Segment Mafic Rocks

• Project started in the discipline "Machine Learning, Theory and Practice in Python" of the postgraduate program PPGE3M-UFRGS;

• It is a routine to train the CNN model for the task of segmenting the minerals that constitute mafic rock in SEM images;

• For this a multi-class segmentation task i used the classic U-Net architecture, designed for cell microscopy image segmentation tasks;

• SEM images were acquired at the University of Leeds during GS Hoerlle's PhD and provided for this work;

• Images were classified manually using the open source software QuPath;

• Today (07/25/2023) this project is in the final stage, changing the data again to improve it and present it at the XII Southern Brazilian Geology Symposium (SSBG)