Zooplankton Classification Using Hierarchical Attention Branch Network

Koichi Ito, Kanta Miura, Takafumi Aoki, Yurie Otake, Wataru Makino, Jotaro Urabe

Abstract: Plankton is recognized as one of the most important indicators of the health of aquatic ecosystems and water quality. Surveys of plankton populations in oceans and lakes have been conducted manually. Plankton classification methods using deep learning have been developed to automatically classify plankton images. These methods do not sufficiently take into account the bias in the species included in the dataset or the similarity of their shapes. In this paper, we propose a hierarchical attention branch network (H-ABN) to utilize that plankton are hierarchically named according to their taxonomic ranks. We demonstrate the effectiveness of the proposed method through experiments using a zooplankton dataset collected from lakes and ponds in Japan.