マスダ　ガク   MASUDA Gaku   益田　岳    所属   医学部 医学科    職種   助教
論文種別	原著
言語種別	英語
査読の有無	査読なし
表題	UAV-Based Mosquito Larval Detection with Minimal Training Data: From Controlled Environments to Turbid Field Conditions
掲載誌名	正式名：Preprints.org
掲載区分	国外
巻・号・頁	pp.1-14
総ページ数	15
国際共著	国際共著
著者・共著者	Gaku Masuda, Masaki Shuzo , Thomson Ngumbira , Dylo Foster Pemba , Hitoshi Kawada
発行年月	2026/03/19
概要	What are the main findings? • A functional mosquito larval detection model was constructed from only 8 UAV-captured training images using a free, no-code cloud AutoML platform, demonstrating that local vector control specialists can build detection models without programming expertise or large annotated datasets. • The model generalised across imaging devices in controlled environments, but Precision collapsed in turbid field conditions, revealing that background diversity in training data—not model architecture or data volume—is the primary barrier to real-world deployment. What are the implications of the main findings? • Free cloud AutoML services lower the barrier for endemic-country practitioners to autonomously build and update larval detection models, making community-owned larval source management technologically feasible. • The characterization of this minimum viable model provides a practical foundation for developing actionable guidelines on constructing minimum yet maximally efficient training datasets, enabling local specialists to achieve field-deployable performance with minimal annotated data.
DOI	10.20944/preprints202603.1543.v1