Construction of a health diagnosis system for rotifers using deep learning

2023 Fiscal Year Final Research Report

• Project/Area Number: 22K14932
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 40030:Aquatic bioproduction science-related
• Research Institution: University of Tsukuba
• Principal Investigator: Ienaga Naoto 筑波大学, システム情報系, 助教 (30899133)
• Project Period (FY): 2022-04-01 – 2024-03-31
• Keywords: 深層学習 / 画像処理 / 物体検出 / 物体追跡 / ワムシ / 種苗生産 / 養殖

Outline of Final Research Achievements

In this study, we first constructed a video dataset of rotifers containing over 30,000 instances. We applied object detection and tracking methods to this dataset and achieved an average accuracy of 83% for the detection of two classes: non-egg-bearing and egg-bearing rotifers.
Next, considering the practical applicability in aquaculture sites, we advanced our validation using rotifer images. We improved the system to perform size measurement simultaneously with detection by using oriented bounding boxes. Although still in the preliminary validation stage, the mAP50, a common evaluation metric for object detection models, is approximately 95%, indicating high detection accuracy.
In the future, we aim to publish these results in scientific journals and work towards the practical implementation of the system, as well as its application to other plankton species.

Free Research Field

機械学習

Academic Significance and Societal Importance of the Research Achievements

本研究で構築した3万以上のインスタンスを含むワムシの動画データセットと，それに適用した物体検出・追跡モデルを，当該領域の研究発展のため公開した。
現在手作業で行われているワムシの測定作業が自動化できれば，時間と労力の大幅な削減につながるだけでなく，誰にでも客観的な測定が可能となる。また本研究成果は，将来的にはワムシの培養自体の自動化や，アルテミアなどワムシ以外の生物餌料への応用にもつながると考えられ，水産業への大きなインパクトが期待される。