Code repository for DAS denoising model described in paper "DAS-N2N: Machine learning Distributed Acoustic Sensing (DAS) signal denoising without clean data" (https://doi.org/10.1093/gji/ggad460).
To install package + core dependencies: clone the repository, navigate to the directory and run:
pip install .
It is recommended that you do this in a new environment (e.g., using virtualenv, conda, etc) to avoid package/Python version compatability issues. If you're having issues, check that your Python version is compatible with your PyTorch version (here).
After, to install optional dependencies (e.g., for running example notebooks), run:
pip install dasn2n[optional]
pip install dasn2n[jupyter] # If jupyter lab also required
from dasn2n import DASN2N # Import main model class
model = DASN2N() # Initialise model
model.load_weights() # Load default weights (currently the ones from the paper)
# model = model.to('cuda') # Uncomment if CUDA-enabled GPU available
# model = model.to('mps') # Uncomment if Apple GPU available
data_denoised = model.denoise_numpy(data) # Denoise 2D numpy array (data) containing DAS data
See notebooks in examples directory for more guidance.
If you would like further guidance on training/implementing a DAS-N2N model for your own DAS data, please feel free to get in touch: [email protected]