SeismoCNN – Seismic Event Classification with Deep Learning

About

This repository contains code and data processing tools for the undergraduate research project: Seismic Event Classification Using Convolutional Neural Networks on Multichannel Waveform Data
Read the full report (PDF)

This repository is a modified fork of the official STEAD (STanford EArthquake Dataset) repository. The original contents of the forked repository have been relocated to the INFO/ directory.

The current focus of this project is the classification of seismic waveform data into four categories: Noise, Low, Medium, and High magnitude earthquakes using a Convolutional Neural Network (CNN).

Prerequisites

To fully utilize this repository on a local machine, follow these steps:

1. Follow the setup instructions from the original STEAD repository (see INFO/).
2. Download and include the full STEAD dataset:
   • Ensure the ~91GB merged/ directory containing the full labeled dataset is placed in the root directory of this repo.

---

Execution Workflow

Once the merged/ directory is in place, run the following scripts in order:

1. GenerateTVTSets.py

• Parses data from the merged/ directory.
• Generates PyTorch datasets and index files for Noise, Low, Medium, and High waveform examples used in this study.
• Output: TXT files of dataset splits.

2. CNNGenerator.py

• Trains the CNN model using predefined hyperparameters.
• Training time (on NVIDIA GeForce RTX 3070 Ti): ~1.7 hours per model over 10 epochs.
• Output: Trained weights saved as seismo_cnn_weights.pth.

3. CNNRun.py

• Loads the trained model from seismo_cnn_weights.pth.
• Evaluates performance on the testing set.
• Output: Evaluation results printed to terminal. A copy is saved in FinalEvaluationPrintout.txt.

---

Additional Files and Directories

File/Directory	Description
Tuning.py	Performs hyperparameter tuning on 8 different configurations to optimize model performance. (on NVIDIA GeForce RTX 3070 Ti): ~14.2 Hours
plotting.py	Generates visualizations (accuracy/loss curves) of tuning results. Outputs are saved to Pictures/.
TuningStats/	Stores .npy and .txt files with training/validation statistics from tuning runs.
Pictures/	Contains saved plots of model performance across hyperparameter settings.
GPUAvailable.py	Utility script to verify CUDA/GPU availability for PyTorch acceleration.
TimeKeeper.py	Custom utility to estimate completion times during model training/tuning.
SeismoCNNImport.py	Defines the CNN model architecture and handles STEAD dataset loading.
viewingCenter.py	Utility script to visualize and verify the contents of the parsed datasets.

---

Citation

If you use this project or any of its components in your own work, please cite this repository and the STEAD dataset as follows:

Project Citation

Laureano, N. (2025). Seismic Event Classification Using Convolutional Neural Networks on Multichannel Waveform Data. GitHub repository: https://github.com/NLaureano/SeismicCNN

STEAD Dataset Citation

Mousavi, S. M., Sheng, Y., Zhu, W., & Beroza, G. C. (2019). STanford EArthquake Dataset (STEAD): A Global Data Set of Seismic Signals for AI. IEEE Access, https://doi.org/10.1109/ACCESS.2019.2947848

---

Note: This project is not a peer-reviewed publication but represents undergraduate research publicly shared for reproducibility and educational purposes.