A hierarchical method for quick and automatic recognition of sunspots

Zhao Ziliang; Liu Jiazhen; Hu Zhen; Jia Yanhao; Wang Yue; Li Qingwei; Zhao Zeyang; Liu Yangyi

doi:10.12086/oee.2020.190342

Article navigation > Opto-Electronic Engineering > 2020 Vol. 47 > No. 7 > 190342

Next Article Previous Article

Zhao Z L, Liu J Z, Hu Z, et al. A hierarchical method for quick and automatic recognition of sunspots[J]. Opto-Electron Eng, 2020, 47(7): 190342. doi: 10.12086/oee.2020.190342

Citation:

Zhao Z L, Liu J Z, Hu Z, et al. A hierarchical method for quick and automatic recognition of sunspots[J]. Opto-Electron Eng, 2020, 47(7): 190342. doi: 10.12086/oee.2020.190342

A hierarchical method for quick and automatic recognition of sunspots

1.
School of Mechanical, Electrical & Information Engineering, Shandong University, Weihai, Shandong 264209, China
2.
Institute of Optics and Electronics, Chinese Academy of Sciences, Chengdu, Sichuan 610209, China
3.
Sichuan Police College, Luzhou, Sichuan 646000, China
4.
Key Laboratory of Adaptive Optics, Chinese Academy of Sciences, Chengdu, Sichuan 610209, China

Fund Project: Supported by National Natural Science Foundation of China (11727805 and 11733005) and College Students' Innovation Practice Training Program of Institute of Optics and Electronics, Chinese Academy of Sciences (20184001188)

More Information

Corresponding author: Liu Yangyi, E-mail:liuyangyi_ioe@163.com

Received Date 21 June 2019

Revised Date 04 November 2019

Published Date 01 July 2020

Abstract

Abstract

The observation and recognition of sunspots is an important task of solar physics. By observing and analyzing sunspots, solar physicists are able to analyze and predict solar activities with higher accuracy. With the continuous progress of observation instruments, solar full-disk image data amount is also on a rapid growth. In order to recognize and label sunspots quickly and accurately, a two-layer sunspot recognition model is proposed in this paper. The first layer model is based on deep learning model YOLO. In order to enhance the ability of YOLO to recognize small sunspots, the parameters of YOLO are optimized by using the k-means algorithm based on intersection-over-union. The final YOLO model can identify most large sunspots and sunspot groups, with only a few isolated small sunspots being unidentified. For the purpose of further improving recognition rate of small sunspots, the second layer model applies AGAST (adaptive and generic accelerated segment test) feature detection algorithm to specifically identify the missing small sunspots. The experimental results on SDO/HMI sunspot data set show that all kinds of sunspots can be recognized effectively with high recognition accuracy by using the model proposed in this paper, thus realizing the real-time sunspot detection task.
- sunspot /
- object detection /
- YOLO /
- convolutional neural network /
- k-means /
- AGAST

FullText(HTML)

References

[1]	Liu Y, Zhang H. Relationship between magnetic field evolution and major flare event on July 14, 2000[J]. Astronomy & Astrophysics, 2001, 372(3): 1019-1029. doi: 10.1051/0004-6361:20010550 CrossRef Google Scholar
[2]	饶长辉, 朱磊, 张兰强, 等.太阳自适应光学技术进展[J].光电工程, 2018, 45(3): 170733. doi: 10.12086/oee.2018.170733 CrossRef Google Scholar Rao C H, Zhu L, Zhang L Q, et al. Development of solar adaptive optics[J]. Opto-Electronic Engineering, 2018, 45(3): 170733. doi: 10.12086/oee.2018.170733 CrossRef Google Scholar
[3]	鲍华, 饶长辉, 田雨, 等.自适应光学图像事后重建技术研究进展[J].光电工程, 2018, 45(3): 170730. doi: 10.12086/oee.2018.170730 CrossRef Google Scholar Bao H, Rao C H, Tian Y, et al. Research progress on adaptive optical image post reconstruction[J]. Opto-Electronic Engineering, 2018, 45(3): 170730. doi: 10.12086/oee.2018.170730 CrossRef Google Scholar
[4]	Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems, Lake Tahoe, Nevada, USA, 2012: 1097-1105. Google Scholar
[5]	Matsugu M, Mori K, Mitari Y, et al. Subject independent facial expression recognition with robust face detection using a convolutional neural network[J]. Neural Networks, 2003, 16(5-6): 555-559. doi: 10.1016/S0893-6080(03)00115-1 CrossRef Google Scholar
[6]	Acharya U R, Oh S L, Hagiwara Y, et al. A deep convolutional neural network model to classify heartbeats[J]. Computers in Biology and Medicine, 2017, 89: 389-396. doi: 10.1016/j.compbiomed.2017.08.022 CrossRef Google Scholar
[7]	Banko M, Brill E. Scaling to very very large corpora for natural language disambiguation[C]//Proceedings of the 39th Annual Meeting on Association for Computational Linguistics, Toulouse, France, 2001: 26-33. Google Scholar
[8]	Noor M H M, Salcic Z, Wang K I K. Adaptive sliding window segmentation for physical activity recognition using a single tri-axial accelerometer[J]. Pervasive and Mobile Computing, 2017, 38: 41-59. doi: 10.1016/j.pmcj.2016.09.009 CrossRef Google Scholar
[9]	Nguyen T T, Willis C P, Paddon D J, et al. A hybrid system for learning sunspot recognition and classification[C]//Proceedings of 2006 International Conference on Hybrid Information Technology, Cheju Island, South Korea, 2006: 257-264. Google Scholar
[10]	Zhao C, Lin G H, Deng Y Y, et al. Automatic recognition of sunspots in HSOS full-disk solar images[J]. Publications of the Astronomical Society of Australia, 2016, 33: e018. doi: 10.1017/pasa.2016.17 CrossRef Google Scholar
[11]	Zharkov S, Zharkova V, Ipson S, et al. Technique for automated recognition of sunspots on full-disk solar images[J]. EURASIP Journal on Advances in Signal Processing, 2005, 2005(15): 318462. doi: 10.1155/ASP.2005.2573 CrossRef Google Scholar
[12]	付小娜, 廖成武, 白先勇, 等.基于LeNet-5卷积神经网络的太阳黑子检测方法[J].天文研究与技术, 2018, 15(3): 340-346. Google Scholar Fu X N, Liao C W, Bai X Y, et al. A detection method for sunspots based on convolutional neural network LeNet-5[J]. Astronomical Research & Technology, 2018, 15(3): 340-346. Google Scholar
[13]	LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. doi: 10.1109/5.726791 CrossRef Google Scholar
[14]	Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 2016: 779-788. Google Scholar
[15]	Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]//Proceedings of 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, San Diego, CA, USA, 2005: 886-893. Google Scholar
[16]	Uijlings J R R, van de Sande K E A, Gevers T, et al. Selective search for object recognition[J]. International Journal of Computer Vision, 2013, 104(2): 154-171. Google Scholar
[17]	Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, 2014: 580-587. Google Scholar
[18]	He K M, Zhang X Y, Ren S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916. doi: 10.1109/TPAMI.2015.2389824 CrossRef Google Scholar
[19]	Girshick R. Fast R-CNN[C]//Proceedings of 2015 IEEE International Conference on Computer Vision, Santiago, Chile, 2015: 1440-1448. Google Scholar
[20]	Ren S Q, He K M, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. doi: 10.1109/TPAMI.2016.2577031 CrossRef Google Scholar
[21]	Szegedy C, Liu W, Jia Y Q, et al. Going deeper with convolutions[C]//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 2015: 1-9. Google Scholar
[22]	Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, 2017: 6517-6525. Google Scholar
[23]	Redmon J, Farhadi A. Yolov3: an incremental improvement[Z]. arXiv: 1804.02767[cs: CV], 2018. Google Scholar
[24]	Baranyi T, Győri L, Ludmány A. On-line tools for solar data compiled at the Debrecen observatory and their extensions with the Greenwich sunspot data[J]. Solar Physics, 2016, 291(9): 3081-3102. doi: 10.1007/s11207-016-0930-1 CrossRef Google Scholar
[25]	Győri L, Ludmány A, Baranyi T. Comparative analysis of Debrecen sunspot catalogues[J]. Monthly Notices of the Royal Astronomical Society, 2017, 465(2): 1259-1273. doi: 10.1093/mnras/stw2667 CrossRef Google Scholar
[26]	Kilcik A, Ozguc A, Yurchyshyn V, et al. Sunspot count periodicities in different Zurich sunspot group classes since 1986[J]. Solar Physics, 2014, 289(11): 4365-4376. doi: 10.1007/s11207-014-0580-0 CrossRef Google Scholar
[27]	Rosten E, Porter R, Drummond T. Faster and better: a machine learning approach to corner detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(1): 105-119. doi: 10.1109/TPAMI.2008.275 CrossRef Google Scholar
[28]	Mair E, Hager G D, Burschka D, et al. Adaptive and generic corner detection based on the accelerated segment test[C]//Proceedings of the 11th European Conference on Computer Vision, Heraklion, Crete, Greece, 2010: 183-196. Google Scholar

Overview

Overview

Overview: Sunspots are small dark spots, patches or regions appearing on the sun's surface where strong magnetic fields converge. As an important solar phenomenon, observation and analysis of sunspots can promote understanding and learning of solar activities. For example, it can help astrologists to study the relevance of sunspots groups with flare eruptions. With the development of solar physics and observation instrument, methods of sunspots detection proposed earlier can not satisfy the rapid growth of data amount and data processing performance. Thus, it is urgent for solar physics to propose new methods to detect sunspots with higher accuracy and efficiency.

Traditional digital image processing methods and algorithms based on the sliding window are usually characterized by a slow speed and cannot achieve high accuracy. For instance, pure digital image processing methods are usually not flexible enough to detect sunspots because of the divergence of colors and patterns of different types of sunspots. In addition, the sliding window algorithm proposed earlier is of high time complexity, which shows poor performance in practical applications. To solve problems mentioned above, this paper aims to rapidly recognize all types of sunspots for real-time detection.

We proposed a hierarchical model composed of two components. The first layer is based on deep learning model YOLO. According to the nature of YOLO network, the raw image data with a size of 4096 pixels would be divided into smaller sub-images because sunspots are relatively tiny compared with the whole solar image. At the same time, sunspots will be compressed and disappear in neural network, which will influence the training process of network. In order to improve the ability of first layer to detect more smaller sunspots, the k-means algorithm is applied to optimize the anchor parameter in YOLO model. After the first layer model, most sunspots and sunspot groups are able to be recognized, with just a few smaller sunspots being unidentified. For the purpose of detecting more unidentified smaller sunspots in the first layer, the second layer utilizes AGAST algorithms for feature detection, in which the smaller sunspots are viewed as speckles.

In the experiment, 700 original images from SDO/HMI data set are used to train YOLO network. After training process, the loss function reduces from 935.12 to 0.22. The detecting results show that all kinds of sunspots can be recognized effectively with intersection-over-union being 73.41%, detecting accuracy being about 98.50%, and error recognition rate being around 0.60%. Therefore, the hierarchical model can be used to complete real-time sunspot detection task, and relevant ideas and models could also be applied to solve other object detection problems.