Method and system for generating meteorological satellite visible light cloud image

Abstract

The application discloses a kind of meteorological satellite visible light cloud chart generation method and system, it is related to meteorological satellite technical field, the method includes: based on the multi-channel imaging equipment containing infrared and visible light satellite observation, collect the infrared cloud chart set of night target to be carried out cloud chart conversion;Carry out cloud chart data block, obtain block infrared cloud chart data;Call seasonal time-domain cloud chart in predetermined time interval;Train difference adaptive calibration model;Train cloud chart conversion model;Combining the difference adaptive calibration model and cloud chart conversion model of front and rear position connection, the independent processing of block cloud chart is carried out;To block conversion cloud chart is spliced, generates daytime target visible light cloud chart set.The application solves the technical problem that the feature difference of existing technology in daytime and night infrared cloud chart causes the visible light cloud chart difference of adjacent time generation is larger, reaches the technical effect of intelligent analysis infrared cloud chart difference feature, improves the accuracy of visible light cloud chart generation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of meteorological satellite technology, in particular to a method and system for generating a visible light cloud image of a meteorological satellite. BACKGROUND

[0002] The imager of a meteorological satellite includes multiple channels, wherein the cloud image of the visible light channel can reflect rich and detailed information of clouds and the earth's surface, but is greatly affected by the angle of solar irradiation, and cannot display the characteristics of clouds and the earth's surface at night; while the infrared channel can be observed at all times, but has poor expression of details and can only show vague outlines. Therefore, appropriate visual enhancement processing of the satellite infrared cloud image at night is conducive to the forecasters to view and grasp the detailed characteristics and classification of the cloud cluster.

[0003] At present, a deep learning algorithm for image translation is used to generate a visible light cloud image from an infrared cloud image, thereby improving the efficiency of generating a visible light cloud image. However, the characteristics of the underlying surface are quite different between daytime and nighttime infrared cloud images, and the visible light cloud image at night is not available, so that the model parameters trained by the daytime data set are used for generating a visible light cloud image at night, resulting in a visible light cloud image with great randomness, and a great difference between the visible light cloud images generated at adjacent times.

[0004] There is a technical problem in the prior art that the difference in characteristics between daytime and nighttime infrared cloud images leads to a large difference between the visible light cloud images generated at adjacent times. SUMMARY

[0005] The present application provides a method and system for generating a visible light cloud image of a meteorological satellite, which is used to solve the technical problem in the prior art that the difference in characteristics between daytime and nighttime infrared cloud images leads to a large difference between the visible light cloud images generated at adjacent times.

[0006] In view of the above problems, the present application provides a method and system for generating a visible light cloud image of a meteorological satellite.

[0007] In a first aspect of the present application, a method for generating a visible light cloud image of a meteorological satellite is provided, the method comprising:

[0008] Satellite observation is performed based on a multi-channel imaging device including infrared and visible light, and a nighttime target infrared cloud image set to be converted into a cloud image is collected, the nighttime target infrared cloud image set corresponding to different channel bands;

[0009] Based on the nighttime target infrared cloud image set, cloud image data is divided into blocks, and block infrared cloud image data is obtained;

[0010] A seasonal time-domain cloud image within a predetermined time interval is called, the seasonal time-domain cloud image including a nighttime infrared cloud image, a daytime infrared cloud image, and a daytime visible light cloud image;

[0011] training a difference adaptive calibration model for difference analysis and processing of night cloud images and day cloud images based on the night infrared cloud image and the day infrared cloud image;

[0012] training a cloud image conversion model for simulation conversion of infrared cloud images and visible light cloud images based on the day infrared cloud image and the day visible light cloud image;

[0013] independent processing of the block cloud images based on the difference adaptive calibration model and the cloud image conversion model in combination, to obtain block conversion cloud images;

[0014] stitching the block conversion cloud images to generate a day target visible light cloud image set.

[0015] In a second aspect of the present application, a meteorological satellite visible light cloud image generation system is provided, which comprises:

[0016] a cloud image set acquisition module configured to acquire a night target infrared cloud image set to be converted based on satellite observation by a multi-channel imaging device containing infrared and visible light, the night target infrared cloud image set corresponding to different channel bands;

[0017] a cloud image data acquisition module configured to acquire block infrared cloud image data based on cloud image data blocking of the night target infrared cloud image set;

[0018] a time domain cloud image calling module configured to call seasonal time domain cloud images in a predetermined time interval, the seasonal time domain cloud images including night infrared cloud images, day infrared cloud images and day visible light cloud images;

[0019] a calibration model training module configured to train a difference adaptive calibration model for difference analysis and processing of night cloud images and day cloud images based on the night infrared cloud image and the day infrared cloud image;

[0020] a conversion model training module configured to train a cloud image conversion model for simulation conversion of infrared cloud images and visible light cloud images based on the day infrared cloud image and the day visible light cloud image;

[0021] a block conversion cloud image acquisition module configured to independently process the block cloud images based on the difference adaptive calibration model and the cloud image conversion model in combination, to obtain block conversion cloud images;

[0022] A visible light cloud atlas generation module is configured to stitch the patch converted cloud images to generate a daytime target visible light cloud atlas.

[0023] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0024] In the present application, satellite observation is performed based on a multi-channel imaging device containing infrared and visible light, a nighttime target infrared cloud atlas to be converted is collected, the nighttime target infrared cloud atlas corresponds to different channel bands, then based on the nighttime target infrared cloud atlas, cloud image data is divided into patches to obtain patch infrared cloud image data, a seasonal time-domain cloud image in a predetermined time interval is called, the seasonal time-domain cloud image includes a nighttime infrared cloud image, a daytime infrared cloud image and a daytime visible light cloud image, then based on the nighttime infrared cloud image and the daytime infrared cloud image, a difference adaptive calibration model for differential analysis and processing of nighttime cloud images and daytime cloud images is trained, and then based on the daytime infrared cloud image and the daytime visible light cloud image, a cloud image conversion model for simulation conversion of infrared cloud images and visible light cloud images is trained, the difference adaptive calibration model and the cloud image conversion model connected in front and back are combined to independently process patch cloud images based on the patch infrared cloud image data, patch converted cloud images are obtained, and by stitching the patch converted cloud images, a daytime target visible light cloud atlas is generated. The technical effect of intelligently generating a visible light cloud image and improving the accuracy of the visible light cloud image is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0026] Figure 1 A flowchart of a method for generating a meteorological satellite visible light cloud image is provided for the embodiments of the present application.

[0027] Figure 2 A flowchart of a method for generating a meteorological satellite visible light cloud image in which multiple groups of patch cloud images are used as patch infrared cloud image data is provided for the embodiments of the present application.

[0028] Figure 3 A flowchart of a method for generating a meteorological satellite visible light cloud image in which double-channel cloud image processing is performed synchronously is provided for the embodiments of the present application.

[0029] Figure 4 A structure diagram of a system for generating a meteorological satellite visible light cloud image is provided for the embodiments of the present application.

[0030] Explanation of reference signs: cloud atlas collection module 11, cloud data acquisition module 12, time domain cloud calling module 13, calibration model training module 14, conversion model training module 15, block conversion cloud acquisition module 16, visible light cloud atlas generation module 17. DETAILED DESCRIPTION

[0031] The present application provides a meteorological satellite visible light cloud atlas generation method and system, which is used to solve the technical problem that the feature difference between daytime and nighttime infrared cloud atlas leads to a large difference between visible light cloud atlas generated at adjacent times.

[0032] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0033] It should be noted that the terms “include” and “have” and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units need not be limited to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices.

[0034] Embodiment one

[0035] As shown in the accompanying drawings, Figure 1 The present application provides a meteorological satellite visible light cloud atlas generation method, wherein the method comprises:

[0036] S1: satellite observation based on a multi-channel imaging device containing infrared and visible light, collection of nighttime target infrared cloud atlas to be converted, the nighttime target infrared cloud atlas corresponding to different channel bands;

[0037] In an embodiment of the present application, the multi-channel imaging device containing infrared and visible light is used to collect a variety of spectral radiation information of the earth's surface and atmosphere, to provide data for weather forecasting and climate monitoring, and has a large number of detection bands, spatial resolution, time resolution, etc., and can realize multi-spectral, high-frequency and quantitative detection of physical parameters of the earth's surface and atmosphere. Although the infrared channel in the multi-channel imaging device containing infrared and visible light can be observed at any time, the features of the underlying surface are quite different between daytime and nighttime infrared cloud atlas, so it is necessary to convert the nighttime target infrared cloud atlas, so as to eliminate the unique feature difference between daytime and nighttime infrared data.

[0038] Preferably, the multi-channel imaging device containing infrared and visible light has 14 channels, including 6 visible / near-infrared bands, 2 mid-infrared bands, 2 water vapor bands, and 4 long-infrared bands. Among them, the night target infrared cloud image set to be converted is the data collected by the infrared channel in the multi-channel imaging device containing infrared and visible light. Each channel is two-dimensional grid data, and each grid saves a positive integer between 0 and 4095. The night target infrared cloud image set corresponds to different channel bands. By collecting the night target infrared cloud image set to be converted, the data for generating a visible light cloud image is provided.

[0039] S2: based on the night target infrared cloud image set, cloud image data is divided into blocks to obtain block infrared cloud image data;

[0040] Further, as shown in the cloud image data is divided into blocks to obtain block infrared cloud image data, the embodiment of the application step S2 includes: Figure 2

[0041] S2-1: build a two-dimensional coordinate system with a cloud coverage domain as a coordinate space;

[0042] S2-2: based on a predetermined segmentation size, combine the two-dimensional coordinate system, and traverse the night infrared cloud image set to perform uniform segmentation of the cloud image to obtain a plurality of groups of block cloud images, the plurality of groups of block cloud images have a neighborhood overlapping area;

[0043] S2-3: the plurality of groups of block cloud images are taken as the block infrared cloud image data, and the plurality of groups of block cloud images correspond one-to-one to the night target infrared cloud image set.

[0044] In one possible embodiment, the images in the night target infrared cloud image set are divided into cloud image data blocks to lay the foundation for subsequent improvement of the efficiency and accuracy of cloud image conversion. Among them, the block infrared cloud image data is the data obtained by dividing the night target infrared cloud image set according to a certain segmentation rule.

[0045] ​A two-dimensional coordinate system for describing the position of the image in the cloud cover domain is built, taking the cloud cover domain as the coordinate space. The cloud cover domain refers to the area covered by the collected set of night target infrared cloud images. The predetermined segmentation size is the segmentation scale when uniformly segmenting the cloud image, which can be 1024x1024. Each night infrared cloud image in the set of night infrared cloud images is uniformly segmented according to the predetermined segmentation size, taking the two-dimensional coordinate system as the basis for division, to obtain the corresponding segmentation result, that is, the plurality of groups of segmented cloud images. Each group of segmented cloud images corresponds to a set of night infrared cloud images. Preferably, in the process of uniformly segmenting the cloud image, the two adjacent segmented cloud images have a neighborhood overlap region, which facilitates the fusion of the segmented infrared cloud image data and reduces the generation of seams. Further, the plurality of groups of segmented cloud images are taken as the segmented infrared cloud image data to provide data for subsequent segmented stitching.

[0046] S3: calling a seasonal time-domain cloud image in a predetermined time interval, the seasonal time-domain cloud image including a night infrared cloud image, a daytime infrared cloud image, and a daytime visible light cloud image;

[0047] In one embodiment, in order to obtain data for training a difference adaptive calibration model for differential analysis and processing of night cloud images and daytime cloud images, the seasonal time-domain cloud image is obtained by calling images observed by a multi-channel imaging device containing infrared and visible light in a predetermined time interval. Preferably, the predetermined time interval can be the cloud images at the whole point in the 10 days before and after the spring equinox, the 10 days before and after the summer solstice, the 10 days before and after the autumnal equinox, and the 10 days before and after the winter solstice, thereby covering the daytime and the night. The seasonal time-domain cloud image in the predetermined time interval includes a night infrared cloud image, a daytime infrared cloud image, and a daytime visible light cloud image. The night infrared cloud image is collected by the infrared channel of the multi-channel imaging device containing infrared and visible light at night. The daytime infrared cloud image is collected by the infrared channel of the multi-channel imaging device containing infrared and visible light during the day. The daytime visible light cloud image is collected by the visible light channel of the multi-channel imaging device containing infrared and visible light during the day.

[0048] S4: training a difference adaptive calibration model for differential analysis and processing of night cloud images and daytime cloud images based on the night infrared cloud image and the daytime infrared cloud image;

[0049] Further, as shown in Figure 3 The training of the difference adaptive calibration model for differential analysis and processing of night cloud images and daytime cloud images includes:

[0050] S4-1: based on a twin network, supervised training of a night infrared feature processing channel and a day infrared feature processing channel capable of cloud image translation and convolution feature extraction;

[0051] S4-2: performing channel parallel arrangement and post-differentiation compensation layer, generating the adaptive calibration model;

[0052] S4-3: based on the day infrared feature processing channel, performing mapping matching of the input infrared cloud image of the night infrared feature processing channel, and synchronously performing dual-channel cloud image processing.

[0053] Further, the step S4 of the embodiment of the application further comprises:

[0054] S4-4: based on the seasonal time-domain cloud image, extracting the night infrared cloud image and the day infrared cloud image, performing analysis and processing to obtain a calibrated converted cloud image, and mapping and determining training data;

[0055] S4-5: based on the training data, training to generate an initial adaptive calibration model;

[0056] S4-6: combining the training data, verifying the initial adaptive calibration model, screening training data that do not meet the deviation threshold for retraining, until the verification results all meet the deviation threshold.

[0057] In one possible embodiment, by taking the night infrared cloud image and the day infrared cloud image as training data, a difference adaptive calibration model for difference analysis and processing of the night cloud image and the day cloud image is supervised trained, so as to achieve the technical effect of intelligently analyzing the difference between the day cloud image and the night cloud image, and improving the analysis efficiency and the analysis accuracy.

[0058] Preferably, a twin network is used for supervised training, a night infrared feature processing channel for cloud image translation and convolution feature extraction of the night infrared cloud image, and a day infrared feature processing channel for cloud image translation and convolution feature extraction of the day infrared cloud image. The night infrared feature processing channel and the day infrared feature processing channel are arranged in parallel, and the output layer of the night infrared feature processing channel and the output layer of the day infrared feature processing channel are respectively connected in communication with the input layer of the differentiating compensation layer, thereby generating the adaptive calibration model.

[0059] Optionally, the night infrared cloud image is input into the night infrared feature processing channel for data processing, and meanwhile, the features extracted from the night infrared cloud image are input into the mapping matching of the day infrared cloud image in the day infrared feature processing channel, the day infrared cloud image matched successfully is used for compensating the pixel points of the night infrared cloud image by using the differential compensation features, so that the difference between the night infrared cloud image and the day infrared cloud image caused by different underlying surfaces is compensated, and the double-channel cloud image processing is performed. The technical effect of improving the reliability of the analysis data is achieved.

[0060] Preferably, the training data is determined by extracting the night infrared cloud image and the day infrared cloud image based on the seasonal time domain cloud image, obtaining the calibrated conversion cloud image through analysis and processing, and constructing the mapping relationship between the calibrated conversion cloud image and the night infrared cloud image and the day infrared cloud image according to the corresponding relationship. Optionally, the training data is divided into a training set and a validation set according to a certain proportion. The proportion of the training set is 30%, and the proportion of the validation set is 70%. The training set is used for supervised training to obtain the initial adaptive calibration model, and then the validation set is used for testing the initial adaptive calibration model. Preferably, the current model parameters are saved after each training cycle, the last model parameters are covered, and the latest model is saved. In the verification process, the training data that does not meet the deviation threshold is retrained until the test results all meet the deviation threshold, the verification is passed, and the difference adaptive calibration model is obtained.

[0061] S5: training a cloud image conversion model for simulating conversion of infrared cloud images and visible light cloud images based on the day infrared cloud image and the day visible light cloud image;

[0062] In one embodiment, the day infrared cloud image and the day visible light cloud image are used as first training data to supervise the training of a conditional generative adversarial network model, which can be pix2pixHD. After training the training set in the first training data for a certain number of times, simulated visible light data is generated from all infrared data in the validation set in the first training data, and PSNR is calculated with all visible light data in the validation set in the first training data. The current model parameters are saved after each training cycle, the last model parameters are covered, and the latest model is obtained. During each verification, the model with higher PSNR than the previous one is the best model, and the best model parameters are saved. The cloud image conversion model is used to convert the day infrared cloud image into the day visible light cloud image.

[0063] S6: combining the difference adaptive calibration model and the cloud image conversion model connected in front and back, performing independent processing on the block cloud image data to obtain a block conversion cloud image.

[0064] Further, the step S6 of the embodiment of the present application further comprises:

[0065] S6-1: performing image translation on the night infrared cloud image and the day infrared cloud image to extract semantic conversion information;

[0066] S6-2: performing differential correction on the semantic conversion information to obtain a cloud image deviation pixel point domain;

[0067] S6-3: locating the deviation pixel point domain of the day infrared cloud image and performing convolution feature extraction, replacing the convolution feature of the night infrared cloud image with the deviation pixel point domain to obtain the block conversion cloud image.

[0068] In one possible embodiment, the difference adaptive correction model and the cloud image conversion model are connected in sequence, and then the block infrared cloud image data is input into the difference adaptive correction model and the cloud image conversion model in sequence to realize independent processing and conversion of the block infrared cloud image data, thereby obtaining the converted block conversion cloud image.

[0069] In one embodiment, the semantic conversion information is obtained by using a deep learning algorithm of image translation to perform image translation on the night infrared cloud image and the day infrared cloud image. Then, the semantic conversion information corresponding to the night infrared cloud image and the day infrared cloud image is differentially corrected to obtain a cloud image deviation pixel point domain. The cloud image deviation pixel point domain is a pixel point region with large deviation.

[0070] The day infrared cloud image is located according to the cloud image deviation pixel point domain, the deviation pixel point domain is determined and convolution feature extraction is performed, the extracted convolution feature is replaced in the night infrared cloud image, and thus a block conversion cloud image that compensates for the difference is obtained.

[0071] S7: stitching the block conversion cloud image to generate a day target visible light cloud image set.

[0072] Further, the step S7 of stitching the block conversion cloud image further comprises:

[0073] S7-1: locating and filling the obtained block conversion cloud image in the two-dimensional coordinate system to obtain a stitched cloud image;

[0074] S7-2: identifying the stitched cloud image to extract an interleaved overlapping region;

[0075] S7-3: performing weighted average fusion of the overlapping cloud image and the second overlapping cloud image in the neighborhood overlapping region to determine a region fusion result.

[0076] In one embodiment, the obtained sub-block converted cloud images are spliced to obtain the set of daytime target visible light cloud images. By weighted average fusion of the overlapping regions of adjacent sub-blocks, the seams between adjacent sub-blocks are avoided.

[0077] Specifically, by locating and filling the obtained sub-block converted cloud images according to the corresponding point positions in the two-dimensional coordinate system, a spliced cloud image is obtained. According to the spliced cloud image, the overlapping regions of two adjacent sub-block converted cloud images are extracted. By using inverse distance linear weighted average to perform weighted average fusion of the overlapping cloud image and the second overlapping cloud image in the overlapping region, a region fusion result is determined. Preferably, when sub-block A and sub-block B overlap, the value of any grid point after overlap is calculated using P = a * (1 - d / L) + b * d / L, where a and b are the values of sub-block A and sub-block B at the grid point, L is the width of the overlapping region, d is the distance of the grid point from the edge of the overlapping region on the side of a, and P is the value of any grid point in the fused overlapping region.

[0078] Further, the step S7 of the embodiment of the present application further comprises:

[0079] S7-4: mapping and calculating the solar elevation angle to determine the cloud image acquisition time point;

[0080] S7-5: determining the cloud image acquisition mode based on the cloud image acquisition time point;

[0081] S7-6: if it is daytime, directly collecting a visible light satellite cloud image as a first cloud image acquisition mode;

[0082] S7-7: if it is nighttime, collecting an infrared satellite cloud image and performing compensation conversion to obtain a visible light satellite cloud image as a second cloud image acquisition mode;

[0083] S7-8: if it is in the vicinity of the twilight line, executing the first cloud image acquisition mode and the second cloud image acquisition mode and performing weighted average of the execution results as a visible light satellite cloud image.

[0084] In one possible embodiment, the solar elevation angle refers to the angle between the incident direction of sunlight at a certain location on Earth and the horizontal plane, and the solar elevation angle changes with the local time (hour angle) and the solar declination. Therefore, the cloud image acquisition time point can be determined by determining the solar elevation angle. The cloud image acquisition time point is the time point at which a multi-channel imaging device containing infrared and visible light performs cloud image acquisition. Preferably, the solar elevation angle at noon = 90° - the difference between the latitude of the location and the solar direct point.

[0085] Further, the satellite monitoring mode for collecting cloud images can be determined according to the cloud image collection time point. When the cloud image collection time point is in the daytime, that is, the solar elevation angle is greater than a threshold (for example, 0 degrees), the visible light satellite cloud image can be directly collected, and the collected satellite cloud image is the visible light satellite cloud image, which is the first cloud image acquisition mode. When the cloud image collection time point is at night, that is, the solar elevation angle is less than a threshold (for example, 0 degrees), the collected infrared satellite cloud image at night cannot be directly used for visible light conversion, and the collected infrared satellite cloud image needs to be compensated and converted to obtain the visible light satellite cloud image, which is the second cloud image acquisition mode. Preferably, when the solar elevation angle is within a specified threshold range (for example, 0 degrees to 5 degrees), it is considered to be in the vicinity of the terminator, the first cloud image acquisition mode and the second cloud image acquisition mode are executed at the same time, and the weighted average of the execution results is taken as the visible light satellite cloud image. For example, the weighted average is performed according to the formula w = a / 5 and V = w*T+(1-w)*F, where a is the solar elevation angle, w is the weight, T is the value of the real visible light (the value obtained by executing the first cloud image acquisition mode), F is the value of the simulated visible light (the value obtained by executing the second cloud image acquisition mode), and V is the value of the weighted average of the visible light.

[0086] In summary, the embodiments of the present application have at least the following technical effects:

[0087] The present application divides the night target infrared cloud image set into data blocks, independently processes and converts the divided infrared cloud image data, and then uses the seasonal time domain cloud image to provide intelligent processing and conversion by using the differential adaptive calibration model and the cloud image conversion model, and then splices the converted cloud image blocks to obtain the daytime target visible light cloud image set. The reliability of the visible light cloud image set is improved, and the generation efficiency and the intelligent degree are improved.

[0088] Embodiment two

[0089] Based on the same inventive concept as the generation method of the meteorological satellite visible light cloud image in the foregoing embodiment, as shown in Figure 4 The present application provides a generation system of a meteorological satellite visible light cloud image, and the system and method embodiments in the present application are based on the same inventive concept. The system comprises:

[0090] A cloud image set collection module 11 is configured to collect a night target infrared cloud image set to be converted by satellite observation based on a multi-channel imaging device comprising infrared and visible light, and the night target infrared cloud image set corresponds to different channel bands.

[0091] The cloud image data acquisition module 12 is configured to perform cloud image data blocking based on the set of night target infrared cloud images to obtain blocked infrared cloud image data.

[0092] The time-domain cloud image calling module 13 is configured to call seasonal time-domain cloud images in a predetermined time interval, the seasonal time-domain cloud images including night infrared cloud images, day infrared cloud images, and day visible light cloud images.

[0093] The calibration model training module 14 is configured to train a difference adaptive calibration model for night cloud image and day cloud image differential analysis and processing based on the night infrared cloud images and the day infrared cloud images.

[0094] The conversion model training module 15 is configured to train a cloud image conversion model for infrared cloud image and visible light cloud image simulation conversion based on the day infrared cloud images and the day visible light cloud images.

[0095] The blocked converted cloud image acquisition module 16 is configured to perform independent processing of the blocked infrared cloud image data by combining the difference adaptive calibration model and the cloud image conversion model in front and back positions to obtain blocked converted cloud images.

[0096] The visible light cloud image set generation module 17 is configured to perform splicing on the blocked converted cloud images to generate a set of day target visible light cloud images.

[0097] Further, the cloud image data acquisition module 12 is configured to perform the following method:

[0098] A two-dimensional coordinate system is built with a cloud image coverage domain as a coordinate space.

[0099] Based on a predetermined segmentation size, the night infrared cloud image set is traversed for cloud image uniform segmentation in combination with the two-dimensional coordinate system to obtain a plurality of groups of blocked cloud images, the plurality of groups of blocked cloud images having a neighborhood overlapping area.

[0100] The plurality of groups of blocked cloud images are taken as the blocked infrared cloud image data, and the plurality of groups of blocked cloud images correspond one-to-one to the set of night target infrared cloud images.

[0101] Further, the calibration model training module 14 is configured to perform the following method:

[0102] Based on a twin network, night infrared feature processing channels and day infrared feature processing channels that can perform cloud image translation and convolution feature extraction are supervised and trained.

[0103] Performing channel parallel arrangement and post-differentiation compensation layer, generating the adaptive calibration model;

[0104] Based on the daytime infrared feature processing channel, the input infrared cloud image of the night infrared feature processing channel is mapped and matched, and the dual-channel cloud image processing is synchronously performed.

[0105] Further, the calibration model training module 14 is used to execute the following method:

[0106] Based on the seasonal time domain cloud image, the night infrared cloud image and the daytime infrared cloud image are extracted, analyzed and processed to obtain a calibrated conversion cloud image, and the training data is mapped and determined;

[0107] Based on the training data, an initial adaptive calibration model is trained and generated;

[0108] In combination with the training data, the initial adaptive calibration model is verified, and the training data that does not meet the deviation threshold is retrained until the verification result meets the deviation threshold.

[0109] Further, the block conversion cloud image acquisition module 16 is used to execute the following method:

[0110] Image translation is performed on the night infrared cloud image and the daytime infrared cloud image, and semantic conversion information is extracted;

[0111] The difference of the semantic conversion information is corrected to obtain a cloud image deviation pixel point domain;

[0112] The deviation pixel point domain of the daytime infrared cloud image is located and convolution feature extraction is performed, and the convolution feature of the deviation pixel point domain is replaced in the night infrared cloud image to obtain the block conversion cloud image.

[0113] Further, the visible light cloud image set generation module 17 is used to execute the following method:

[0114] In the two-dimensional coordinate system, the obtained block conversion cloud image is positioned and filled to obtain a spliced cloud image;

[0115] The spliced cloud image is identified, and an interleaved overlapping area is extracted;

[0116] The overlapping cloud image and the second overlapping cloud image are weighted and averaged to fuse the adjacent overlapping area, and a regional fusion result is determined.

[0117] Further, the visible light cloud image set generation module 17 is used to execute the following method:

[0118] The sun elevation angle is measured and calculated to determine the cloud image acquisition time point;

[0119] Based on the cloud image acquisition time point, a cloud image acquisition mode is determined;

[0120] If it is daytime, a visible light satellite cloud image is directly collected as a first cloud image acquisition mode;

[0121] If it is nighttime, an infrared satellite cloud image is collected and compensated and converted to obtain a visible light satellite cloud image as a second cloud image acquisition mode;

[0122] If it is in the vicinity of the terminator, the first cloud image acquisition mode and the second cloud image acquisition mode are executed, and a weighted average of the execution results is performed as a visible light satellite cloud image.

[0123] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned description is made for specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that in the embodiments and still achieve the desired result. In addition, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or advantageous.

[0124] The above-mentioned is only the preferred embodiment of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0125] The specification and drawings are merely exemplary of the present application, and any and all modifications, variations, combinations or equivalents that are within the scope of the present application should be considered. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method of generating a meteorological satellite visible cloud picture, characterized in that, The method comprises: Satellite observation based on a multi-channel imaging device containing infrared and visible light, collecting a set of night target infrared cloud images to be converted into cloud images, the set of night target infrared cloud images corresponding to different channel bands; Based on the set of night target infrared cloud images, cloud image data is divided into blocks to obtain divided infrared cloud image data; Seasonal time-domain cloud images in a predetermined time interval are called, the seasonal time-domain cloud images including night infrared cloud images, daytime infrared cloud images and daytime visible light cloud images; Based on the night infrared cloud images and the daytime infrared cloud images, a difference adaptive calibration model for differential analysis and processing of night cloud images and daytime cloud images is trained; Based on the daytime infrared cloud images and the daytime visible light cloud images, a cloud image conversion model for simulation conversion of infrared cloud images and visible light cloud images is trained; Combined with the difference adaptive calibration model and the cloud image conversion model connected in front and back positions, independent processing of the divided infrared cloud image data is performed to obtain divided conversion cloud images; The divided conversion cloud images are spliced to generate a set of daytime target visible light cloud images.

2. The method of claim 1, wherein, The cloud image data is divided into blocks to obtain divided infrared cloud image data, and the method comprises: A two-dimensional coordinate system is built with a cloud image coverage domain as a coordinate space; Based on a predetermined segmentation size, the set of night infrared cloud images is traversed for uniform segmentation of cloud images combined with the two-dimensional coordinate system to obtain a plurality of groups of divided cloud images, the plurality of groups of divided cloud images having a neighborhood overlapping area; The plurality of groups of divided cloud images are taken as the divided infrared cloud image data, and the plurality of groups of divided cloud images correspond one-to-one to the set of night target infrared cloud images.

3. The method of claim 1, wherein, The method for training the difference adaptive calibration model for differential analysis and processing of night cloud images and daytime cloud images comprises: Based on a twin network, night infrared feature processing channels and daytime infrared feature processing channels for executable cloud image translation and convolution feature extraction are supervised and trained; The channels are arranged in parallel and a post-differential compensation layer is arranged to generate the adaptive calibration model; Based on the daytime infrared feature processing channels, mapping matching of input infrared cloud images of the night infrared feature processing channels is performed, and double-channel cloud image processing is simultaneously performed.

4. The method of claim 3, wherein, The method comprises: Based on the seasonal time-domain cloud images, the night infrared cloud images and the daytime infrared cloud images are extracted for analysis and processing to obtain calibrated conversion cloud images, and training data is determined by mapping; Based on the training data, an initialized adaptive calibration model is trained and generated; Combined with the training data, the initialized adaptive calibration model is tested, and training data that does not meet a deviation threshold is filtered for retraining until the test results all meet the deviation threshold.

5. The method of claim 3, wherein, The method comprises: Image translation is performed on the night infrared cloud images and daytime infrared cloud images to extract semantic conversion information; Differential proofreading of the semantic conversion information is performed to obtain a cloud image deviation pixel point domain; The deviation pixel point domain of the daytime infrared cloud images is located and convolution feature extraction is performed, the convolution feature of the deviation pixel point domain is replaced in the night infrared cloud images to obtain the divided conversion cloud images.

6. The method of claim 1, wherein, The method for splicing the divided conversion cloud images comprises: In a two-dimensional coordinate system, the obtained sub-block conversion cloud map is positioned and filled to obtain a spliced cloud map; The spliced cloud map is identified, and an interleaved overlapping area is extracted; The overlapping cloud map and the second overlapping cloud map are weighted and averaged to fuse the adjacent overlapping area, and a regional fusion result is determined.

7. The method of claim 1, wherein, The method comprises: Surveying and calculating the solar elevation angle to determine the cloud map acquisition time point; Based on the cloud map acquisition time point, the cloud map acquisition mode is determined; If it is daytime, visible light satellite cloud map is directly collected as the first cloud map acquisition mode; If it is night, infrared satellite cloud map is collected and compensated to obtain visible light satellite cloud map as the second cloud map acquisition mode; If it is the vicinity of the terminator, the first cloud map acquisition mode and the second cloud map acquisition mode are executed, and the execution results are weighted and averaged to obtain the visible light satellite cloud map.

8. A system for generating a meteorological satellite visible cloud picture, characterized in that The system comprises: A cloud map set collection module is used for satellite observation based on a multi-channel imaging device containing infrared and visible light, collecting a night target infrared cloud map set to be converted into a cloud map, and the night target infrared cloud map set corresponds to different channel bands; A cloud map data acquisition module is used for cloud map data blocking based on the night target infrared cloud map set to obtain sub-block infrared cloud map data; A time domain cloud map calling module is used for calling seasonal time domain cloud maps in a predetermined time interval, including night infrared cloud maps, daytime infrared cloud maps and daytime visible light cloud maps; A calibration model training module is used for training a difference adaptive calibration model for differential analysis and processing of night cloud maps and daytime cloud maps based on the night infrared cloud map and the daytime infrared cloud map; A conversion model training module is used for training a cloud map conversion model for simulating conversion of infrared cloud maps and visible light cloud maps based on the daytime infrared cloud map and the daytime visible light cloud map; A sub-block conversion cloud map acquisition module is used for independent processing of sub-block cloud maps of the sub-block infrared cloud map data by combining the difference adaptive calibration model and the cloud map conversion model connected in front and back, to obtain sub-block conversion cloud maps; A visible light cloud map set generation module is used for splicing the sub-block conversion cloud maps to generate a daytime target visible light cloud map set.

Images