Earthquake and explosion identification based on Deep Learning residual network model
-
摘要: 为加强对地震台网记录的天然地震与人工爆破事件进行准确的性质识别，本文基于深度学习技术中的残差网络模型，提出了一种新的爆破识别方法，并根据北京数字遥测地震台网及国家数字测震台网中心记录的波形数据及其发布的包含事件性质的地震报告，选取河北三河采石场的93次爆破事件和54次周边地震事件的波形功率谱，分别采用不同的训练样本比例进行了100次和1 000次独立的随机抽样子试验以及 “留一交叉验证法” 试验，对人工爆破与天然地震进行了识别研究。试验结果表明，深度学习残差网络模型在天然地震与爆破事件的性质识别中具有很高的识别率且效果稳定，具有较好的应用前景。Abstract: In order to enhance the property identification of earthquakes and explosions recorded by seismic network, this paper proposed a new technology of explosion discrimination based on the residual network model in Deep Learning technology, and utilized it to identifying explosion and surrounding earthquakes in Sanhe Quarry of Hebei Province. According to the waveform data recorded by the Beijing Digital Telemetry Seismic Network and China Center of Digital Seismic Network, and the released seismic phase reports, we analyzed the waveform power spectrum of 93 explosion events and 54 surrounding seismic events in Sanhe Quarry of Hebei Province. Moreover, 100 independent random sampling sub-tests, 1 000 independent random sampling sub-tests and leave-one-out-cross-validation test were conducted by adopting different training sample proportions, respectively. The test results show that the residual network model in Deep Learning has a high recognition rate and a stable effect in identifying the property of earthquakes and explosions, therefore it has a wonderfully potential application.
-
-
![]()
图 2 本文应用的残差网络模型
图中省略了4层卷积核为64和128的卷积层，加上全连接层及分类器层，共计14层，图中上方数字2×2为卷积核大小，下方数字为卷积核的数量
Figure 2. The residual network model applied in this paper
The four convolutional layers with 64 and 128 convolution kernels was omitted；coupled with the fully connected layer and the classifier layer，there are 14 layers in total。2×2 on the upper in the figures is the size of the convolution kernel，and on the bottom is the number of the convolution kernel
![]()
图 4 波形筛选示意图
（a） 原始波形；（b） 带通滤波去噪结果；（c） TKEO短长时窗比
Figure 4. Schematic diagram of waveform selection
（a） Original waveform；（b） De-noising result by band-pass filtering；（c） TKEO short-length time-to-window ratio
![]()
图 5 MS2.0天然地震（a）和人工爆破（b）经1—25 Hz带通滤波的波形及功率谱
Figure 5. Waveforms with 1−25 Hz bandpass filter and power spectra of MS2.0 earthquake （a） and explosion （b）
![]()
图 6 基于台站记录的试验结果
图（a）和（b）为第一组试验中的100次随机子试验和1 000次随机子试验，每次子试验从地震和爆破中分别抽取20次事件进行训练，余下的事件作为预测；图（c）和（d）分别表示第二组试验中100次随机子试验和1 000次随机子试验，每次子试验从地震和爆破中分别抽取 27和46次事件进行训练，余下的事件作为预测
Figure 6. Test results based on station records
Figs. （a） and （b） are the results for the 100 random subtests and 1 000 random subtests in the first group， each subtest selects 20 events from earthquakes and explosions for training，and the remaining events are used as predictions；Figs. （c） and （d） are the results for 100 random subtests and 1 000 random subtests in the second group，each subtest selects 27 and 46 events from earthquakes and explosions for training，and the remaining events are used as predictions
![]()
图 7 基于事件的100次随机子试验结果
图（a）为第一组试验中的100次随机子试验，每次的子试验从地震和爆破中分别抽取20次事件进行训练，余下的事件作为预测；图（b）为第二组试验中的100次随机子试验，每次子试验从地震和爆破中分别抽取27和46次事件进行训练，余下的事件作为预测。图中红、蓝线条分别代表基于50%，60%台站记录被正确判别为识别标准（基于事件）的识别率
Figure 7. Results of 100 random subtests based on the events
Figs. （a） and （b） represent 100 random subtests in the first group and the second group，respectively. Fig. （a） indicates that each subtest takes 20 events from the blasting and earthquakes for training，and the remaining events are used as predictions；Fig. （b） indicates that each subtest takes 46 and 27 events from the blasting and earthquakes for training，and the remaining events are used as predictions.
![]()
表 1 各卷积层输入及输出的通道数量
Table 1 The number of channels for input and output of each convolution layer
卷积层 输入通道数 输出通道数 卷积层 输入通道数 输出通道数 卷积层 输入通道数 输出通道数 1 1 8 6 32 64 11 256 256 2 8 16 7 64 64 12 256 512 3 16 16 8 64 128 13 512 512 4 16 32 9 128 128 14 全连接 5 32 32 10 128 256 
下载: 导出CSV
表 2 基于事件的3组试验结果对比
Table 2 Comparison of the three groups of event-based test results
试验组 随机子
试验次数基于50%记录被正确
判别的平均识别率基于50%记录被正确
判别的地震识别率基于50%记录被正确
判别的爆破识别率平均值 最大值 最小值 平均值 最大值 最小值 平均值 最大值 最小值 第一组 100 97.7% 100% 93.5% 99.4% 100% 97.1% 96.9% 100% 91.8% 1 000 97.6% 100% 92.5% 99.3% 100% 97.1% 96.8% 100% 90.4% 第二组 100 98.4% 100% 95.9% 98.7% 100% 96.3% 98.3% 100% 93.6% 1 000 98.3% 100% 94.6% 98.8% 100% 90.5% 98.1% 100% 93.6% 留一交叉验证 97.3% 96.3% 97.8%
下载: 导出CSV
-
郝春月,郑重,张爽. 2012. 玉树地震前后当地的噪声变化研究[J]. 地球物理学进展,27(6):2418–2428. doi: 10.6038/j.issn.1004-2903.2012.06.016 Hao C Y,Zheng Z,Zhang S. 2012. Research about noise variation around YUS station before and after the Yushu earthquake[J]. Progress in Geophysics,27(6):2418–2428 (in Chinese).
刘晗,张建中. 2014. 微震信号自动检测的STA/LTA算法及其改进分析[J]. 地球物理学进展,29(4):1708–1714. doi: 10.6038/pg20140429 Liu H,Zhang J Z. 2014. STA/LTA algorithm analysis and improvement of microseismic signal automatic detection[J]. Progress in Geophysics,29(4):1708–1714 (in Chinese).
山世光,阚美娜,刘昕,刘梦怡,邬书哲. 2016. 深度学习:多层神经网络的复兴与变革[J]. 科技导报,34(14):60–70. Shan S G,Kan M N,Liu X,Liu M Y,Wu S Z. 2016. Deep Learning:The revival and transformation of multi layer neural networks[J]. Science&Technology Review,34(14):60–70 (in Chinese).
孙志军,薛磊,许阳明,王正. 2012. 深度学习研究综述[J]. 计算机应用研究,29(8):2806–2810. doi: 10.3969/j.issn.1001-3695.2012.08.002 Sun Z J,Xue L,Xu Y M,Wang Z. 2012. Overview of Deep Learning[J]. Application Research of Computers,29(8):2806–2810 (in Chinese).
王婷婷,边银菊,张博. 2013. 地震和爆破的综合识别方法研究[J]. 地球物理学进展,28(5):2433–2443. doi: 10.6038/pg20130522 Wang T T,Bian Y J,Zhang B. 2013. The comprehensive identification methods between earthquakes and explosions[J]. Progress in Geophysics,28(5):2433–2443 (in Chinese).
杨宏,贾维敏. 2000. 基于神经网络的综合评判在核爆模式识别中的应用[J]. 核电子学与探测技术,20(4):279–283. doi: 10.3969/j.issn.0258-0934.2000.04.009 Yang H,Jia W M. 2000. Recognition of underground nuclear explosion and natural earthquake based on neural network[J]. Nuclear Electronics&Detection Technology,20(4):279–283 (in Chinese).
AitLaasri E H,Akhouayri E S,Agliz D,Atmani A. 2013. Seismic signal classification using multi-layer perceptron neural network[J]. Int J Comput Appl,79(15):35–43.
Baer M,Kradolfer U. 1987. An automatic phase picker for local and teleseismic events[J]. Bull Seismol Soc Am,77(4):1437–1445.
Bengio Y,Simard P,Frasconi P. 2002. Learning long-term dependencies with gradient descent is difficult[J]. IEEE Trans Neural Netw,5(2):157–166.
Bennett T J,Murphy J R. 1986. Analysis of seismic discrimination capabilities using regional data from western United States events[J]. Bull Seismol Soc Am,76(4):1069–1086.
Dahl G E, Yu D, Deng L, Acero A. 2011. Large vocabulary continuous speech recognition with context-dependent DBN-HMMS[C]//IEEE International Conference on Acoustics Speech and Signal Processing. Prague: IEEE: 4688−4691.
Esposito A M,Giudicepietro F,Scarpetta S,D’Auria L,Marinaro M,Martini,M. 2006. Automatic discrimination among landslide,explosion-quake,and microtremor seismic signals at Stromboli Volcano using neural networks[J]. Bull Seismol Soc Am,96(4A):1230–1240. doi: 10.1785/0120050097
Falsaperla S,Graziani S,Nunnari G,Spampinato S. 1996. Automatic classification of volcanic earthquakes by using multi-layered neural networks[J]. Nat Hazards,13(3):205–228.
He K M, Zhang X Y, Ren S Q, Sun J. 2016. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE: 770−778.
Hong T K. 2013. Seismic discrimination of the 2009 North Korean nuclear explosion based on regional source spectra[J]. J Seismol,17(2):753–769. doi: 10.1007/s10950-012-9352-1
Kingma D P, Ba J. 2015. Adam: A method for stochastic optimization[C]//Proceedings of the 3rd International Conference for Learning Representations. San Diego: The Institute for Catastrophic Loss Redu Ction.
Kohavi R. 1995. A study of cross-validation and bootstrap for accuracy estimation and model selection[C]//Proceedings of the 14th International Joint Conference on Artificial Intelligence. Montreal: Morgan Kaufmann Publishers Inc, 14: 1137−1143.
Kortström J,Uski M,Tiira T. 2016. Automatic classification of seismic events within a regional seismograph network[J]. Comput Geosci,87:22–30. doi: 10.1016/j.cageo.2015.11.006
Krizhevsky A, Sutskever I, Hinton G E. 2012. ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems. Lake Tahoe: Curran Associates Inc, 1097−1105.
Kuyuk H S,Yildirim E,Dogan E,Horasan G. 2011. An unsupervised learning algorithm:Application to the discrimination of seismic events and quarry blasts in the vicinity of Istanbul[J]. Nat Hazards Earth Syst Sci,11(1):93–100. doi: 10.5194/nhess-11-93-2011
Lee H, Largman Y, Pham P, Ng A Y. 2009. Unsupervised feature learning for audio classification using convolutional deep belief networks[C]//Proceedings of the 22nd International Conference on Neural Information Processing Systems. Vancouver: Curran Associates Inc: 1096−1104.
Rabin N,Bregman Y,Lindenbaum O,Ben-Horin Y,Averbuch A. 2016. Earthquake-explosion discrimination using diffusion maps[J]. Geophys J Int,207(3):1484–1492. doi: 10.1093/gji/ggw348
Ren S Q,He K M,Girshick R,Sun J. 2015. Faster R-CNN:Towards real-time object detection with region proposal networks[J]. IEEE Trans Pattern Anal Mach Intell,39(6):1137–1149.
Reynen A,Audet P. 2017. Supervised machine learning on a network scale:Application to seismic event classification and detection[J]. Geophys J Int,210(3):1394–1409. doi: 10.1093/gji/ggx238
Riggelsen C,Ohrnberger M. 2014. A machine learning approach for improving the detection capabilities at 3C seismic stations[J]. Pure Appl Geophys,171(3/4/5):395–411. doi: 10.1007/s00024-012-0592-3
Simonyan K, Zisserman A. 2015. Very deep convolutional networks for large-scale image recognition[C]//Proceedings of the 3rd International Conference for Learning Representations. San Diego: The Institute for Catastrophic Loss Redu Ction.
Stevenson P R. 1976. Microearthquakes at Flathead Lake,Montana:A study using automatic earthquake processing[J]. Bull Seismol Soc Am,66(1):61–80.
Stump B W,Hedlin M A H,Pearson D C,Hsu V. 2002. Characterization of mining explosions at regional distances[J]. Rev Geophys,40(4):2–1.
Szegedy C, Liu W, Jia Y Q, Sermanet P, Reed S, Anguelov D, Erhan D, Vanhoucke V, Rabinovich A. 2015. Going deeper with convolutions[C]//IEEE Conference on Computer Vision and Pattern Recognition. Boston: IEEE: 1−9.
Tiira T. 1995. Discrimination of nuclear explosions and earthquakes from teleseismic distances with a local network of short period seismic stations using artificial neural networks[J]. Phys Earth Planet Inter,97(1/2/3/4):247–268.
Vallejos J A,McKinnon S D. 2013. Logistic regression and neural network classification of seismic records[J]. Int J Rock Mech Min Sci,62:86–95. doi: 10.1016/j.ijrmms.2013.04.005
Wiemer S. 2000. Mapping and removing quarry blast events from seismicity catalogs[J]. Bull Seismol Soc Am,90(2):525–530. doi: 10.1785/0119990104
计量
- 文章访问数: 1725
- HTML全文浏览量: 981
- PDF下载量: 215
