Pytorch library for seismic data augmentation
pip install --upgrade git+https://github.com/IMGW-univie/seismic-augmentation.git
import torch
from seismic_augmentation.composition import Compose
from seismic_augmentation.augmentations import *
aug = Compose([
FlipChannels(init_channel_order='ZNE'),
AddRandomNoise(snr_level_db=-10),
RandomLowPassFilter(cutoff_freq_range=[1,10]),
RandomHighPassFilter(cutoff_freq_range=[3,14]),
Taper(max_percentage=0.5, max_length=10),
PolarityChange(),
Normalize()
],
p=0.5)
transformed = aug(data=waveform, sample_rate=30)
Contributors welcome!
For now this library is very simple
FlipChannels(init_channel_order='ZNE')
'''
Swaps N and E channels. Easiest way to change azimuth of a signal
init_channel_order - ordering of the channels of your seismic data
'''
AddRandomNoise(snr_level_db=-10)
'''
Adds random noise with desired SNR
snr_level_db - desired signal to noise ratio after augmentation
'''
RandomLowPassFilter(cutoff_freq_range=[1,10])
'''
Applies Low Pass Filter with a random cutoff frequency
cutoff_freq_range - range of possible cutoff frequencies
'''
RandomHighPassFilter(cutoff_freq_range=[1,10])
'''
Applies High Pass Filter with a random cutoff frequency
cutoff_freq_range - range of possible cutoff frequencies
'''
LowPassFilter(cutoff_freq=9.)
'''
Applies Low Pass Filter with a desired cutoff frequency
cutoff_freq - desired cutoff frequency
'''
HighPassFilter(cutoff_freq=9.)
'''
Applies High Pass Filter with a desired cutoff frequency
cutoff_freq - desired cutoff frequency
'''
Taper(max_percentage=0.5, max_length=10)
'''
Applies a taper with specified parameters
max_percentage - how strongly the signal is suppresed
max_length - maximum length of a taper in samples
'''
PolarityChange()
'''
Flips polarity of the signal
'''
Normalize()
'''
Global normalization of 3-channel signal
'''
RotateWaveforms(angle_range=[0, 360])
'''
Rotates waveform to a specified angle
angle_range - range of angles in degrees
'''
p
- probability that an augmentation would be applied
Highly inspired by Facebook Augly