Repository for Characterizing Marine Magnetic Anomalies

This repository contains the codes and data associated with the manuscript “Characterizing Marine Magnetic Anomalies: A Machine Learning Approach to Advancing the Understanding of Oceanic Crust Formation”.

Repository Structure

• CVAETraining.ipynb

Code for training the magnetic anomaly generator, implemented using a Conditional Variational Autoencoder (CVAE). This generator creates realistic magnetic data based on specified linearity and orientation parameters.

• CVAELoad.ipynb

Code for using the trained CVAE-based generator to produce synthetic magnetic data with user-defined linearity and orientation.

• CNNTraining.ipynb

Code for training the characterization network (a Convolutional Neural Network, CNN) to determine the linearity and orientation of magnetic anomalies. The training data is generated using the trained CVAE.

Folders:

• CVAETrainingData/

Contains the training data for the CVAE, derived from the EMAG2v3 database with manually-labeled linearity and algorithmically derived orientation

• CNNTrainingData/

Contains the training data for the CNN, derived from the data generated by the CVAE.

• NetSave/

Stores the two well-trained networks:

        • CVAE (for magnetic anomaly generation)

        • CNN (for characterizing linearity and orientation)

• OriginalMagData/

Contains the original magnetic data for seven study areas, extracted from the EMAG2v3 database.