A Python package for anomaly detection in distributed acoustic sensing (DAS) datasets using an autoencoder-based deep learning algorithm.
- Python >= 3.10
pip
It would be great to use a virtual environment or a fresh conda environment for better dependency management.
To install the package in editable mode with the required dependencies, run the following when you are in the das-anomaly directory:
pip install -e .To install the package in editable mode with all optional dependencies, run:
pip install -e '.[all]'The main steps are as follows:
- Using the plot_psd scripts, create power spectral density (PSD) plots in RGB format. We average the energy over a desired time window and stack all channels together to create a PSD with channels on the X-axis and frequency on the Y-axis. We create PSD of normal images (images without any anomaly or seismic event) and known seismic events. We can use MPI to distribute plotting PSDs over CPUs.
- Using the train_model scripts, randomly select train and test PSD images and train the model on normal PSD images.
- Using the validate_and_plot_density jupyter notebook, validate the trained model and find an appropriate density score as a threshold for anomaly detection.
- Using the detect_anomalies scripts, detect anomalies in PSD images with the trained model and write their information.
- Using the count_anomalies scripts, count the number of detected anomalies.
Optional:
Ahmad Tourei, Colorado School of Mines tourei@mines.edu | ahmadtourei@gmail.com