Real-Time Seismic Event Recognition with Semantic Segmentation

Code Implementation of the article "A Framework for Real-Time Volcano-Seismic Event Recognition Based on Multi-Station Seismograms and Semantic Segmentation Models" (under review).

Structure

• utils/: Contains utility functions to perform the proposed framework.
• models/: Includes model architectures. UNet and SwinUNet Implementations are used out-of-the-box from the codes at https://github.com/mateuszbuda/brain-segmentation-pytorch and https://github.com/HuCaoFighting/Swin-Unet, respectively. Pre-trained weights are available at .
• examples/: Example scripts for data exploration, demonstration of the patch stacking procedure, and running the model on pre-segmented windows and continuous data streams.

Setup

Clone this repository:

git clone https://github.com/camilo-espinosa/volcano-seismic-segmentation.git
cd volcano-seismic-segmentation

Install dependencies in requirements.txt

pip install -r requirements.txt

A version of PyTorch, with CUDA compatibility is also necessary to use GPU: https://pytorch.org/get-started/locally/.

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

Usage examples (Notebooks):

from the examples folder:

cd examples

Explore the data:

explore_data.ipynb

Patch Stacking Demonstration:

patch_stacking_demo.ipynb

Window Level Detection Demo:

window_level_demo.ipynb

Continuous Data Detection Demo:

continuous_detection_demo.ipynb

Processing Times:

processing_times.ipynb

Data and weights availability:

Datasets are freely available at:

Pre-trained weights for the four evaluated models (UNet, UNet++, DeepLabV3+ and SwinUNet) are also freely available at: .