Skip to content

Latest commit

 

History

History
32 lines (23 loc) · 2.65 KB

README.md

File metadata and controls

32 lines (23 loc) · 2.65 KB
Raw

ARRU phase picker: Attention-Recurrent-Residual-U-Net-for-earthquake-detection

We're working on a more stable model on processing continuous seismograms as well as an useful repository!
Here are just the simple scripts for model training and prediction using STandford Earthquake Dataset (STEAD) dataset.

march_MWA.mp4

Script piplines

Below describes the workflow from data generation, model training, making predictions, to model evaluation.

  1. Prepare the seismic recordings from STEAD data : P01_make_stream_STEAD.py
    This script simply generates sac files as well as TFRecord in length of 20 seconds. This would require the STEAD dataset , please download and place the file 'merge.hdf5' (you could retreive this entire STEAD dataset here in the directory './data'.
    You can change the variable 'csv_type' in line 22 with [train, test, val] to generate dataset we used in our study according to the files stored in './data/partition_csv'. Noted that you have to make './data/partition_csv/train_STEAD.csv' by yourselves according to uploaded lists of STEAD dataset for data partition, if needed.

    Output directory: (1) './data/sac_data_STEAD_20s' (2) './data/input_TFRecord_STEAD_20s'

  2. Model training:
    In this repository we provide two pretrained models,
    ./pretrained_model/paper_model_ARRU_20s and ./pretrained_model/multitask_ARRU_20s

    The former works as seismic phase picker described in our paper, and the latter one provides an additional mask associating the P and S arrivals, which could also be treated as the earthquake event detector. Both of these models were trained with local earthquake events that the maximum separation between P and S arrival is 12 seconds.

    ./P02_train_codes/P01_Unet_train_gpus_STEAD.py
    ./P02_train_codes/P01_Unet_train_detect_gpus_STEAD.py

  3. Making predictions on STEAD dataset

  4. Model evaluation

  5. Quick example of making predictions using pretrained model
    $ python quick_ex.py

Binder link

Binder

Reference

Wu‐Yu Liao, En‐Jui Lee, Dawei Mu, Po Chen, Ruey‐Juin Rau; ARRU Phase Picker: Attention Recurrent‐Residual U‐Net for Picking Seismic P‐ and S‐Phase Arrivals. Seismological Research Letters 2021; doi: https://doi.org/10.1785/0220200382