Benchmark on the accuracy and efficiency of several neural network based phase pickers using datasets from China Seismic Network

作     者：Ziye Yu Weitao Wang Yini Chen Ziye Yu;Weitao Wang;Yini Chen

作者机构：Institute of GeophysicsChina Earthquake AdministrationBeijing 100081China Key Laboratory of Earthquake Source PhysicsChina Earthquake AdministrationBeijing 100081China Zhejiang Earthquake AgencyHangzhou 310013China 

出 版 物：《Earthquake Science》 (地震学报（英文版）)

年 卷 期：2023年第36卷第2期

页      面：113-131页

学科分类：0709[理学-地质学] 0819[工学-矿业工程] 07[理学] 070801[理学-固体地球物理学] 0707[理学-海洋科学] 0818[工学-地质资源与地质工程] 0708[理学-地球物理学] 0815[工学-水利工程] 0816[工学-测绘科学与技术] 0813[工学-建筑学] 0825[工学-航空宇航科学与技术] 0704[理学-天文学] 0814[工学-土木工程] 

基　　金：jointly supported by the National Natural Science Foundation of China (No. 42074060) the Special Fund, Institute of Geophysics, China Earthquake Administration (CEA-IGP) (Nos. DQJB19B29, DQJB20B15, and DQJB22Z01) supported by XingHuo Project, CEA (No. XH211103) 

主　　题：neural network deep learning seismic phase picking earthquake detection open-source science 

摘      要：Seismic phase pickers based on deep neural networks have been extensively used recently,demonstrating their advantages on both performance and ***,these pickers are trained with and applied to different data.A comprehensive benchmark based on a single dataset is therefore ***,using the recently released DiTing dataset,we analyzed performances of seven phase pickers with different network structures,the efficiencies are also evaluated using both CPU and GPU *** based on F1-scores reveal that the recurrent neural network(RNN)and EQTransformer exhibit the best performance,likely owing to their large receptive *** performances are observed among PhaseNet(UNet),UNet++,and the lightweight phase picking network(LPPN).However,the LPPN models are the most *** RNN and EQTransformer have similar speeds,which are slower than those of the LPPN and ***++requires the most computational effort among the *** all of the pickers perform well after being trained with a large-scale dataset,users may choose the one suitable for their *** beginners,we provide a tutorial on training and validating the pickers using the DiTing *** also provide two sets of models trained using datasets with both 50 Hz and 100 Hz sampling rates for direct application by *** of our models are open-source and publicly accessible.