A Novel Network Architecture for Microplankton Classification in Digital Holographic Images

Abstract

Planktons are the building blocks of marine food webs and key indicators of ocean health. Monitoring of plankton populations help study the biological diversity of microbial eukaryotes. Recent years have witnessed the wide usage of digital holographic microscopes (DHM) for in situ detection of underwater microplanktons. Holography has an edge over other imaging techniques due to its unique ability to provide a 3D hologram of the microplankton without disturbing its orientations. In this paper, a novel network architecture with 5.29 GFLOPs is developed for the classification of microplanktons in digital holographic images. The proposed method achieved a class-wise F1-scores above 80 % at a lower computational cost. The proposal provided competitive performance with respect to six baseline network architectures. This technique has the potential to be appealing for future applications of in situ classification of microplanktons.