Automatic Detection of Occulted Hard X-ray Flares Using Deep-Learning Methods
Shin-nosuke Ishikawa, Hideaki Matsumura, Yasunobu Uchiyama, Lindsay Glesener
aa r X i v : . [ a s t r o - ph . S R ] J a n Solar PhysicsDOI: 10.1007/ ••••• - ••• - ••• - •••• - • Automatic Detection of Occulted Hard X-ray FlaresUsing Deep-Learning Methods
Shin-nosuke Ishikawa · Hideaki Matsumura · Yasunobu Uchiyama · Lindsay Glesener © Springer ••••
Abstract
We present a concept for a machine-learning classification of hardX-ray (HXR) emissions from solar flares observed by the
Reuven Ramaty HighEnergy Solar Spectroscopic Imager (RHESSI), identifying flares that are eitherocculted by the solar limb or located on the solar disk. Although HXR obser-vations of occulted flares are important for particle-acceleration studies, HXRdata analyses for past observations were time consuming and required specializedexpertise. Machine-learning techniques are promising for this situation, and weconstructed a sample model to demonstrate the concept using a deep-learningtechnique. Input data to the model are HXR spectrograms that are easily pro-duced from RHESSI data. The model can detect occulted flares without theneed for image reconstruction nor for visual inspection by experts. A techniqueof convolutional neural networks was used in this model by regarding the inputdata as images. Our model achieved a classification accuracy better than 90 %,and the ability for the application of the method to either event screening or foran event alert for occulted flares was successfully demonstrated.
Keywords:
Flares, Energetic Particles; Energetic Particles, Acceleration; Spec-trum, X-Ray; X-Ray Bursts, Hard; Corona, Active B S. [email protected] Strategic Digital Business Unit, Mamezou Co., Ltd., 2-1-1 Nishi-Shinjuku, Shinjuku,Tokyo 163-0434, Japan Graduate School of Artificial Intelligence and Science, Rikkyo University, 3-34-1Nishi-Ikebukuro, Toshima, Tokyo 171-8501, Japan Galaxies Inc., 1-1-11 Minami-Ikebukuro, Toshima, Tokyo 171-0022, Japan School of Physics and Astronomy, University of Minnesota, Minneapolis, MN 55455,USA
SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 1 . Ishikawa et al.
1. Introduction
Hard X-ray (HXR) emissions from solar flares provide important informationon how particles are accelerated to high energies. In particular, non-thermalHXR emissions from the solar corona are important because they are emittednear energy-release sites just after acceleration. In many flares, HXR sourcesat footpoints of magnetic-loop structures are prominent compared to coronalsources (Krucker et al., 2008). Therefore, it is challenging to observe weak HXRsources in the corona, with stronger HXR sources that are relatively near.This problem requires a high imaging dynamic range, but the dynamic rangeis limited with past instruments for solar HXR observations. Unlike visible-lightoptics, it is technically difficult to focus HXRs using optics (mirrors and lenses),so indirect imaging methods are typically used. Applications of directly focusingoptics for solar HXR observations have been made but are so far limited toshort-term observations (Krucker et al., 2014; Christe et al., 2016; Grefenstetteet al., 2016). However, to study particle acceleration, which is usually detectedin medium to large flares, long-term observation is necessary. The most thor-ough available data set to date for solar HXRs comes from the
Reuven RamatyHigh Energy Solar Spectroscopic Imager (RHESSI: Lin et al., 2002) spacecraft,which operated from 2002 to 2018. The imaging technique used by RHESSI is arotating modulation collimator, and it is not optimized for high dynamic rangeobservations (Hurford et al., 2002).One of the ways to investigate coronal HXR sources with RHESSI data is toselect flares whose footpoints are behind the solar limb. In those flares, footpointHXR emissions are absorbed by the lower layers of the Sun and are not visible tothe observing instrument. In such a case, there is the possibility to more easilyobserve coronal HXR sources, which are at, or near, the places where particleacceleration is thought to occur. Those flares are often referred to as partiallyocculted flares, and are thought to be important events for investigating featuresof accelerated particles. In this article, we define a partially occulted flare, or justan “occulted flare”, as a flare that occurred near the limb without a detectionof a HXR source at its footpoint(s).Occulted flares observed by RHESSI have been published as catalogs. Kruckerand Lin (2008) studied 55 events, and Effenberger et al. (2017) added to thatcatalog 61 more events that occurred later. Both studies identified the occultedflares by checking HXR images of candidate flares by eye for every event. To per-form such a search, two major steps need to be performed by an experienced HXRimaging analyst (upper path in Figure 1): image reconstruction and visual in-spection. Obtaining HXR images with good quality from the RHESSI data is nota trivial effort with modulation collimators. A complicated image-reconstructionmethod is necessary, and knowledge and experience are required to obtain re-liable images. It is also difficult to determine which features of a reconstructedimage are real and which are artificially generated due to noise in the recon-struction algorithm. As a result, statistical studies of occulted flares were timeconsuming and not available to non-experts. We note that the RHESSI missionarchive (https://hesperia.gsfc.nasa.gov/rhessi3/mission-archive/index.html) haspartially alleviated one of these problems, by providing automatically generated
SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 2 utomatic Detection of Occulted Flares Using Deep Learning
Raw Data ImageReconstructionSpectrogramPreparation Model ApplicationVisual Inspection Occulted Flare
Before this studyThis study Experience necessary
Automation Possible
Figure 1.
Diagram comparing two methods of searching for occulted flares. The upper pathshows a method using manual image reconstruction and visual inspection by an experiencedresearcher. The lower path shows the method suggested in this study using a machine-learningmodel. images in several energy ranges for all
RHESSI flares. However, identifyingocculted flares using traditional methods would still require visual inspectionof all of these events by an expert, so examining the entire RHESSI database inthat way would be impossible.A better situation would be if a broad range of researchers can determinewhether flares are partially occulted or on the solar disk without spending toomuch time, and application of artificial intelligence (AI) technologies, especiallymachine-learning techniques, is a candidate to overcome that difficulty. By treat-ing data without image reconstruction as explanatory variables, we can formulatethis problem as a two-class classification problem. If we construct and train amachine-learning model with a reasonable accuracy, we can reduce the timeand effort to find occulted events and focus more on scientific analysis of targetevents.In the field of solar physics, several recent studies use machine-learning tech-niques for space weather prediction or image classification. Specifically, deeplearning (DL) using models of deep neural networks (DNNs) are used for com-plicated data sets including image data. Nishizuka et al. (2018) and Panos andKleint (2020) each use DNNs for their solar flare prediction models. A convo-lutional neural network (CNN) is a kind of DNN especially applied for imagerecognition, and it is used for classifying solar images in categories of filaments,flare ribbons, prominences, sunspots, and no feature regions (Armstrong andFletcher, 2019). In addition, CNN is also used for instrumental calibration of so-lar telescopes (Neuberg et al., 2019). The generative adversarial network (GAN)technique is an application of DNN used for generating images from other data.GAN is used for generation of solar images from magnetograms (Park et al.,2019), and solar image deconvolution (Xu et al., 2020).So far, in most of the classification studies using machine learning in solarphysics, input data were images understandable to humans, such as UV imagesand magnetograms. While it is true that those automations are natural appli-cations of machine-learning techniques, another advantage of machine-learningalgorithms is the ability to judge data that humans cannot recognize with oureyes only. The input data are not necessarily in an understandable form. Oneexample of non-understandable image classification is a malware detection algo-rithm (Nataraj et al., 2011). In this study, they made “images” from binary data
SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 3 . Ishikawa et al. files, and they classified whether each image contained malware or not based onthe observed patterns. Of course, humans cannot understand an image producedfrom binary data inside a data file, but we can use a CNN model since the dataare represented as a two-dimensional image.In this study, we use RHESSI spectrograms as input data and present aconcept of a CNN model to judge whether test flares are occulted or not. TheRHESSI spectrograms can be systematically produced without specialized ex-pertise, and we can search for occulted flares automatically without an expertand without the need for individual inspection (lower path in Figure 1). Amajor goal of this work is to liberate the extensive database of indirect solarHXR observations from the realm of specialists and make its scientific contentmore easily available. This will induce more active researchers to study particleacceleration in the Sun. This effort supports and enhances the use of the RHESSIlegacy mission archive.
2. Data Set and Modeling
To train and test a machine-learning model to detect occulted flares for su-pervised learning, we prepared HXR spectrograms observed by RHESSI duringflares as input data. A RHESSI spectrogram is a two-dimensional histogram withaxes of time and energy, and the value in each bin indicating the number of countsobserved by RHESSI at that time and energy. Since RHESSI’s imaging techniqueworks by modulating the photon flux, the intensity changes over time containboth temporal and positional information for the HXR source. Imaging analysesare typically performed using data binned in intervals that are an integral multi-ple of the modulation period, which corresponds to the spacecraft rotation periodof approximately four seconds. Rather than separating spectral and positionaldata, we use the spectrograms themselves as input data; we can then applymodels that are well developed for image recognition with little modification.We defined an input data format as a 100 pixel ×
100 pixel histogram coveringone rotation period for the time axis ( ≈ ×
100 pixels.There are 116 occulted-flare events published by Krucker and Lin (2008) andEffenberger et al. (2017) that occurred between 2002 and 2015, and we used all ofthese published events for the data set of occulted flares. Since 116 data are notenough to train a CNN model in most cases, we made ten spectrograms for eachevent using ten consecutive time intervals. In the context of machine learning,
SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 4 utomatic Detection of Occulted Flares Using Deep Learning E ne r g y [ k e V ] On-disk flareOcculted flare
Figure 2.
Sample spectrograms for the input data shown with a log scale, including anexample of (left) an occulted flare and (right) an on-disk flare. this is related to the concept of data augmentation to obtain enough trainingdata from a limited sized data set. The time intervals are selected from the timerange between five rotation periods before the flare peak time to five periods afterit, and there is no time overlap between the intervals. The peak time values arefrom the RHESSI flare list (https://hesperia.gsfc.nasa.gov/rhessi3/data-access/rhessi-data/flare-list/index.html). We therefore obtained 1160 spectrograms offlares that are known to be occulted.For the data set of flares not occulted by the solar limb, we randomly selected1000 events from the RHESSI flare list that had a radial distance from the solardisk center of < ′′ . This ensures that this set of flares occurred on the disk.In addition, we selected only flares with a peak count rate of >
100 counts s − to ensure sufficient statistics for each spectrogram. We did not set a criterion onthe time of the flare observation. One spectrogram for each event with a timeinterval starting from the peak time was produced, and 1000 spectrograms intotal were used for the on-disk data set.We added labels of either of two classes - “occulted” or “on-disk” - to all of thespectrograms. Examples of the spectrograms for the occulted and on-disk flaresare shown in Figure 2. It is difficult to tell by visual inspection which featuresin the spectrograms correspond to the labels. The problem to be solved in thisarticle is formulated to classify this kind of difficult-to-understand images, andwe expect that the machine-learning model can distinguish them even thoughthe human eye cannot.We split each of the occulted and on-disk data set into 80 and 20 % at randomfor training and testing the machine-learning model. 24 occulted flares are pickedand 240 spectrograms of those flares were used for the test data set. The other920 spectrograms were used for the model training. We note that we did not usespectrograms from any of the same flares for both the training and test data sets.Spectrograms from the same flare are thought to have similar features, so there SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 5 . Ishikawa et al.
ResNet-50 Fully-connectedlayer, 256 outputs 50% Dropout layer Fully-connectedlayer, 2 outputsFlatten layer
Figure 3.
The DL model presented in this article based on the ResNet-50 model (He et al.,2015a). is a risk to overestimate the performance if we include them for both the trainingand test data sets. 800 and 200 randomly sampled on-disk flare spectrogramswere used for the training and test sets.
We constructed the DL model with CNNs based on the residual net (ResNet:He et al., 2015a) model originally developed to classify a large set of images(ImageNet). A feature of the ResNet is a concept to train only residuals of eachlayer of neural networks against the input data for better training of neuralnetworks deeper than before. Accuracy of ImageNet data classification with amachine-learning model outperformed that of a human for the first time byHe et al. (2015b) with a new training parameter initializing method (known asHe initialization), and the same group improved the accuracy further with theResNet models in the same year.We added a few layers to a 50 layer ResNet model (ResNet-50) to make atwo-class classifier. A flatten layer was added to convert the ResNet-50 outputto a one-dimensional array. Two fully connected layers with 256 and 2 outputswere added to reduce the output size gradually, and a dropout layer (Srivastavaet al., 2014) was added between the fully connected layers to avoid overfitting.The model configuration is summarized in Figure 3. This model has 31,923,970trainable parameters. We trained the model with cross entropy loss functionand Adam optimizer (Kingma and Ba, 2014) with a learning rate of 3 × .Mini-batch training was performed with a batch size of 64.We implemented the model using the machine-learning framework Tensor-Flow version 1.14.0 (Abadi et al., 2015) in combination with the neural networkwrapper library Keras version 2.2.4 (Chollet, 2015). The model construction,training, and test are performed using Anaconda3 5.3.0 with Python version3.6.8. The ResNet-50 model with the trained parameter set by ImageNet isavailable with the Keras library, and we utilized it as the starting point forbuilding our model.
3. Result
We trained the CNN model for the flare event classification using the trainingdata described in the previous section. Initial parameters for the ResNet-50part of the model were the parameters obtained by training ImageNet. Theparameters are for color images with red, green, and blue channels, and we putidentical two-dimensional spectrogram data for all three color channels of themodel input. This concept corresponds to transfer learning or fine tuning of an
SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 6 utomatic Detection of Occulted Flares Using Deep Learning existing trained machine-learning model. Although it is usually used for new datasets with similarities with the original data used for the model training, we foundthe model training with the HXR spectrograms got better results initializing withthe ImageNet parameters than when not. The ImageNet parameters are onlyinitial values; we did not fix any parameters from them. Even so, the performancewas better than using random initialization. The trained model is available aselectronic supplementary material and in the Data Repository for University ofMinnesota: https://doi.org/10.13020/wtbm-2258We then tested the model with the test data set described in the previoussection. The model classified correctly for &
90 % of the cases. We made severaltrials with different splits for training and test data sets with the same optimizer,loss function, and hyper-parameters, and the results ranged within a few percent.We show detailed results for one of these trials here. In this trial, 226 out of the240 occulted flare test data were successfully categorized as occulted. For theon-disk events, 179 out of the 200 data were successfully categorized. Focusingon the occulted-event detection, the true positive (TP) and true negative (TN)cases are the 226 and 179 events. The accuracy is defined byAccuracy = TP + TN[Total Cases] , (1)and it is calculated to be 92 %. Although the accuracy varied depending ontraining and test data set selections in the several trials, it was >
90 % for mostcases and reached a maximum of 94 %.The precision and recall, which are defined asPrecision = TPTP + FP , (2)Recall = TPTP + FN , (3)where FP and FN are false positive and false negative cases, were 91 % and94 % for the occulted events in the trial shown in the previous paragraph. Theprecision corresponds to the ratio of the actually occulted events out of theevents classified as occulted, and the recall corresponds to the coverage of theocculted event detections out of the total occulted events. The result of thosemeasures are similarly high, and it means we do not have many FN cases andwe do not miss many occulted flares. The F score, which is the harmonic meanof the precision and recall described as[ F -score] = 2[Precision][Recall][Precision] + [Recall] = 2TP2TP + FP + FN , (4)was calculated to be 92 % for the occulted flares. Therefore, it is confirmed thatthe model has an ability to detect occulted flares from systematically generatedHXR spectrograms. The statistical measures are summarized in Table 1 for theocculted and on-disk events.Since the objective of this work is to demonstrate the concept of automaticdetection of occulted flares without specialized expertise, we purposefully did not SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 7 . Ishikawa et al.
Table 1.
Summary of the statistical measures for onemodel test. The total accuracy was 92 %.Label Precision [%] Recall [%] F score [%]Occulted 91 94 92On-disk 93 89 91 fine-tune the model. The classification accuracy could be much higher if hyper-parameter tuning and/or model architecture modification were made. We foundthat the ResNet-50 model achieved better performance than other models basedon built-in models in the Keras library. We did adjust some hyper-parameterssuch as learning rate and batch size to achieve better results with this model.Not much computer resources were necessary to train this model. We per-formed the model training not with a GPU server but with a CPU in a laptopcomputer (MacBook 2017 model with 1.4 GHz dual core Intel Core i7 and 16 GBmemory). It took less ≈
300 seconds per epoch, and we trained the model for10 epochs since initial tests did not show any improvement after 10 epochs.Applying the classification using the already-trained model is even easier; onecan apply this model to a set of spectrograms in much less than an hour with alaptop CPU. With our MacBook, it took less than a minute to classify the 648events described in the following section.
4. Discussion
The accuracy of ≈
90 % or better for the occulted versus on-disk flare classifica-tion shows that this technique could be useful for event screening in the archivedata, or for future observations. The accuracy from this initial demonstration isnot high enough to identify all of the occulted flares from the whole RHESSI flarelist, but it would identify the vast majority ( ≈
90 %) of them, providing plenty ofevents for statistical studies. We can utilize the model in combination with otherinformation such as radial distance from the Sun center, flare size, and energyrange detected. If the radial distance is small, the probability of misdetection ishigh even if it is classified as occulted by the model. But these flares are easyto exclude from the analysis and would not affect studies for occulted flares.If the energy range detected for a flare only included thermal emission (suchas up to ≈
12 keV or less) and not higher-energy, nonthermal emission, or theflare was faint, the probability of a misdetection is higher. But these flares arenot often the ones desired to study for particle acceleration. Therefore, oneusage for this model is to select the set of events to analyze in detail. We wouldobtain a reasonable number of events by using additional criteria along with theclassification result, such as basic information from the flare list, and we cantake more time for scientific analyses for more suspect events. We note that themodel presented in this article was trained with the on-disk events with < ′′ from the disk center, and there is a possibility that the performance to classify SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 8 utomatic Detection of Occulted Flares Using Deep Learning events close to the limb improves if we train the model using the on-disk eventscloser to the limb. In addition to that, it is also possible to use the model toperform an onboard event alert for occulted flares for future observations (e.g.
Solar Orbiter , in which other instruments could respond to a flare trigger by thehard X-ray instrument).We have tested the model for events from the RHESSI flare catalog withcriteria of a peak count rate to be >
100 counts s − , a radial distance to be > ′′ and detection energy range to be up to 25 keV or more. 648 events meetthese criteria, and 536 events are classified as occulted events by our model. The648 events include 78 occulted events from the lists published by Krucker andLin (2008) and Effenberger et al. (2017), and 76 of those events were success-fully classified as occulted events. In addition to these are the other 460 eventsclassified as occulted but not on the previously published lists. These includethe rather famous X-class flare that occurred on September 10 2017, and wecan confirm that the HXR footpoint sources were actually occulted over thesolar limb from previously published results (Gary et al., 2018; Ovchinnikova,Charikov, and Shabalin, 2019). As a result, we successfully demonstrated theability to find an occulted event not in our training or test data sets.In general, for standard DL models, it is difficult to interpret which features ofthe input data are important for the result. The is true for the model presentedin this article, and we cannot tell which feature in the HXR spectrograms isimportant for identifying the occulted flares. We can guess at possibilities forthe differences in HXR spectrograms of occulted and on-disk flares. In general,intensities of coronal HXR sources are weaker than those of footpoint HXRsources, and source areas are generally larger for coronal sources. In HXR ob-servations with both coronal and footpoint source detections, coronal sourcestend to have steeper spectral indices (Masuda et al., 1995; Ishikawa et al., 2011;Krucker et al., 2014). This might make a difference in spectra for the occultedflares, which show only a coronal source in the HXR range, in contrast to theon-disk flares with strong footpoint sources. Future work will examine spatialand spectral characteristics of the false positives in order to determine whethersuch trends are important for the DL model identification.. As described in theprevious section, we did not consume too much time in optimizing the model toachieve the highest possible accuracy. This means that there remains potentialfor future work to construct an optimized model that can detect the occultedflares with even better accuracy. Even in that case, we note that this modelshould be used primarily for event screening, not for directly deriving scientificresults.In this article, we focused only on a two-class classification of the HXR sources,i.e. whether they are occulted or on-disk. For future work, classification in morethan two classes would be possible based on the successful result of this model.Also, a model to classify, detect, or predict features of solar phenomena wouldbe possible using a combination of observations in HXR and other wavelengths. SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 9 . Ishikawa et al.
5. Summary
We constructed a DL model to classify HXR observations of solar flares as oc-culted or on-disk, and have demonstrated a successful concept to detect occultedHXR flares automatically without too much time or presence of an imaginganalysis expert. An accuracy of ≈
90 % was achieved for the test data set, andsuccessfully demonstrated the ability to find occulted flares by detecting anotherflare not on the existing list of occulted flares. We can directly apply this modelfor event screening, and there exists a wide range of application for the DLtechnique in this scientific area.
Acknowledgments
We would like to thank Tomoe Hoshi for introducing an example ofnon-understandable image analysis.
Disclosure of Potential Conflicts of Interest
The authors declare that they have noconflicts of interest.
References
Abadi, M., Agarwal, A., Barham, P., Brevdo, E., Chen, Z., Citro, C., Corrado, G.S., Davis,A., Dean, J., Devin, M., Ghemawat, S., Goodfellow, I., Harp, A., Irving, G., Isard, M., Jia,Y., Jozefowicz, R., Kaiser, L., Kudlur, M., Levenberg, J., Man´e, D., Monga, R., Moore,S., Murray, D., Olah, C., Schuster, M., Shlens, J., Steiner, B., Sutskever, I., Talwar, K.,Tucker, P., Vanhoucke, V., Vasudevan, V., Vi´egas, F., Vinyals, O., Warden, P., Wattenberg,M., Wicke, M., Yu, Y., Zheng, X.: 2015,
TensorFlow: Large-scale machine learning onheterogeneous systems . Software available from tensorflow.org. http://tensorflow.org/.Armstrong, J.A., Fletcher, L.: 2019, Fast Solar Image Classification Using Deep Learning andIts Importance for Automation in Solar Physics.
Solar Phys. , 80. DOI. ADS.Chollet, F.: 2015, keras , GitHub. https://github.com/fchollet/keras.Christe, S., Glesener, L., Buitrago-Casas, C., Ishikawa, S.-N., Ramsey, B., Gubarev, M., Kilaru,K., Kolodziejczak, J.J., Watanabe, S., Takahashi, T., Tajima, H., Turin, P., Shourt, V.,Foster, N., Krucker, S.: 2016, FOXSI-2: Upgrades of the Focusing Optics X-ray Solar Imagerfor its Second Flight.
J. Astron. Instrum. , 1640005. DOI. ADS.Effenberger, F., Rubio da Costa, F., Oka, M., Saint-Hilaire, P., Liu, W., Petrosian, V., Glesener,L., Krucker, S.: 2017, Hard X-Ray Emission from Partially Occulted Solar Flares: RHESSIObservations in Two Solar Cycles. Astrophys. J. , 124. DOI. ADS.Gary, D.E., Chen, B., Dennis, B.R., Fleishman, G.D., Hurford, G.J., Krucker, S., McTiernan,J.M., Nita, G.M., Shih, A.Y., White, S.M., Yu, S.: 2018, Microwave and Hard X-RayObservations of the 2017 September 10 Solar Limb Flare.
Astrophys. J. , 83. DOI.ADS.Grefenstette, B.W., Glesener, L., Krucker, S., Hudson, H., Hannah, I.G., Smith, D.M., Vogel,J.K., White, S.M., Madsen, K.K., Marsh, A.J., Caspi, A., Chen, B., Shih, A., Kuhar, M.,Boggs, S.E., Christensen, F.E., Craig, W.W., Forster, K., Hailey, C.J., Harrison, F.A.,Miyasaka, H., Stern, D., Zhang, W.W.: 2016, The First Focused Hard X-ray Images of theSun with NuSTAR.
Astrophys. J. , 20. DOI. ADS.He, K., Zhang, X., Ren, S., Sun, J.: 2015a, Deep Residual Learning for Image Recognition,arXiv:1512.03385. ADS.He, K., Zhang, X., Ren, S., Sun, J.: 2015b, Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification, arXiv:1502.01852. ADS.Hurford, G.J., Schmahl, E.J., Schwartz, R.A., Conway, A.J., Aschwanden, M.J., Csillaghy, A.,Dennis, B.R., Johns-Krull, C., Krucker, S., Lin, R.P., McTiernan, J., Metcalf, T.R., Sato,J., Smith, D.M.: 2002, The RHESSI Imaging Concept.
Solar Phys. , 61.Ishikawa, S., Krucker, S., Takahashi, T., Lin, R.P.: 2011, On the Relation of Above-the-loopand Footpoint Hard X-Ray Sources in Solar Flares.
Astrophys. J. , 48. DOI. ADS.
SOLA: ishikawa2021_sol_phys.tex; 28 January 2021; 1:47; p. 10 utomatic Detection of Occulted Flares Using Deep LearningKingma, D.P., Ba, J.: 2014, Adam: A Method for Stochastic Optimization, arXiv:1412.6980.ADS.Krucker, S., Lin, R.P.: 2008, Hard X-Ray Emissions from Partially Occulted Solar Flares.
Astrophys. J. , 1181. DOI. ADS.Krucker, S., Battaglia, M., Cargill, P.J., Fletcher, L., Hudson, H.S., MacKinnon, A.L., Masuda,S., Sui, L., Tomczak, M., Veronig, A.L., Vlahos, L., White, S.M.: 2008, Hard X-ray emissionfrom the solar corona.
Astron. Astrophys. Rev. , 155.Krucker, S., Christe, S., Glesener, L., Ishikawa, S.-n., Ramsey, B., Takahashi, T., Watanabe, S.,Saito, S., Gubarev, M., Kilaru, K., Tajima, H., Tanaka, T., Turin, P., McBride, S., Glaser,D., Fermin, J., White, S., Lin, R.: 2014, First Images from the Focusing Optics X-Ray SolarImager. Astrophys. J. Lett. , L32. DOI. ADS.Lin, R.P., Dennis, B.R., Hurford, G.J., Smith, D.M., Zehnder, A., Harvey, P.R., Curtis, D.W.,Pankow, D., Turin, P., Bester, M., Csillaghy, A., Lewis, M., Madden, N., van Beek, H.F.,Appleby, M., Raudorf, T., McTiernan, J., Ramaty, R., Schmahl, E., Schwartz, R., Krucker,S., Abiad, R., Quinn, T., Berg, P., Hashii, M., Sterling, R., Jackson, R., Pratt, R., Campbell,R.D., Malone, D., Landis, D., Barrington-Leigh, C.P., Slassi-Sennou, S., Cork, C., Clark,D., Amato, D., Orwig, L., Boyle, R., Banks, I.S., Shirey, K., Tolbert, A.K., Zarro, D., Snow,F., Thomsen, K., Henneck, R., McHedlishvili, A., Ming, P., Fivian, M., Jordan, J., Wanner,R., Crubb, J., Preble, J., Matranga, M., Benz, A., Hudson, H., Canfield, R.C., Holman,G.D., Crannell, C., Kosugi, T., Emslie, A.G., Vilmer, N., Brown, J.C., Johns-Krull, C.,Aschwanden, M., Metcalf, T., Conway, A.: 2002, The Reuven Ramaty High-Energy SolarSpectroscopic Imager (RHESSI).
Solar Phys. , 3.Masuda, S., Kosugi, T., Hara, H., Sakao, T., Shibata, K., Tsuneta, S.: 1995, Hard X-RaySources and the Primary Energy-Release Site in Solar Flares.
Pub. Astron. Soc. Japan ,677. ADS.Nataraj, L., Karthikeyan, S., Jacob, G., Manjunath, B.S.: 2011, Malware images: Visualizationand automatic classification. In: Proc. 8th Internat. Symp. on Visualization for CyberSecurity , VizSec ’11 , Association for Computing Machinery, New York, NY, USA. ISBN9781450306799. DOI.Neuberg, B., Bose, S., Salvatelli, V., dos Santos, L.F.G., Cheung, M., Janvier, M., GunesBaydin, A., Gal, Y., Jin, M.: 2019, Auto-Calibration of Remote Sensing Solar Telescopeswith Deep Learning, arXiv:1911.04008. ADS.Nishizuka, N., Sugiura, K., Kubo, Y., Den, M., Ishii, M.: 2018, Deep Flare Net (DeFN) Modelfor Solar Flare Prediction.
Astrophys. J. , 113. DOI. ADS.Ovchinnikova, E.P., Charikov, Y.E., Shabalin, A.N.: 2019, X-ray of the 2017 September 10Solar Flare.
J. Phys. Conf. Ser. , 022028. DOI. ADS.Panos, B., Kleint, L.: 2020, Real-time Flare Prediction Based on Distinctions between Flaringand Non-flaring Active Region Spectra.
Astrophys. J. , 17. DOI. ADS.Park, E., Moon, Y.-J., Lee, J.-Y., Kim, R.-S., Lee, H., Lim, D., Shin, G., Kim, T.: 2019,Generation of Solar UV and EUV Images from SDO/HMI Magnetograms by Deep Learning.
Astrophys. J. Lett. , L23. DOI. ADS.Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R.: 2014, Dropout: Asimple way to prevent neural networks from overfitting.
J. Machine Learn. Res. , 1929.https://jmlr.org/papers/v15/srivastava14a.html.Xu, L., Sun, W., Yan, Y., Zhang, W.: 2020, Solar Image Deconvolution by GenerativeAdversarial Network, arXiv:2001.03850. ADS., 1929.https://jmlr.org/papers/v15/srivastava14a.html.Xu, L., Sun, W., Yan, Y., Zhang, W.: 2020, Solar Image Deconvolution by GenerativeAdversarial Network, arXiv:2001.03850. ADS.