Coronal mass ejection area automatic detection method based on velocity field

Abstract

The invention relates to an automatic detection method for a coronal mass ejection (CME) area based on a velocity field, and belongs to the field of image processing. In order to improve the accuracy of CME area automatic detection and consider the advantage that the velocity field is not influenced by CME observation data brightness and subjective factors, the invention provides a velocity field-based CME area automatic detection method. Experimental research on C2 observation data provided by SOHO satellites of the sun and the sunlight layer shows that the method provided by the invention has a good effect in the aspect of accurate identification of the CME area.

Description

Technical Field

[0001] This invention relates to an automatic detection method for coronal mass ejection (CME) regions based on velocity fields, specifically involving the acquisition of velocity fields from coronal mass ejection observation data and the application of threshold filtering, orientation filtering, outlier filtering, convex hull extraction, and CME region extraction based on the velocity fields, belonging to the field of image processing technology. Background Technology

[0002] A coronal mass ejection (CME) is a violent eruption phenomenon that occurs on the surface of the Sun, ejecting coronal material from the solar corona into interplanetary space. If the CME propagates toward Earth, it will disturb the environment near Earth, triggering various geophysical effects, seriously threatening the operation of facilities such as satellites, remote sensing, and power, and even affecting human health and safety [1]. Therefore, accurately detecting CME eruption events to provide early warning signals to humans is of great practical significance for avoiding or reducing the harm caused by CMEs.

[0003] Currently, the most commonly used corona image data is obtained by the LASCO coronagraph carried by the SOHO satellite, which provides three different observation radii of corona observation images: C1, C2, and C3. Due to the failure of the C1 observation equipment and the official discontinuation of the download permission for the C3 dataset, only the C2 dataset is currently available. The observation range of the C2 image is 1.5R-6R, and the observation range of the C3 image is 3.7R-30R (R is the solar radius) [2]. Researchers can calculate the physical characteristics of CME based on the observation data, such as linear velocity, acceleration, central angle, and angular width, and then explore the evolution and propagation mechanism of CME [3], providing a scientific basis for the prediction of CME occurrence. The premise of calculating the physical characteristics of CME is to accurately detect the CME region. Therefore, it is necessary to conduct research on the detection of CME regions.

[0004] Existing CME region detection methods can be divided into two main categories: manual annotation and computer-automated detection [4]. Manual annotation relies on researchers observing corona images daily and recording relevant feature information to compile a CME catalog. Currently, commonly used CME catalogs include the CDAW catalog and the NRL catalog. Each burst process is recorded as a CME event, and these catalogs record the start time, linear velocity, acceleration, central angle, angular width, and the occurrence time, height, central angle, etc. of each CME in each frame of the event [5]. Manual annotation is easily affected by the observer's subjectivity and is labor-intensive and time-consuming, but it also provides a benchmark and reference for further research. Computer-automated detection methods can avoid the influence of subjective factors to a certain extent and save human resources, and have gradually become a research hotspot in the field of CME detection.

[0005] Computer-based automatic detection methods include texture feature-based methods, gray-scale feature-based methods, learning-based methods, inter-frame difference methods, and background difference methods. Texture feature-based methods mainly identify CMEs by recognizing significant differences in texture between the CME region and other regions of the corona image. In 2010, Goussies et al. [6] proposed a non-parametric supervised method (CORonal Segmentation Technique, CORSET), which uses the gray-scale co-occurrence matrix to obtain texture features representing different regions to achieve CME detection. Some scholars have also proposed a processing method that converts to the frequency domain. This method uses wavelet and multi-scale image processing techniques to enhance the visibility of CMEs and suppress noise, taking into account the multi-scale characteristics of corona images. However, the lack of clear boundary information of CMEs makes the implementation of this method complex and the detection effect poor. Gray-scale feature-based methods essentially use the fact that corona images observed under white light have a brighter structure when CMEs occur to detect CMEs. In 2002, Robbrecht and Berghmans et al. [7] developed an automatic CME detection method called the Computer-Aided Tracking Software Package (CACTus). This method uses the characteristic physical parameters of the coronal mass eruption to form a special data structure and uses Hough transform to complete the CME detection. In 2005, Olmedo et al. [8] proposed the Solar Eruption Event Detection System (SEEDS). This method uses image segmentation technology to realize CME detection in the coronal image in polar coordinates. However, both CACTus and SEEDS use brightness enhancement to highlight the suspected CME region in the coronal image, which enhances the intensity of both the CME region and the non-eruption region, resulting in more false detections. The learning-based method regards the detection of CME as a binary classification problem and learns whether there is a CME in each frame of the coronal image based on the labeled samples. Based on the Adaptive Boosting (AdaBoost) classification algorithm, this method uses the AdaBoost algorithm to integrate multiple weak classifiers into a strong classifier to classify the CME region from the coronal image. However, this method requires a large amount of manually labeled solar corona image datasets, which is subject to subjective interference and leads to certain errors in the detection results. The inter-frame difference method selects several frames in an image sequence, performs inter-frame difference calculations, calculates the pixel difference at corresponding positions in the images, and determines whether the pixel difference exceeds a certain threshold to obtain the contour of the moving target. However, this method cannot cope with sudden changes in illumination, and the detection results generally exhibit a "hole" phenomenon. The background difference method models the background and obtains the foreground target by continuously detecting the differences between the video image and the background model, but this method requires the image or video state to remain stable.

[0006] When a CME occurs, it moves outward from the surface of the sun and its shape changes continuously, exhibiting the characteristics of a moving target. The CME region is essentially a high-energy particle ejected in the form of plasma. The velocity of the particles varies significantly at different locations in the corona image [9]. The moving particles in the CME region of the corona image are fast and outward. There are also moving particles in the non-burst region, but their velocity is smaller. The motion of these particles is called a disturbance. CME images are characterized by complex observation backgrounds, disturbances, and weak observation signals. These characteristics make it difficult for the above research methods to automatically detect the CME region. Considering the difference in velocity between the CME region and the non-burst region in the corona image, this paper introduces the velocity field into the automatic detection of CME and proposes an automatic detection method for CME based on the velocity field. The velocity field is a physical field composed of velocity vectors at multiple times and multiple points. The velocity field can calculate the motion of each pixel in the corona image and can also visualize the constantly evolving structure of the plasma in the CME during the eruption process

[10] . Therefore, compared with other methods, the automatic detection method of CME based on velocity field has the advantage of not being affected by the brightness of CME observation data and subjective factors. The CME region can be extracted by analyzing the CME velocity field. Therefore, it is necessary to carry out research on automatic detection of CME region based on velocity field. Summary of the Invention

[0007] This invention proposes a novel automatic detection method for cavity-mesh-like structures (CMEs) based on velocity fields, solving the problem of low detection accuracy in existing technologies due to brightness variations or human factors. Since the velocity field is unaffected by brightness, it can more accurately detect CME regions and extract unique features such as cavities and high-velocity areas.

[0008] The technical solution of this invention is: an automatic detection method for coronal mass ejection (CME) regions based on velocity fields, the specific steps of which are as follows:

[0009] Step 1: Construct an optical flow calculation network to obtain the optical flow velocity field of coronal mass ejection observation data. This step is defined as the acquisition of the velocity field of coronal mass ejection observation data.

[0010] The specific steps for obtaining the velocity field of the coronal mass ejection observation data in Step 1 are as follows:

[0011] This invention constructs an unsupervised, visual Transformer-based multi-scale optical flow network (UTFlowNet) to solve the calculation of the velocity field in CME observation data. The network model consists of two modules: a feature extraction module and an optical flow calculation module. The feature extraction module extracts compressed high-dimensional features at multiple scales from the C2 image. After passing through the feature extraction module, the CME observation data is compressed into feature maps of different scales. These feature maps are then input into the optical flow calculation module in ascending order of scale to calculate the optical flow velocity field.

[0012] Step 2: Apply the K-Means clustering method to cluster the velocity field of a set of CME events into two classes: one with larger velocity and one with smaller velocity. Then, set a threshold according to the 3σ principle to filter out the part of the velocity field with smaller velocity values ​​in the CME observation data. This step is defined as threshold filtering.

[0013] The specific steps of threshold filtering in Step 2 are as follows:

[0014] To address the characteristic of fast pixel velocity in the CME region and slow velocity in perturbed pixels in corona images, this paper employs the K-Means clustering method to categorize all velocity values ​​of a set of CME event velocity fields into two classes: one class, consisting of pixels with larger velocity values ​​(called the large velocity class), and the other class, consisting of pixels with smaller velocity values ​​(called the small velocity class). Let μ1 be the average velocity value of all pixels in the small velocity class, and σ1 be the standard deviation. The calculation formula is as follows:

[0015]

[0016] In the formula, n represents the number of pixels in the small velocity class, and v xi v yi and v i These represent the x-direction velocity value, y-direction velocity value, and actual velocity value of the i-th pixel in the small velocity class, respectively.

[0017] Based on the 3σ principle, the speed threshold is set as v. threshold As shown in formula (4), the threshold filtering method is as shown in formula (5). If the velocity value of a pixel in the velocity field of the CME observation data is less than v... threshold If the value is positive, the pixel is considered to be part of the disturbance and its speed is set to 0; otherwise, the original value is retained.

[0018] v threshold =μ1+3σ1 (4)

[0019]

[0020] Step 3: Determine the velocity direction of each pixel in the velocity field of the CME observation data after threshold filtering, find pixels whose motion direction is not from the center of the sun outward, and set the velocity of the pixel to 0. This step is defined as direction filtering.

[0021] The specific steps of directional filtering in Step 3 are as follows:

[0022] Given that the velocity of pixels in the CME region is directed outward from the center of the sun, the velocity direction of each pixel in the velocity field after threshold filtering is determined. If the velocity direction of a pixel is outward from the center of the sun, the pixel is considered to belong to the CME region; otherwise, the pixel is considered to belong to the perturbation part, and the velocity of the pixel is set to 0. The specific approach is as follows.

[0023] Taking a CME observation image as an example, the top left corner of the image is taken as the origin of the coordinate system, the horizontal direction to the right is the positive x-axis, and the vertical direction downwards is the positive y-axis. Point C(C1, C2) is the center point of the image coordinate system, and point A(A1, A2) is any point in the image. Let A represent the vector pointing from point A to point C. The coordinates are represented as in formula (6). Let v represent the velocity vector at point A, and let the velocity of point A in the x-direction be v. xA The velocity in the y-direction is v yA ,but The coordinates are represented as in formula (7). ∠θ represents... and The angle between them, according to the vector angle formula, can be calculated using formula (8).

[0024]

[0025]

[0026] The method for determining the direction of the filter is shown in formula (9). If ∠θ is an obtuse angle, it means... The direction is outward from the center of the sun; otherwise, if If the direction is towards the center of the sun, it means that point A belongs to the disturbed part, so the velocity of point A in the velocity field is set to 0.

[0027]

[0028] Step 4: Obtain the velocity field of the CME observation data after direction filtering. Set the value of all non-zero pixels in the velocity field to 255. Let the processed velocity field matrix be D. Calculate the sum of the distances between any non-zero pixel in D and other non-zero pixels to obtain the distance sum matrix. Set the threshold using the 3σ principle to filter out outliers in the velocity field. Define this step as outlier filtering.

[0029] The specific steps for outlier filtering in Step 4 are as follows:

[0030] After the threshold and direction filtering steps described above, the interference in the CME velocity field has been removed. Pixels with non-zero velocity values ​​all exhibit high velocity, with the velocity direction pointing outwards from the center of the sun. The focus now shifts to extracting the CME region. Next, the CME velocity field is binarized by changing all non-zero velocity values ​​to 255. Visualizing the CME velocity field reveals that pixels with velocity are white foregrounds, while the rest are black backgrounds. Most of these foreground pixels are concentrated in one region, the CME region, but some may be more scattered. These outliers are considered outliers and need to be filtered out. To address this, this paper calculates the sum of the distances between each non-zero velocity pixel and all other non-zero velocity pixels, forming a distance matrix. Then, using the 3σ outlier handling method, pixels with larger distance sums are considered discrete points, and their velocities are set to 0. The specific outlier filtering process is as follows:

[0031] The binarized CME velocity field matrix is ​​D, with row height h and column width w. First, initialize the distance and matrix dm as a matrix of all zeros. Then, calculate as shown in formula (10).

[0032]

[0033] In the formula, (i,j)_ represents the coordinates of the pixel in the i-th row and j-th column, and (p,q) is similar; dm i,j Z represents the distance and the value of the pixel in the i-th row and j-th column of the matrix dm; Z represents the set of coordinates of all pixels in matrix D whose pixel values ​​are not 0.

[0034] Let the mean of the dm matrix be μ² and the standard deviation be σ², then,

[0035]

[0036] In the formula, m represents the number of non-zero points in the dm matrix.

[0037] Based on the 3σ principle, the distance threshold is set to d. thresholdAs shown in formula (13), the method for judging outlier filtering is shown in formula (14). If the value of a pixel in the distance matrix dm is greater than d... threshold If the value is found to be an outlier, the pixel value at the same position in the velocity field matrix D is set to 0; otherwise, the original value is retained.

[0038] d threshold =μ² + 3σ² (13)

[0039]

[0040] Step 5: Extract the convex hull of all non-zero pixels in the matrix D after outlier filtering, and set the value of all pixels in the convex hull to 255. This step is defined as convex hull extraction.

[0041] The specific steps for convex hull extraction in Step 5 are as follows:

[0042] After the above steps, the points in matrix D with non-zero pixel values ​​are concentrated in a relatively concentrated area. To extract the CME region to the maximum extent, this section further extracts the convex hull of the points in matrix D with non-zero pixel values. The convex hull is the smallest convex polygon on a plane that can contain all given points. Based on matrix D, the convex hull of the points in matrix D with non-zero pixel values ​​is first extracted, and the value of all pixels within the convex hull is set to 255.

[0043] Step 6: Set the pixels in the region where the light-shielding plate is located in matrix D after convex hull extraction to 0. The remaining non-zero parts in matrix D are the CME region. This step is defined as CME region extraction.

[0044] The specific steps for CME region extraction in Step 6 are as follows:

[0045] Since the central part of the corona observation image is the region where the sun is located, no signal will be observed in this part after it is blocked by the shield. Therefore, the values ​​of the pixels in the shielded part of matrix D should be set to 0. At this time, the region formed by all the pixels in matrix D with non-zero pixel values ​​is the CME region detected in this paper.

[0046] The beneficial effects of this invention are:

[0047] A velocity field is a physical field composed of velocity vectors at multiple times and points. It can be used to calculate the motion of each pixel in a CME observation image. Since there is a significant difference in velocity between the CME region and the non-burst region in a CME observation image, analyzing the velocity field can accurately identify the CME region in the CME observation data. Furthermore, the automatic CME region detection method based on the velocity field has the advantage of being unaffected by the brightness of the CME observation data and subjective factors, especially in cases where the background of daily CME observation images is complex and disturbed, and the observation signal is weak, it can still provide reliable detection results. Attached Figure Description

[0048] Figure 1 A flowchart for automatic detection of coronal mass ejection regions based on velocity fields;

[0049] Figure 2 Diagram of the network structure for calculating the optical flow velocity field;

[0050] Figure 3 This is a visualization of the automatic detection process of the coronal mass ejection area from 15:12:08 to 17:12:05 on January 1, 2014.

[0051] Figure 4 The error distribution of the CME physical characteristics calculated by the method proposed in this invention is shown below. Figure 4 (a) is a histogram of the error frequency distribution of the center angle; Figure 4 (b) is the histogram of the error frequency distribution of the angular width; Figure 4 (c) is the histogram of the error frequency distribution of linear velocity. Detailed Implementation

[0052] like Figure 1 As shown, an automatic detection method for coronal mass ejection (CME) regions based on velocity fields is described, and the specific implementation steps are as follows:

[0053] Step 1: Construct an optical flow calculation network to obtain the optical flow velocity field of coronal mass ejection observation data. This step is defined as the acquisition of the velocity field of coronal mass ejection observation data.

[0054] The specific steps for obtaining the velocity field of the coronal mass ejection observation data in Step 1 are as follows:

[0055] This invention constructs an unsupervised, visual Transformer-based multi-scale optical flow network (UTFlowNet) to solve the calculation of the velocity field in CME observation data. The network model consists of two modules: a feature extraction module and an optical flow calculation module. The feature extraction module extracts compressed high-dimensional features at multiple scales from the C2 image. After passing through the feature extraction module, the CME observation data is compressed into feature maps of different scales. These feature maps are then input into the optical flow calculation module in ascending order of scale to calculate the optical flow velocity field.

[0056] Step 2: Apply the K-Means clustering method to cluster the velocity field of a set of CME events into two classes: one with larger velocity and one with smaller velocity. Then, set a threshold according to the 3σ principle to filter out the part of the velocity field with smaller velocity values ​​in the CME observation data. This step is defined as threshold filtering.

[0057] The specific steps of threshold filtering in Step 2 are as follows:

[0058] To address the characteristic of fast pixel velocity in the CME region and slow velocity in perturbed pixels in corona images, this paper employs the K-Means clustering method to categorize all velocity values ​​of a set of CME event velocity fields into two classes: one class, consisting of pixels with larger velocity values ​​(called the large velocity class), and the other class, consisting of pixels with smaller velocity values ​​(called the small velocity class). Let μ1 be the average velocity value of all pixels in the small velocity class, and σ1 be the standard deviation. The calculation formula is as follows:

[0059]

[0060] In the formula, n represents the number of pixels in the small velocity class, and v xi v yi and v i These represent the x-direction velocity value, y-direction velocity value, and actual velocity value of the i-th pixel in the small velocity class, respectively.

[0061] Based on the 3σ principle, the speed threshold is set as v. threshold As shown in formula (4), the threshold filtering method is as shown in formula (5). If the velocity value of a pixel in the velocity field of the CME observation data is less than v... threshold If the value is positive, the pixel is considered to be part of the disturbance and its speed is set to 0; otherwise, the original value is retained.

[0062] v threshold =μ1+3σ1 (4)

[0063]

[0064] Step 3: Determine the velocity direction of each pixel in the velocity field of the CME observation data after threshold filtering, find pixels whose motion direction is not from the center of the sun outward, and set the velocity of the pixel to 0. This step is defined as direction filtering.

[0065] The specific steps of directional filtering in Step 3 are as follows:

[0066] Given that the velocity of pixels in the CME region is directed outward from the center of the sun, the velocity direction of each pixel in the velocity field after threshold filtering is determined. If the velocity direction of a pixel is outward from the center of the sun, the pixel is considered to belong to the CME region; otherwise, the pixel is considered to belong to the perturbation part, and the velocity of the pixel is set to 0. The specific approach is as follows.

[0067] Taking a CME observation image as an example, the top left corner of the image is taken as the origin of the coordinate system, the horizontal direction to the right is the positive x-axis, and the vertical direction downwards is the positive y-axis. Point C(C1, C2) is the center point of the image coordinate system, and point A(A1, A2) is any point in the image. Let A represent the vector pointing from point A to point C. The coordinates are represented as in formula (6). Let v represent the velocity vector at point A, and let the velocity of point A in the x-direction be v. xA The velocity in the y-direction is v yA ,but The coordinates are represented as in formula (7). ∠θ represents... and The angle between them, according to the vector angle formula, can be calculated using formula (8).

[0068]

[0069]

[0070] The method for determining the direction of the filter is shown in formula (9). If ∠θ is an obtuse angle, it means... The direction is outward from the center of the sun; otherwise, if If the direction is towards the center of the sun, it means that point A belongs to the disturbed part, so the velocity of point A in the velocity field is set to 0.

[0071]

[0072] Step 4: Obtain the velocity field of the CME observation data after direction filtering. Set the value of all non-zero pixels in the velocity field to 255. Let the processed velocity field matrix be D. Calculate the sum of the distances between any non-zero pixel in D and other non-zero pixels to obtain the distance sum matrix. Set the threshold using the 3σ principle to filter out outliers in the velocity field. Define this step as outlier filtering.

[0073] The specific steps for outlier filtering in Step 4 are as follows:

[0074] After the threshold and direction filtering steps described above, the interference in the CME velocity field has been removed. Pixels with non-zero velocity values ​​all exhibit high velocity, with the velocity direction pointing outwards from the center of the sun. The focus now shifts to extracting the CME region. Next, the CME velocity field is binarized by changing all non-zero velocity values ​​to 255. Visualizing the CME velocity field reveals that pixels with velocity are white foregrounds, while the rest are black backgrounds. Most of these foreground pixels are concentrated in one region, the CME region, but some may be more scattered. These outliers are considered outliers and need to be filtered out. To address this, this paper calculates the sum of the distances between each non-zero velocity pixel and all other non-zero velocity pixels, forming a distance matrix. Then, using the 3σ outlier handling method, pixels with larger distance sums are considered discrete points, and their velocities are set to 0. The specific outlier filtering process is as follows:

[0075] The binarized CME velocity field matrix is ​​D, with row height h and column width w. First, initialize the distance and matrix dm as a matrix of all zeros. Then, calculate as shown in formula (10).

[0076]

[0077] In the formula, (i,j)_ represents the coordinates of the pixel in the i-th row and j-th column, and (p,q) is similar; dm i,j Z represents the distance and the value of the pixel in the i-th row and j-th column of the matrix dm; Z represents the set of coordinates of all pixels in matrix D whose pixel values ​​are not 0.

[0078] Let the mean of the dm matrix be μ² and the standard deviation be σ², then,

[0079]

[0080] In the formula, m represents the number of non-zero points in the dm matrix.

[0081] Based on the 3σ principle, the distance threshold is set to d. thresholdAs shown in formula (13), the method for judging outlier filtering is shown in formula (14). If the value of a pixel in the distance matrix dm is greater than d... threshold If the value is found to be an outlier, the pixel value at the same position in the velocity field matrix D is set to 0; otherwise, the original value is retained.

[0082] d threshold =μ² + 3σ² (13)

[0083]

[0084] Step 5: Extract the convex hull of all non-zero pixels in the matrix D after outlier filtering, and set the value of all pixels in the convex hull to 255. This step is defined as convex hull extraction.

[0085] The specific steps for convex hull extraction in Step 5 are as follows:

[0086] After the above steps, the points in matrix D with non-zero pixel values ​​are concentrated in a relatively concentrated area. To extract the CME region to the maximum extent, this section further extracts the convex hull of the points in matrix D with non-zero pixel values. The convex hull is the smallest convex polygon on a plane that can contain all given points. Based on matrix D, the convex hull of the points in matrix D with non-zero pixel values ​​is first extracted, and the value of all pixels within the convex hull is set to 255.

[0087] Step 6: Set the pixels in the region where the light-shielding plate is located in matrix D after convex hull extraction to 0. The remaining non-zero parts in matrix D are the CME region. This step is defined as CME region extraction.

[0088] The specific steps for CME region extraction in Step 6 are as follows:

[0089] Since the central part of the corona observation image is the region where the sun is located, no signal will be observed in this part after it is blocked by the shield. Therefore, the values ​​of the pixels in the shielded part of matrix D should be set to 0. At this time, the region formed by all the pixels in matrix D with non-zero pixel values ​​is the CME region detected in this paper.

[0090] This invention first obtains the velocity field of coronal mass ejection (CME) observation data by constructing an optical flow computational network. Then, based on the velocity field, it performs threshold filtering, orientation filtering, outlier filtering, convex hull extraction, and CME region extraction. The specific process of the proposed method is as follows: Figure 1 As shown. This invention considers the velocity difference between CME regions and non-burst regions in CME observation images, and leverages the advantage that velocity fields are unaffected by the brightness of CME observation data and subjective factors, thereby improving the accuracy of automatic CME region detection.

[0091] Current research on CME is mostly based on white light data of the corona and its surroundings obtained from observations by upper-air coronagraphs and ground-based coronagraphs. This invention selects Level 1 white light fits images transmitted from the C2 instrument of the LASCO upper-air coronagraph. This invention first utilizes... Figure 2 The optical flow computing network converts CME's FTS format data into optical flow FLO format data, and then follows... Figure 1 The demonstrated subsequent process performs automatic detection of the CME area.

[0092] like Figure 3 As shown, this paper uses the CME event spanning from 15:12:08 to 17:12:05 on January 1, 2014, for automatic CME region detection. Five frames from this event are presented here. From top to bottom, the figure shows the C2 observation image, dense optical flow map, image after threshold filtering, image after orientation filtering, and the CME region image. It can be seen that after processing through the steps of the proposed method, the burst region of the CME event can be accurately detected, consistent with the burst region shown in the C2 observation image.

[0093] Twelve groups of CME events were selected as shown in Table 1. First, CME region images for each group of CME events were obtained using the method proposed in this invention. Then, the center angle (CPA), angular width (AW), and linear velocity of each group of CME events were calculated as follows. To verify the accuracy of the Ours method (the method proposed in this invention), the CME physical characteristic values ​​corresponding to each group of CME events provided on the CDAW website were used as a reference and compared with the CME physical characteristic values ​​calculated using the method of this invention.

[0094] Table 1 Calculation Results of Physical Characteristics of CME Events

[0095]

[0096] Based on the calculation results of the CME event center angle, angular width, and linear velocity in Table 1, the absolute values ​​of the errors between the Ours method and the eigenvalues ​​from the CDAW official website are statistically analyzed, and the frequency distribution histograms of the absolute values ​​of the errors for the center angle, angular width, and linear velocity are visualized as follows: Figure 4 As shown.

[0097] Figure 4 (a) shows that the absolute center position angle errors for nine CME events are between 0° and 5°, while the absolute center position angle errors for the other three CME events are between 8° and 20°. This small error distribution of CPA indicates that the method proposed in this invention is effective in accurately locating CMEs. Figure 4(b) shows that the angular width detection error for nine CME events was between 0° and 10°, while the error for the other three CME events was between 12° and 23°. The angular width detection error is concentrated in a narrow range of 10°, further confirming the accuracy of this method for CME region size. Finally, Figure 4 (c) shows that for the eight CME events, the absolute linear velocity error is between 0 and 30 km / s, and for the other four events, the error is between 30 and 60 km / s. It can be seen that the linear velocity error calculated by the method in this paper is mainly distributed within the acceptable range of 30 km / s, and the calculation results are accurate.

[0098] References

[0099] [1].Webb,DF,&Howard,TA(2012).Coronal mass ejections:Observations.Living Reviews in Solar Physics,9(1),1-83.

[0100] [2].Schumacher,G.,&Gay,J.(2001).An attempt to detect Vulcanoids withSOHO / LASCO images-I.Scale relativity and quantization of the solarsystem.Astronomy&Astrophysics,368(3),1108-1114.

[0101] [3].Wang,Y.,Chen,C.,Gui,B.,Shen,C.,Ye,P.,&Wang,S.(2011).Statisticalstudy of coronal mass ejection source locations:Understanding CMEs viewed incoronagraphs.Journal of Geophysical Research:Space Physics,116(A4).

[0102] [4].Yashiro,S.,Michalek,G.,&Gopalswamy,N.(2008,October).A comparisonof coronal mass ejections identified by manual and automatic methods.InAnnales geophysicae(Vol.26,No.10,pp.3103-3112).Copernicus GmbH.

[0103] [5].St.Cyr,O.C.,Howard,R.A.,Sheeley Jr,N.R.,Plunkett,S.P.,Michels,D.J.,Paswaters,S.E.,...&Lamy,P.L.(2000).Properties of coronal mass ejections:SOHO LASCO observations from January 1996 to June 1998.Journal of GeophysicalResearch:Space Physics,105(A8),18169-18185.

[0104] [6].Goussies,N.,Stenborg,G.,Vourlidas,A.,&Howard,R.(2010).Tracking ofcoronal white-light events by texture.Solar Physics,262,481-494.

[0105] [7].Bonte,K.,Jacobs,C.,Robbrecht,E.,De Groof,A.,Berghmans,D.,&Poedts,S.(2011).Validation of CME detection software(CACTus)by means of simulateddata,and analysis of projection effects on CME velocity measurements.SolarPhysics,270,253-272.

[0106] [8].Olmedo,O.,Zhang,J.,Wechsler,H.,Poland,A.,&Borne,K.(2009).Automatic detection and tracking of coronal mass ejections in coronagraphtime series.Solar image analysis and visualization,275-289.

[0107] [9].Chattopadhyay,A.,Khondekar,M.H.&Bhattacharjee,A.K.Stationarityand periodicities of linear speed of coronal mass ejection:a statisticalsignal processing approach.Astrophys Space Sci 362,179(2017).

[0108]

[10] .Colaninno,R.C.,&Vourlidas,A.(2006).Analysis of the velocityfield of CMEs using optical flow methods.The Astrophysical Journal,652(2),174。

Claims

1. An automatic detection method for coronal mass ejection (CME) regions based on velocity fields, comprising the following steps: Step 1: Construct an optical flow calculation network to obtain the optical flow velocity field of coronal mass ejection observation data. This step is defined as the acquisition of the velocity field of coronal mass ejection observation data. Step 2: Apply the K-Means clustering method to cluster the velocity field of a set of CME events into two classes: one with larger velocity and one with smaller velocity. Then, set a threshold according to the 3σ principle to filter out the part of the velocity field with smaller velocity values ​​in the CME observation data. This step is defined as threshold filtering. Step 3: Determine the velocity direction of each pixel in the velocity field of the CME observation data after threshold filtering, find pixels whose motion direction is not from the center of the sun outward, and set the velocity of the pixel to 0. This step is defined as direction filtering. Step 4: Obtain the velocity field of the CME observation data after direction filtering. Set the value of all non-zero pixels in the velocity field to 255. Let the processed velocity field matrix be D. Calculate the sum of the distances between any non-zero pixel in D and other non-zero pixels to obtain the distance sum matrix. Set the threshold using the 3σ principle to filter out outliers in the velocity field. Define this step as outlier filtering. Step 5: Extract the convex hull of all non-zero pixels in the matrix D after outlier filtering, and set the value of all pixels in the convex hull to 255. This step is defined as convex hull extraction. Step 6: Set the pixels in the region where the light-shielding plate is located in matrix D after convex hull extraction to 0. The remaining non-zero parts in matrix D are the CME region. This step is defined as CME region extraction.

2. The automatic detection method for coronal mass ejection (CME) regions based on velocity fields according to claim 1, characterized in that: In Step 1, the specific steps for obtaining the velocity field of coronal mass ejection observation data are as follows: This invention constructs an unsupervised, visual Transformer-based multi-scale optical flow network (UTFlowNet) to solve the calculation of the velocity field in CME observation data. The network model consists of two modules: a feature extraction module and an optical flow calculation module. The feature extraction module extracts compressed high-dimensional features at multiple scales from the C2 image. After passing through the feature extraction module, the CME observation data is compressed into feature maps of different scales. These feature maps are then input into the optical flow calculation module in ascending order of scale to calculate the optical flow velocity field.

3. The automatic detection method for coronal mass ejection (CME) regions based on velocity fields according to claim 1, characterized in that: In Step 2, the specific steps of threshold filtering are as follows: To address the characteristic of fast pixel velocity in the CME region and slow velocity in perturbed pixels in corona images, this paper employs the K-Means clustering method to categorize all velocity values ​​of a set of CME event velocity fields into two classes: one class, consisting of pixels with larger velocity values ​​(called the large velocity class), and the other class, consisting of pixels with smaller velocity values ​​(called the small velocity class). Let μ1 be the average velocity value of all pixels in the small velocity class, and σ1 be the standard deviation. The calculation formula is as follows: In the formula, n represents the number of pixels in the small velocity class, and v xi v yi and v i These represent the x-direction velocity value, y-direction velocity value, and actual velocity value of the i-th pixel in the small velocity class, respectively. Based on the 3σ principle, the speed threshold is set as v. threshold As shown in formula (4), the threshold filtering method is as shown in formula (5). If the velocity value of a pixel in the velocity field of the CME observation data is less than v... threshold If the value is positive, the pixel is considered to be part of the disturbance and its speed is set to 0; otherwise, the original value is retained. v threshold =μ1+3σ1 (4) 4. The automatic detection method for coronal mass ejection (CME) regions based on velocity fields according to claim 1, characterized in that: The specific steps of directional filtering in Step 3 are as follows: Given that the velocity of pixels in the CME region is directed outward from the center of the sun, the velocity direction of each pixel in the velocity field after threshold filtering is determined. If the velocity direction of a pixel is outward from the center of the sun, the pixel is considered to belong to the CME region; otherwise, the pixel is considered to belong to the perturbation part, and the velocity of the pixel is set to 0. The specific approach is as follows. Taking a CME observation image as an example, the top left corner of the image is taken as the origin of the coordinate system, the horizontal direction to the right is the positive x-axis, and the vertical direction downwards is the positive y-axis. Point C(C1, C2) is the center point of the image coordinate system, and point A(A1, A2) is any point in the image. Let A represent the vector pointing from point A to point C. The coordinates are represented as in formula (6). Let v represent the velocity vector at point A, and let the velocity of point A in the x-direction be v. xA The velocity in the y-direction is v yA ,but The coordinates are represented as in formula (7). ∠θ represents... and The angle between them, according to the vector angle formula, can be calculated using formula (8). The method for determining the direction of the filter is shown in formula (9). If ∠θ is an obtuse angle, it means... The direction is outward from the center of the sun; otherwise, if If the direction is towards the center of the sun, it means that point A belongs to the disturbed part, so the velocity of point A in the velocity field is set to 0.

5. The automatic detection method for coronal mass ejection (CME) regions based on velocity fields according to claim 1, characterized in that: The specific steps for outlier filtering in Step 4 are as follows: After the threshold and direction filtering steps described above, the interference in the velocity field of the CME observation data has been removed. Pixels with non-zero velocity values ​​all exhibit high velocity, with the velocity direction pointing outward from the center of the sun. The focus now shifts to extracting the CME region. Next, the velocity field of the CME observation data is binarized by changing all non-zero velocity values ​​to 255. Visualizing the velocity field reveals that pixels with velocity are white foregrounds, while the rest are black backgrounds. Most of these foreground pixels are concentrated in one region, the CME region, but some may be more scattered. These outliers are considered outliers and need to be filtered out. To address this, this paper calculates the sum of the distances between each non-zero velocity pixel and all other non-zero velocity pixels, forming a distance matrix. Then, using the 3σ outlier handling method, pixels with larger distance sums are considered discrete points, and their velocities are set to 0. The specific outlier filtering process is as follows: Let the binarized CME velocity field matrix be D, with row height h and column width w. First, initialize the distance and sum matrix dm as a matrix of all zeros. Then, the calculation is as shown in formula (10). In the formula, (i,j)_ represents the coordinates of the pixel in the i-th row and j-th column, and (p,q) is similar; dm i,j Z represents the distance and the value of the pixel in the i-th row and j-th column of the matrix dm; Z represents the set of coordinates of all pixels in matrix D whose pixel values ​​are not 0. Let the mean of the dm matrix be μ² and the standard deviation be σ², then, In the formula, m represents the number of non-zero points in the dm matrix. Based on the 3σ principle, the distance threshold is set to d. threshold As shown in formula (13), the method for judging outlier filtering is shown in formula (14). If the value of a pixel in the distance matrix dm is greater than d... threshold If the value is found to be an outlier, the pixel value at the same position in the velocity field matrix D is set to 0; otherwise, the original value is retained. d threshold =μ2+3σ2 (13) 6. The automatic detection method for coronal mass ejection (CME) regions based on velocity fields according to claim 1, characterized in that: The specific steps for convex hull extraction in Step 5 are as follows: After the above steps, the points in matrix D with non-zero pixel values ​​are concentrated in a relatively concentrated area. To extract the CME region to the maximum extent, this section further extracts the convex hull of the points in matrix D with non-zero pixel values. The convex hull is the smallest convex polygon on a plane that can contain all given points. Based on matrix D, the convex hull of the points in matrix D with non-zero pixel values ​​is first extracted, and the value of all pixels within the convex hull is set to 255.

7. The automatic detection method for coronal mass ejection (CME) regions based on velocity fields according to claim 1, characterized in that: The specific steps for CME region extraction in Step 6 are as follows: Since the central part of the coronal mass ejection (CME) observation image is the region containing the Sun, and no signal will be observed in this part after being blocked by the shield, the values ​​of the pixels in the shielded area of ​​matrix D should be set to 0. At this time, the region formed by all the pixels in matrix D with non-zero pixel values ​​is the CME region detected in this paper.

Images