太阳射电频谱爆发识别的元学习方法

doi:10.12305/j.issn.1001-506X.2022.08.04

系统工程与电子技术 ›› 2022, Vol. 44 ›› Issue (8): 2410-2418.doi: 10.12305/j.issn.1001-506X.2022.08.04

太阳射电频谱爆发识别的元学习方法

郭军成^1,³, 万刚¹, 胡欣杰¹, 严发宝², 王帅^1,*

1. 航天工程大学航天信息学院, 北京 101416
2. 山东大学空间科学研究院空间电磁探测技术实验室, 山东威海 264200
3. 中国人民解放军66444部队, 北京 100042

收稿日期:2021-08-31 出版日期:2022-08-01 发布日期:2022-08-24
通讯作者: 王帅
作者简介:郭军成(1996—), 男, 硕士研究生, 主要研究方向为智能信息处理、深度学习|万刚(1976—), 男, 教授, 博士研究生导师, 主要研究方向为空间环境仿真、本体构建|胡欣杰(1972—), 女, 教授, 博士研究生导师, 主要研究方向为智能信息处理、数据库系统|严发宝(1984—), 男, 副教授, 硕士研究生导师, 学历主要研究方向为太阳射电观测设备|王帅(1990—), 男, 讲师, 博士, 主要研究方向为空间环境、深度学习
基金资助:
国家自然科学基金(41904158);国家自然科学基金(11703017);中国博士后科学基金(2019M652385)

Meta-learning method for solar radio spectrum burst recognition

Juncheng GUO^1,³, Gang WAN¹, Xinjie HU¹, Fabao YAN², Shuai WANG^1,*

1. Department of Aerospace Information, Space Engineering University, Beijing 101416, China
2. Laboratory of Space Electromagnetic Detection Technology, Shandong University, Weihai 264200, China
3. Unit 66444 of the PLA, Beijing 100042, China

Received:2021-08-31 Online:2022-08-01 Published:2022-08-24
Contact: Shuai WANG

摘要/Abstract

摘要：

太阳射电频谱图像在太阳活动和空间天气的观测、研究和预报中有着重要的作用。太阳射电宽带动态频谱仪是国内观测太阳射电信号的主要设备, 但受到窗口时间、观测设备和太阳活动规律的影响, 其所采集到的频谱数据存在有效样本量少的问题。针对这一现状, 提出了一种基于元学习和迁移学习的少样本学习方法, 用于改善太阳射电频谱图像的分类性能。首先模型在元学习基准数据集上进行元知识的学习, 然后对射电频谱图像进行小样本识别的模型定义, 最后将元知识迁移到频谱图像数据集的分类任务中。通过对多种元学习方法进行实验分析和性能比较, 证明了本文方法的先进性和有效性。

关键词: 太阳射电频谱, 少样本学习, 元学习, 知识迁移

Abstract:

Solar radio spectrum images play an important role in the observation, research and prediction of solar activity and space weather. The solar radio broadband dynamic spectrometer is the main equipment for observing solar radio signals in China. However, it is affected by the window time, observation equipment and the law of solar activities, and the collected spectrum data has the problem of few effective samples. In view of this situation, a few-shot learning method based on meta-learning and transfer-learning is proposed to improve the classification performance of solar radio spectrum images. First, the model learns meta-knowledge on the meta-learning benchmark dataset, then defines the model for a few sample recognition of radio spectrum images, and finally transfers the meta-knowledge to the classification task of the spectrum image dataset. Through experimental analysis and performance comparison of multiple meta-learning methods, it proves that the method in this paper is advanced and effective.

Key words: solar radio spectrum, few-shot learning, meta-learning, knowledge transfer

中图分类号:

TP391.1

郭军成, 万刚, 胡欣杰, 严发宝, 王帅. 太阳射电频谱爆发识别的元学习方法[J]. 系统工程与电子技术, 2022, 44(8): 2410-2418.

Juncheng GUO, Gang WAN, Xinjie HU, Fabao YAN, Shuai WANG. Meta-learning method for solar radio spectrum burst recognition[J]. Systems Engineering and Electronics, 2022, 44(8): 2410-2418.

图/表 15

图1

图2

图3

图4

图5

图6

图7

表1

表2

表3

表4

表5

表6

图8

图9

参考文献 33

1	陈耀. 错位的频谱: 太阳爆发的奥秘[EB/OL]. [2021-08-13]. http://blog.sciencenet.cn/blog-685476-826124.html.
	CHEN Y. Misplaced spectrum: the mystery of the solar eruption[EB/OL]. [2021-08-13]. http://blog.sciencenet.cn/blog-685476-826124.html.
2	谭程明. 太阳活动的射电爆发研究及多波段观测综合分析[D]. 北京: 中国科学院国家天文台, 2007.
	TAN C M. Study on radio burst of solar activity and comprehensive analysis of multi band observation[D]. Beijing: National Astronomical Observatory, Chinese Academy of Sciences, 2007.
3	FU Q J , JI H R , QIN Z H , et al. A new solar broadband radio spectrometer (SBRS) in China[J]. Solar Physics, 2004, 222 (1): 167- 173. doi: 10.1023/B:SOLA.0000036876.14446.dd
4	SALMANE H , WEBER R , ABED-MERAIM K , et al. A method for the automated detection of solar radio bursts in dynamic spectra[J]. Journal of Space Weather and Space Climate, 2018, 8 (15): 123- 134.
5	FU Q J , QIN Z H , JI H R , et al. A broadband spectrometer for decimeter and microwave radio bursts[J]. Solar Physics, 1995, 160 (1): 97- 103. doi: 10.1007/BF00679098
6	LOBZIN V V , CAIRNS I H , ROBINSON P A , et al. Automatic recognition of type Ⅲ solar radio bursts: automated radio burst identification system method and first observations[J]. Space Weather the International Journal of Research & Applications, 2009, 7 (4): 67- 85.
7	GU B , SHENG V S , TAY K Y , et al. Incremental support vector learning for ordinal regression[J]. IEEE Trans.on Neural Networks & Learning Systems, 2015, 26 (7): 1403- 1416.
8	CHEN Z , MA L , XU L , et al. Imaging and representation learning of solar radio spectrums for classification[J]. Multimedia Tools & Applications, 2016, 75 (5): 2859- 2875.
9	陈思思. 基于卷积神经网络的太阳射电频谱图的分类算法研究[D]. 深圳: 深圳大学, 2018.
	CHEN S S. Research on classification algorithm of solar radio spectrum based on convolutional neural network[D]. Shenzhen: Shenzhen University, 2018.
10	崔泽潇. 基于异常值检测和K均值聚类的太阳射电频谱图像自动检测方法研究[D]. 昆明: 云南大学, 2019.
	CUI Z X. Research on automatic detection method of solar radio spectrum image based on outlier detection and K-means clustering[D]. Kunming: Yunnan University, 2019.
11	张巧曼. 太阳射电爆发事件的分类与定位检测方法研究[D]. 威海: 山东大学, 2020.
	ZHANG Q M. Research on classification and location detection methods of solar radio burst events[D]. Weihai: Shandong University, 2020.
12	JANKOWSKI N , DUCH W , GRABCZEWSKI K . Meta-learning in computational intelligence[M]. Berlin: Springer Science and Business Media, 2011.
13	LI F F , FERGUS R , PERONA P . One-shot learning of object categories[J]. IEEE Trans.on Pattern Analysis and Machine Intelligence, 2006, 28 (4): 594- 611. doi: 10.1109/TPAMI.2006.79
14	SHEN X Y , HERTZMANN A , JIA J Y , et al. Automatic portrait segmentation for image stylization[J]. Computer Graphics Forum, 2016, 35 (2): 93- 102. doi: 10.1111/cgf.12814
15	GARCIA-GARCIA A , ORTS-ESCOLANO S , OPREA S , et al. A survey on deep learning techniques for image and video semantic segmentation[J]. Applied Soft Computing, 2018, 70 (4): 41- 65.
16	LAKE B M, SALAKHUTDINOV R, TENENBAUM J B. One-shot learning by inverting a compositional causal process[C]//Proc. of the Conference and Workshop on Neural Information Processing Systems, 2013: 2526-2534.
17	SALAKHUTDINOV R, TENENBAUM J, TORRALBA A. One-shot learning with a hierarchical nonparametric Bayesian Model[C]//Proc. of the ICML Workshops, 2012: 195-207.
18	MUNKHDALAI T, YU H. Meta networks[C]//Proc. of the Machine Learning Machine, 2017: 2554-2563.
19	LI X Y , LONG S P , ZHU J . Survey of few-shot learning based on deep neural network[J]. Application Research of Computers, 2020, 37 (8): 2241- 2247.
20	SUNG F, ZHANG L, XIANG T, et al. Learning to learn: Meta-Critic networks for sample efficient learning[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1706.09529.
21	LU J, GONG P H, YE J P, et al. Learning from very few samples: a survey[EB/OL]. [2021-08-20]. https://arXiv.org/abs/2009.02653v2.
22	SNELL J, SWERSKY K, ZEMEL R S. Prototypical networks for few-shot learning[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1703.05175.
23	ORIAL V, CHARLES B, TIMOTHY L, et al. Matching networks for one shot learning[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1606.04080v2.
24	SUNG F, YANG Y X, ZHANG L, et al. Learning to compare: relation network for few-shot learning[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1711.06025.
25	SANTORO A, BARTUNOV S, BOTVINICK M, et al. One-shot learning with memory-augmented neural networks[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1605.06065.
26	GRAVES A, WAYNE G. Neural turing, machines[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1410.5401.
27	FINN C, ABBEEL P, LEVINE S. model-agnostic meta-learning for fast adaptation of deep networks[C]//Proc. of International Conference on Machine Learning, 2017: 1126-1135.
28	SACHIN R, HUGO L. Optimization as a model for few-shot learning[C]//Proc. of the International Learning Representations, 2017: 3981-3989.
29	MARCIN A, MISHA D, SERGIO G C, et al. Learning to learn by gradient descent by gradient descent[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1606.04474.
30	LI Z G, ZHOU F W, CHEN F, et al. Meta-SGD: learning to learn quickly for few-shot learning[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1707.09835.
31	ALEX N, JOSHUA A, JOHN S, et al. On first-order meta-learning algothrim[EB/OL]. [2021-08-20]. https://arXiv.org/abs/1803.02999.
32	HOCHREITER S , SCHMIDHUBER J . Long short-term memory[J]. Neural Computation, 1997, 9 (8): 1735- 1780. doi: 10.1162/neco.1997.9.8.1735
33	CHEN Z, MA L, XU L, et al. Multimodal learning for classification of solar radio spectrum[C]//Proc. of the IEEE International Conference on Systems, Man, and Cybernetics, 2015: 1035-1040.

参数	类型
参数	Non-burst	Burst	Calibration	总计
数量	3 335	579	494	4 408
尺寸	240×2 520	240×2 520	240×2 520	240×2 520

参数	类型
参数	MiniImageNet	Omniglot	射电频谱数据
原始尺寸	84×84×3	105×105×1	240×2 520×1
实验尺寸	60×60×3	60×60×1	60×60×3(1)
类别数量	64	964	3
每类数量	600	20	600

GPU	CPU	主板	内存
NVIDIA TITAN RTX 24G	Intel i9-9900k	Intel Z390 UD	64G

模型	base class	K	dist	inner_train/ val_steps	layerr_size	batch_size (episodes)	inner_lr/ meta_lr
Reptile^[31]	MiniImageNet	1、5、20	N/A	1/3	576	32	0.4/0.001
Prototypical Nets^[22]	MiniImageNet	1、5、20	L2	N/A	N/A	100	N/A
Matching Nets^[23]	MiniImageNet	1、5、20	Cosine	N/A	1 600	100	N/A
Meta-learner LSTM^[28]	MiniImageNet	1、5、20	N/A	N/A	512	64	0.5/0.001
MAML^[27]	MiniImageNet	1、5	N/A	1/3	576	32	0.4/0.001
MAML	MiniImageNet	20	N/A	1/3	576	16	0.4/0.005
Prototypical Nets	Omniglot	1、5	L2	N/A	N/A	100	N/A
Matching Nets	Omniglot	1、5	Cosine	N/A	64	100	N/A

模型	3-way accracy/%
模型	1-shot	5-sot	20-shot
Reptile	60.28±0.66	69.59±1.53	70.52±1.64
Prototypical Nets	54.68±0.94	73.20±0.01	80.56±0.33
Matching Nets	53.39±0.58	63.80±0.46	68.69±0.45
MAML	75.19±0.55	88.57±0.55	85.72±0.89
Meta-learner LSTM	72.95±0.69	85.46±0.84	91.60±0.73

太阳射电频谱爆发识别的元学习方法

Meta-learning method for solar radio spectrum burst recognition

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 33

相关文章 4

编辑推荐

Metrics

本文评价

[1]	孙晶明, 虞盛康, 孙俊. 基于元学习的雷达小样本目标识别方法及改进[J]. 系统工程与电子技术, 2022, 44(6): 1839-1845.
[2]	刘旗, 张新禹, 刘永祥. 基于门控多尺度匹配网络的小样本SAR目标识别[J]. 系统工程与电子技术, 2022, 44(11): 3346-3356.
[3]	李睿峰, 许爱强, 孙伟超, 王树友. 基于元学习的航空电子设备特征选择算法推荐方法[J]. 系统工程与电子技术, 2021, 43(7): 2011-2020.
[4]	晏媛, 孙俊, 孙晶明, 于俊朋. 雷达小样本目标识别方法及应用分析[J]. 系统工程与电子技术, 2021, 43(3): 684-692.