Method and device for acquiring super-resolution magnetic map

By performing local magnetic map segmentation and magnetic flux classification on the solar magnetogram, and super-resolution reconstruction using different neural network structures, the problem of difficulty in combining magnetic map data and large amount of calculation in the existing technology is solved, and high quality and efficient Super resolution magnetogram acquisition.

CN119090720APending Publication Date: 2024-12-06NINGBO UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411169003.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

When the prior art combines different solar earthquakes and magnetic map data provided by magnetic imagers, it is difficult to effectively study the small-scale magnetic field structure of the sun, and the super-resolution reconstruction process is computationally large, which increases the application cost.

Method used

By segmenting the local magnetic map of the solar magnetic map, classifying the magnetic flux of each overlapping image block, and super-resolution reconstruction is used using different neural network structures to obtain high-quality super-resolution magnetic maps.

Benefits of technology

It improves the quality of the reconstruction image, makes more efficient use of computing resources, reduces the computational complexity, and enhances the accuracy and efficiency of super-resolution reconstruction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090720A_ABST
    Figure CN119090720A_ABST
Patent Text Reader

Abstract

The invention discloses an acquisition method and an acquisition device for a super-resolution magnetic map. The method comprises the following steps: acquiring a local magnetic map of a solar magnetic map; segmenting the local magnetic image to obtain N overlapped image blocks; the N overlapped image blocks are classified according to the magnetic flux of each overlapped image block, at least one subset is obtained, each subset comprises at least one overlapped image block, and N is a positive integer larger than 1; obtaining a corresponding super-resolution sub-image block according to each subset and the corresponding neural network structure; and splicing the super-resolution sub-image blocks corresponding to each subset to obtain a super-resolution magnetic image. Therefore, according to the method, the N overlapped image blocks are classified based on the magnetic flux, and different types of overlapped image blocks are subjected to super-resolution reconstruction by using different neural network structures, so that the quality of the reconstructed image can be improved, and computing resources can be more effectively utilized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic image acquisition, and in particular to a method for acquiring a super-resolution magnetic image and a device for acquiring a super-resolution magnetic image. Background Art

[0002] Space-based instruments provide us with a large number of solar magnetic field observation images, which are called magnetograms. These magnetograms measure the intensity and polarity of the solar magnetic field in detail, providing an important basis for studying the source and evolution of the solar magnetic field. Magnetograms are mainly provided by the Helioseismic and Magnetic Imager (HMI) and the Michelson Doppler Imager (MDI). Because they have differences in resolution, spectral inversion technology, instrument noise level or other instrument characteristics, it is still difficult to combine the data of various instruments to study the small-scale magnetic field structure of the sun despite the large amount of magnetogram data.

[0003] In the related art, a low-resolution magnetic map (LR) is reconstructed as a whole into a high-quality high-resolution (HR) magnetic map through magnetic map super-resolution (SR). However, this technical solution has a large amount of calculation and increases the application cost. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems in the related art to a certain extent. To this end, the first object of the present invention is to propose a method for obtaining a super-resolution magnetic map, classifying N overlapping image blocks based on magnetic flux, and super-reconstructing overlapping image blocks of different categories using different neural network structures, which can not only improve the quality of the reconstructed image, but also more effectively utilize computing resources.

[0005] The second object of the present invention is to provide a device for acquiring a super-resolution magnetic image.

[0006] To achieve the above-mentioned purpose, the first aspect of the embodiment of the present invention proposes a method for obtaining a super-resolution magnetic map, the method comprising: obtaining a local magnetic map of a solar magnetic map; segmenting the local magnetic map to obtain N overlapping image blocks; classifying the N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, and N is a positive integer greater than 1; obtaining corresponding super-resolution sub-blocks according to each subset and its corresponding neural network structure; splicing the super-resolution sub-blocks corresponding to each subset to obtain a super-resolution magnetic map.

[0007] According to the method for obtaining a super-resolution magnetic map of an embodiment of the present invention, firstly, a local magnetic map of the solar magnetic map is obtained, and the local magnetic map is segmented to obtain N overlapping image blocks, and the N overlapping image blocks are classified according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, and N is a positive integer greater than 1; the corresponding super-resolution sub-image blocks are obtained according to each subset and its corresponding neural network structure, and then the super-resolution sub-image blocks corresponding to each subset are spliced ​​to obtain a super-resolution magnetic map. Therefore, the method classifies N overlapping image blocks based on magnetic flux, and uses different neural network structures to perform super-resolution reconstruction on overlapping image blocks of different categories, which can not only improve the quality of the reconstructed image, but also more effectively utilize computing resources.

[0008] In addition, the method for acquiring a super-resolution magnetic map according to the above embodiment of the present invention may also have the following additional technical features:

[0009] According to one embodiment of the present invention, obtaining a local magnetic map of a solar magnetic map includes: obtaining multiple MDI magnetic maps and multiple HMI magnetic maps; cutting the multiple MDI magnetic maps and multiple HMI magnetic maps according to the center position of the sun to obtain the local magnetic maps of the multiple solar magnetic maps.

[0010] According to one embodiment of the present invention, N overlapping image blocks are classified according to the magnetic flux of each overlapping image block to obtain at least one subset, including: according to the size relationship between the magnetic flux of each overlapping image block and a preset M magnetic flux threshold, the N overlapping image blocks are divided into M+1 types of subsets, where M is a positive integer greater than 1.

[0011] According to one embodiment of the present invention, when M is 2, N overlapping image blocks are classified according to the magnetic flux of each overlapping image block to obtain at least one subset, including: when the magnetic flux of the overlapping image block is less than a first preset magnetic flux threshold, determining that the overlapping image block is a first subset; when the magnetic flux of the overlapping image block is greater than or equal to the first preset magnetic flux threshold and less than a second preset magnetic flux threshold, determining that the overlapping image block is a second subset; when the magnetic flux of the overlapping image block is greater than or equal to the second preset magnetic flux threshold, determining that the overlapping image block is a third subset.

[0012] According to one embodiment of the present invention, the number of channels of the neural network structure corresponding to the first subset is the first number of channels, the number of channels of the neural network structure corresponding to the second subset is the second number of channels, and the number of channels of the neural network structure corresponding to the third subset is the third number of channels, wherein the first number of channels is smaller than the second number of channels, and the second number of channels is smaller than the third number of channels.

[0013] According to one embodiment of the present invention, a corresponding super-resolution sub-image block is obtained according to each subset and its corresponding neural network structure, including: performing image alignment on the MDI overlapping image blocks and the HMI overlapping image blocks in the subset to obtain aligned overlapping image blocks; and using the corresponding neural network structure according to the aligned overlapping image blocks to obtain the corresponding super-resolution sub-image block.

[0014] According to an embodiment of the present invention, the method for acquiring a super-resolution magnetic map further includes: performing pixel adjustment on a plurality of MDI magnetic maps and a plurality of HMI magnetic maps.

[0015] According to one embodiment of the present invention, the method for obtaining a super-resolution magnetic map also includes: dividing the super-resolution magnetic map into multiple small blocks; obtaining the average value and variance of the data points of each small block and the real average value and real variance of the data points of the corresponding real small block; obtaining an evaluation index of the super-resolution magnetic map based on the average value and variance of each small block and the corresponding real average value and real variance; when the evaluation index is greater than a preset threshold, determining that the super-resolution magnetic map is valid.

[0016] According to one embodiment of the present invention, the evaluation index is determined by the following formula:

[0017]

[0018]

[0019] Among them, MFSS(x,y) represents the evaluation index, σ xy The variance of the data points of the small blocks divided by the super-resolution magnetic map and the corresponding real small blocks, σ x The variance of the data points representing the small blocks of the super-resolution magnetic map, σ y represents the variance of the corresponding real small patch of data points, represents the average value of the data points of the small block divided by the super-resolution magnetic map, xi represents the i-th data point of the small block divided by the super-resolution magnetic map, Represents the average value of the corresponding real small block data points, y i represents the i-th data point of the corresponding small block, c1 represents a constant, and N represents the maximum value of the magnetic field strength.

[0020] To achieve the above-mentioned purpose, the second embodiment of the present invention proposes a device for acquiring a super-resolution magnetic map, which includes: a first acquisition module, used to acquire a local magnetic map of a solar magnetic map; a segmentation module, used to segment the local magnetic map to obtain N overlapping image blocks; a classification module, used to classify the N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, and N is a positive integer greater than 1; a second acquisition module, used to obtain corresponding super-resolution sub-blocks according to each subset and its corresponding neural network structure; a splicing module, used to splice the super-resolution sub-blocks corresponding to each subset to obtain a super-resolution magnetic map.

[0021] According to the device for acquiring a super-resolution magnetic map of an embodiment of the present invention, a local magnetic map of a solar magnetic map is acquired by a first acquisition module, the local magnetic map is segmented by a segmentation module to obtain N overlapping image blocks, the N overlapping image blocks are classified by a classification module according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, N is a positive integer greater than 1, a corresponding super-resolution sub-image block is obtained by a second acquisition module according to each subset and its corresponding neural network structure, and a splicing module splices the super-resolution sub-image blocks corresponding to each subset to obtain a super-resolution magnetic map. Thus, the device classifies N overlapping image blocks based on magnetic flux, and uses different neural network structures to perform super-resolution reconstruction on overlapping image blocks of different categories, which can not only improve the quality of the reconstructed image, but also more effectively utilize computing resources.

[0022] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a method for acquiring a super-resolution magnetic map according to an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of an overall network framework according to a specific embodiment of the present invention;

[0025] Figure 3 A schematic diagram of main characteristic points of an MDI magnetic map and an HMI magnetic map according to a specific embodiment of the present invention;

[0026] Figure 4 A schematic diagram of the distribution of the magnetic flux of each sample in the training set according to a specific embodiment of the present invention;

[0027] Figure 5 A flowchart of a method for acquiring a super-resolution magnetic map according to a specific embodiment of the present invention;

[0028] Figure 6 Schematic diagram of the connection of a device for acquiring a super-resolution magnetic image according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0030] The following describes a method for acquiring a super-resolution magnetic map and a device for acquiring a super-resolution magnetic map according to an embodiment of the present invention with reference to the accompanying drawings.

[0031] Space-based instruments provide us with a large number of solar magnetic field observation images, which are called magnetograms. These magnetograms measure the intensity and polarity of the sun's magnetic field in detail, providing an important basis for studying the source and evolution of the sun's magnetic field. Not only that, these magnetograms are also indispensable in detecting solar mechanisms, understanding the corona, and predicting space weather events. In the process of acquiring magnetograms, the resolution of the full-disk solar magnetogram provided by the Michelson Doppler Imager (MDI) is 2 arc seconds, and the resolution of the full-disk solar magnetogram provided by the Helioseismic and Magnetic Imager (HMI) is 0.5 arc seconds. Despite the large amount of magnetogram data, it is still difficult to combine the data of various instruments to study the small-scale magnetic field structure of the sun because of differences in resolution, spectral inversion technology, instrument noise level or other instrument characteristics between MDI and HMI.

[0032] In related technologies, low-resolution magnetic maps are reconstructed into high-quality high-resolution magnetic maps through magnetic map super-resolution, so that the resolution of the magnetic maps provided by MDI and HMI are the same, thereby constructing a consistent high-resolution image data set to compensate for the inhomogeneity and discontinuity between different instruments. The spatial resolution of the MDI magnetic map can also be improved to a higher level, thereby obtaining more detailed and accurate magnetic field data, which in turn helps to study solar flare forecasts and space weather events.

[0033] With the powerful capabilities of neural networks in feature representation and model fitting, image super-resolution based on deep learning has become a hot topic of research and has achieved remarkable results on natural images. In addition, this technology has also received widespread attention from scholars in the field of physical science because it can simultaneously achieve super-resolution (adding scientifically accurate details to images) and cross-calibration (correcting systematic differences between instruments). For example, two methods based on convolutional neural networks (CNNs), namely the residual attention model and the progressive GAN model, were used to super-resolution the HMI magnetogram, but the model was only trained and tested on the low-resolution HMI magnetogram without using the MDI magnetogram; for another example, the HighRes-Net method was used to super-resolution the full-disk MDI magnetogram to match the resolution of the HMI magnetogram; for another example, the SPSR-GAN model was used to reconstruct the magnetic map of the MDI active region. However, although the deep learning-based solar magnetogram super-resolution methods have achieved significant improvements in performance, most of them borrowed the SR technology of natural images and lacked the ability to discern the complexity of the magnetic field.

[0034] To this end, the present invention proposes a method for acquiring a super-resolution magnetic map. The method classifies N overlapping image blocks according to the magnetic flux of each overlapping image block, and processes the image blocks based on the corresponding neural network structure. The method can more accurately handle the magnetic field complexity of different regions, and adopts different neural network structures to process the active region (AR) and the quiet region (QR), which more effectively utilizes computing resources and improves the accuracy and efficiency of super-resolution. It can provide more accurate results for solar magnetic map processing, help to better understand the magnetic field structure and activity laws of the sun, and is of great significance for solar physics research and space weather forecasting.

[0035] The method for obtaining a super-resolution magnetic map of the present application is described in detail below with reference to the accompanying drawings.

[0036] Figure 1 The figure is a flow chart of a method for acquiring a super-resolution magnetic map according to an embodiment of the present invention.

[0037] like Figure 1 As shown, the method for obtaining a super-resolution magnetic map according to an embodiment of the present invention may include:

[0038] S1, obtain the local magnetic map of the solar magnetogram;

[0039] S2, segmenting the local magnetic map to obtain N overlapping image blocks;

[0040] S3, classifying the N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, and N is a positive integer greater than 1;

[0041] S4, obtaining the corresponding super-resolution sub-block according to each subset and its corresponding neural network structure;

[0042] S5, splicing the super-resolution sub-image blocks corresponding to each subset to obtain a super-resolution magnetic map.

[0043] Specifically, in order to solve the problems faced by neural network models in image super-resolution, the characteristics of the dataset must be fully considered when designing the neural network model to ensure a balance between model complexity and computing resources during image reconstruction. Generally, lightweight networks with fewer parameters are unable to cope with images with higher complexity, while complex networks with too many parameters are prone to overfitting when processing background images, resulting in inefficient use of computing resources. The increase in the number of model parameters increases the required computing resource rate, which also limits the performance of super-resolution.

[0044] To this end, the present application first divides the local magnetic map of the solar magnetic map into N overlapping image blocks, calculates the magnetic flux of each overlapping image block, and classifies the N overlapping image blocks according to the magnetic flux. Taking the setting of two magnetic flux thresholds as an example, three subsets are obtained by comparing the magnetic flux of each overlapping image block with the two magnetic flux thresholds. The three subsets correspond to the three types of simple, medium and complex. In addition, since the solar magnetic map is significantly different from the natural image and has a wider dynamic range, the present application uses the physical parameter magnetic field strength value of the solar magnetic map for classification.

[0045] Then, for overlapping image blocks of different categories (i.e., subsets), neural network structures of different complexity are used to reconstruct the target, thereby achieving adaptive processing of images of different complexity. Through such a neural network structure and classification method, image blocks of different complexity can be effectively processed, improving the accuracy and efficiency of super-resolution reconstruction.

[0046] Furthermore, a multi-branch super-resolution network can be established corresponding to the acquisition method of the super-resolution magnetic map, such as Figure 2As shown in Figure 1, the network consists of two main parts: an image classification module and a multi-branch SR module. In the classification module, the input magnetic map is segmented into small overlapping image patches. These image patches are then classified into three categories based on the unsigned total magnetic flux: simple (weak magnetic field, such as quiet areas), medium (medium magnetic field), and complex (strong magnetic field, such as active areas). Through this classification method, we can use networks of different complexities for different types of image patches for super-resolution reconstruction, which not only improves the quality of the reconstructed image, but also makes more efficient use of computing resources, thereby improving the performance of super-resolution.

[0047] In one embodiment of the present invention, obtaining a local magnetic map of a solar magnetic map includes: obtaining multiple MDI magnetic maps and multiple HMI magnetic maps; cutting the multiple MDI magnetic maps and multiple HMI magnetic maps according to the center position of the sun to obtain the local magnetic maps of the multiple solar magnetic maps.

[0048] Specifically, the MDI magnetic map is a magnetic map provided by the Michelson Doppler imager, and the HMI magnetic map is a magnetic map provided by the solar seismic and magnetic imager.

[0049] In actual applications, the solar earthquake and magnetic imager takes a line-of-sight magnetic map every 45 seconds, while the Michelson Doppler imager takes one every 96 minutes, so there is time overlap. The random position, rotation, and time differences between the MDI and HMI magnetic maps will affect the model's ability to learn mapping. Therefore, before training, we need to eliminate these differences. The specific steps are:

[0050] 1) Decentralization: Determine the solar radius according to the RSUN_OBS parameter in the fits header file, find the CDELTI conversion factor, and convert the solar radius to heliospheric coordinates. Next, determine the solar center position according to the IMCRPIX1 and IMCRPIX2 parameters. Finally, according to the solar radius, use the circumscribed rectangle to cut out the full solar disk magnetic map.

[0051] 2) Rotation processing: Apply the instrument pointing information contained in the header file to rotate the north side of the sun to the north side of the image, so that the direction of the sun in each image is normalized (Map.rotate(order=3)).

[0052] 3) Normalization: Convert the magnetic field intensity data into image data in the range of [0, 255]. This processing not only speeds up the training of the network model, but also prevents the model gradient from exploding and reduces the possibility of the model falling into the local optimum.

[0053] After the above processing steps, the difference between the MDI magnetic map and the HMI magnetic map is weakened, providing a more accurate and reliable data basis for subsequent super-resolution reconstruction.

[0054] In one embodiment of the present invention, N overlapping image blocks are classified according to the magnetic flux of each overlapping image block to obtain at least one subset, including: according to the size relationship between the magnetic flux of each overlapping image block and a preset M magnetic flux threshold, the N overlapping image blocks are divided into M+1 types of subsets, where M is a positive integer greater than 1.

[0055] That is to say, assuming that two magnetic flux thresholds are preset, the magnetic flux of each overlapping image block is compared with the two magnetic flux thresholds, thereby dividing the N overlapping image blocks into three subsets. For another example, if four magnetic flux thresholds are preset, the magnetic flux of each overlapping image block is compared with the four magnetic flux thresholds, thereby dividing the N overlapping image blocks into five subsets. The number M of magnetic flux thresholds can be set according to actual conditions.

[0056] Thus, the N overlapping image blocks are classified by the pre-set M magnetic flux thresholds, and M+1 subsets are obtained, and the classified overlapping image blocks are respectively entered into the corresponding neural network structure, so as to realize adaptive processing of image blocks of different complexities, and can effectively process overlapping image blocks with different complexities, thereby improving the accuracy and efficiency of super-resolution reconstruction. In addition, after calculating the magnetic flux threshold of each overlapping image block, these values ​​are sorted in ascending order, and then the magnetic flux threshold is determined based on the magnetic flux of the N overlapping image blocks to classify the N overlapping image blocks, and there is no specific limitation.

[0057] In one embodiment of the present invention, when M is 2, N overlapping image blocks are classified according to the magnetic flux of each overlapping image block to obtain at least one subset, including: when the magnetic flux of the overlapping image block is less than a first preset magnetic flux threshold, determining that the overlapping image block is a first subset; when the magnetic flux of the overlapping image block is greater than or equal to the first preset magnetic flux threshold and less than a second preset magnetic flux threshold, determining that the overlapping image block is a second subset; when the magnetic flux of the overlapping image block is greater than or equal to the second preset magnetic flux threshold, determining that the overlapping image block is a third subset.

[0058] That is to say, when M is 2, the two pre-set magnetic flux thresholds are respectively the first preset magnetic flux threshold and the second magnetic flux threshold, and the first preset magnetic flux threshold is smaller than the second magnetic flux threshold, thereby dividing the N overlapping image blocks into three subsets, namely the first subset, the second subset and the third subset.

[0059] Assume that the magnetic flux distribution of each sample is as follows Figure 3 As shown, let the first magnetic flux threshold t1 = 5.7 × 10 5 , the second magnetic flux threshold t2 = 8.9 × 10 5. Compare the magnetic flux of each overlapping image block with the first magnetic flux threshold t1 and the second magnetic flux threshold t2 respectively. If the magnetic flux of the overlapping image block is less than t1, the overlapping image block is divided into the first subset; if t1≤the magnetic flux of the overlapping image block is less than t2, the overlapping image block is divided into the second subset; if the magnetic flux of the overlapping image block is greater than or equal to t2, the overlapping image block is divided into the third subset. Then the magnetic flux range corresponding to the first subset is [0, t1), and the overlapping blocks in the first subset are of simple type; the magnetic flux range corresponding to the second subset is [t1, t2), and the overlapping blocks in the second subset are of medium type; the magnetic flux range corresponding to the third subset is [t2, +∞), and the overlapping blocks in the third subset are of complex type.

[0060] In one embodiment of the present invention, the number of channels of the neural network structure corresponding to the first subset is the first number of channels, the number of channels of the neural network structure corresponding to the second subset is the second number of channels, and the number of channels of the neural network structure corresponding to the third subset is the third number of channels, wherein the first number of channels is smaller than the second number of channels, and the second number of channels is smaller than the third number of channels.

[0061] In other words, the neural network is a multi-branch neural network structure that can efficiently process tiles of different complexities.

[0062] When N overlapping tiles are divided into three subsets, the neural network structure includes three branches, each branch having a different number of channels, for example, 32, 48 and 64. That is, the number of channels of the neural network structure corresponding to the first subset is 32, the number of channels of the neural network structure corresponding to the second subset is 48, and the number of channels of the neural network structure corresponding to the third subset is 64.

[0063] As a result, the neural network structure can meet the processing requirements of different overlapping tiles, ensuring that each branch can efficiently process specific types of overlapping tiles, improving the accuracy and efficiency of super-resolution reconstruction. At the same time, the multi-branch neural network structure makes it easier and more comprehensive for each branch to be trained, especially for branches that process complex data, ensuring that the true details of the magnetic map are preserved while ensuring the accuracy of the scientific data.

[0064] In one embodiment of the present invention, a corresponding super-resolution sub-image block is obtained according to each subset and its corresponding neural network structure, including: performing image alignment on the MDI overlapping image blocks and the HMI overlapping image blocks in the subset to obtain aligned overlapping image blocks; and using the corresponding neural network structure according to the aligned overlapping image blocks to obtain the corresponding super-resolution sub-image block.

[0065] Specifically, since the shooting time of the MDI magnetic map and the HMI magnetic map is different, it is necessary to screen out the HMI magnetic map that is closest to the shooting time of the MDI magnetic map. At the same time, due to the difference in shooting positions, pixel misalignment may occur. In order to match these features more accurately, the classic ORB feature matching algorithm is used to align the MDI overlapping image blocks and the HMI overlapping image blocks in the subset to obtain the aligned overlapping image blocks. The core of this algorithm is to find multiple key feature points, match the features of the reference image (HMI image) with the sensor image (MDI image), and realize the alignment of the two images by converting the sensor image. Figure 4 As shown in the figure, the key feature points of the MDI and HMI magnetograms are shown, which are used to ensure accurate alignment between the images. Through the image alignment processing step, the subtle changes in the solar magnetic field structure can be analyzed more accurately, providing a reliable basis for subsequent super-resolution reconstruction.

[0066] Then, the corresponding neural network structure is used according to the aligned overlapping image blocks to obtain the corresponding super-resolution sub-blocks. Figure 2 As shown in the figure, the neural network structure of each branch consists of two parts: the gradient path and the content path to balance detail recovery and magnetic field fidelity. Among them, the gradient path plays a vital role in recovering the fine features of the magnetic field structure. Its input is the gradient map of the overlapping image, which is convolved through a series of 3×3 convolution kernels. There are 5 residual dense blocks in total, which help to extract subtle changes and local features in the image. The content path focuses on ensuring the fidelity of the magnetic field. Its input is the overlapping image itself, which is also convolved through 5 residual dense blocks and uses a 3×3 convolution kernel. After the processing is completed, the feature map is output by upsampling, and then feature fusion is performed through a residual dense block and a 1×1 convolution layer. This step aims to integrate the information of the gradient branch and the content branch to produce a richer and more accurate super-resolution magnetic field map.

[0067] Furthermore, in the data processing stage of overlapping image blocks, the collected sample data needs to be screened and sorted, and damaged or poor quality data is eliminated to ensure the accuracy and reliability of the data. For example, in order to ensure the effectiveness of neural network model training, 6742 HMI-MDI local magnetic maps around the center of the solar disk were selected from the aligned full-disk magnetic map dataset. These local magnetic maps are randomly cropped from the full-disk magnetic map without a fixed step size and size. They contain various strong / weak magnetic field areas and can provide rich feature information for the model. When dividing the data set, in order to maintain the temporal continuity and authenticity of the data, 6282 pairs of data from May 1, 2010 to November 15, 2010 were used as the training data set to ensure that the model can learn the evolution law of the magnetic field over a period of time; 170 pairs of data from November 15, 2010 to November 30, 2010 were used as the validation data set to adjust the model parameters and prevent overfitting; 290 pairs of data from December 1, 2010 to December 31, 2010 were used as the test data set to evaluate the final performance of the model. The purpose of dividing the data set by month is to retain some temporal evolution characteristics of the magnetic field during training. If the images are randomly assigned, the training data set and the test data set may differ by only a few minutes, which will cause the model to face similar magnetic field states during training and testing, making it impossible to accurately evaluate the generalization ability of the model. Therefore, dividing the data set by month can better simulate the evolution of the magnetic field in different time periods and provide a better training environment for the model. The detailed information of the training set, validation set, and test set is shown in Table 1.

[0068] Table 1

[0069] Training set Validation set Test Set time 2010 / 05-2010 / 11 / 15 2010 / 11 / 15-2010 / 11 / 30 2010 / 12 Sample size 6282 170 290 Number of sub-graphs 70674 1912 6830

[0070] In one embodiment of the present invention, the method for acquiring a super-resolution magnetic map further includes: performing pixel adjustment on a plurality of MDI magnetic maps and a plurality of HMI magnetic maps.

[0071] That is to say, in order to more intuitively reveal the details of the solar magnetic field structure, first, the numerical range of the magnetic map is adjusted to [-500, +500] to highlight the actual range of variation of the magnetic field intensity. Secondly, considering the elliptical characteristics of the Earth's orbit, the preset MDI magnetic map can be adjusted to 465 pixels and the HMI magnetic map can be adjusted to 1860 pixels. The specific pixel setting can be made according to the actual situation.

[0072] In one embodiment of the present invention, the method for obtaining a super-resolution magnetic map also includes: dividing the super-resolution magnetic map into multiple small blocks; obtaining the average value and variance of the data points of each small block and the real average value and real variance of the data points of the corresponding real small block; obtaining an evaluation index of the super-resolution magnetic map based on the average value and variance of each small block and the corresponding real average value and real variance; when the evaluation index is greater than a preset threshold, determining that the super-resolution magnetic map is valid.

[0073] In one embodiment of the present invention, the evaluation index is determined by the following formula:

[0074]

[0075] Among them, MFSS(x,y) represents the evaluation index, σ xy The variance of the data points of the small blocks divided by the super-resolution magnetic map and the corresponding real small blocks, σ x The variance of the data points representing the small blocks of the super-resolution magnetic map, σ y represents the variance of the corresponding real small patch of data points, represents the average value of the data points of the small blocks divided by the super-resolution magnetic map, x i represents the ith data point of a small block of super-resolution magnetic map, Represents the average value of the corresponding real small block data points, y i represents the i-th data point of the corresponding small block, c1 represents a constant, and N represents the maximum value of the magnetic field strength.

[0076] Among them, c1 is a constant, and its value is (k*M), which is to avoid the situation where the denominator is 0. Among them, k and M represent two parameters respectively, and their specific values ​​need to be set according to the actual situation.

[0077] In order to evaluate the effectiveness of the generated super-resolution magnetic map, this embodiment evaluates the effectiveness of the magnetic map based on the evaluation index MFSS (Magnetic Field Structure Similarity) of the super-resolution magnetic map. This index is based on the concept of covariance and is used to measure structural similarity.

[0078] Specifically, the super-resolution magnetic map needs to be divided into multiple small blocks and the average value of the data points in each small block is obtained. and variance σ y And the corresponding real average value of the data points of the small block and the true variance σ y , and then put it into formula (1) to calculate the evaluation index MFSS of super-resolution magnetic map. Among them, the variance is used to balance the degree of dispersion between the data points and their mean.

[0079] Small blocks of the magnetic map are the basic units for evaluating structural similarity. By comparing the structural characteristics of different small blocks, the similarity between the generated super-resolution magnetic map and the real magnetic map can be more accurately evaluated.

[0080] In addition, in actual operation, the super-resolution magnetic map can be divided into small blocks of various sizes, and by comparing the MFSS values ​​calculated from small blocks of various sizes, the structural characteristics of the generated super-resolution magnetic map can be more comprehensively understood, thereby obtaining more accurate evaluation results. For example, three different sizes of small blocks, 7, 11 and 15, were tested respectively, and the structural characteristics of the generated super-resolution magnetic map were evaluated by comparing the MFSS values ​​calculated from these three different sizes of small blocks. This multi-scale evaluation method can more comprehensively reveal the details and characteristics of the generated magnetic map structure, and provide a more accurate evaluation basis for the performance of the super-resolution reconstruction algorithm. At the same time, this flexible way of adjusting the size of small blocks also makes the research more versatile and adaptable, and can better adapt to magnetic field data of different scales.

[0081] This embodiment can quantify the structural similarity between the generated magnetic map and the real magnetic map through MFSS. The closer the MFSS value is to 1, the more similar the structure of the generated magnetic map is to the real magnetic map, thereby evaluating the performance of the generated magnetic map structure. The introduction of this evaluation index helps to more accurately evaluate the quality of the generated magnetic map and provide an objective evaluation basis for the performance of the super-resolution reconstruction algorithm.

[0082] As a specific embodiment of the present invention, Figure 5 As shown, the method for obtaining the super-resolution magnetic map may include the following steps:

[0083] S101, obtaining multiple MDI magnetic images and multiple HMI magnetic images;

[0084] S102, performing pixel adjustment on the multiple MDI magnetic images and the multiple HMI magnetic images.

[0085] S103, cutting the multiple MDI magnetic maps and the multiple HMI magnetic maps according to the center position of the sun to obtain local magnetic maps of the multiple solar magnetic maps.

[0086] S104, segmenting the local magnetic map to obtain N overlapping image blocks.

[0087] S105 , classifying the N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain three subsets.

[0088] S106, obtaining a corresponding super-resolution sub-block according to each subset and its corresponding neural network structure.

[0089] S107, splicing the super-resolution sub-image blocks corresponding to each subset to obtain a super-resolution magnetic map.

[0090] S108, dividing the super-resolution magnetic map into a plurality of small blocks.

[0091] S109, obtaining the mean value and variance of the data points of each small block and the real mean value and real variance of the data points of the corresponding real small block.

[0092] S110, obtaining evaluation indicators of the super-resolution magnetic map according to the average value and variance of each small block and the corresponding true average value and true variance.

[0093] This method treats active regions and quiet regions differently, allocating more computing resources to active regions and less computing resources to quiet regions. In this way, the magnetic field complexity of different regions can be processed more accurately, thereby improving the accuracy and efficiency of super-resolution, providing more accurate results for solar magnetic map processing, and helping to better understand the magnetic field structure and activity patterns of the sun.

[0094] A multi-branch super-resolution network is established corresponding to the acquisition method of the super-resolution magnetic map, such as Figure 2 As shown. The network is divided into two main parts: the classification module and the super-resolution module. First, the classification module is responsible for the detailed segmentation of the local magnetic map, dividing it into N overlapping image blocks. These overlapping image blocks are further classified into three types: "simple", "medium" or "complex" according to their magnetic flux values. Subsequently, these sub-blocks are input into the corresponding branches for super-resolution reconstruction. In the multi-branch super-resolution module, three different branches are designed, each branch targets a type of overlapping image block, and each branch adopts a different number of network parameters to adapt to sub-blocks of different complexities. Finally, the outputs of the three branches are merged to form the final super-resolution magnetic map.

[0095] The network model can super-resolve the low-resolution MDI magnetic map into a high-resolution magnetic map with the same resolution as the HMI, ensuring adaptive processing for different types of overlapping image blocks, improving the quality and accuracy of the reconstructed image, and effectively improving the utilization efficiency of computing resources. Through reasonable allocation of computing resources, the computational complexity is effectively reduced.

[0096] In summary, according to the method for obtaining a super-resolution magnetic map in an embodiment of the present invention, firstly, a local magnetic map of the solar magnetic map is obtained, and the local magnetic map is segmented to obtain N overlapping image blocks, and the N overlapping image blocks are classified according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, and N is a positive integer greater than 1; a corresponding super-resolution sub-image block is obtained according to each subset and its corresponding neural network structure, and then the super-resolution sub-image blocks corresponding to each subset are spliced ​​to obtain a super-resolution magnetic map. Therefore, the method classifies N overlapping image blocks based on magnetic flux, and uses different neural network structures to perform super-resolution reconstruction on overlapping image blocks of different categories, which can not only improve the quality of the reconstructed image, but also make more efficient use of computing resources.

[0097] Corresponding to the above embodiment, the present invention also proposes a device for acquiring a super-resolution magnetic image.

[0098] like Figure 6 As shown, the apparatus for acquiring a super-resolution magnetic image according to an embodiment of the present invention may include: a first acquisition module 10 , a segmentation module 20 , a classification module 30 , a second acquisition module 40 and a splicing module 50 .

[0099] Among them, the first acquisition module 10 is used to obtain the local magnetic map of the solar magnetic map. The segmentation module 20 is used to segment the local magnetic map to obtain N overlapping image blocks. The classification module 30 is used to classify the N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, and N is a positive integer greater than 1. The second acquisition module 40 is used to obtain the corresponding super-resolution sub-image block according to each subset and its corresponding neural network structure. The splicing module 50 is used to splice the super-resolution sub-image blocks corresponding to each subset to obtain the super-resolution magnetic map.

[0100] According to one embodiment of the present invention, the first acquisition module 10 acquires a local magnetic map of the solar magnetic map, and is specifically used to: acquire multiple MDI magnetic maps and multiple HMI magnetic maps; cut the multiple MDI magnetic maps and multiple HMI magnetic maps according to the center position of the sun to obtain local magnetic maps of the multiple solar magnetic maps.

[0101] According to one embodiment of the present invention, the classification module 30 classifies N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain at least one subset, and is specifically used to: divide the N overlapping image blocks into M+1 categories of subsets according to the size relationship between the magnetic flux of each overlapping image block and a preset M magnetic flux threshold, where M is a positive integer greater than 1.

[0102] According to one embodiment of the present invention, when M is 2, the classification module 30 classifies N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain at least one subset, which is specifically used to: when the magnetic flux of the overlapping image block is less than a first preset magnetic flux threshold, determine that the overlapping image block is a first subset; when the magnetic flux of the overlapping image block is greater than or equal to the first preset magnetic flux threshold and less than the second preset magnetic flux threshold, determine that the overlapping image block is a second subset; when the magnetic flux of the overlapping image block is greater than or equal to the second preset magnetic flux threshold, determine that the overlapping image block is a third subset.

[0103] According to one embodiment of the present invention, the number of channels of the neural network structure corresponding to the first subset is the first number of channels, the number of channels of the neural network structure corresponding to the second subset is the second number of channels, and the number of channels of the neural network structure corresponding to the third subset is the third number of channels, wherein the first number of channels is smaller than the second number of channels, and the second number of channels is smaller than the third number of channels.

[0104] According to one embodiment of the present invention, the second acquisition module 40 obtains the corresponding super-resolution sub-image block according to each subset and its corresponding neural network structure, and is specifically used for: performing image alignment on the MDI overlapping image blocks and the HMI overlapping image blocks in the subset to obtain aligned overlapping image blocks; and using the corresponding neural network structure according to the aligned overlapping image blocks to obtain the corresponding super-resolution sub-image block.

[0105] According to an embodiment of the present invention, the acquisition module 10 is further used to: perform pixel adjustment on the multiple MDI magnetic images and the multiple HMI magnetic images.

[0106] According to one embodiment of the present invention, the stitching module 50 is also used to: divide the super-resolution magnetic map into multiple small blocks; obtain the average value and variance of the data points of each small block and the real average value and real variance of the data points of the corresponding real small block; obtain an evaluation index of the super-resolution magnetic map based on the average value and variance of each small block and the corresponding real average value and real variance; when the evaluation index is greater than a preset threshold, determine that the super-resolution magnetic map is valid.

[0107] According to one embodiment of the present invention, the evaluation index is determined by the following formula:

[0108]

[0109]

[0110] Among them, MFSS(x,y) represents the evaluation index, σ xy The variance of the data points of the small blocks divided by the super-resolution magnetic map and the corresponding real small blocks, σ xThe variance of the data points representing the small blocks of the super-resolution magnetic map, σ y represents the variance of the corresponding real small patch of data points, represents the average value of the data points of the small blocks divided by the super-resolution magnetic map, x i represents the ith data point of a small block of super-resolution magnetic map, Represents the average value of the corresponding real small block data points, y i represents the i-th data point of the corresponding small block, c1 represents a constant, and N represents the maximum value of the magnetic field strength.

[0111] It should be noted that for details not disclosed in the apparatus for acquiring a super-resolution magnetic image in an embodiment of the present invention, please refer to the details disclosed in the method for acquiring a super-resolution magnetic image in the above embodiment of the present invention, and no further details will be given here.

[0112] According to the device for acquiring a super-resolution magnetic map of an embodiment of the present invention, a local magnetic map of a solar magnetic map is acquired by a first acquisition module, the local magnetic map is segmented by a segmentation module to obtain N overlapping image blocks, the N overlapping image blocks are classified by a classification module according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, N is a positive integer greater than 1, a corresponding super-resolution sub-image block is obtained by a second acquisition module according to each subset and its corresponding neural network structure, and a splicing module splices the super-resolution sub-image blocks corresponding to each subset to obtain a super-resolution magnetic map. Thus, the device classifies N overlapping image blocks based on magnetic flux, and uses different neural network structures to perform super-resolution reconstruction on overlapping image blocks of different categories, which can not only improve the quality of the reconstructed image, but also more effectively utilize computing resources.

[0113] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0114] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0115] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0116] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0117] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0118] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for obtaining a super-resolution magnetic map, characterized in that: The method comprises: Get local magnetic maps of the solar magnetogram; Segmenting the local magnetic map to obtain N overlapping image blocks; Classifying the N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, and N is a positive integer greater than 1; Obtaining a corresponding super-resolution sub-block according to each of the subsets and its corresponding neural network structure; The super-resolution sub-image blocks corresponding to each of the subsets are spliced ​​to obtain the super-resolution magnetic map.

2. The method for obtaining a super-resolution magnetic map according to claim 1, characterized in that: The method of obtaining a local magnetic map of a solar magnetic map comprises: Get multiple MDI magnetic images and multiple HMI magnetic images; The multiple MDI magnetic maps and the multiple HMI magnetic maps are cut according to the central position of the sun to obtain local magnetic maps of the multiple solar magnetic maps.

3. The method for obtaining a super-resolution magnetic map according to claim 1, characterized in that: The N overlapping image blocks are classified according to the magnetic flux of each overlapping image block to obtain at least one subset, including: According to the magnitude relationship between the magnetic flux of each overlapping image block and a preset M magnetic flux threshold, the N overlapping image blocks are divided into M+1 subsets, where M is a positive integer greater than 1.

4. The method for obtaining a super-resolution magnetic map according to claim 3, characterized in that: When M is 2, the N overlapping image blocks are classified according to the magnetic flux of each overlapping image block to obtain at least one subset, including: In a case where the magnetic flux of the overlapping image blocks is less than a first preset magnetic flux threshold, determining that the overlapping image blocks are a first subset; In a case where the magnetic flux of the overlapping image block is greater than or equal to a first preset magnetic flux threshold and less than a second preset magnetic flux threshold, determining that the overlapping image block is a second subset; When the magnetic flux of the overlapping image blocks is greater than or equal to the second preset magnetic flux threshold, the overlapping image blocks are determined to be a third subset.

5. The method for obtaining a super-resolution magnetic map according to claim 4, characterized in that: The number of channels of the neural network structure corresponding to the first subset is the first number of channels, the number of channels of the neural network structure corresponding to the second subset is the second number of channels, and the number of channels of the neural network structure corresponding to the third subset is the third number of channels, wherein the first number of channels is smaller than the second number of channels, and the second number of channels is smaller than the third number of channels.

6. The method for obtaining a super-resolution magnetic map according to claim 1, characterized in that: According to each of the subsets and its corresponding neural network structure, a corresponding super-resolution sub-block is obtained, including: Performing image alignment on the MDI overlapping image blocks and the HMI overlapping image blocks in the subset to obtain the aligned overlapping image blocks; The corresponding neural network structure is adopted according to the aligned overlapping image blocks to obtain corresponding super-resolution sub-blocks.

7. The method for obtaining a super-resolution magnetic map according to claim 2, characterized in that: The method further comprises: Pixel adjustment is performed on the multiple MDI magnetic images and the multiple HMI magnetic images.

8. The method for obtaining a super-resolution magnetic map according to claim 1, characterized in that: The method further comprises: Dividing the super-resolution magnetic map into a plurality of small blocks; Get the mean value and variance of each small block of data points and the true mean value and true variance of the corresponding real small block of data points; Obtaining evaluation indicators of the super-resolution magnetic map according to the average value and variance of each small block and the corresponding true average value and true variance; When the evaluation index is greater than a preset threshold, it is determined that the super-resolution magnetic map is valid.

9. The method for obtaining a super-resolution magnetic map according to claim 8, characterized in that: The evaluation index is determined by the following formula: Among them, MFSS(x,y) represents the evaluation index, σ xy represents the variance of the data points of the small blocks divided by the super-resolution magnetic map and the corresponding real small blocks, σ x represents the variance of the data points of the small blocks of the super-resolution magnetic map, σ y represents the variance of the corresponding real small patch of data points, represents the average value of the data points of the small blocks divided by the super-resolution magnetic map, x i represents the ith data point of the small block divided by the super-resolution magnetic map, Represents the average value of the corresponding real small block data points, y i represents the i-th data point of the corresponding small block, c1 represents a constant, and N represents the maximum value of the magnetic field strength.

10. A device for acquiring a super-resolution magnetic map, characterized in that: The device comprises: A first acquisition module is used to acquire a local magnetic map of the solar magnetic map; A segmentation module, used for segmenting the local magnetic map to obtain N overlapping image blocks; a classification module, configured to classify the N overlapping image blocks according to the magnetic flux of each overlapping image block to obtain at least one subset, wherein each subset includes at least one overlapping image block, and N is a positive integer greater than 1; A second acquisition module is used to obtain a corresponding super-resolution sub-block according to each of the subsets and its corresponding neural network structure; The splicing module is used to splice the super-resolution sub-image blocks corresponding to each of the subsets to obtain the super-resolution magnetic map.