A method for detecting coronal mass ejections based on a low-rank recovery model

Abstract

The application discloses a kind of coronal mass ejection detection methods based on low rank restoration model, comprising the following steps: step 1, to the LASCO C2 image downloaded from SOHO homepage is carried out image preprocessing, to coronal image is denoised, screening processing;Step 2, the LASCO C2 image obtained after step 1 is carried out image gray, image rotation and flip operation;Step 3, the coronal image after processing in step 2 is converted to polar coordinates and is represented;Step 4, the moving target in coronal image is extracted using the method based on low rank restoration model;Step 5: fine-tuning low rank restoration model parameters, coronal mass ejection is distinguished from coronal stream, using image three frame difference is carried out moving target extraction, finally obtains coronal mass ejection.The method detection speed is fast, and precision is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of astronomical image target detection, and in particular to a coronal mass ejection detection method based on a low-rank restoration model. BACKGROUND

[0002] Coronal mass ejection (CME) is a large-scale and severe explosive phenomenon. Since this explosive phenomenon can cause serious interference to the Earth environment, the detection of coronal mass ejection is of great significance for forecasting disastrous space weather.

[0003] CME is a dynamic event that ejects plasma from the corona layer of the sun to the interplanetary space. It was first observed in the Skylab mission in the early 1970s, is the largest and most active explosive phenomenon originating from the sun, and can be observed in the extended corona by a white light coronagraph. Since CME is a huge magnetized plasma eruption, it can propagate at a speed of up to several thousand kilometers per second. Given that the solar high-energy particle events and geomagnetic storms caused by such explosive activities can affect aviation safety, satellite operation, communication systems, power facilities, etc., it is necessary to detect CME to provide warning signals for space weather forecasters, so that appropriate measures can be taken to avoid unnecessary losses. In addition, statistical information about CME is also very important for better understanding its nature.

[0004] Although previous studies have developed many methods for detecting CME, most of these methods are based on artificial selection of features such as gray scale, texture, etc., and use simple threshold segmentation techniques for detection, so the detection effect of weak CME is not good, and the CME result directory obtained is very different from the CDAW (Coordinated Data Analysis Workshop) center directory. According to the occurrence state of CME, the relatively mature moving target detection method is applied to CME detection. Unlike general natural images, coronal images are a special kind of solar observation images, and CME phenomena are complex and variable, with many similarities between foreground and background, and the influence of dynamic coronal flow background. The traditional moving target detection method needs to be improved in terms of detection effect and detection speed. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a coronal mass ejection detection method based on a low-rank restoration model, which has fast detection speed and greatly improved accuracy.

[0006] To solve the above technical problems, the technical scheme of the present application is as follows: a coronal mass ejection detection method based on a low-rank restoration model, comprising the following steps:

[0007] Step 1: image preprocessing is performed on the LASCO C2 image downloaded from the SOHO homepage, and the coronal image is denoised and filtered;

[0008] Step 2: the LASCO C2 image obtained in step 1 is subjected to image graying, image rotation and flipping operations;

[0009] Step 3: the coronal image processed in step 2 is converted to polar coordinates for representation;

[0010] Step 4: a method based on a low-rank restoration model is used to extract moving targets in the coronal image;

[0011] Step 5: the low-rank restoration model parameters are fine-tuned, the coronal mass ejection is distinguished from the coronal streamer, the image three-frame difference is used for moving target extraction, and finally the coronal mass ejection is obtained.

[0012] As a further description of the above scheme: in step 2, in order to compare the detection results with the CDAW, CACTu and SEEDS catalog methods, the coronal image sequence downloaded from the SOHO homepage is uniformly selected with the east (E) of the coronal image as the starting point, i.e. 0 degrees, and rotated clockwise to 360 degrees; then the corresponding rotation and flipping operations are performed.

[0013] As a further description of the above scheme: in step 3, the polar coordinate conversion is a geometric transformation of the processed coronal observation image, the pixel gray value is kept unchanged, and only the geometric position of the pixel is transformed; the LASCO C2 coronal image of 512x512 pixels is polarized and the sun center baffle part is cut off, and finally a 360x156 pixel image is obtained.

[0014] As a further description of the above scheme: steps 4 and 5 are operated as follows: the coronal image processed in step 4 is input into the low-rank restoration model, and after model calculation, a binary image is obtained, which contains the coronal mass ejection (moving part) and the coronal stream (stationary part). In order to detect the coronal mass ejection, the three-frame difference method is used to difference the continuous coronal images, and since the coronal stream is relatively slow, the coronal stream can be filtered out, and finally the coronal mass ejection is left, i.e. the coronal mass ejection is detected.

[0015] As a further description of the above scheme: the three-frame difference method used in step 5 is as follows:

[0016]

[0017] In formula (4), D m，i，t and D m，i，t+1 respectively represent the difference results of the tth frame and the t-1th frame and the difference results of the tth frame and the t+1th frame; Sm，i，t represents the differential image obtained by performing logical AND operation on the results of two difference terms, and then performing S m，i，t After threshold segmentation, a binary image BS m，i，t :

[0018]

[0019] In formula (5), T is a threshold value selected by experiment, and T=2 is taken after a large number of experiments; as can be seen from equation (5), when the value of BS m，i，t is 1, it represents the foreground, that is, CMEs; and when the value of BS m，i，t is 0, it represents the background, that is, no CMEs.

[0020] The working principle of the present application is as follows: the present application obtains the coronal image required for the experiment by denoising, flipping, rotating and polar coordinate transformation of the coronal image, then inputs the experimental image into a low-rank restoration model to obtain the region with obvious features in the coronal image, and then extracts the moving target, that is, the coronal mass ejection, by using the three-frame difference method, so as to automatically complete the detection of the coronal mass ejection.

[0021] Compared with the prior art, the present application has the following beneficial effects: the saliency target detection model based on low-rank matrix recovery (LRMR) decomposes the image features into a low-rank component associated with the background and a sparse component associated with the saliency target, and obtains the saliency target from the sparse component. The coronal mass ejection is a foreground target relative to the entire coronal image, and the present application uses the saliency target detection and refinement method based on low-rank matrix recovery to detect the saliency target in the image and enhance the difference between the foreground and the background. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is the overall flowchart of the method of the present application;

[0023] Figure 2 is an example of the detection result pattern of the method of the present application. DETAILED DESCRIPTION

[0024] The technical solutions of the present application will be further described in detail below in combination with the drawings and specific embodiments, but the present application is not limited to the following technical solutions.

[0025] Example 1

[0026] A coronal mass ejection detection method based on a low-rank restoration model, the method comprising the following steps, and the flowchart is as shown in Figure 1

[0027] ​Step 1, image preprocessing is performed on the LASCO C2 image downloaded from the SOHO homepage, and the coronal image is denoised and screened;

[0028] Step 2, the LASCO C2 image obtained in step 1 is subjected to image graying, image rotation and flipping and the like;

[0029] Step 3, the coronal image processed in step 2 is converted to polar coordinates for representation;

[0030] Step 4, a method based on a low-rank restoration model is used to extract the moving target in the coronal image;

[0031] Step 5, the low-rank restoration model parameters are fine-tuned, the coronal mass ejection is distinguished from the coronal streamer, the image three-frame difference is used for moving target extraction, and finally the coronal mass ejection is obtained, and the detection result is as shown in Figure 2

[0032] The low-rank restoration model-based coronal mass ejection detection method, wherein in step 1, the coronal image is first subjected to necessary preprocessing, and the stars, comets and other interference sources in the coronal image are denoised to reduce the interference of these interference sources on the low-rank restoration model.

[0033] The low-rank restoration model-based coronal mass ejection detection method has the following specific implementation manners:

[0034] 1. Coronal image preprocessing

[0035] In this paper, the LASCO C2 512x512 image on the SOHO homepage is selected as the experimental data. Since the satellite-mounted LASCO coronagraph is occasionally disturbed during imaging, some images with poor coronagraph shooting effect need to be manually removed. The observation range of the LASCO C2 image is 2.3R~6.5R (assuming that the center of the C2 image is the radius of the baffle R). Then the coronal image is rotated and flipped, and the polar coordinate transformation method is to rotate the east (E) of the coronal image to 360 degrees clockwise from the zero degree. Since the sun is a circularly symmetric structure, and the CME eruption process and the coronal flow movement process are both along the radial direction of the sun, in order to more effectively process the coronal image, it is necessary to convert the image from the rectangular coordinate system to the polar coordinate system. This polar coordinate conversion converts the circular image into a rectangular image which is easier to process. In the polar coordinate transformation, the east (E) of the gray image is taken as the zero degree starting point, i.e. the circular image is cut off from the zero degree position and stretched into a rectangular image, and the polar coordinate is obtained by rotating clockwise to 360 degrees.

[0036] 2. Spatial region detection based on low-rank restoration model

[0037] ​The low-rank restoration model-based coronal mass ejection detection method adopts a structured matrix decomposition-based method to perform spatial saliency detection on the moving target in the coronal image to obtain a spatial saliency map, that is, a coronal mass ejection region. The low-rank restoration model refers to structured decomposition of a feature observation matrix of an image to obtain a low-rank matrix and a sparse matrix, wherein the low-rank part obtained by the decomposition represents a non-salient part in the scene, that is, a spatial background; and the sparse part represents a salient part, that is, a foreground target. Finally, restoration of the sparse matrix after the structured decomposition can obtain the salient target in the scene. The coronal mass ejection is a sparse part relative to the coronal image, and the low-rank restoration model can be used to decompose the coronal image into a low-rank part and a sparse part, and then filter out the low-rank part to obtain the sparse part, that is, the coronal mass ejection region.

[0038] For a given coronal image sequence frame, it is first segmented into N image blocks, and a D-dimensional feature is extracted from each image block to form a feature matrix F, which can be represented as:

[0039] F=L+S (1)

[0040] Wherein L represents the low-rank part of the feature matrix, and S represents the sparse part.

[0041]

[0042] Wherein Ψ(L) is a low-rank constraint, which is used to identify the intrinsic feature subspace of the redundant background fragments to constrain the background region; Ω(S) is a structured sparse constraint, which adopts a tree-structured sparse inducing norm to constrain the foreground matrix S, so as to consider the connection in the saliency image block space and the similarity in the feature into the matrix problem, and solve the problems of separation difficulty and incomplete foreground separation of the traditional low-rank model; Θ(L, S) is a Laplace regularization rule introduced, which is specifically represented as follows:

[0043] Θ(F, S)=Θ(L+S, S)=Θ(L, S) (3)

[0044] The Laplace regularization term Θ(L, S) smoothes the vectors in S according to the local invariance of the feature matrix (F) to increase the distance between the sparse matrix and the low-rank matrix, and reduce the similarity between the low-rank matrix and the structured sparse matrix, so that the sparse part can be better decomposed when the structured decomposition is performed.

[0045] 3. Static and dynamic separation

[0046] The low-rank restoration model-based coronal mass ejection detection method, after calculation by the low-rank restoration model, obtains a coronal mass ejection region, but also obtains a coronal flow region similar to the coronal mass ejection. Thus, in the image obtained by the model, one image contains the coronal mass ejection region and also contains a solar burst structure similar to the coronal mass ejection. Generally, the coronal flow motion is not obvious after the coronal mass ejection burst, that is, the motion is not significant, and it can be considered to be relatively static, that is, it is considered to be stationary. Based on this understanding, the method of the present application uses a three-frame difference method to subtract the relatively stationary coronal flow from the image.

[0047] The improved three-frame difference method is as follows:

[0048]

[0049] Here, D m，i，t and D m，i，t+1 respectively represent the difference results of the tth frame and the t-1th frame and the difference results of the tth frame and the t+1th frame. S m，i，t represents a difference image obtained by performing logical AND operation on the results of the two difference items, and S m，i，t After threshold segmentation, a binary image BS m，i，t is obtained.

[0050]

[0051] wherein T is a threshold value selected by experiment, and T=2 is obtained through a large number of experiments. As can be seen from equation (5), when the value of BS m，i，t is 1, it indicates a foreground, that is, a CME; and when the value of BS m，i，t is 0, it indicates a background, that is, no CME.

[0052] The low-rank restoration model-based coronal mass ejection method is not limited to the above description, and it should be pointed out that, for ordinary skilled persons in the technical field, a number of improvements and replacements can be made without departing from the technical principles of the present patent, and these improvements and replacements should also be considered as the protection scope of the present patent.

Claims

1. A method for detecting coronal mass ejections based on a low-rank recovery model, characterized in that, Includes the following steps: Step 1: Perform image preprocessing on the LASCO C2 image downloaded from the SOHO homepage, including noise reduction and filtering of the corona image; Step 2: Perform image grayscale conversion, image rotation, and image flipping operations on the LASCO C2 image obtained in Step 1; Step 3: Convert the corona image processed in Step 2 to polar coordinates for representation; Step 4: Extract moving targets from the corona image using a method based on a low-rank recovery model; Step 5: Fine-tune the parameters of the low-rank recovery model to distinguish between coronal mass ejections and coronal flow, and use the three-frame difference of the image to extract moving targets, finally obtaining the coronal mass ejection. The three-frame difference method used in step 5 is shown below: In formula (4), D m，i，t With D m，i,t+1 These represent the difference results between frame t and frame (t-1) and between frame t and frame (t+1), respectively. S m,i,t This represents the difference image obtained by performing a logical AND operation on the results of the two difference terms, and then using S... m,i,t The binary image BS is obtained after threshold segmentation. m，i,t : In formula (5), T is the threshold selected in the experiment. After a large number of experiments, T = 2 was chosen. From equation (5), it can be seen that when BS m,i,t When the value is 1, it indicates the foreground, i.e., CMEs. When BS m，i，t When the value is 0, it indicates background, that is, no CMEs.

2. The method for detecting coronal mass ejections based on a low-rank recovery model as described in claim 1, characterized in that, In step 2, to facilitate comparison of the detection results with CDAW, CACTu, and SEEDS catalog methods, the corona image sequence downloaded from the SOHO homepage will be uniformly selected with the eastern (E) side of the corona image as the starting point, i.e., 0 degrees, and rotated clockwise to 360 degrees; then, the corresponding rotation and flipping operations will be performed.

3. The method for detecting coronal mass ejections based on a low-rank recovery model according to claim 1, characterized in that, In step 3, polar coordinate transformation is a geometric transformation performed on the processed corona observation image, keeping the pixel grayscale values ​​unchanged and only changing the geometric position of the pixels; the 512×512 pixel LASCO C2 corona image is polarized and the central bulge of the sun is cropped to finally obtain a 360×156 pixel image.

4. The method for detecting coronal mass ejections based on a low-rank recovery model as described in claim 1, characterized in that, Steps 4 and 5 are performed as follows: The corona image processed in step 3 is input into the low-rank recovery model. After the model calculates, a binarized image is obtained, which contains coronal mass ejections and coronal flow. In order to detect coronal mass ejections, the three-frame difference method is used to perform difference on the continuous corona images to filter out the coronal flow. What remains is the coronal mass ejection, that is, the coronal mass ejection is detected.

Images