted by the CNN model combined with microwave channel data to distinguish deep convective clouds;

[0134] Precipitation clouds: BTD (10.8-11.2) ≈ 0, COT>30, CTT>-10℃, heavy precipitation clouds with significant albedo and higher water path, distinguished by microwave channels;

[0135] Cirrostratus: COT<10, CTT<-20℃, use the convolutional layer of the deep learning model to identify the characteristics of thin clouds;

[0136] Aerosol and pollution cloud: Apparent reflectivity after removing the contribution of Rayleigh scattering of air molecules from the satellite channel with a central wavelength of around 0.47 μm And the ratio of its apparent reflectivity to the satellite channel with a central wavelength near 2.1 μm BTD (11.2-12.0)>0;

[0137] Dust roll: Daytime: VIS1.6 / VIS0.87>1, e ((BT12 -1.0) / BT10.8-1.0) >1, (BT12-BT10.8) / (BT12+BT10.8)+0.001>0; Night: e ((BT12 -1.0) / BT10.8-1.0) >1, (BT12-BT10.8) / (BT12+BT10.8)+0.001>0.

[0138] e. Output cloud classification products according to cloud components

[0139] After the above classification steps, a classification product containing 8 types of clouds is generated and verified by observation data. Quantitative indicators (such as accuracy, recall, and F1 value) are used to evaluate the classification results.

[0140] Scenario 4: Cloud classification based on cloud shape:

[0141] From the cloud basic product data set corresponding to the specified star source, the target satellite observation data is extracted as the target cloud basic product data corresponding to the cloud classification request; two-dimensional satellite images are extracted from the target satellite observation data in the target cloud basic product data, and the two-dimensional satellite images of multiple consecutive time frames are stacked into a three-dimensional data block; through the second neural network, the cloud classification response result in the cloud classification scenario based on the cloud shape is generated based on the three-dimensional data block.

[0142] Specifically, clouds are divided according to their shapes because clouds of different shapes are closely related to meteorological phenomena, weather forecasts, and climate change. Clouds of different shapes usually reflect different meteorological conditions and atmospheric stability. For example, layered clouds are usually associated with mild precipitation and stable weather systems, while pileus clouds are often associated with unstable weather and heavy precipitation. Cirrus clouds are generally not accompanied by precipitation. Based on this, clouds are divided into five categories: layered clouds, pileus clouds, stratocumulus clouds, thin clouds, and rain clouds. By classifying clouds by shape, it can help meteorologists more accurately predict the weather, assess the risk of meteorological disasters, and provide important meteorological information for agriculture, shipping, and other industries. Therefore, this classification has practical application value.

[0143] Considering the multi-scale variation characteristics of clouds in space and time, especially rain clouds and stratus clouds have significant variation characteristics in the time dimension, a 3D convolutional network (3D-CNN) is used to extract the spatiotemporal characteristics of clouds, and then realize the classification of cloud types.

[0144] The implementation methods are as follows:

[0145] a. Data preprocessing:

[0146] Obtain multi-temporal satellite image data of the same area from Fengyun satellites. For each time point, extract the two-dimensional satellite image (height H and width W) of the corresponding area, stack T consecutive time frames to form a three-dimensional data block, and the input shape of each sample is (T, H, W, C), where C is the number of spectral channels (such as visible light, short-wave infrared, long-wave infrared, etc.). Enhance the training set through rotation, flipping, scaling and other technologies to provide the generalization ability of the model, and normalize the image data to the [0,1] interval to accelerate the training process.

[0147] b.3D Convolutional Network Architecture:

[0148] Input layer: The input data shape is (T, H, W, C);

[0149] Convolutional layer: Use convolution kernel size (k T ,k H ,k W ) with an output shape of (T′,H′,W′,N), where N is the number of output feature maps;

[0150]

[0151] Activation layer: Apply activation function (such as ReLU) to perform nonlinear transformation on the convolution result.

[0152] Pooling layer: A 3D max pooling layer is added to reduce the size of the feature map.

[0153]

[0154] Fully connected layer: The extracted features are flattened and connected to the fully connected layer.

[0155] Output layer: Use the softmax activation function to output the probability distribution of each cloud type.

[0156] c. Model training:

[0157] The entire dataset is divided into training set (70%), validation set (15%) and test set (15%) according to the ratio.

[0158] Training parameter settings: Batch size is set to 64; the initial learning rate of learning rate is 0.001, and learning rate scheduling (such as ReduceLROnPlateau) is used; the maximum number of iterations (Epochs) is set to 100, and the early stopping method is used to monitor the validation loss.

[0159] Training process;

[0160] Forward propagation:

[0161] Loss calculation:

[0162] Backward Propagation:

[0163] At the end of each epoch, the model performance is evaluated using the validation set, and the validation loss and accuracy are recorded.

[0164] d. Model evaluation:

[0165] Test set evaluation: Load the model weights that perform best on the validation set, use the test set to make predictions, and obtain the performance of the model on unseen data.

[0166]

[0167] Evaluation indicator calculation:

[0168] Accuracy:

[0169] Accuracy:

[0170] Recall:

[0171] F1-score:

[0172] The trained model is applied to real-time cloud image data to classify cloud types.

[0173] Scenario 5: Customized cloud classification scenario:

[0174] Create a custom cloud classification tool. Users can define the name of the output cloud type in the front-end interactive interface and draw samples of each type of cloud by themselves. Use a deep learning model, consider multiple factors (such as CLM, CTT, CTP, COT, etc.) as variables, and output the corresponding cloud classification according to the sample definition to realize the production of custom cloud classification products.

[0175] The implementation methods are as follows:

[0176] a. Front-end user interface design, including: receiving cloud classification definition requests sent by users to determine custom cloud classifications, receiving cloud classification label annotation requests sent by users to determine custom cloud classification labels corresponding to custom cloud classifications, and receiving cloud basic product data selection requests sent by users to determine custom cloud basic product data from the cloud basic product data sets corresponding to Star Source.

[0177] The details are as follows:

[0178] Cloud type definition: Users can enter a custom cloud type name, multiple types are supported, and a drop-down list is provided to select existing cloud types (such as cumulus, stratus, etc.).

[0179] Sample sketching: Provides image upload function, users can upload satellite images. Use visualization tools (such as rectangle selection tool, polygon drawing tool) to sketch samples. Provides sample label input box, allowing users to enter cloud type labels for the sketched samples.

[0180] Factor selection: Multiple factor options are provided (such as satellite channels, CLM, CTT, COT, CTP, etc.), and users can select the required factors as model input.

[0181] Users can start model training and monitor the training progress. It provides one-click inference function, and users can input new data into the model for cloud classification.

[0182] b. Data processing and feature extraction:

[0183] Data collection: Obtain Fengyun satellite data (such as FY-3D, FY-3F, FY-4B) and extract factor data selected by the user (such as CLM, CTT, CTP, COT, etc.).

[0184] Sample preparation: Generate training sets, validation sets, and test sets based on the sample areas and labels defined by the user.

[0185] c. Deep learning model construction: Use custom cloud basic product data, custom cloud classification and its corresponding custom cloud classification labels to train the third neural network to generate cloud classification response results in the custom cloud classification scenario through the trained third neural network. Specifically, use a convolutional neural network or a 3D convolutional network (3D-CNN) to extract temporal and spatial features, where 3D-CNN requires time series data. The model construction, training, and evaluation methods have been mentioned in scenarios three and four, so I will not repeat them here.

[0186] d. Output cloud classification results: Output cloud classification results, including model performance indicators, etc., and generate visual thematic maps to display cloud classification effects.

[0187] (4) Automatically generate cloud classification reports:

[0188] For the cloud classification response results of any of the above scenarios, a cloud classification result report is automatically generated. The report includes thematic maps of cloud classification results, the area and proportion of each type of cloud. If machine learning or deep learning models are used for cloud classification, the report also includes model evaluation results, showing the classification accuracy, confusion matrix, etc., clearly demonstrating the performance of the model in each category.

[0189] In summary, the purpose of the present invention is to construct a multi-source satellite cloud classification method and device suitable for multiple scenarios to solve the problems and defects of current cloud classification products. In view of the fact that the cloud classification results in the prior art are fixed and cannot meet the needs of different application scenarios, the embodiment of the present invention aims to design a flexible and multifunctional cloud classification method, which can accurately identify and classify the characteristics of the cloud such as height, composition and morphology based on multi-source satellite data according to the needs of different industries or fields, and realize dynamic monitoring and tracking of cloud characteristics to adapt to the changes in clouds in different regions, seasons and weather conditions. In addition, the embodiment of the present invention is completely based on the domestic Fengyun series of satellites, combining the high frequency advantages of the Fengyun-4 geostationary meteorological satellite with the high spatial resolution advantages of the Fengyun-3 polar-orbiting meteorological satellite, breaking the foreign technology monopoly, and improving the independent capabilities in the field of cloud classification and remote sensing monitoring, especially in response to emergencies and natural disasters, to improve the emergency response capabilities of satellite remote sensing data. The satellite cloud classification method suitable for multiple scenarios provided by the embodiment of the present invention has at least the following characteristics:

[0190] (I)Flexible adaptation to various application scenarios:

[0191] The embodiment of the present invention adopts a flexible cloud classification method and device design, so that cloud classification can be customized according to different industries and application requirements, and achieves adaptability to different application scenarios. The method can accurately identify the height, composition and morphology of clouds according to requirements, thereby meeting the special needs of multiple fields such as agriculture, weather forecasting, and environmental monitoring.

[0192] (II) Dynamically monitor changes in cloud characteristics:

[0193] By using the multi-source satellite data of the Fengyun series, the embodiments of the present invention can realize dynamic monitoring and tracking of clouds and accurately depict the spatiotemporal variation characteristics of clouds. The system combines the high-frequency observation of the Fengyun-4 geostationary meteorological satellite and the high spatial resolution of the Fengyun-3 polar-orbiting satellite, and can cope with the changes in clouds in different regions, seasons and weather conditions, effectively improving the accuracy of capturing changes in cloud characteristics.

[0194] (III) Improving emergency response capabilities:

[0195] This method fully relies on the data of the domestic Fengyun series satellites, realizes rapid response in dealing with sudden meteorological events and natural disasters, and enhances the emergency application capability of satellite remote sensing data. Under extreme weather conditions such as typhoons and heavy rains, the device can quickly and accurately provide cloud data, providing timely support for disaster warning and emergency response.

[0196] (IV) Enhance independent innovation and break dependence on foreign technology:

[0197] The embodiment of the present invention breaks the dependence on foreign technology by making full use of domestic satellite data resources, and improves independent innovation capabilities in the fields of cloud classification and remote sensing monitoring. At the same time, based on the unique advantages of the Fengyun satellite series data, the device ensures high accuracy and timeliness of cloud classification, providing support for establishing a stronger voice in the field of international remote sensing technology.

[0198] Based on the above embodiments, the present invention provides an implementation of a satellite cloud classification device applicable to multiple scenarios. The device is applied to a satellite cloud classification system. The satellite cloud classification system is configured with cloud classification algorithms corresponding to multiple cloud classification scenarios. Figure 5 The structure diagram of a satellite cloud classification device suitable for multiple scenarios is shown in FIG. 1 , and the device mainly includes the following parts:

[0199] The request receiving module 502 is used to receive a cloud classification request sent by a user for a specified cloud classification scene and a specified star source, where the cloud classification request carries an identifier of the specified cloud classification scene and an identifier of the specified star source;

[0200] The data determination module 504 is used to determine the target cloud basic product data corresponding to the cloud classification request from the cloud basic product data sets corresponding to the multiple star sources obtained in advance based on the identifier of the specified cloud classification scene and the identifier of the specified star source;

[0201] The cloud classification module 506 is used to call the cloud classification algorithm corresponding to the specified cloud classification scenario based on the identifier of the specified cloud classification scenario, so as to process the target cloud basic product data using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result.

[0202] The satellite cloud classification device suitable for multiple scenarios provided in the embodiment of the present invention utilizes cloud classification algorithms corresponding to multiple cloud classification scenarios and generates cloud classification response results under multiple cloud classification scenarios based on cloud basic product data sets corresponding to multiple satellite sources, so as to alleviate the limitation of the current cloud classification products in terms of single application scenarios, thereby meeting the needs of different industries or fields for accurate identification and quantitative analysis of cloud features.

[0203] In one embodiment, it further includes a data processing module for:

[0204] Obtain satellite observation data and satellite cloud detection products from multiple star sources;

[0205] For any star source, the numerical model data and static data corresponding to the satellite observation data of the star source are obtained, and the satellite observation data, satellite cloud detection products, numerical model data and static data of the star source are preprocessed, quality controlled, time matched and normalized to obtain the target satellite observation data, target satellite cloud detection products, target numerical model data and target static data corresponding to the star source;

[0206] Through the inversion model corresponding to the star source, based on the target satellite observation data, target satellite cloud detection products, target numerical model data and target static data, the cloud basic product data set corresponding to the star source is inverted;

[0207] Among them, the cloud basic product data set includes target satellite observation data, target satellite cloud detection products, cloud optical thickness inversion products, cloud top pressure inversion products and cloud top temperature inversion products.

[0208] In one embodiment, the cloud classification scenario includes a cloud classification scenario based on the ISCCP classification standard, and the cloud classifications in the cloud classification scenario include cirrus, cirrostratus, deep convection, altocumulus, altostratus, nimbostratus, cumulus, stratocumulus, and stratus;

[0209] The data determination module 504 is specifically used to: extract cloud optical thickness inversion products and cloud top pressure inversion products from the cloud basic product data set corresponding to the specified star source as target cloud basic product data corresponding to the cloud classification request;

[0210] The cloud classification module 506 is specifically used to compare the cloud optical thickness inversion product and the cloud top pressure inversion product in the target cloud basic product data with the thresholds set in the pre-set ISCCP classification standard to obtain the cloud classification response results in the cloud classification scenario based on the ISCCP classification standard.

[0211] In one embodiment, the cloud classification scenario includes a cloud classification scenario based on cloud top height, and the cloud classification in the cloud classification scenario includes high clouds, middle clouds, low clouds, and straight clouds;

[0212] The data determination module 504 is specifically used to extract target satellite observation data, cloud optical thickness inversion products, cloud top pressure inversion products and cloud top temperature inversion products from the cloud basic product data set corresponding to the specified star source as target cloud basic product data corresponding to the cloud classification request;

[0213] The cloud classification module 506 is specifically used to compare the cloud top pressure inversion product in the target cloud basic product data with the threshold values ​​corresponding to high clouds, middle clouds and low clouds respectively, and to compare the brightness temperature data, cloud optical thickness inversion product and cloud top temperature inversion product of the target satellite observation data in the target cloud basic product data with the threshold values ​​corresponding to straight-spread clouds, so as to obtain the cloud classification response result in the cloud classification scenario based on the cloud top height.

[0214] In one embodiment, the cloud classification scenario includes a cloud classification scenario based on cloud components, and the cloud classifications in the cloud classification scenario include water clouds, ice clouds, mixed phase clouds, supercooled water clouds, deep convective clouds, precipitation clouds, aerosol and pollution clouds, and dust rolls;

[0215] The data determination module 504 is specifically used to: extract target satellite observation data, cloud optical thickness inversion products and cloud top temperature inversion products from the cloud basic product data set corresponding to the specified star source as target cloud basic product data corresponding to the cloud classification request;

[0216] The cloud classification module 506 is specifically used for: performing radiation calibration and cloud mask processing on the target satellite observation data in the target cloud basic product data to obtain multi-band brightness temperature, and obtaining a first coarse classification result based on the difference between the brightness temperatures in different bands, the cloud optical thickness inversion product and the cloud top temperature inversion product; and obtaining a second coarse classification result based on the target satellite observation data, the cloud optical thickness inversion product and the cloud top temperature inversion product in the target cloud basic product data through a first neural network model; and correcting the first coarse classification result and the second coarse classification result based on the classification rules of the fusion spectral characteristics and microphysical parameters to obtain a cloud classification response result in a cloud classification scenario based on cloud components.

[0217] In one embodiment, the cloud classification scenario includes a cloud classification scenario based on cloud shape, and the cloud classifications in the cloud classification scenario include layered clouds, pile clouds, stratocumulus clouds, thin clouds and rain clouds;

[0218] The data determination module 504 is specifically used to: extract the target satellite observation data from the cloud basic product data set corresponding to the specified star source as the target cloud basic product data corresponding to the cloud classification request;

[0219] The cloud classification module 506 is specifically used to: extract two-dimensional satellite images from the target satellite observation data in the target cloud basic product data, and stack the two-dimensional satellite images of multiple consecutive time frames into a three-dimensional data block; through a second neural network, generate a cloud classification response result in a cloud classification scenario based on cloud shape based on the three-dimensional data block.

[0220] In one embodiment, the cloud classification scenario includes a custom cloud classification scenario and a custom module for:

[0221] Receive a cloud classification definition request sent by a user to determine a custom cloud classification, and receive a cloud classification label annotation request sent by a user to determine a custom cloud classification label corresponding to the custom cloud classification;

[0222] Receive a cloud basic product data selection request sent by a user to determine the custom cloud basic product data from the cloud basic product data set corresponding to the star source;

[0223] The third neural network is trained using the custom cloud basic product data, the custom cloud classification and its corresponding custom cloud classification labels, so as to generate a cloud classification response result in the custom cloud classification scenario through the trained third neural network.

[0224] The device provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0225] An embodiment of the present invention provides a server. Specifically, the server includes a processor and a storage device. The storage device stores a computer program, and when the computer program is executed by the processor, it executes the method described in any one of the above-mentioned embodiments.

[0226] Figure 6 A structural diagram of a server provided in an embodiment of the present invention, the server 100 includes: a processor 60, a memory 61, a bus 62 and a communication interface 63, wherein the processor 60, the communication interface 63 and the memory 61 are connected via the bus 62; the processor 60 is used to execute an executable module stored in the memory 61, such as a computer program.

[0227] The memory 61 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 63 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0228] The bus 62 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0229] Among them, the memory 61 is used to store programs, and the processor 60 executes the program after receiving the execution instruction. The method executed by the device for flow process definition disclosed in any embodiment of the above-mentioned embodiment of the present invention can be applied to the processor 60 or implemented by the processor 60.

[0230] The processor 60 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 60. The above processor 60 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 61, and the processor 60 reads the information in the memory 61 and completes the steps of the above method in combination with its hardware.

[0231] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be referred to the previous method embodiments, which will not be repeated here.

[0232] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0233] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A satellite cloud classification method applicable to multiple scenarios, characterized in that: The method is applied to a satellite cloud classification system, wherein the satellite cloud classification system is configured with cloud classification algorithms corresponding to multiple cloud classification scenarios, and the method comprises: Receiving a cloud classification request sent by a user for a specified cloud classification scene and a specified star source, wherein the cloud classification request carries an identifier of the specified cloud classification scene and an identifier of the specified star source; Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, determining target cloud basic product data corresponding to the cloud classification request from cloud basic product data sets corresponding to multiple star sources obtained by pre-inversion; The cloud classification algorithm corresponding to the specified cloud classification scenario is called based on the identifier of the specified cloud classification scenario, so as to process the target cloud basic product data using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result.

2. The satellite cloud classification method applicable to multiple scenarios according to claim 1, characterized in that: Before receiving a cloud classification request sent by a user for a specified cloud classification scene and a specified star source, the method further includes: Obtain satellite observation data and satellite cloud detection products from multiple star sources; For any of the star sources, the numerical model data and static data corresponding to the satellite observation data of the star source are obtained, and the satellite observation data, the satellite cloud detection product, the numerical model data and the static data of the star source are preprocessed, quality controlled, time matched and normalized to obtain the target satellite observation data, target satellite cloud detection product, target numerical model data and target static data corresponding to the star source; Through the inversion model corresponding to the star source, based on the target satellite observation data, the target satellite cloud detection product, the target numerical model data and the target static data, the cloud basic product data set corresponding to the star source is inverted; Among them, the cloud basic product data set includes the target satellite observation data, the target satellite cloud detection product, cloud optical thickness inversion product, cloud top pressure inversion product and cloud top temperature inversion product.

3. The satellite cloud classification method applicable to multiple scenarios according to claim 2, characterized in that: The cloud classification scenario includes a cloud classification scenario based on the ISCCP classification standard, and the cloud classifications under the cloud classification scenario include cirrus, cirrostratus, deep convection, altocumulus, altostratus, nimbostratus, cumulus, stratocumulus, and stratus; Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, determining target cloud basic product data corresponding to the cloud classification request from cloud basic product data sets corresponding to multiple star sources obtained by pre-inversion, including: extracting the cloud optical thickness inversion product and the cloud top pressure inversion product from the cloud basic product data set corresponding to the designated star source as the target cloud basic product data corresponding to the cloud classification request; The target cloud basic product data is processed by using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result, including: comparing the cloud optical thickness inversion product and the cloud top pressure inversion product in the target cloud basic product data with the thresholds set in the preset ISCCP classification standard to obtain the cloud classification response result under the cloud classification scenario based on the ISCCP classification standard.

4. The satellite cloud classification method applicable to multiple scenarios according to claim 2, characterized in that: The cloud classification scenario includes a cloud classification scenario based on cloud top height, and the cloud classifications in the cloud classification scenario include high clouds, middle clouds, low clouds, and straight clouds; Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, the target cloud basic product data corresponding to the cloud classification request is determined from the cloud basic product data sets corresponding to the multiple star sources obtained by pre-inversion, including: extracting the target satellite observation data, the cloud optical thickness inversion product, the cloud top pressure inversion product and the cloud top temperature inversion product from the cloud basic product data set corresponding to the designated star source as the target cloud basic product data corresponding to the cloud classification request; The target cloud basic product data are processed by using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result, including: comparing the cloud top pressure inversion product in the target cloud basic product data with the thresholds corresponding to the high clouds, the middle clouds, and the low clouds, respectively, and comparing the brightness temperature data of the target satellite observation data, the cloud optical thickness inversion product, and the cloud top temperature inversion product in the target cloud basic product data with the thresholds corresponding to the straight-spread clouds, so as to obtain a cloud classification response result in a cloud classification scenario based on cloud top height.

5. The satellite cloud classification method applicable to multiple scenarios according to claim 2, characterized in that: The cloud classification scenario includes a cloud classification scenario based on cloud components, and the cloud classifications in the cloud classification scenario include water clouds, ice clouds, mixed phase clouds, supercooled water clouds, deep convective clouds, precipitation clouds, aerosols and pollution clouds, and dust rolls; Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, determining the target cloud basic product data corresponding to the cloud classification request from the cloud basic product data sets corresponding to the multiple star sources obtained by pre-inversion, including: extracting the target satellite observation data, the cloud optical thickness inversion product and the cloud top temperature inversion product from the cloud basic product data set corresponding to the designated star source as the target cloud basic product data corresponding to the cloud classification request; The target cloud basic product data are processed by using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result, including: performing radiation calibration and cloud mask processing on the target satellite observation data in the target cloud basic product data to obtain multi-band brightness temperature, and obtaining a first coarse classification result based on the difference between the brightness temperatures in different bands, the cloud optical thickness inversion product and the cloud top temperature inversion product; and obtaining a second coarse classification result based on the target satellite observation data, the cloud optical thickness inversion product and the cloud top temperature inversion product in the target cloud basic product data through a first neural network model; and correcting the first coarse classification result and the second coarse classification result based on a classification rule that integrates spectral features and microphysical parameters to obtain a cloud classification response result in a cloud classification scenario based on cloud components.

6. The satellite cloud classification method applicable to multiple scenarios according to claim 2, characterized in that: The cloud classification scenario includes a cloud classification scenario based on cloud shape, and the cloud classifications in the cloud classification scenario include layered clouds, pile clouds, stratocumulus clouds, thin clouds and rain clouds; Based on the identifier of the designated cloud classification scene and the identifier of the designated star source, determining target cloud basic product data corresponding to the cloud classification request from cloud basic product data sets corresponding to multiple star sources obtained by pre-inversion, including: extracting the target satellite observation data from the cloud basic product data set corresponding to the designated star source as the target cloud basic product data corresponding to the cloud classification request; The target cloud basic product data is processed by using the cloud classification algorithm corresponding to the specified cloud classification scenario to obtain a cloud classification response result, including: extracting a two-dimensional satellite image from the target satellite observation data in the target cloud basic product data, and stacking the two-dimensional satellite images of multiple consecutive time frames into a three-dimensional data block; through a second neural network, generating a cloud classification response result in a cloud classification scenario based on cloud shape based on the three-dimensional data block.

7. The satellite cloud classification method applicable to multiple scenarios according to claim 2, characterized in that: The cloud classification scenario includes a custom cloud classification scenario, and the method includes: receiving a cloud classification definition request sent by the user to determine a custom cloud classification, and receiving a cloud classification tag annotation request sent by the user to determine a custom cloud classification tag corresponding to the custom cloud classification; receiving a cloud basic product data selection request sent by the user to determine custom cloud basic product data from the cloud basic product data set corresponding to the star source; The third neural network is trained using the custom cloud basic product data, the custom cloud classification and its corresponding custom cloud classification label, so as to generate a cloud classification response result in a custom cloud classification scenario through the trained third neural network.

8. A satellite cloud classification device suitable for multiple scenarios, characterized in that: The device is applied to a satellite cloud classification system, wherein the satellite cloud classification system is configured with cloud classification algorithms corresponding to multiple cloud classification scenarios, and the device comprises: A request receiving module, used to receive a cloud classification request sent by a user for a specified cloud classification scene and a specified star source, wherein the cloud classification request carries an identifier of the specified cloud classification scene and an identifier of the specified star source; A data determination module, configured to determine target cloud basic product data corresponding to the cloud classification request from cloud basic product data sets corresponding to a plurality of star sources obtained by pre-inversion based on the identifier of the specified cloud classification scene and the identifier of the specified star source; The cloud classification module is used to call the cloud classification algorithm corresponding to the specified cloud classification scenario based on the identifier of the specified cloud classification scenario, so as to use the cloud classification algorithm corresponding to the specified cloud classification scenario to process the target cloud basic product data to obtain a cloud classification response result.

9. A server, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images