Recognittion Rate Improvement of Injurious Bird Recognition System by Increasing CNN Learning Image using Data Augmentation

Abstract

In this paper, we discuss injurious bird recognition system that we have developed. Among injurious bird, the damage of Plecoglossus altivelis and Oncorhynchus masou by Phalacrocorax carbo are especially large. In recent years, some researchers have been trying to automatically identify this injurious bird using a surveillance system. However, it was difficult to identify the Phalacrocorax carbo from images including background and other wild birds. Therefore, our research grope examined a method of identification using a convolutional neural network. In order to improve recognition accuracy, learning images were increased by realizing data augmentation of 3 stages. As a result of investigating about this effect, it was able to improve to about 80% of recognition rate.