Localising Honeybees with Deep-Learning Based Methods

Bio-hybrid systems are a field of study that combines biology, engineering and computer science to create a new, more effective solution to a variety of problems. The RoboRoyale project seeks to combine honeybee colonies and modern digital technology to help these bees thrive in their environment. Man-made changes to this environment, such as global-warming and pests introduced from other areas of the world have had a large impact on the ability of the European Honeybee’s (Apis Melifera) ability to survive without human intervention. This study explores how deep-learning based object detection models can be trained to detect honeybees in low-light environments and how these methods compare to traditional ones. Additionally the deep-learning models are trained using simple images tagged by these traditional methods, in order to prevent manual tagging of data. This was done using images collected for the RoboRoyale project for the purposes of investigating the thermotactic aggregation of honeybees, particularly when a group of bees includes a queen.