Automatic Tracking of Marine Life by Multirotor Type UAV Using Image Recognition and Deep Neural Network

Japan has been facing serious challenges in its fishing industry, including a decline in the number of fishermen and an aging workforce, coupled with the depletion of marine resources. In response, this paper aims to address these concerns by developing an object recognition AI and UAV combined system to survey marine life, specifically focusing on “Risso’s dolphin”, aiming for increased automation and efficiency. The developed AI-based tracking system achieved an score of 0.971, indicating a satisfactory level of accuracy. Furthermore, the system was successfully deployed on an edge device, achieving real-time tracking with a frame processing time of 116ms per frame, ensuring sufficient real-time capabilities for tracking marine organisms.