Proceedings of the 7th International Conference on Biological Science (ICBS 2021)

Application of Deep (Machine) Learning for Phytoplankton Identification Using Microscopy Images

Authors
Arief Rachman1, *, Aulia Salsabella Suwarno2, Susanna Nurdjaman3
1Biological Oceanography Research Group, Research Center for Oceanography, National Research and Innovation Agency (RCO-BRIN), Indonesia
2Faculty of Earth Science and Technology, Bandung Institute of Technology (ITB), Indonesia
3Department of Oceanography, Faculty of Earth Sciences and Technology, Bandung Institute of Technology (ITB), Indonesia
*Corresponding author. Email: arief_rachman1987@yahoo.com
Corresponding Author
Arief Rachman
Available Online 2 May 2022.
DOI
10.2991/absr.k.220406.032How to use a DOI?
Keywords
Computer vision; CNNs; Phytoplankton diversity; Taxonomic identification; VGG-16
Abstract

As a hot spot of marine diversity, between 150 – 400 phytoplankton species have been reported in various Indonesian marine ecosystems. However, phytoplankton identification in Indonesia is mainly made manually by a human expert, which is a time-consuming process with many limitations. Thus, this study aimed to develop automatic phytoplankton identification using Deep Machine Learning algorithms, such as Convolutional Neural Networks (CNNs), to help the identification process of the Indonesian phytoplankton. A pre-trained VGG-16 model was used to build a CNN model to identify phytoplankton up to genus level under five different model scenarios (S) based on curated phytoplankton images from the Plankton Image Database of RCO-BRIN. The cross-entropy loss analysis and confusion matrix showed the simple model (S1) and genus-level model (S4) have the best performance with low classification errors. In the application trial, the S1 model could differentiate diatoms and dinoflagellates group with up to 78% accuracy, while the S4 model could differentiate the target genus of Ceratium, Chaetoceros, Coscinodiscus, Protoperidinium, and Rhizosolenia up to 79% accuracy. However, the S4 model suffers from forced classification problems due to its inability to identify images of any non-target genus. Unfortunately, the S5 model created to solve the S4 problems has a much lower accuracy at 54% due to highly diverse data stored in the ‘Others’ category, which confuses the model. Although the CNNs models in this study can automatically identify phytoplankton up to genus level at accuracy >75%, the current limitations in all scenarios need to be solved before the model can be used in a real-world research scenario.

Copyright
© 2022 The Authors. Published by Atlantis Press International B.V.
Open Access
This is an open access article distributed under the CC BY-NC 4.0 license.

Download article (PDF)

Volume Title
Proceedings of the 7th International Conference on Biological Science (ICBS 2021)
Series
Advances in Biological Sciences Research
Publication Date
2 May 2022
ISBN
978-94-6239-573-2
ISSN
2468-5747
DOI
10.2991/absr.k.220406.032How to use a DOI?
Copyright
© 2022 The Authors. Published by Atlantis Press International B.V.
Open Access
This is an open access article distributed under the CC BY-NC 4.0 license.

Cite this article

TY  - CONF
AU  - Arief Rachman
AU  - Aulia Salsabella Suwarno
AU  - Susanna Nurdjaman
PY  - 2022
DA  - 2022/05/02
TI  - Application of Deep (Machine) Learning for Phytoplankton Identification Using Microscopy Images
BT  - Proceedings of the 7th International Conference on Biological Science (ICBS 2021)
PB  - Atlantis Press
SP  - 213
EP  - 224
SN  - 2468-5747
UR  - https://doi.org/10.2991/absr.k.220406.032
DO  - 10.2991/absr.k.220406.032
ID  - Rachman2022
ER  -