CubeNet: Array-based Seismic Phase Picking with Deep Learning

A multi-trace seismic phase picking method for both P- and S- arrivals which is particularly suitable for dense array.

By Guoyi Chen @ USTC, Email: billychen@mail.ustc.edu.cn

1. Install (using conda)

conda create -n CubeNet python=3.6
conda activate CubeNet
pip install -r requirements.txt

2. Prediction

import sys; sys.path.append('PATH'); # add the path of CubeNet package if necessary
picker = CubeNet.picker
picker.RegCube(sta) # Put the station locations in, shape: (Station number, 2), unit: km
_,pick_result,fs = picker.pick(data) # Put the raw data in, shape: (TraceID, Components, Sampling pts)

Three demos are prepared to give more details on prediction, which are located in directory: demo/

Demo1: Predict P/S arrival phases with given raw waveforms and station locations

Demo2: Direct Pick P/S arrivals from a given cube

Demo3: Usage of data resampling in CubeNet

3. Training

import sys; sys.path.append('PATH'); # add the path of CubeNet package if necessary
import CubeNet
CubeNet.start_train(CubeNet.network,CubeNet.config.train_info)

Please set the training parameter before training by config.py in the CubeNet package

The demo data cubes for training are located in the directory of npz files: demo/demo_data

The npz file contains four variable: "waveform", "label", "label_mat", "fs"

"waveform" and "label" have a shape of 3 $\times$ 8 $\times$ 8 $\times$ 6000

"label_mat" are marks with a shape of 3 $\times$ 8 $\times$ 8 indicating that traces in the cubes are labeled (marked as True) or not (marked as False)

"fs" is the sampling rate of a cube.

4. Parameter Setup

Some parameters (like GPU acceleration) for prediction and training can be set by config.py in the CubeNet package.

A GPU with more than 4GB of memory is recommended to run CubeNet.

Citation

If you find CubeNet useful, please cite the following reference:

Guoyi Chen, Junlun Li; CubeNet: Array‐Based Seismic Phase Picking with Deep Learning. Seismological Research Letters 2022; doi: https://doi.org/10.1785/0220220147

License

The CubeNet package is distributed under the MIT license (free software).

Billy-Chen0327 / CubeNet