MagNet: Automated Magnetic Mineral Grain Morphometry Using Convolutional Neural Network

Morphometry (i.e., the quantitative determination of grain size and shape information) is an essential component of all rock and environmental magnetic studies. Electron microscopy is often used to image magnetic mineral grains, but the current lack of systematic image processing tools makes it challenging to quantify key morphological features of magnetic minerals in natural samples. Here, we present an easy-to-use machine learning framework MagNet for automated morphological recognition of magnetic mineral grains in microscopic images. This framework, based on a convolutional neural network, performs well in the recognition and classification of magnetofossil nanoparticles in transmission electron microscopy images after training and testing. MagNet is open-source and can easily be extended to process different types of mineral images. This tool has the potential, therefore, to extract key quantitative information of magnetic mineral populations within heterogeneous terrestrial and meteoritic samples for the interpretations of Earth and planetary processes.

Automated summary

This study presents a machine learning framework called MagNet for automated recognition and classification of magnetic mineral grains in microscopic images. The framework performs well in identifying and classifying magnetofossil nanoparticles and can be extended to process different types of mineral images. This tool is important for extracting key quantitative information of magnetic mineral populations within diverse samples for interpreting Earth and planetary processes.