IDENTIFICATION OF ENVIRONMENTALLY RELEVANT BENTHIC FORAMINIFERA FROM THE SKAGERRAK FJORDS BY DEEP LEARNING IMAGE MODELING

Abstract

Over the several past decades, there has been increasing interest in using foraminifera as environmental indicators for coastal marine environments. As compared to macrofauna, which are currently used in environmental studies, foraminifera offer several distinct advantages as bioindicators, including short generation times, a high number of individuals per small sample volume, hard and durable tests with high preservation potential, and low cost of sample extraction. One of the main problems with foraminifera identification is reliance on manual identification and expert judgement, which is a tedious and slow process prone to errors and subjectivity. Deep learning, a subfield of machine learning, has emerged as a promising solution to this challenge, since a neural network can learn to recognize subtle differences in shell morphology that may be difficult for the human eye to distinguish. Benthic foraminifera mounted on microslides from several Skagerrak fjords including Gullmar Fjord, Hakefjord, and Idefjord were imaged using a Nikon SMZ-10 stereomicroscope and DeltaPix DP450 microscope camera. Images were then processed in Roboflow API, where individual foraminifera were labelled and classified. This resulted in 3003 images and 22 138 labelled individuals. Using the labeled images, a dataset was created to be used for deep learning training. YOLO (You Only Look Once) v7 model implemented in the PyTorch framework was used in this work, which has demonstrated state-of-the-art speed and performance for object detection as of the time of writing. Models were trained using a Nvidia RTX A4000 GPU (graphical processing unit). The models are able to distinguish 29 species, while preliminary results show a 90.3% mAP (mean average precision) and 78.8% mAP on the best and the worst performing models, respectively. Even though the imaging and labelling was done in a short amount of time, the results look promising and show that even a relatively small dataset can be used for training a reliable deep learning species identification model.