Method for detecting and classifying multi-class events of solar activity
Through deep learning algorithm combined with multimodal solar image data, a multi-category event detection and classification model for solar activity was established, which solved the problem of uncommon detection of multiple solar events and insufficient real-time performance, and achieved efficient automated detection and classification.
Patent Information
- Application Number
- CN202510752053.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-06
- Publication Date
- 2025-09-02
AI Technical Summary
The existing solar activity detection algorithm cannot handle multiple solar events at the same time, and the traditional methods lack real-time performance in the face of massive data, so they cannot realize automated detection.
Deep learning algorithms are used to establish detection and classification models, use multimodal solar image data for training, combine traditional algorithms to generate tags, establish data sets of multi-class solar activity event detection and classification tasks, and use deep learning networks such as YOLOv5 for training to realize automated detection and classification of multi-class solar activity events.
It realizes the automation, real-time detection and classification of multiple types of solar activity events, reduces the degree of manual intervention, and improves the processing speed and accuracy of massive data.
Smart Images
Figure CN120580504A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of solar image processing, and in particular to a method for detecting and classifying multiple types of solar activity events. Background Art
[0002] Solar activity is a key area of solar physics research. Solar activity is a dynamic and complex system encompassing a range of events from static to active regions on the solar surface, including sunspots, flares, prominences, and the corona. These events reflect the magnetic field activity, material movement, and energy release on the solar surface, with causal relationships linking their occurrence and development. For example, during peak solar activity, sunspot regions exhibit vibrant and complex magnetic field activity, generating intense electromagnetic radiation known as solar flares, and potentially triggering large coronal mass ejections (CMEs). When a CME event occurs, a large amount of charged particles and high-energy magnetic fields will be released into interstellar space. Especially when the CME is heading towards the Earth, it will cause strong disturbances to the Earth's space environment, triggering space weather phenomena such as geomagnetic storms and auroras, which may cause catastrophic consequences such as the collapse of communication terminals and high-latitude power systems, damage to space satellite equipment, and even threaten the personal safety of astronauts (Palmerio E, Lee CO, Richardson IG, et al. CME evolution in the structured heliosphere and effects at Earth and Mars during solar minimum[J]. Space Weather, 2022, 20(9): e2022SW003215). Therefore, in order to study the precursor activities of solar eruptions and reveal the laws of their interaction with the Earth's space environment, it is necessary to monitor and analyze a series of solar events, so as to analyze the complex dynamic characteristics of solar activities, predict and prevent space weather disasters, and ensure the safety of human scientific and technological infrastructure and space activities.
[0003] Initially, solar event detection algorithms used predefined detection rules based on the appearance and activity characteristics of the solar atmospheric structure (Henney CJ, Harvey J W. Automated coronal hole detection using He I 1083 nmspectro-heliograms and photo-spheric magnetograms[J]. arXiv preprint astro-ph / 0701122, 2007; Pérez-Suárez D, Higgins PA, Bloomfield DS, et al. Automated solar feature detection for space weather applications[J]. ImageProcessing: Concepts, Methodologies, Tools, and Applications, 2013: 979-997.). However, because different solar events typically exhibit distinct image features in space and time, using generally defined rules cannot capture the unique structures of all solar events. Therefore, scientists have gradually introduced more advanced mathematical algorithms to optimize detection algorithms based on the characteristics of each solar event. In 2009, the SMART algorithm smoothed and thresholded the magnetic image, then performed masking and dilation to detect and track solar active regions (P. Higgins, P. Gallagher, R. McAteer, and D. Bloomfield. The solar monitoractive region tracking (smart) algorithm: Variation of magnetic feature properties through solar cycle 23. In AGU Fall Meeting Abstracts, 2009.). The CHIMERA algorithm based on multi-thermal emission identification was proposed and applied to the detection of coronal hole events on the solar surface (Garton TM, Gallagher PT, Murray S A. Automated coronal hole identification via multi-thermal intensity segmentation[J]. Journal of Space Weather and SpaceClimate, 2018, 8: A02).The SPoCA algorithm, based on spatial probability clustering, uses multi-wavelength images to extract features of solar active regions and coronal holes and clusters these features using an iterative minimization algorithm. This enables the automatic detection and classification of these two events (Verbeeck C, Delouille V, Mampaey B, et al. TheSPoCA-suite: Software for extraction, characterization, and tracking ofactive regions and coronal holes on EUV images[J]. Astronomy&Astrophysics, 2014, 561: A29.). Furthermore, corresponding detection algorithms have been developed for other solar events. For example, the Sigmoid Sniffer algorithm uses multiple thresholds to identify persistent high-resolution structures in full-disk solar images, enabling the automatic detection of S-type sigmoid events on the solar surface and determining their chirality and other characteristics. The Advanced Automated Solar Filament Detection and Characterization Code algorithm can detect solar filament events from full-disk Ha images (Raouafi N, Bernasconi PN, Georgoulis M K. The “Sigmoid Sniffer” and the “Advanced Automated Solar Filament Detection and Characterization Code” Modules[C] / / American Astronomical Society MeetingAbstracts. 216. 2010, 216: 402.32.). While these traditional solar event detection algorithms can detect specific events, they suffer from the following problems: While each algorithm processes a specific target event, it is unable to apply to other solar events. Furthermore, with the explosive growth of observational data, traditional solar event detection algorithms are unable to cope, reducing the real-time performance of solar event detection.
[0004] In recent years, with advances in computer vision technology, machine learning-based methods have been gradually applied to solar image event detection. Some traditional machine learning algorithms, such as support vector machines, decision trees, and random forests, have been shown to be effective for detecting coronal holes. These algorithms, by learning from the features of extreme ultraviolet solar images, have advantages for detecting coronal holes, but there is still significant room for improvement in detection accuracy (Reiss MA, Hofmeister SJ, De Visscher R, et al. Improvements on coronal hole detection in SDO / AIA images using supervised classification[J]. Journal of SpaceWeather and Space Climate, 2015, 5: A23; Delouille V, Hofmeister SJ, Reiss MA, et al. Coronal holes detection using supervised classification[M] / / Machinelearning techniques for space weather. Elsevier, 2018: 365-395). In 2017, Kucuk et al. used the Faster R-CNN model to detect solar active regions (Kucuk A, Aydin B, Angryk R. Multi-wavelength solar event detection using faster R-CNN[C] / / 2017 IEEE International Conference on Big Data (Big Data). IEEE, 2017: 2552-2558). However, these machine learning-based solar event detection techniques are unable to simultaneously process and detect multiple types of solar events. Therefore, faced with the massive increase in solar observation data, there is an urgent need for an automated detection and classification method for multiple types of solar activity events.
[0005] In response to the above problems, the present invention proposes a method for detecting and classifying multiple types of solar activity events, which can better solve the problem of universality and automatic detection of multiple types of solar activities. Summary of the Invention
[0006] (1) The technical problem solved by the present invention is: Aiming at the problem that the detection algorithms for multiple types of solar events are not universal and the demand for automated detection algorithms due to the massive growth of data, a detection and classification method for multiple types of solar activity events is proposed.
[0007] (2) To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: A method for detecting and classifying multiple types of solar activity events is proposed. This method proposes using a deep learning algorithm to establish a detection and classification model, and uses multimodal solar image data as a basis for learning and training, ultimately achieving automated detection and classification of multiple types of solar activity events. The method includes: Step 1: Collect multimodal solar image data, process the image data to ensure that the spatial dimensions of the image data are the same, and use the processed images as the images of the dataset; Step 2: Use a variety of traditional solar image detection algorithms to detect multiple types of solar activity events, generate the detection box position and category of the solar activity event as the label of the dataset; Step 3: Classify and integrate the images obtained in step 1 and the event labels obtained in step 2 to establish a dataset for solar activity multi-class event detection and classification tasks; Step 4: Build a multi-class solar event detection and classification model based on deep learning. Train the model to learn the characteristics of various events in the dataset to detect multiple types of solar events. Step 5: Use the trained model to detect and classify the active events involved in the input solar image, and output the detection and classification results.
[0008] Among them, in step 1, the multimodal image data is processed as follows: the sun's position is corrected based on the solar north pole to ensure that the solar disk is centered; the solar image is cropped to ensure that the multimodal data has the same solar disk size; and the solar image is up- / down-sampled to ensure the same spatial pixel scale.
[0009] Among them, in step 2, the multiple types of solar activity events refer to sunspots, flares, coronal holes, active regions, solar prominences and other events; the multiple traditional solar image detection algorithms include but are not limited to SPoCA, CHIMERA, SMART, SigmoidSniffer and other algorithms.
[0010] In step three, the image and the event tags involved are integrated according to the image shooting time, and the tags include the detection frame of the event location and the event type.
[0011] Among them, in step 4, the input of the deep learning network model is set to multimodal image data, and the output nodes are the locations and types of various events involved in the image; the deep learning method includes but is not limited to SSD, CNN, R-CNN, Faster R-CNN, U-Net, YOLO and other models; based on the data set obtained in step 3, the initial network is trained to obtain effective model parameters, and then the said solar activity multi-class event detection and classification model is obtained.
[0012] (3) Beneficial effects: This paper proposes combining multimodal data from solar images with a deep learning-based approach to explore the spatial characteristics of multiple types of solar events, resulting in a method that automates the simultaneous detection and classification of multiple solar events. This method effectively detects and classifies solar events in solar images. This end-to-end processing significantly reduces manual intervention and demonstrates superior real-time processing speed, even with massive amounts of solar observation data. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 Flowchart of the method for detecting and classifying multiple types of solar activity events.
[0014] Figure 2 It is a flow chart of a specific design example in the present invention.
[0015] Figure 3 A deep learning model and input-output diagram of a specific design example in the present invention. DETAILED DESCRIPTION
[0016] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.
[0017] Figure 1 Shown is a flow chart of the method of the present invention, which includes the following steps: Step 1: Collect multimodal solar image data, process the image data to ensure that the spatial dimensions of the image data are the same, and use the processed images as the images of the dataset; Step 2: Use a variety of traditional solar image detection algorithms to detect multiple types of solar activity events, generate the detection box position and category of the solar activity event as the label of the dataset; Step 3: Classify and integrate the images obtained in step 1 and the event labels obtained in step 2 to establish a dataset for solar activity multi-class event detection and classification tasks; Step 4: Build a multi-class solar event detection and classification model based on deep learning. Train the model to learn the characteristics of various events in the dataset to detect multiple types of solar events. Step 5: Use the trained model to detect and classify the active events involved in the input solar image, and output the detection and classification results.
[0018] Figure 2 The flowchart of a specific embodiment of the present invention is a method for detecting and classifying multiple types of solar activity events. The specific process is as follows: Step 1: Collect the solar full-disk observation image data from the SDO satellite. There are five types of solar activity events involved, namely: sunspots, flares, dark stripes, coronal holes, and prominences; solar activity is usually active in different atmospheric layers, and different types of events usually have specific observation bands. There are five observation bands involved, namely: HMI, 171 Å, 131 Å, 193 Å, and 304 Å; the observation time of the collected data is from 2010 to 2019. Convert the solar image data format to JPEG format; perform direction correction on the multimodal image data: correct the solar position based on the solar north pole to ensure that the solar disk is centered; process the solar size in the multimodal image data: crop the solar image to ensure that the multimodal data has the same solar disk size; process the size of the multimodal image data: up / downsample the solar image so that the image size of each event in each band is 512×512 pixels; Step 2: Use a variety of traditional algorithms to detect solar activity events and generate labels. Based on the EGSO-SFC algorithm, extract sunspot events in HMI solar images and generate labels for the events involved; based on the spatial probability clustering SPoCA algorithm, extract solar flares and coronal hole events in multi-wavelength images and generate labels for the events involved. The imaging bands involved include 171 Å and 193 Å; based on the Sigmoid Sniffer algorithm, extract solar dark stripe events in 131 Å wavelength images and generate labels for the events involved; based on the Chunming algorithm, extract solar prominence events in 304 Å wavelength images and generate labels for the events involved. There are a total of 4 traditional detection algorithms involved; the labels involved include the pixel position of the rectangular box where the solar event occurs and the event category; Step 3: Integrate the image obtained in step 1 and the event label obtained in step 2 according to the image shooting time, and establish a dataset for solar activity multi-class event detection and classification tasks. The number of sunspot, flare, dark stripe, coronal hole, and solar prominence event labels in the dataset are: 8766, 59069, 18319, 94693, and 5858, respectively. The event category ratio of the training set and the test set of each type of event in the dataset is about 4:1. Specifically, the number of sunspot, flare, dark stripe, coronal hole, and solar prominence event labels in the training set of the dataset are: 7000, 47200, 14600, 75700, and 4680, respectively; the number of sunspot, flare, dark stripe, coronal hole, and solar prominence event labels in the test set of the dataset are: 1766, 11869, 3719, 18993, and 1178, respectively; Step 4: Build a multi-class solar activity event detection and classification model based on the YOLOv5 deep learning algorithm. Five-band full-disk solar images from SDO serve as input to the deep learning algorithm. The YOLOv5-based deep learning model consists of four modules: input module, backbone module, neck module, and prediction module. For example: Figure 3 It is a schematic diagram of a deep learning model of a specific design example in the present invention. The input module uses the Mosaic data enhancement operation to improve the training speed of the model and the accuracy of the network. The Backbone module uses the CSP structure and the Focus structure to extract the characteristics of various solar events. The Neck module uses the FPN+PAN structure to improve the diversity and robustness of the characteristics of various solar events. The prediction module uses GIOU-Loss to improve the detection accuracy of the algorithm. The features of various events in the training set are learned using the YOLOv5-based deep learning algorithm, and the parameter model is saved as the solar activity multi-class event detection and classification algorithm described in the present invention. The test was carried out based on the data of the test set, and the detection accuracy mAP (@0.5IoU) indicators of sunspots, flares, dark stripes, coronal holes, and prominence events were 81%, 79%, 80%, 77%, and 70%, respectively; Step 5: Output the evaluation results. For example, for a set of five-band full disk image data, the solar activity multi-class event detection and classification model is used to give the following Figure 3 The output results are shown. In the example described, the time between detection input and output is only approximately 2.1 seconds, while traditional algorithms require over an hour to process. Compared to the minimum solar activity cycle of a few minutes, the present invention enables automated real-time detection and classification of multiple types of solar activity events. The present invention can effectively detect and classify multiple types of solar activity events involved in solar images. This end-to-end processing approach significantly reduces the degree of manual intervention and improves the real-time processing speed of massive solar observation data.
[0019] The above are specific embodiments disclosed in the present invention. Any portions not described in detail are known in the art. However, the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by any person skilled in the art within the technical scope disclosed in the present invention are intended to be covered by the scope of protection of the present invention.
Claims
1. A method for detecting and classifying multiple types of solar activity events, characterized by: This method proposes combining solar multimodal image data with deep learning methods to mine the spatial characteristics of multiple types of solar activity events and design a method that can automatically detect and classify multiple types of solar activity events. The method includes the following steps: Step 1: Collect multimodal solar image data, process the image data to ensure that the spatial dimensions of the image data are the same, and use the processed images as the images of the dataset; Step 2: Use a variety of traditional solar image detection algorithms to detect multiple types of solar activity events, generate the detection box position and category of the solar activity event as the label of the dataset; Step 3: Classify and integrate the images obtained in step 1 and the event labels obtained in step 2 to establish a dataset for solar activity multi-class event detection and classification tasks; Step 4: Build a multi-class solar event detection and classification model based on deep learning. Train the model to learn the characteristics of various events in the dataset to detect multiple types of solar events. Step 5: Use the trained model to detect and classify the active events involved in the input solar image, and output the detection and classification results.
2. A method for detecting and classifying multiple types of solar activity events according to claim 1, characterized in that: In step 1, the multimodal image data is processed by: correcting the sun's position based on the solar north pole to ensure that the solar disk is centered; cropping the solar image to ensure that the multimodal data has the same solar disk size; and up / down sampling the solar image to ensure the same spatial pixel scale.
3. The method for detecting and classifying multiple types of solar activity events according to claim 2, wherein: In step 2, the multiple types of solar activity events include sunspots, flares, coronal holes, active regions, solar prominences and other events; the multiple traditional solar image detection algorithms include but are not limited to SPoCA, CHIMERA, SMART, Sigmoid Sniffer and other algorithms.
4. A method for detecting and classifying multiple types of solar activity events according to claim 3, characterized in that: In step three, the image and the label of the event involved are integrated according to the image shooting time. The label includes the detection box of the event location and the event type.
5. The method for detecting and classifying multiple types of solar activity events according to claim 4, characterized in that: In step 4, the input of the deep learning network model is set to multimodal image data, and the output nodes are the locations and types of various events involved in the image; the deep learning method includes but is not limited to SSD, CNN, R-CNN, FasterR-CNN, U-Net, YOLO and other models; based on the data set obtained in step 3, the initial network is trained to obtain effective model parameters, and then the solar activity multi-class event detection and classification model is obtained.