YOLOv8 for Zebrafish –Object Detection for an Autonomous Fish Feeder

Sammanfattning

Zebrafish are involved in a wide range of research as a test subject, this is also the case in ongoing research about PFAS conducted at the Man-Technology-Environment (MTM) research centre at Örebro University. The zebrafish are housed in several smaller tanks with a varying number of fish. They require manual feeding every day of the week, thus an autonomous fish feeder with object detection capabilities was proposed to ease the current manual labor. The state-of-the-art object detection model You Only Look Once (YOLOv8m) was chosen and fine-tuned after extensive research in the field of object detection, to accurately detect and count zebrafish. The models were trained using custom datasets to achieve this specific task. The thesis briefly outlines how the fish feeder system is developed and discuss potential integration of the object detection model. The two resulting models YOLO-Zebrafish1 and YOLOZebrafish2, were both analyzed thoroughly and received an Average Precision when evaluated on a test set of 82.1% and 84.1% respectively.