Deep Learning for Extracting Dispersion Curves

High‐Resolution Ambient Noise Imaging of Geothermal Reservoir Using 3C Dense Seismic Nodal Array and Ultra‐Short Observation

“…Finally, a task pool with more than 23k dispersion measurements is prepared for dispersion curves picking. We utilize a deep learning model, named DCNet developed by Dai et al (2020), for full automatic dispersion curve picking by regrading dispersion curves extracted as an instance segmentation task. To help the machine to distinguish the target dispersion curves in this work, we set a confidence region based on the reference dispersion curves picked from the bin-stacked virtual-source gathers.…”

Improved High‐Resolution 3D Vs Model of Long Beach, CA: Inversion of Multimodal Dispersion Curves From Ambient Noise of a Dense Array

“…Ambient noise technology (Sabra et al., 2005; Shapiro et al., 2005) uses cross‐correlation to extract the empirical Green's function between station pairs and has become a common tool to obtain underground structures. It is widely used in continental (Bensen et al., 2008; Ritzwoller et al., 2011; Ward et al., 2013; Yang et al., 2007), regional (Behm et al., 2016; Stehly et al., 2009; Wilson et al., 2002), and urban (Dai et al., 2021; Li et al., 2016; Mordret et al., 2013) geophysical studies. Inverting the dispersion curves extracted from ambient noise or earthquake events is used to image the shear wave velocity (Vs) structure.…”

Domain Adaptation in Automatic Picking of Phase Velocity Dispersions Based on Deep Learning

“…For ambient noise tomography, a large number of dispersion curves need to be picked, especially for seismic network like USArray. Traditionally, manual picking of dispersion curves from phase velocity‐frequency diagrams requires considerable time and energy; therefore, several researchers have recently attempted to pick dispersion curves using deep learning (Dai et al., 2021; Dong et al., 2021; Song et al., 2021; Zhang et al., 2020) in an attempt to save time and energy. However, sample‐inadaptability occurs while automatically selecting dispersions from some testing phase velocity‐frequency diagrams, that is, in computer vision, the generalization or performance of a neural network is high for the training phase velocity‐frequency diagrams but low for some testing diagrams (testing set) of other geological areas that differ from the training set.…”