MoCoProto: Enhancing few-shot pest image classification with self-supervised representation learning
Dong Jin, Helin Yin, Xianghua Piao, Yeong Hyeon Gu · 2026
Read MoCoProto: Enhancing few-shot pest image classification with self-supervised representation learning by Dong Jin, Helin Yin, Xianghua Piao, Yeong Hyeon Gu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13505369.txt.
Description
Dong Jin, Helin Yin, Xianghua Piao, and Yeong Hyeon Gu tackle pest recognition when only a handful of labelled photos exist for each species. Their method, MoCoProto, first learns visual features from unlabelled images using a self-supervised contrast technique. That prepared network is then plugged into a prototype-based learner that classifies new pests from one or five example shots.
Pretraining used a broad farm-image collection of more than 357,000 pictures covering sixty-five pest species and many plant diseases, with all labels stripped away. Final testing ran on the IP102 insect dataset. The method scored about forty-nine percent accuracy with a single example per class and about seventy percent with five examples, beating the rival approaches tried. The authors argue that strong unlabelled pretraining is the lever that makes few-example pest identification practical for field use.
Catalog record
| CREATOR | Dong Jin, Helin Yin, Xianghua Piao, Yeong Hyeon Gu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55454 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |