Instance Segmentation of Microscopic Foraminifera

Foraminifera are single-celled marine organisms that construct shells remain as fossils in the sediments. Classifying and counting these important paleo-oceanographic -climatological research. However, identification process has been performed manually since 1800s is laborious time-consuming. In this work, we present a deep learning-based instance segmentation model for classifying, detecting, segmenting microscopic foraminifera. Our based on Mask R-CNN architecture, using weight parameters have learned COCO detection dataset. We use fine-tuning approach to adapt novel object dataset of more than 7000 foraminifera sediment grains. The achieves (COCO-style) average precision 0.78 classification task, 0.80 task. When evaluated without challenging grain images, both tasks increases 0.84 0.86, respectively. Prediction results analyzed quantitatively qualitatively discussed. Based our findings propose several directions future work conclude proposed an step towards automating