Cloud detection method and device, electronic equipment and readable storage medium

Abstract

The invention relates to the technical field of meteorological services, in particular to a cloud detection method and device, electronic equipment and a readable storage medium. The method comprises the following steps: obtaining a plurality of cloud detection models with different structures obtained by pre-training; obtaining cloud feature data, taking the cloud feature data as input, and inputting the cloud feature data into each cloud detection model to obtain a cloud detection result output by each cloud detection model; and obtaining a target cloud detection result based on cloud detection results output by the plurality of cloud detection models. Wherein the accuracy and the stability of the target cloud detection result are high, and the cloud in the remote sensing image observation data can be effectively detected and identified based on the target cloud detection result.

Description

Technical Field

[0001] The present disclosure relates to the technical field of meteorological services, and in particular to a cloud detection method, device, electronic device and readable storage medium. Background Art

[0002] Satellite remote sensing technology is widely used in various fields due to its advantages of obtaining large-scale observation data in a short period of time, all-weather monitoring, and not being easily affected by weather and terrain. However, global cloud cover data shows that more than 66% of the earth's surface is often covered with clouds, and the distribution of clouds varies in shape, with varying degrees of obstruction to the surface. Considering that most satellite sensors cannot penetrate clouds, the presence of clouds will hinder optical satellites from obtaining useful information on the earth's surface and affect the availability of remote sensing image observation data to varying degrees, resulting in missing and blurred remote sensing image observation data. Therefore, effective detection and identification of clouds in remote sensing image observation data is particularly important for subsequent research and application.

[0003] In the related art, the method for detecting clouds in remote sensing image observation data mainly includes a method based on spectral features. Specifically, this method needs to manually set spectral features, thresholds, etc. based on experience or statistical analysis data, and perform cloud detection based on the set spectral features, thresholds, etc.

[0004] However, the applicant found that there is a complex nonlinear relationship between the spectral characteristics of each channel and the cloud detection results. Relying solely on empirical values ​​and traditional statistical analysis cannot fully reveal the cloud distribution characteristics, resulting in low accuracy and stability of cloud detection results, and it is impossible to effectively detect and identify clouds in remote sensing image observation data based on cloud detection results. Summary of the invention

[0005] In order to solve the problems in the related art, the embodiments of the present disclosure provide a cloud detection method, device, electronic device and readable storage medium.

[0006] In a first aspect, an embodiment of the present disclosure provides a cloud detection method, including:

[0007] Obtain multiple pre-trained cloud detection models with different structures;

[0008] Obtain cloud feature data, and use the cloud feature data as input to input each cloud detection model respectively, so as to obtain cloud detection results output by each cloud detection model;

[0009] The target cloud detection result is obtained based on the cloud detection results output by multiple cloud detection models.

[0010] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection type;

[0011] Obtain target cloud detection results based on cloud detection results output by multiple cloud detection models, including:

[0012] Among the cloud detection types indicated by the cloud detection results output by the multiple cloud detection models, determine the target cloud detection type with the largest number;

[0013] A target cloud detection result indicating a target cloud detection type is generated.

[0014] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection probability;

[0015] Obtain target cloud detection results based on cloud detection results output by multiple cloud detection models, including:

[0016] Get the weight value corresponding to each cloud detection model;

[0017] A weighted average calculation is performed based on the weight value corresponding to each cloud detection model and the cloud detection probability indicated by the cloud detection result output by each cloud detection model to obtain the target cloud detection probability;

[0018] A target cloud detection result indicating a target cloud detection probability is generated.

[0019] In one embodiment of the present disclosure, the method further comprises:

[0020] Obtain the underlying surface type of the area corresponding to the cloud feature data;

[0021] The cloud feature data is used as input to each cloud detection model to obtain the cloud detection results output by each cloud detection model, including:

[0022] Determining a plurality of target cloud detection models corresponding to underlying surface types among the plurality of cloud detection models;

[0023] The cloud feature data is taken as input and input into each target cloud detection model respectively to obtain the cloud detection result output by each target cloud detection model.

[0024] In a second aspect, the present disclosure provides a cloud detection model training method, comprising:

[0025] Obtaining first historical cloud feature data;

[0026] Acquire historical CALIOP data that matches the first historical cloud feature data, wherein a collection time of the first historical cloud feature data matches a collection time of the historical CALIOP data, and a collection range of the first historical cloud feature data matches a collection range of the historical CALIOP data;

[0027] Obtaining cloud detection tags for first historical cloud feature data based on historical CALIOP data;

[0028] Obtain multiple initial cloud detection models with different structures;

[0029] Inputting the first historical cloud feature data into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model;

[0030] Based on the initial cloud detection results and cloud detection labels output by each initial cloud detection model, the model parameters of each initial cloud detection model are adjusted to train multiple cloud detection models.

[0031] In one embodiment of the present disclosure, the method further comprises:

[0032] Acquire the second historical cloud characteristic data and the historical MODIS data corresponding to the second historical cloud characteristic data, wherein the collection time of the second historical cloud characteristic data matches the collection time of the historical MODIS data, and the collection range of the second historical cloud characteristic data matches the collection range of the historical MODIS data;

[0033] The second historical cloud feature data is used as input and input into each trained cloud detection model respectively to obtain a model detection cloud detection result output by each trained cloud detection model;

[0034] Based on the model detection cloud detection results output by the multiple cloud detection models, obtain the target model detection cloud detection result corresponding to the second historical cloud feature data;

[0035] Determine the number of first detection sub-areas where both the target model detection cloud detection result and the historical MODIS data identify as having clouds, the number of second detection sub-areas where the historical MODIS data identify as clouds and the target model detection cloud detection result identify as clear sky, the number of third detection sub-areas where the historical MODIS data identify as clear sky and the target model detection cloud detection result identify as having clouds, and the number of fourth detection sub-areas where both the historical MODIS data and the target model detection cloud detection result identify as clear sky;

[0036] Based on the first detection sub-region number, the second detection sub-region number, the third detection sub-region number, and the fourth detection sub-region number, it is determined whether the multiple cloud detection models meet the preset accuracy requirement.

[0037] In a third aspect, the present disclosure provides a cloud detection device, comprising:

[0038] A model acquisition module is configured to acquire a plurality of pre-trained cloud detection models with different structures;

[0039] A cloud detection module is configured to obtain cloud feature data, and use the cloud feature data as input to input each cloud detection model to obtain a cloud detection result output by each cloud detection model;

[0040] The target detection module is configured to obtain a target cloud detection result based on the cloud detection results output by multiple cloud detection models.

[0041] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection type;

[0042] The target detection module is specifically configured as follows:

[0043] Among the cloud detection types indicated by the cloud detection results output by the multiple cloud detection models, determine the target cloud detection type with the largest number;

[0044] A target cloud detection result indicating a target cloud detection type is generated.

[0045] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection probability;

[0046] The target detection module is specifically configured as follows:

[0047] Get the weight value corresponding to each cloud detection model;

[0048] A weighted average calculation is performed based on the weight value corresponding to each cloud detection model and the cloud detection probability indicated by the cloud detection result output by each cloud detection model to obtain the target cloud detection probability;

[0049] A target cloud detection result indicating a target cloud detection probability is generated.

[0050] In one embodiment of the present disclosure, the device further comprises:

[0051] An underlying surface acquisition module is configured to acquire the underlying surface type of the area corresponding to the cloud feature data;

[0052] There is a cloud detection module, which is specifically configured as follows:

[0053] Determining a plurality of target cloud detection models corresponding to underlying surface types among the plurality of cloud detection models;

[0054] The cloud feature data is taken as input and input into each target cloud detection model respectively to obtain the cloud detection result output by each target cloud detection model.

[0055] In a fourth aspect, the present disclosure provides a cloud detection model training device, comprising:

[0056] A historical data acquisition module, configured to acquire first historical cloud feature data;

[0057] a CALIOP data acquisition module configured to acquire historical CALIOP data matching the first historical cloud feature data, wherein a collection time of the first historical cloud feature data matches a collection time of the historical CALIOP data, and a collection range of the first historical cloud feature data matches a collection range of the historical CALIOP data;

[0058] A detection tag acquisition module configured to acquire a cloud detection tag of the first historical cloud feature data based on the historical CALIOP data;

[0059] An initial model acquisition module is configured to acquire a plurality of initial cloud detection models with different structures;

[0060] A first detection result acquisition module is configured to input the first historical cloud feature data into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model;

[0061] The detection model training module is configured to adjust the model parameters of each initial cloud detection model based on the initial cloud detection results and cloud detection labels output by each initial cloud detection model, so as to train multiple cloud detection models.

[0062] In one embodiment of the present disclosure, the device further comprises:

[0063] A MODIS data acquisition module is configured to acquire second historical cloud characteristic data and historical MODIS data corresponding to the second historical cloud characteristic data, wherein a collection time of the second historical cloud characteristic data matches a collection time of the historical MODIS data, and a collection range of the second historical cloud characteristic data matches a collection range of the historical MODIS data;

[0064] A second detection result acquisition module is configured to take the second historical cloud feature data as input and input it into each trained cloud detection model respectively to obtain a model detection cloud detection result output by each trained cloud detection model;

[0065] A target detection result acquisition module is configured to acquire a target model detection cloud detection result corresponding to the second historical cloud feature data based on the model detection cloud detection results output by the multiple cloud detection models;

[0066] a detection sub-region number acquisition module, configured to determine the number of first detection sub-regions for which both the target model detection cloud detection result and the historical MODIS data identify as having clouds, the number of second detection sub-regions for which the historical MODIS data identify as clouds and the target model detection cloud detection result identify as clear sky, the number of third detection sub-regions for which the historical MODIS data identify as clear sky and the target model detection cloud detection result identify as having clouds, and the number of fourth detection sub-regions for which both the historical MODIS data and the target model detection cloud detection result identify as clear sky;

[0067] The accuracy determination module is configured to determine whether multiple cloud detection models meet preset accuracy requirements based on the first detection sub-area number, the second detection sub-area number, the third detection sub-area number, and the fourth detection sub-area number.

[0068] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement a method as described in any one of the first aspects.

[0069] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed by a processor, the method according to any one of the first aspects is implemented.

[0070] According to the technical solution provided by the embodiments of the present disclosure, multiple cloud detection models with different structures obtained by pre-training are obtained; cloud feature data is obtained, and the cloud feature data is used as input and input into each cloud detection model respectively to obtain the cloud detection result output by each cloud detection model; and the target cloud detection result is obtained based on the cloud detection results output by multiple cloud detection models. Among them, the cloud detection model can be understood as a model that has learned the law between cloud feature data and accurate cloud detection results (i.e., cloud distribution characteristics). At the same time, since the structures of multiple cloud detection models are different, it can be considered that they have learned the above-mentioned cloud distribution characteristics from different angles. Therefore, the target cloud detection results obtained based on the cloud detection results output by multiple cloud detection models have high accuracy and stability, and clouds in remote sensing image observation data can be effectively detected and identified based on the target cloud detection results.

[0071] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Other features, objectives and advantages of the present disclosure will become more apparent through the following detailed description of non-limiting embodiments in conjunction with the accompanying drawings. In the accompanying drawings:

[0073] Figure 1A flow chart of a cloud detection method according to an embodiment of the present disclosure is shown.

[0074] Figure 2 A flowchart of a cloud detection model training method according to an embodiment of the present disclosure is shown.

[0075] Figure 3 A structural block diagram of a cloud detection device according to an embodiment of the present disclosure is shown.

[0076] Figure 4 A structural block diagram of a cloud detection model training device according to an embodiment of the present disclosure is shown.

[0077] Figure 5 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0078] Figure 6 A schematic diagram showing the structure of a computer system suitable for implementing the method according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0079] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.

[0080] In the present disclosure, it should be understood that terms such as "include" or "have" are intended to indicate the presence of features, numbers, steps, behaviors, components, parts, or a combination thereof disclosed in the present specification, and are not intended to exclude the possibility that one or more other features, numbers, steps, behaviors, components, parts, or a combination thereof exist or are added.

[0081] It should also be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0082] In the present disclosure, if it involves operations of obtaining user information or user data or displaying user information or user data to others, the operations are all authorized and confirmed by the user, or actively selected by the user.

[0083] In the related art, the method for detecting clouds in remote sensing image observation data mainly includes a method based on spectral features. Specifically, this method needs to manually set spectral features, thresholds, etc. based on experience or statistical analysis data, and perform cloud detection based on the set spectral features, thresholds, etc.

[0084] However, the applicant found that there is a complex nonlinear relationship between the spectral characteristics of each channel and the cloud detection results. Relying solely on empirical values ​​and traditional statistical analysis cannot fully reveal the cloud distribution characteristics, resulting in low accuracy and stability of cloud detection results, and it is impossible to effectively detect and identify clouds in remote sensing image observation data based on cloud detection results.

[0085] In order to solve the above problems, embodiments of the present disclosure provide a cloud detection method, device, electronic device and readable storage medium.

[0086] According to the technical solution provided by the embodiments of the present disclosure, multiple cloud detection models with different structures obtained by pre-training are obtained; cloud feature data is obtained, and the cloud feature data is used as input and input into each cloud detection model respectively to obtain the cloud detection result output by each cloud detection model; and the target cloud detection result is obtained based on the cloud detection results output by multiple cloud detection models. Among them, the cloud detection model can be understood as a model that has learned the law between cloud feature data and accurate cloud detection results (i.e., cloud distribution characteristics). At the same time, since the structures of multiple cloud detection models are different, it can be considered that they have learned the above-mentioned cloud distribution characteristics from different angles. Therefore, the accuracy and stability of the target cloud detection results obtained based on the cloud detection results output by multiple cloud detection models are both high, and clouds in remote sensing image observation data can be effectively detected and identified based on the target cloud detection results.

[0087] Figure 1 FIG. 1 is a flow chart showing a cloud detection method according to an embodiment of the present disclosure. Figure 1 As shown, the cloud detection method includes the following steps S101-S103:

[0088] In step S101, a plurality of pre-trained cloud detection models with different structures are obtained;

[0089] In step S102, cloud feature data is obtained, and the cloud feature data is used as input to each cloud detection model to obtain a cloud detection result output by each cloud detection model;

[0090] In step S103, a target cloud detection result is obtained based on the cloud detection results output by the multiple cloud detection models.

[0091] In one implementation of the present disclosure, cloud feature data can be understood as data collected by meteorological satellites. Exemplarily, cloud feature data can include full disk data and L1 data of a specified area collected by the Advanced Geostationary Radiation Imager (AGRI) on the Fengyun-4B satellite, where AGRI includes at least 14 channels including visible light, near infrared and infrared bands, and can simultaneously use the on-board black body for high-frequency infrared calibration to ensure the accuracy of the observation data. Historical cloud feature data can be understood as cloud feature data of a specified area at a certain time in history.

[0092] In one implementation of the present disclosure, multiple cloud detection models with different structures may include at least two of a support vector machine (SVM) model, a logistic regression (LR) model, a Naive Bayes (NB) model, a decision tree (DT) model, a random forest (RF) model, and a multi-layer perceptron (MLP) model.

[0093] In one implementation of the present disclosure, a target cloud detection result is obtained based on the cloud detection result output by each cloud detection model. This can be understood as a result obtained by comprehensive calculation based on the cloud detection results output by multiple cloud detection models. For example, if the cloud detection result output by each cloud detection model is a cloud probability, then the target cloud detection result is the average of the cloud probabilities output by multiple cloud detection models, and the average result is determined as the target cloud detection result. Of course, in other embodiments, other calculations such as weighted summation and averaging can be performed on the cloud detection results output by multiple cloud detection models based on other pre-set algorithms to obtain the target cloud detection result.

[0094] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection type, wherein the cloud detection type may include at least two of the following: cloud, possible cloud, possible clear sky, and clear sky;

[0095] Obtaining the target cloud detection result based on the cloud detection results output by multiple cloud detection models can be achieved in the following ways:

[0096] Among the cloud detection types indicated by the cloud detection results output by the multiple cloud detection models, a target cloud detection type with the largest number is determined, and a target cloud detection result indicating the target cloud detection type is generated.

[0097] Exemplarily, if five cloud detection models are pre-trained, and the cloud detection types indicated by the cloud detection results output by the five cloud detection models are respectively "clear sky", "clouds", "clouds", "clouds", and "clear sky", among which "clouds" has the largest number, then a target cloud detection result indicating that the target cloud detection type is "clouds" can be generated.

[0098] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection probability;

[0099] Obtaining the target cloud detection result based on the cloud detection results output by multiple cloud detection models can be achieved in the following ways:

[0100] Get the weight value corresponding to each cloud detection model;

[0101] A weighted average calculation is performed based on the weight value corresponding to each cloud detection model and the cloud detection probability indicated by the cloud detection result output by each cloud detection model to obtain the target cloud detection probability;

[0102] A target cloud detection result indicating a target cloud detection probability is generated.

[0103] For example, if the weight values ​​of the five cloud detection models obtained by pre-training are 1, 1.2, 1.1, 0.9, and 0.8 respectively, and the cloud detection probabilities indicated by the cloud detection results output by the five cloud detection models are "75%", "82%", "86%", "71%", and "84%" respectively, then the target cloud detection probability p can be calculated based on the following formula:

[0104]

[0105] According to the technical solution provided by the embodiments of the present disclosure, multiple cloud detection models with different structures obtained by pre-training are obtained; cloud feature data is obtained, and the cloud feature data is used as input and input into each cloud detection model respectively to obtain the cloud detection result output by each cloud detection model; and the target cloud detection result is obtained based on the cloud detection results output by multiple cloud detection models. Among them, the cloud detection model can be understood as a model that has learned the law between cloud feature data and accurate cloud detection results (i.e., cloud distribution characteristics). At the same time, since the structures of multiple cloud detection models are different, it can be considered that they have learned the above-mentioned cloud distribution characteristics from different angles. Therefore, the accuracy and stability of the target cloud detection results obtained based on the cloud detection results output by multiple cloud detection models are both high, and clouds in remote sensing image observation data can be effectively detected and identified based on the target cloud detection results.

[0106] In one embodiment of the present disclosure, the method further comprises:

[0107] Obtain the underlying surface type of the area corresponding to the cloud feature data;

[0108] The cloud feature data is used as input to each cloud detection model to obtain the cloud detection results output by each cloud detection model, including:

[0109] Determining a plurality of target cloud detection models corresponding to underlying surface types among the plurality of cloud detection models;

[0110] The cloud feature data is taken as input and input into each target cloud detection model respectively to obtain the cloud detection result output by each target cloud detection model.

[0111] In one implementation of the present disclosure, the underlying surface type of the area corresponding to the cloud feature data can be obtained by obtaining the navigation file of the Fengyun-4B meteorological satellite to preliminarily determine the underlying surface type of the corresponding area, and combined with the near-real-time daily global synthetic 5km ice and snow product file (AVH10B1_EOSMLT_NISE) provided by the Advanced Very High Resolution Radiometer (AVHRR) on relevant satellites (such as NOAA14, NOAA16, NOAA17, EOSMLT, MetOpA, etc.) to update the area belonging to snow in the underlying surface type in real time, thereby improving the underlying surface type of the area corresponding to the cloud feature data. The underlying surface types include deep sea, shallow water, land without snow, snow and desert.

[0112] In one implementation of the present disclosure, determining multiple target cloud detection models corresponding to the underlying surface type from multiple cloud detection models can be understood as determining multiple target cloud detection models corresponding to the underlying surface type based on a preset correspondence between the cloud detection model and the underlying surface type. The preset correspondence between the cloud detection model and the underlying surface type can be understood as obtaining cloud feature data corresponding to the corresponding underlying surface type in advance, inputting the cloud feature data corresponding to the underlying surface type into different cloud detection models for processing, obtaining the accuracy of the cloud detection result, and determining the cloud detection model with an accuracy greater than a preset accuracy threshold as the cloud detection model corresponding to the corresponding underlying surface type.

[0113] According to the technical solution provided by the embodiments of the present disclosure, by obtaining the underlying surface type of the area corresponding to the cloud feature data, multiple target cloud detection models corresponding to the underlying surface type are determined in multiple cloud detection models; the cloud feature data is used as input and input into each target cloud detection model respectively to obtain the cloud detection results output by each target cloud detection model, which can further improve the accuracy of the obtained cloud detection results.

[0114] Figure 2FIG. 1 is a flow chart showing a cloud detection model training method according to an embodiment of the present disclosure. Figure 2 As shown, the cloud detection model training method includes the following steps S201-S206:

[0115] In step S201, first historical cloud feature data is obtained;

[0116] In step S202, historical CALIOP data matching the first historical cloud feature data is obtained;

[0117] wherein the collection time of the first historical cloud feature data matches the collection time of the historical CALIOP data, and the collection range of the first historical cloud feature data matches the collection range of the historical CALIOP data;

[0118] In step S203, a cloud detection tag of the first historical cloud feature data is obtained based on the historical CALIOP data;

[0119] In step S204, a plurality of initial cloud detection models with different structures are obtained;

[0120] In step S205, the first historical cloud feature data is input into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model;

[0121] In step S206, based on the initial cloud detection results and cloud detection labels output by each initial cloud detection model, the model parameters of each initial cloud detection model are adjusted to train and obtain multiple cloud detection models.

[0122] In one implementation of the present disclosure, the collection time of the first historical cloud feature data matches the collection time of the historical CALIOP data. It can be understood that the collection time of the first historical cloud feature data is consistent with the collection time of the historical CALIOP data, or it can be understood that the time difference between the collection time of the first historical cloud feature data and the collection time of the historical CALIOP data is less than or equal to a preset time difference threshold.

[0123] In one implementation of the present disclosure, the collection range of the first historical cloud feature data matches the collection range of the historical CALIOP data. It can be understood that the collection range of the first historical cloud feature data is consistent with the collection range of the historical CALIOP data, or it can be understood that the ratio of the overlapping part of the collection range of the first historical cloud feature data and the collection range of the historical CALIOP data is greater than or equal to a preset ratio threshold.

[0124] Exemplarily, determining whether the collection range of historical cloud feature data is consistent with the collection range of historical CALIOP data can be understood as calculating the spherical distance between the pixel in the historical cloud feature data and the corresponding pixel in the historical CALIOP data. If the spherical distance is less than or equal to a preset spherical distance threshold, it is determined that the collection ranges of the two are consistent.

[0125] In one implementation of the present disclosure, the multiple initial cloud detection models with different structures may include at least two of a support vector machine model, a logistic regression model, a naive Bayes model, a decision tree model, a random forest model, and a multilayer perceptron model.

[0126] In one implementation of the present disclosure, polarized cloud-aerosol detection lidar (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation, CALIOP) data is data collected by the CALIOP payload on the Calypso (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation, CALIPSO) satellite. The CALIOP payload can measure the two orthogonal polarization components of 532nm backscatter and the total intensity of 1064nm backscatter. The physical interval between lidar footprints (center to center) is 335m. On average, a 1km cloud product can be obtained for every three CALIOP footprints. The cloud product provides cloud detection information at a horizontal resolution of 1km, including the number of cloud distribution layers in the vertical direction, the height of the layered cloud tops and other information. Historical CALIOP data can be understood as CALIOP data for a specified area at a certain moment in history.

[0127] In one implementation of the present disclosure, the acquisition range corresponding to the first historical cloud feature data may include multiple detection sub-areas, and the acquisition range corresponding to the historical CALIOP data may include multiple CALIOP sub-areas. It is understandable that the CALIOP sub-area may be a pixel in the historical CALIOP data or a region corresponding to multiple adjacent pixels, and each data in the historical CALIOP data carries attribute information of whether the pixel corresponding to the data has clouds; the detection sub-area may be a pixel in the first historical cloud feature data or a region corresponding to multiple adjacent pixels.

[0128] In one implementation of the present disclosure, obtaining a cloud detection label for first historical cloud feature data based on historical CALIOP data can be understood as, based on the position information of the detection sub-area in the collection range of the first historical cloud feature data, determining at least one CALIOP sub-area whose collection range is consistent with the collection range of the detection sub-area, or whose collection range overlaps with the collection range of the detection sub-area at a ratio greater than or equal to an overlap ratio threshold, then determining partial data corresponding to the at least one CALIOP sub-area in the historical CALIOP data, and based on attribute information carried by the partial data (the attribute information identifies information on whether the corresponding data has clouds), determining whether each CALIOP sub-area in the at least one CALIOP sub-area has clouds, and then based on position conversion, determining whether the detection sub-area corresponding to the at least one CALIOP sub-area has clouds, and then obtaining a cloud detection label for the corresponding data of the corresponding detection sub-area in the first historical cloud feature data.

[0129] According to the technical solution provided by the embodiment of the present disclosure, first historical cloud feature data is obtained; historical CALIOP data matching the first historical cloud feature data is obtained; cloud detection labels of the first historical cloud feature data are obtained based on the historical CALIOP data; multiple initial cloud detection models with different structures are obtained; the first historical cloud feature data is input into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model; based on the initial cloud detection result and cloud detection label output by each initial cloud detection model, model parameters of each initial cloud detection model are adjusted to train multiple cloud detection models. Among them, based on the historical CALIOP data, the cloud detection label of the corresponding data of the corresponding detection sub-area in the first historical cloud feature data can be determined more accurately. The cloud detection label can be used to verify the correctness of the cloud detection result. Therefore, the multiple cloud detection models obtained by training can be understood as models that have learned the rules between cloud feature data and cloud distribution characteristics. At the same time, due to the different structures of the multiple cloud detection models, it can be considered that the multiple cloud detection models have learned the above-mentioned cloud distribution characteristics from different angles. Therefore, the above-mentioned scheme can ensure that the multiple cloud detection models obtained by training can output relatively accurate cloud detection results, which helps to improve the accuracy of the target cloud detection results.

[0130] In one embodiment of the present disclosure, the method further comprises:

[0131] Acquire the second historical cloud characteristic data and the historical MODIS data corresponding to the second historical cloud characteristic data, wherein the collection time of the second historical cloud characteristic data matches the collection time of the historical MODIS data, and the collection range of the second historical cloud characteristic data matches the collection range of the historical MODIS data;

[0132] The second historical cloud feature data is used as input and input into each trained cloud detection model respectively to obtain a model detection cloud detection result output by each trained cloud detection model;

[0133] Based on the model detection cloud detection results output by the multiple cloud detection models, obtain the target model detection cloud detection result corresponding to the second historical cloud feature data;

[0134] Determine the number of first detection sub-areas where both the target model detection cloud detection result and the historical MODIS data identify as having clouds, the number of second detection sub-areas where the historical MODIS data identify as clouds and the target model detection cloud detection result identify as clear sky, the number of third detection sub-areas where the historical MODIS data identify as clear sky and the target model detection cloud detection result identify as having clouds, and the number of fourth detection sub-areas where both the historical MODIS data and the target model detection cloud detection result identify as clear sky;

[0135] Based on the first detection sub-region number, the second detection sub-region number, the third detection sub-region number, and the fourth detection sub-region number, it is determined whether the multiple cloud detection models meet the preset accuracy requirement.

[0136] In one implementation of the present disclosure, the Moderate-resolution Imaging Spectroradiometer (MODIS) data is data collected by the Moderate-resolution Imaging Spectroradiometer, which is carried on the Earth Observation System (EOS) Terra AM satellite and the Earth Observation System Aqua PM satellite. It is used to understand the changes in the global climate and the impact of human activities on the climate. The Moderate-resolution Imaging Spectroradiometer can capture data in 36 mutually aligned spectral bands, covering from visible light to infrared bands, and provide Earth surface observation data once every 1 to 2 days. The entire Earth surface can be repeatedly observed to obtain observation data in 36 bands. The Moderate-resolution Imaging Spectroradiometer is designed to provide dynamic measurements of large-scale global data, including changes in cloud cover, changes in Earth energy radiation, changes in oceans, land, and low-altitude processes. Historical MODIS data can be understood as MODIS data for a specified area at a certain time in history.

[0137] In one implementation of the present disclosure, the collection time of the second historical cloud characteristic data matches the collection time of the historical MODIS data, which can be understood as the collection time of the second historical cloud characteristic data is consistent with the collection time of the historical MODIS data, or it can be understood as the time difference between the collection time of the second historical cloud characteristic data and the collection time of the historical MODIS data is less than or equal to a preset time difference threshold.

[0138] In one implementation of the present disclosure, the collection range of the second historical cloud feature data matches the collection range of the historical MODIS data, which can be understood as the collection range of the second historical cloud feature data is consistent with the collection range of the historical MODIS data, or it can be understood as the ratio of the overlapping part of the collection range of the second historical cloud feature data and the collection range of the historical MODIS data is greater than or equal to a preset ratio threshold.

[0139] In one implementation of the present disclosure, the acquisition range corresponding to the second historical cloud feature data may include multiple detection sub-areas, and the acquisition range corresponding to the historical MODIS data may include multiple MODIS sub-areas. It is understandable that the MODIS sub-area may be a pixel in the historical MODIS data or a region corresponding to multiple adjacent pixels, and each data in the historical MODIS data carries attribute information of whether the pixel corresponding to the data has clouds; the detection sub-area may also be a pixel in the second historical cloud feature data or a region corresponding to multiple adjacent pixels.

[0140] In one implementation of the present disclosure, determining the detection sub-region identified as having clouds or clear sky by MODIS data can be understood as, based on the location information of the detection sub-region in the collection range of the second historical cloud feature data, determining at least one MODIS sub-region whose collection range is consistent with the collection range of the detection sub-region, or whose collection range overlaps with the collection range of the detection sub-region by a ratio greater than or equal to the overlap ratio threshold, then determining partial data corresponding to the at least one MODIS sub-region in the historical MODIS data, and based on the attribute information carried by the partial data, determining whether each MODIS sub-region in the at least one MODIS sub-region is identified as having clouds, and then determining whether the detection sub-region corresponding to the at least one MODIS sub-region is identified as having clouds based on the position conversion. If it is determined that the corresponding detection sub-region is identified as having clouds, the corresponding detection sub-region is a detection sub-region identified as having clouds by the historical MODIS data; if it is determined that the corresponding detection sub-region is identified as clear sky, the corresponding detection sub-region is a detection sub-region identified as having clear sky by the historical MODIS data.

[0141] In one implementation of the present disclosure, determining whether multiple cloud detection models meet preset accuracy requirements based on the first detection sub-region number, the second detection sub-region number, the third detection sub-region number, and the fourth detection sub-region number can be implemented as follows:

[0142] Substitute the first detection sub-region number a, the second detection sub-region number b, the third detection sub-region number c and the fourth detection sub-region number d into the following formula to calculate;

[0143] Calculate cloud accuracy based on PODcld=a / (a+b) POD c ld ;

[0144] Calculate the clear sky accuracy based on PODclr=d / (d+c) POD c l r ;

[0145] Calculate the cloud false alarm rate based on FARcld=c / (a+c) FAR c ld ;

[0146] Calculate the clear sky false alarm rate FARclr based on FARclr = b / (b+d);

[0147] Calculate the overall cloud recognition accuracy HR based on HR = (a + b) / (a ​​+ b + c + d);

[0148] If the cloud accuracy PODcld is greater than the cloud accuracy threshold, the clear sky accuracy PODclr is greater than the clear sky accuracy threshold, the cloud false alarm rate FARcld is less than the cloud false alarm rate threshold, the clear sky false alarm rate FARclr is less than the clear sky false alarm rate threshold and the overall cloud recognition accuracy HR is greater than the overall cloud recognition accuracy threshold, then it is determined that multiple cloud detection models meet the preset accuracy requirements.

[0149] According to the technical solution provided by the embodiment of the present disclosure, by obtaining the second historical cloud feature data and the historical MODIS data corresponding to the second historical cloud feature data, the second historical cloud feature data is used as input and input into each trained cloud detection model respectively to obtain the model detection cloud detection result output by each trained cloud detection model; based on the model detection cloud detection results output by multiple cloud detection models, the target model detection cloud detection result corresponding to the second historical cloud feature data is obtained; the number of first detection sub-regions, the number of second detection sub-regions, the number of third detection sub-regions and the number of fourth detection sub-regions are determined; based on the number of first detection sub-regions, the number of second detection sub-regions, the number of third detection sub-regions and the number of fourth detection sub-regions, it is determined whether multiple cloud detection models meet the preset accuracy requirements. Based on the above scheme, it can be further ensured that multiple cloud detection models that meet the preset accuracy requirements can output relatively accurate cloud detection results, which is helpful to improve the accuracy of the target cloud detection results obtained subsequently.

[0150] Figure 3 The following is a structural block diagram of a cloud detection device according to an embodiment of the present disclosure, wherein the device can be implemented as part or all of an electronic device through software, hardware, or a combination of both.

[0151] like Figure 3 As shown, the cloud detection device 300 includes:

[0152] A model acquisition module 301 is configured to acquire a plurality of pre-trained cloud detection models with different structures;

[0153] A cloud detection module 302 is configured to obtain cloud feature data, and use the cloud feature data as input to input each cloud detection model to obtain a cloud detection result output by each cloud detection model;

[0154] The target detection module 303 is configured to obtain a target cloud detection result based on the cloud detection results output by the multiple cloud detection models.

[0155] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection type;

[0156] The target detection module 303 is configured as follows:

[0157] Among the cloud detection types indicated by the cloud detection results output by the multiple cloud detection models, determine the target cloud detection type with the largest number;

[0158] A target cloud detection result indicating a target cloud detection type is generated.

[0159] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection probability;

[0160] The target detection module 303 is specifically configured as follows:

[0161] Get the weight value corresponding to each cloud detection model;

[0162] A weighted average calculation is performed based on the weight value corresponding to each cloud detection model and the cloud detection probability indicated by the cloud detection result output by each cloud detection model to obtain the target cloud detection probability;

[0163] A target cloud detection result indicating a target cloud detection probability is generated.

[0164] In one embodiment of the present disclosure, the device further comprises:

[0165] An underlying surface acquisition module is configured to acquire the underlying surface type of the area corresponding to the cloud feature data;

[0166] There is a cloud detection module 302, which is specifically configured as follows:

[0167] Determining a plurality of target cloud detection models corresponding to underlying surface types among the plurality of cloud detection models;

[0168] The cloud feature data is taken as input and input into each target cloud detection model respectively to obtain the cloud detection result output by each target cloud detection model.

[0169] According to the technical solution provided by the embodiments of the present disclosure, multiple cloud detection models with different structures obtained by pre-training are obtained; cloud feature data is obtained, and the cloud feature data is used as input and input into each cloud detection model respectively to obtain the cloud detection result output by each cloud detection model; and the target cloud detection result is obtained based on the cloud detection results output by multiple cloud detection models. Among them, the cloud detection model can be understood as a model that has learned the law between cloud feature data and accurate cloud detection results (i.e., cloud distribution characteristics). At the same time, since the structures of multiple cloud detection models are different, it can be considered that they have learned the above-mentioned cloud distribution characteristics from different angles. Therefore, the accuracy and stability of the target cloud detection results obtained based on the cloud detection results output by multiple cloud detection models are both high, and clouds in remote sensing image observation data can be effectively detected and identified based on the target cloud detection results.

[0170] The cloud detection device in this embodiment corresponds to the cloud detection method in the above embodiment. For specific details, please refer to the description of the cloud detection method above, which will not be repeated here.

[0171] Figure 4 The structural block diagram of the cloud detection model training device according to the embodiment of the present disclosure is shown. The device can be implemented as part or all of the electronic device through software, hardware or a combination of both.

[0172] like Figure 4 As shown, the cloud detection model training device 400 includes:

[0173] The historical data acquisition module 401 is configured to acquire first historical cloud feature data;

[0174] A CALIOP data acquisition module 402 is configured to acquire historical CALIOP data that matches the first historical cloud feature data, wherein a collection time of the first historical cloud feature data matches a collection time of the historical CALIOP data, and a collection range of the first historical cloud feature data matches a collection range of the historical CALIOP data;

[0175] A detection tag acquisition module 403 is configured to acquire a cloud detection tag of the first historical cloud feature data based on the historical CALIOP data;

[0176] An initial model acquisition module 404 is configured to acquire a plurality of initial cloud detection models with different structures;

[0177] The first detection result acquisition module 405 is configured to input the first historical cloud feature data into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model;

[0178] The detection model training module 406 is configured to adjust the model parameters of each initial cloud detection model based on the initial cloud detection results and cloud detection labels output by each initial cloud detection model, so as to train and obtain multiple cloud detection models.

[0179] In one embodiment of the present disclosure, the cloud detection model training device 400 further includes:

[0180] A MODIS data acquisition module is configured to acquire second historical cloud characteristic data and historical MODIS data corresponding to the second historical cloud characteristic data, wherein a collection time of the second historical cloud characteristic data matches a collection time of the historical MODIS data, and a collection range of the second historical cloud characteristic data matches a collection range of the historical MODIS data;

[0181] A second detection result acquisition module is configured to take the second historical cloud feature data as input and input it into each trained cloud detection model respectively to obtain a model detection cloud detection result output by each trained cloud detection model;

[0182] A target detection result acquisition module is configured to acquire a target model detection cloud detection result corresponding to the second historical cloud feature data based on the model detection cloud detection results output by the multiple cloud detection models;

[0183] a detection sub-region number acquisition module, configured to determine the number of first detection sub-regions for which both the target model detection cloud detection result and the historical MODIS data identify as having clouds, the number of second detection sub-regions for which the historical MODIS data identify as clouds and the target model detection cloud detection result identify as clear sky, the number of third detection sub-regions for which the historical MODIS data identify as clear sky and the target model detection cloud detection result identify as having clouds, and the number of fourth detection sub-regions for which both the historical MODIS data and the target model detection cloud detection result identify as clear sky;

[0184] The accuracy determination module is configured to determine whether multiple cloud detection models meet preset accuracy requirements based on the first detection sub-area number, the second detection sub-area number, the third detection sub-area number, and the fourth detection sub-area number.

[0185] According to the technical solution provided by the embodiment of the present disclosure, first historical cloud feature data is obtained; historical CALIOP data matching the first historical cloud feature data is obtained; cloud detection labels of the first historical cloud feature data are obtained based on the historical CALIOP data; multiple initial cloud detection models with different structures are obtained; the first historical cloud feature data is input into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model; based on the initial cloud detection result and cloud detection label output by each initial cloud detection model, model parameters of each initial cloud detection model are adjusted to train multiple cloud detection models. Among them, based on the historical CALIOP data, the cloud detection label of the corresponding data of the corresponding detection sub-area in the first historical cloud feature data can be determined more accurately. The cloud detection label can be used to verify the correctness of the cloud detection result. Therefore, the multiple cloud detection models obtained by training can be understood as models that have learned the rules between cloud feature data and cloud distribution characteristics. At the same time, due to the different structures of the multiple cloud detection models, it can be considered that the multiple cloud detection models have learned the above-mentioned cloud distribution characteristics from different angles. Therefore, the above-mentioned scheme can ensure that the multiple cloud detection models obtained by training can output relatively accurate cloud detection results, which helps to improve the accuracy of the target cloud detection results.

[0186] The cloud detection model training device in this embodiment corresponds to the cloud detection model training method in the above embodiment. For specific details, please refer to the description of the cloud detection model training method above, which will not be repeated here.

[0187] The present disclosure also discloses an electronic device, Figure 5 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0188] like Figure 5 As shown, the electronic device includes a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method according to an embodiment of the present disclosure.

[0189] In a first aspect, an embodiment of the present disclosure provides a cloud detection method, including:

[0190] Obtain multiple pre-trained cloud detection models with different structures;

[0191] Obtain cloud feature data, and use the cloud feature data as input to input each cloud detection model respectively, so as to obtain cloud detection results output by each cloud detection model;

[0192] The target cloud detection result is obtained based on the cloud detection results output by multiple cloud detection models.

[0193] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection type;

[0194] Obtain target cloud detection results based on cloud detection results output by multiple cloud detection models, including:

[0195] Among the cloud detection types indicated by the cloud detection results output by the multiple cloud detection models, determine the target cloud detection type with the largest number;

[0196] A target cloud detection result indicating a target cloud detection type is generated.

[0197] In one embodiment of the present disclosure, the cloud detection result is used to indicate the cloud detection probability;

[0198] Obtain target cloud detection results based on cloud detection results output by multiple cloud detection models, including:

[0199] Get the weight value corresponding to each cloud detection model;

[0200] A weighted average calculation is performed based on the weight value corresponding to each cloud detection model and the cloud detection probability indicated by the cloud detection result output by each cloud detection model to obtain the target cloud detection probability;

[0201] A target cloud detection result indicating a target cloud detection probability is generated.

[0202] In one embodiment of the present disclosure, the method further comprises:

[0203] Obtain the underlying surface type of the area corresponding to the cloud feature data;

[0204] The cloud feature data is used as input to each cloud detection model to obtain the cloud detection results output by each cloud detection model, including:

[0205] Determining a plurality of target cloud detection models corresponding to underlying surface types among the plurality of cloud detection models;

[0206] The cloud feature data is taken as input and input into each target cloud detection model respectively to obtain the cloud detection result output by each target cloud detection model.

[0207] In a second aspect, the present disclosure provides a cloud detection model training method, comprising:

[0208] Obtaining first historical cloud feature data;

[0209] Acquire historical CALIOP data that matches the first historical cloud feature data, wherein a collection time of the first historical cloud feature data matches a collection time of the historical CALIOP data, and a collection range of the first historical cloud feature data matches a collection range of the historical CALIOP data;

[0210] Obtaining cloud detection tags for first historical cloud feature data based on historical CALIOP data;

[0211] Obtain multiple initial cloud detection models with different structures;

[0212] Inputting the first historical cloud feature data into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model;

[0213] Based on the initial cloud detection results and cloud detection labels output by each initial cloud detection model, the model parameters of each initial cloud detection model are adjusted to train multiple cloud detection models.

[0214] In one embodiment of the present disclosure, the method further comprises:

[0215] Acquire the second historical cloud characteristic data and the historical MODIS data corresponding to the second historical cloud characteristic data, wherein the collection time of the second historical cloud characteristic data matches the collection time of the historical MODIS data, and the collection range of the second historical cloud characteristic data matches the collection range of the historical MODIS data;

[0216] The second historical cloud feature data is used as input and input into each trained cloud detection model respectively to obtain a model detection cloud detection result output by each trained cloud detection model;

[0217] Based on the model detection cloud detection results output by the multiple cloud detection models, obtain the target model detection cloud detection result corresponding to the second historical cloud feature data;

[0218] Determine the number of first detection sub-areas where both the target model detection cloud detection result and the historical MODIS data identify as having clouds, the number of second detection sub-areas where the historical MODIS data identify as clouds and the target model detection cloud detection result identify as clear sky, the number of third detection sub-areas where the historical MODIS data identify as clear sky and the target model detection cloud detection result identify as having clouds, and the number of fourth detection sub-areas where both the historical MODIS data and the target model detection cloud detection result identify as clear sky;

[0219] Based on the first detection sub-region number, the second detection sub-region number, the third detection sub-region number, and the fourth detection sub-region number, it is determined whether the multiple cloud detection models meet the preset accuracy requirement.

[0220] Figure 6 A schematic diagram showing the structure of a computer system suitable for implementing the method according to an embodiment of the present disclosure is shown.

[0221] like Figure 6 As shown, the computer system includes a processing unit, which can perform the various methods in the above-mentioned embodiments according to the program stored in the read-only memory (ROM) or the program loaded from the storage part into the random access memory (RAM). In the RAM, various programs and data required for the operation of the computer system are also stored. The processing unit, ROM and RAM are connected to each other through a bus. The input / output (I / O) interface is also connected to the bus.

[0222] The following components are connected to the I / O interface: an input part including a keyboard, a mouse, etc.; an output part including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage part including a hard disk, etc.; and a communication part including a network interface card such as a LAN card, a modem, etc. The communication part performs a communication process via a network such as the Internet. The drive is also connected to the I / O interface as needed. Removable media, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on the drive as needed so that the computer program read therefrom is installed into the storage part as needed. Among them, the processing unit can be implemented as a processing unit such as a CPU, a GPU, a TPU, an FPGA, an NPU, etc.

[0223] In particular, according to an embodiment of the present disclosure, the method described above can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the above method. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium.

[0224] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0225] The units or modules involved in the embodiments described in the present disclosure may be implemented by software or programmable hardware. The units or modules described may also be set in a processor, and the names of these units or modules do not constitute limitations on the units or modules themselves in some cases.

[0226] As another aspect, the present disclosure further provides a computer-readable storage medium, which may be a computer-readable storage medium included in the electronic device or computer system in the above embodiment; or a computer-readable storage medium that exists independently and is not assembled into a device. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the method described in the present disclosure.

[0227] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present disclosure is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other.

Claims

1. A cloud detection method, characterized in that: include: Obtain multiple pre-trained cloud detection models with different structures; Acquire cloud feature data, and use the cloud feature data as input to each cloud detection model, respectively, to obtain a cloud detection result output by each cloud detection model; The target cloud detection result is obtained based on the cloud detection results output by multiple cloud detection models.

2. The cloud detection method according to claim 1, characterized in that: The cloud detection result is used to indicate the cloud detection type; The step of obtaining a target cloud detection result based on the cloud detection results output by the multiple cloud detection models includes: Determine the target cloud detection type with the largest number among the cloud detection types indicated by the cloud detection results output by the multiple cloud detection models; A target cloud detection result indicating the target cloud detection type is generated.

3. The cloud detection method according to claim 1, characterized in that: The cloud detection result is used to indicate the cloud detection probability; The step of obtaining a target cloud detection result based on the cloud detection results output by the multiple cloud detection models includes: Get the weight value corresponding to each cloud detection model; A weighted average calculation is performed based on the weight value corresponding to each cloud detection model and the cloud detection probability indicated by the cloud detection result output by each cloud detection model to obtain the target cloud detection probability; A target cloud detection result indicating a probability of detecting the target cloud is generated.

4. The cloud detection method according to claim 1, characterized in that: The method further comprises: Obtaining the underlying surface type of the area corresponding to the cloud feature data; The cloud feature data is used as input and input into each cloud detection model respectively to obtain a cloud detection result output by each cloud detection model, including: Determining a plurality of target cloud detection models corresponding to the underlying surface type from among the plurality of cloud detection models; The cloud feature data is used as input and input into each target cloud detection model respectively to obtain a cloud detection result output by each target cloud detection model.

5. A cloud detection model training method, characterized in that: include: Obtaining first historical cloud feature data; Acquire historical CALIOP data matching the first historical cloud feature data, wherein a collection time of the first historical cloud feature data matches a collection time of the historical CALIOP data, and a collection range of the first historical cloud feature data matches a collection range of the historical CALIOP data; Acquire cloud detection tags of the first historical cloud feature data based on historical CALIOP data; Obtain multiple initial cloud detection models with different structures; Inputting the first historical cloud feature data into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model; Based on the initial cloud detection result output by each initial cloud detection model and the cloud detection label, the model parameters of each initial cloud detection model are adjusted to train and obtain multiple cloud detection models.

6. The cloud detection model training method according to claim 5, characterized in that: The method further comprises: Acquire second historical cloud characteristic data and historical MODIS data corresponding to the second historical cloud characteristic data, wherein a collection time of the second historical cloud characteristic data matches a collection time of the historical MODIS data, and a collection range of the second historical cloud characteristic data matches a collection range of the historical MODIS data; The second historical cloud feature data is used as input and input into each trained cloud detection model respectively to obtain a model detection cloud detection result output by each trained cloud detection model; Based on the model detection cloud detection results output by the multiple cloud detection models, obtaining the target model detection cloud detection result corresponding to the second historical cloud feature data; Determine the number of first detection sub-areas where both the target model detection cloud detection result and the historical MODIS data identify as having clouds, the number of second detection sub-areas where the historical MODIS data identify as clouds and the target model detection cloud detection result identify as clear sky, the number of third detection sub-areas where the historical MODIS data identify as clear sky and the target model detection cloud detection result identify as having clouds, and the number of fourth detection sub-areas where both the historical MODIS data and the target model detection cloud detection result identify as clear sky; Based on the first number of detection sub-areas, the second number of detection sub-areas, the third number of detection sub-areas, and the fourth number of detection sub-areas, it is determined whether the multiple cloud detection models meet a preset accuracy requirement.

7. A cloud detection device, characterized in that: include: A model acquisition module is configured to acquire a plurality of pre-trained cloud detection models with different structures; a cloud detection module configured to obtain cloud feature data, and use the cloud feature data as input to input each cloud detection model respectively, so as to obtain a cloud detection result output by each cloud detection model; The target detection module is configured to obtain a target cloud detection result based on the cloud detection results output by multiple cloud detection models.

8. A cloud detection model training device, characterized in that: include: A historical data acquisition module, configured to acquire first historical cloud feature data; a CALIOP data acquisition module configured to acquire historical CALIOP data matching the first historical cloud feature data, wherein a collection time of the first historical cloud feature data matches a collection time of the historical CALIOP data, and a collection range of the first historical cloud feature data matches a collection range of the historical CALIOP data; A detection tag acquisition module, configured to acquire a cloud detection tag of the first historical cloud feature data based on historical CALIOP data; An initial model acquisition module is configured to acquire a plurality of initial cloud detection models with different structures; A first detection result acquisition module is configured to input the first historical cloud feature data into each initial cloud detection model to obtain an initial cloud detection result output by each initial cloud detection model; The detection model training module is configured to adjust the model parameters of each initial cloud detection model based on the initial cloud detection result output by each initial cloud detection model and the cloud detection label, so as to train multiple cloud detection models.

9. An electronic device, characterized in that: The method comprises a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method steps described in any one of claims 1 to 6.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the method steps described in any one of claims 1 to 6 are implemented.

Images