Flora-YOLO: A lightweight framework for real-time rose grading on edge devices via structural optimization and feature refinement
Wenwei Liu, Jinyu Xu, Guoao Wang, Shuo Ding, Yuanbo Zhang, Runze Tian, Zejiang Li, Yuanzhe Ji · 2026
Read Flora-YOLO: A lightweight framework for real-time rose grading on edge devices via structural optimization and feature refinement by Wenwei Liu, Jinyu Xu, Guoao Wang, Shuo Ding, Yuanbo Zhang, Runze Tian, Zejiang Li, Yuanzhe Ji (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/plos-pone-0358742.txt.
Description
Wenwei Liu leads a team of eight shrinking a vision model until it can grade roses on a cheap device at the farm, in PLOS ONE in 2026. A rose's price depends on grade. Human grading is slow and inconsistent. The small computers in packing sheds cannot run heavyweight software. Their framework, Flora-YOLO, rebuilds a compact detector with four changes. The changes sharpen faint petal textures and protect detail under leaf clutter. They amplify weak signals from young buds and respect the irregular shapes of flowers.
Across cross validation on their rose dataset, the model reached accuracy near eighty-seven percent on the standard overlap measure. It beat its baseline while cutting computing load by more than a quarter. It needs under two million parameters, small enough for edge hardware. The work points toward grading lines where a camera, not a tired human eye, makes the call on every stem.
Catalog record
| CREATOR | Wenwei Liu, Jinyu Xu, Guoao Wang, Shuo Ding, Yuanbo Zhang, Runze Tian, Zejiang Li, Yuanzhe Ji |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 94077 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |