A clear sky image acquisition method based on multi-temporal data of remote sensing images of geostationary meteorological satellites

Abstract

The application provides a clear sky image acquisition method based on multi-temporal data of remote sensing images of stationary meteorological satellites, comprising: acquiring remote sensing image data of multiple times of the same area and sorting the image data according to time to obtain an image set; converting DN values of relevant original image pixels in the image set into apparent reflectivity; correcting relevant images in the image set to make geographical positions and spatial resolutions consistent; distinguishing relevant images in the image set as clear sky categories or cloud categories according to apparent reflectivity pixel by pixel; and constructing a clear sky image according to pixels distinguished as clear sky categories in each image of the image set. The method can provide a comprehensive clear sky synthesis algorithm based on multi-temporal visible spectral band data for features of different surfaces, and can solve the problems of "patches" and cloud shadows in existing synthesized images by using multi-temporal stationary satellite remote sensing data, combining fast-changing cloud systems within a single day with relatively unchanged underlying surfaces, and acquiring surface texture features under clear sky conditions.

Description

Technical Field

[0001] This application relates to the field of satellite remote sensing technology, specifically to a method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images. Background Technology

[0002] In the field of Earth remote sensing, the average cloud cover on Earth exceeds 60%, which severely limits the use of satellite remote sensing in the optical band. For example, in single-scene remote sensing images, if the cloud cover is too high, the data often cannot be used for application services, resulting in low data resource utilization. Therefore, effectively reducing the impact of clouds and obtaining clear-sky images with no or few clouds is of great significance for improving the utilization rate of visible light remote sensing image data.

[0003] Multi-temporal image compositing is one of the processing techniques to address the impact of clouds on visible light remote sensing images. Multi-temporal clear-sky compositing refers to using remote sensing data from the same region at different times, replacing cloud-covered areas with cloudless surface information to ultimately obtain a cloudless or minimally clouded remote sensing image. Multi-temporal clear-sky compositing fully utilizes spatially or temporally redundant remote sensing data, processing it according to certain rules or algorithms to generate a new composite image. This highlights effective surface information, eliminates or suppresses cloud interference, improves the image environment for ground feature identification, thereby increasing the reliability of ground feature interpretation and improving the application scope and effectiveness.

[0004] Geostationary satellite remote sensors, with their high revisit timeliness, play a vital role in fields such as weather and surface monitoring. For example, the Fengyun-4 (FY-4) is my country's new-generation geostationary meteorological satellite. Its onboard Fast Imager (GHI) can image an area of ​​one million square kilometers every minute, providing enhanced monitoring capabilities for small- and medium-scale weather systems. How to fully leverage the application of payloads like the FY-4 geostationary meteorological satellite GHI in surface monitoring information and address the problem of cloud cover obscuring surface imagery is a pressing technical challenge.

[0005] Currently, the commonly used remote sensing image synthesis algorithms internationally mainly include the following four types: (1) Maximum Normalized Difference Vegetation Index Composite (MNC) algorithm. MNC is one of the earliest methods used for image synthesis. It selects the best observation pixel from multi-temporal remote sensing data based on the maximum NDVI value (Normalized Difference Vegetation Index) for pixel-level synthesis. The MNC method has been successfully applied to cloudless synthesis of data such as AVHRR, MODIS and Landsat. This method is for monitoring vegetation information. After synthesis, it will affect the monitoring of other ground cover information. For dense vegetation areas, the maximum NDVI value is less sensitive to vegetation and tends to saturate. When using the MNC algorithm, the synthesis result is not ideal; (2) Harmonic Analysis of Time (HID) algorithm. Series (HANTS), Roerink proposed this algorithm in 2000. It solves the data missing band in MNC synthesized images by time interpolation. However, the adjustable algorithm parameters in the HANTS algorithm can only be set by experience and experiment, which cannot achieve objectivity and automation. Moreover, the final result is not the actual observation value. (3) Multi-dimensional analogue of the median (Medoid), Flood et al. synthesize LandsatTM / ETM+ quarterly images through the Mediod algorithm. Since each pixel of the quarterly image requires at least 3 days of cloudless input to obtain the predicted value, the Mediod synthesis result still has a large number of invalid missing pixels. (4) Pixel Based Composite (PBC) algorithm. The technical idea of ​​PBC is to score all pixels of multi-temporal images by setting weight rules. For pixels at the same position, the highest scorer is identified as the optimal pixel for pixel-level image synthesis.

[0006] The methods described above can effectively remove the influence of cloud-covered areas, but they cannot effectively remove shadows at the edges of cloud areas. Cloud shadows are caused by the obstruction of sunlight by various clouds (thick clouds or cumulonimbus clouds), resulting in abnormally low reflectivity in some areas, making it difficult to extract useful information. The spectral characteristics of cloud-shadowed areas are also less than ideal due to lower incident energy. Compared to the identification of high-brightness cloud areas, the reflectance spectral characteristics of cloud shadows are weaker, making them difficult to identify using simple channel thresholding methods. Furthermore, the reflectivity of cloud shadows is very similar to that of water bodies, making it difficult to distinguish between cloud shadows and water bodies in the final composite image, severely interfering with water remote sensing information. Therefore, while removing the influence of cloud cover, the influence of cloud shadow areas should also be removed. Due to cloud movement, the underlying surface is sometimes unobstructed and sometimes obstructed; when unobstructed, the satellite observes the reflectivity of the underlying surface, while when obstructed, the satellite can only observe the reflectivity of the clouds, and the reflectivity of the underlying surface cannot be observed. In continuous observation images of this region, the reflectance of a single pixel exhibits a bimodal distribution: one distribution represents the surface reflectance under cloudless conditions, while the other represents the reflectance distribution through clouds at that pixel. Therefore, how to obtain a composite image of the entire region under clear skies by estimating the cloudless reflectance pixel by pixel is the problem this application aims to solve. Summary of the Invention

[0007] To address the issue of lost ground target information due to cloud cover and cloud shadows, this application provides a method for acquiring clear-sky images based on multi-temporal data from geostationary meteorological satellite remote sensing. This method offers a comprehensive algorithm for clear-sky synthesis based on multi-temporal visible spectral data, utilizing multi-temporal geostationary satellite remote sensing data and combining rapidly changing cloud systems with relatively unchanging underlying surfaces to obtain surface texture features under clear-sky conditions. This solves problems such as "patches" and cloud shadows present in existing synthesized images.

[0008] The technical solution adopted by this application to solve its technical problem is: a method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images, including:

[0009] Remote sensing image data of the same area at multiple times are acquired and sorted according to time to obtain an image set;

[0010] Convert the DN values ​​of relevant original image pixels in the image set into apparent reflectance;

[0011] Correct relevant images in the image set to ensure consistency in geographical location and spatial resolution;

[0012] For each pixel of a related image in the image set, the apparent reflectance is used to determine whether it is a clear sky or a cloudy sky category.

[0013] A clear sky image is constructed based on the pixels in each image set that are classified as clear sky.

[0014] In one specific implementation, the step of determining whether a pixel-by-pixel related image in the image set is classified as clear sky or cloudy based on apparent reflectance includes the following steps:

[0015] S1. Under the selected band, calculate the mean value μ of the apparent reflectance B of each pixel in the first image of the image set. img For apparent reflectance B less than μ img The corresponding pixels are identified as clear sky category and the mean value μ is calculated. l-img ,variance For apparent reflectance B not less than μ img The corresponding pixels are identified as having cloud cover and the mean value μ is calculated. h-img ,variance

[0016] S2, the clear sky category count value of the corresponding pixel position in the image set when the cumulative judgment reaches the Mth image. Cloud category count value The mean apparent reflectance of clear sky category is The variance of apparent reflectance for clear sky categories is The average apparent reflectance of the cloud category is The variance of apparent reflectance for cloud categories is Assign initial values ​​to the first image when M=1. And based on the apparent reflectance B of each pixel in the first image, we obtain:

[0017]

[0018]

[0019] S3. Determine the apparent reflectance B of each pixel in the M≥2th image:

[0020] when or And f l (B)>f h (B) indicates a clear sky category, and the corresponding pixel is updated.

[0021]

[0022]

[0023] when or And f l (B)≤f h (B) indicates a cloud cover category, and the corresponding pixels are updated.

[0024]

[0025]

[0026] Where f l (B), f h (B) represents the probability density function of apparent reflectance distribution for clear sky and cloud cover, respectively:

[0027]

[0028] In one specific implementation, when constructing a clear sky image based on pixels identified as clear sky in each image of the image set, the average apparent reflectance of pixels at the corresponding pixel positions in the image set under multiple bands is calculated and then synthesized into a clear sky image.

[0029] In one specific implementation, the apparent reflectance B of the pixels in each image of the image set is:

[0030] L = gain × DN + bias

[0031] In the formula, L is the irradiance, DN is the gray value of the original image pixel, gain and bias are the gain and bias of the sensor that acquired the remote sensing image in the corresponding band, d is the Earth-Sun distance, E0 is the solar irradiance in the corresponding band, and θ is the solar zenith angle of the corresponding image.

[0032] In one specific implementation, the remote sensing image data is a visible and near-infrared remote sensing image, including spectral band data, the gain and bias of the sensor that acquired the image in different bands, the solar elevation angle of the image, and the geographic information of the image.

[0033] In one specific implementation, in step S1, the selected band is the blue light band.

[0034] In one specific implementation, when calculating the average apparent reflectance of pixels at corresponding pixel locations in the image set under the clear sky category in multiple bands, the discrimination result of the relevant image pixel under the blue light band is used when determining the clear sky category or the cloudy category of the relevant image pixels in the corresponding image set of the multiple bands.

[0035] In one specific implementation, the device for the method of acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images includes:

[0036] One or more processors;

[0037] A storage device is used to store one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images.

[0038] In one specific implementation, the computer-readable storage medium stores a program that, when executed by a processor, implements the method for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images.

[0039] The advantages of this application are:

[0040] 1. A method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images: Clear sky images are acquired by classifying and identifying clear sky or cloud cover in remote sensing image data pixel by pixel, effectively removing the influence of cloud-covered areas.

[0041] 2. A method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images: Calculate the probability that the apparent reflectance of the corresponding pixel belongs to the clear sky category and the probability that it belongs to the cloudy category. Based on the magnitude of the probability value, the pixel is classified into the cloudless category or the cloudy category, thereby achieving accurate classification of the corresponding pixel category of a single scene image.

[0042] 3. The method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images uses pixel image processing to directly fuse the measured physical parameters on the acquired raw data layer, thereby preserving more of the original true information of the image based on the most original image data.

[0043] 4. The method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images improves the utilization rate of original image data and the accuracy of processing results by classifying pixels of clear sky and cloudy categories and updating the corresponding parameters in a timely manner during the iterative calculation of multi-image data. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the process for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images according to this application;

[0045] Figure 2 This is a schematic diagram of a portion of the image set constituting a method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images according to this application.

[0046] Figure 3 This is a schematic diagram of a clear sky image after processing the first image of a method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images according to this application.

[0047] Figure 4 This is a schematic diagram of the clear sky image synthesized by the method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images according to this application. Detailed Implementation

[0048] This application provides a method for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images, solving the problem of partial loss of ground target information due to cloud cover and cloud shadows. The overall approach is as follows:

[0049] Please see Figure 1 This application provides a method for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images. The method includes: acquiring remote sensing image data from multiple time periods for the same area and sorting them according to time to obtain an image set; converting the DN values ​​of relevant original image pixels in the image set into apparent reflectance; correcting relevant images in the image set to ensure consistency in geographical location and spatial resolution; classifying each pixel of the relevant images in the image set into either clear-sky or cloudy categories based on apparent reflectance; and constructing a clear-sky image based on the pixels classified as clear-sky in each image of the image set. This method effectively removes the influence of cloud-covered areas by classifying remote sensing image data from multiple time periods for the same area into clear-sky or cloudy categories based on apparent reflectance, and constructing a clear-sky image based on the pixels classified as clear-sky.

[0050] In this example, for the same area, remote sensing image data from multiple times within a day are acquired, and the image set is obtained by sorting the image data according to the imaging time. For example... Figure 2 The image set comprises images, specifically visible and near-infrared remote sensing images. These images include spectral band data, sensor gain and bias for different bands, solar elevation angle, and geographic information. The geographic information may include projection and geometric information for geographic correction. Radiometric correction is performed on each image in the set, converting the original image's DN value into the apparent reflectance of the upper atmospheric boundary of the observed target, which has physical dimensions. Specifically, based on the gain, bias, and solar elevation angle of different bands in the visible light remote sensing image data, the pixel grayscale values ​​of each band in the remote sensing image data are converted into apparent reflectance values. The apparent reflectance B corresponding to each pixel in each image in the set is:

[0051] L = gain × DN + bias

[0052] In the formula, L is the irradiance, DN is the gray value of the original image pixel, gain and bias are the gain and bias of the sensor that acquired the remote sensing image in the corresponding band, d is the Earth-Sun distance in 1 AU, E0 is the solar irradiance in the corresponding band, and θ is the solar zenith angle of the corresponding image.

[0053] In processing single-scene images, except for snow and ice, the apparent reflectance of clouds is higher than that of the underlying surface; moreover, for the same location, the variation in apparent reflectance is smaller, while the variation in apparent reflectance of transiting clouds is larger. That is, the distribution of surface apparent reflectance shows a lower mean and smaller variance (relatively concentrated samples), while the distribution of transiting cloud apparent reflectance shows a higher mean and larger variance (more dispersed samples). If there are no clouds at a corresponding pixel, its apparent reflectance is probabilistically closer to the center of the surface apparent reflectance distribution; if there are clouds at a pixel, its apparent reflectance is probabilistically closer to the center of the cloud apparent reflectance distribution. Therefore, specifically, when determining the clear sky or cloudy category for each pixel of related images in an image set based on apparent reflectance, the steps include:

[0054] S1. Under a selected band, for example, in this case, the selected band is the blue light band, calculate the mean value μ of the apparent reflectance B of each pixel in the first image of the image set. img For apparent reflectance B less than μ img The corresponding pixels are identified as clear sky category and the mean value μ is calculated. l-img ,variance For apparent reflectance B not less than μ img The corresponding pixels are identified as having cloud cover and the mean value μ is calculated. h-img ,variance

[0055] S2, the clear sky category count value of the corresponding pixel position in the image set when the cumulative judgment reaches the Mth image. Cloud category count value The mean apparent reflectance of clear sky category is The variance of apparent reflectance for clear sky categories is The average apparent reflectance of the cloud category is The variance of apparent reflectance for cloud categories is Assign initial values ​​to the first image when M=1. And based on the apparent reflectance B of each pixel in the first image, we obtain:

[0056]

[0057]

[0058] S3. Determine the apparent reflectance B of each pixel in the M≥2th image:

[0059] when or And f l (B)>f h (B) indicates a clear sky category, and the corresponding pixel is updated.

[0060]

[0061]

[0062] when or And f l (B)≤f h (B) indicates a cloud cover category, and the corresponding pixels are updated.

[0063]

[0064]

[0065] Where f l (B), f h (B) represents the probability density function of apparent reflectance distribution for clear sky and cloud cover, respectively:

[0066]

[0067]

[0068] The above steps calculate the probability of a pixel's apparent reflectance belonging to the clear sky category and the probability of it belonging to the cloudy category, respectively. Based on the probability values, the pixel is classified into either the cloudless or cloudy category, thus achieving accurate classification of the corresponding pixel category for a single image. In the iterative calculation of multi-image data, classifying pixels into clear sky and cloudy categories and updating the corresponding parameters in a timely manner further improves the utilization rate of the original image data information and the accuracy of the processing results.

[0069] Based on the above classification method for clear sky and cloudy categories of corresponding image pixels in the blue light band, image pixels in other bands can be processed in the same way to obtain clear sky and cloudy images of the corresponding bands. Preferably, when calculating the average apparent reflectance of pixels at corresponding pixel positions in the image set under multiple bands for clear sky category, the classification result of the relevant image pixel under the blue light band is used when classifying the clear sky or cloudy category of related image pixels in the corresponding image set of multiple bands. For example, the wavelength range of the main channel of the GHI rapid imager of Fengyun-4B satellite (FY-4B) in this example is shown in the table below:

[0070]

[0071] The clear sky image synthesized by processing the first image using channels 4, 6, and 3 is shown below. Figure 3 As shown.

[0072] Furthermore, in this embodiment, when constructing a clear sky image based on pixels identified as clear sky in each image of the image set, the average apparent reflectance of pixels at the corresponding pixel positions in the image set under multiple bands is calculated and synthesized into a clear sky image, such as... Figure 4 As shown.

[0073] This embodiment also provides an apparatus for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images, including one or more processors and a storage device. The storage device is used to store one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the clear-sky image acquisition method based on multi-temporal data of geostationary meteorological satellite remote sensing images.

[0074] Additionally, this embodiment may also include a computer-readable storage medium storing a program that, when executed by a processor, implements a method for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images.

[0075] In summary, this application provides a comprehensive algorithm for clear sky image acquisition based on multi-temporal data of geostationary meteorological satellite remote sensing images. This algorithm can synthesize clear sky images based on multi-temporal visible spectral data to capture the characteristics of different land surfaces. By utilizing multi-temporal geostationary satellite remote sensing data and combining rapidly changing cloud systems within a single day with relatively unchanging underlying surfaces, the algorithm can acquire the surface texture features under clear sky conditions, thereby solving problems such as "patches" and cloud shadows in existing synthesized images.

[0076] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating this application and are not intended to limit the implementation. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A method for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing imagery, characterized in that, include: Remote sensing image data of the same area at multiple times are acquired and sorted according to time to obtain an image set; Convert the DN values ​​of relevant original image pixels in the image set into apparent reflectance; Correct relevant images in the image set to ensure consistency in geographical location and spatial resolution; For each pixel of a related image in the image set, the apparent reflectance is used to determine whether it is a clear sky or a cloudy sky category. A clear sky image is constructed based on the pixels in each image set that are classified as clear sky. The step of determining whether a pixel-by-pixel related image in the image set is classified as clear sky or cloudy based on apparent reflectance includes the following steps: S1. Under the selected band, calculate the mean value μ of the apparent reflectance B of each pixel in the first image of the image set. img For apparent reflectance B less than μ img The corresponding pixels are identified as clear sky category and the mean value μ is calculated. l-img ,variance For apparent reflectance B not less than μ img The corresponding pixels are identified as having cloud cover and the mean value μ is calculated. h-img ,variance S2, the clear sky category count value of the corresponding pixel position in the image set when the cumulative judgment reaches the Mth image. Cloud category count value The mean apparent reflectance of clear sky category is The variance of apparent reflectance for clear sky categories is The average apparent reflectance of the cloud category is The variance of apparent reflectance for cloud categories is Assign initial values ​​to the first image when M=1. And based on the apparent reflectance B of each pixel in the first image, we obtain: S3. Determine the apparent reflectance B of each pixel in the M≥2th image: when or And f l (B)>f h (B) indicates a clear sky category, and the corresponding pixel is updated. when or And f l (B)≤f h (B) indicates a cloud cover category, and the corresponding pixels are updated. Where f l (B), f h (B) represents the probability density function of apparent reflectance distribution for clear sky and cloud cover, respectively: When constructing a clear sky image based on pixels identified as clear sky category in each image of the image set, the average apparent reflectance of pixels at the corresponding pixel positions in the image set under multiple bands for the clear sky category is calculated and then synthesized into a clear sky image. The apparent reflectance B of each pixel in the image set is: L = gain × DN + bias, In the formula, L is the irradiance, DN is the gray value of the original image pixel, gain and bias are the gain and bias of the sensor that acquired the remote sensing image in the corresponding band, d is the Earth-Sun distance, E0 is the solar irradiance in the corresponding band, and θ is the solar zenith angle of the corresponding image.

2. The method for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing imagery as described in claim 1, characterized in that, The remote sensing image data is visible and near-infrared remote sensing imagery, including spectral band data, the gain and bias of the sensors that acquired the images in different bands, the solar elevation angle of the images, and the geographic information of the images.

3. The method for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images as described in claim 2, characterized in that, In step S1, the selected band is the blue light band.

4. The method for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing imagery as described in claim 3, characterized in that, When calculating the average apparent reflectance of pixels at corresponding pixel locations in the image set under the clear sky category in multiple bands, the discrimination result of the relevant image pixel under the blue light band is used when determining the clear sky category or the cloudy category of the relevant image pixels in the corresponding image set of the multiple bands.

5. An apparatus for acquiring clear-sky images based on multi-temporal data of geostationary meteorological satellite remote sensing imagery, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the clear sky image acquisition method based on multi-temporal data of geostationary meteorological satellite remote sensing imagery as described in any one of claims 1-4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method for acquiring clear sky images based on multi-temporal data of geostationary meteorological satellite remote sensing images as described in any one of claims 1-4.

Images