Development of Harmful Bird Recognition System using Object Detection YOLO

In this paper, we discuss the Phalacrocorax carbo recognition system that we have developed. Phalacrocorax carbo is known as harmful bird in Japan. In recent years, many researchers have been studying to eliminate this harmful bird using drone. However, they had to control the drone manually. There was a problem that could not be dealt with without a good pilot. Our research group aims to develop a system that automatically recognizes these harmful birds using drone. One of deep learning, YOLO is used as a recognition method. The recognition rate has been improved by learning the 3D shape of the bird. To create the 3D model, we use an application called Smoothie-3D. Data augmentation was realized by preparing many two-dimensional learning images from the 3D model. As a result of the recognition experiment using Phalacrocorax carbo images, the fit ratio was precise at 98% or more. The precision of the recognition rate also more than 91%. From these results, we have realized the based system for automatic harmful bird elimination system using drone.