AI drone fusion tracks rice growth stages with far fewer flights
Wang H, Guo W, Mu Y, Zhang Y, Wang H
Crop Improvement
If you've ever wondered how plant breeders keep tabs on thousands of individually-timed crop varieties without walking every row by hand, this is the kind of tool making that possible at scale.
Rice breeders grow hundreds of different varieties side by side, and each one moves through its growth stages, like flowering or ripening, on its own schedule. Tracking all of that by eye is slow, so researchers built a system that blends daily low-detail drone photos with occasional sharp high-detail ones, letting a computer model fill in the gaps. It correctly identified growth stages most of the time while needing far less drone flight time than constant high-detail scanning would require.
Key Findings
The fusion model achieved 87.3% overall accuracy and F1-score with a Kappa coefficient of 0.84 across roughly 500 rice cultivars and over 100,000 images from two growing seasons.
A hybrid sampling strategy (daily medium-resolution plus weekly high-resolution images) cut required weekly drone flight time from 28 hours to about 6 hours without sacrificing accuracy.
The model generalized well across years, maintaining 77.4% overall accuracy and 73.8% F1-score when trained on 2024 data and tested on 2023 data, with maturity-stage recall consistently above 0.96.
chevron_right Technical Summary
Researchers combined frequent drone photos with occasional high-detail drone scans and AI to track exactly what growth stage hundreds of rice varieties are at, cutting the drone flight time needed by breeders from 28 hours a week to about 6 while keeping accuracy high.
Abstract Preview
Original paper
Multi-scale spatial-temporal remote sensing fusion for phenology identification in rice germplasm resources.
Crop phenology is a critical determinant for yield prediction and germplasm evaluation. However, precise phenological monitoring in large-scale rice breeding trials faces significant challenges due...
open_in_new Read full abstractAbstract copyright held by the original publisher.
Species Mentioned
Was this useful?
Want to tell us more? (optional)
Thanks for the note!
Something went wrong — please try again.
Too many submissions. Try again in an hour.
Gene editing removes 97% of celiac-triggering proteins from bread wheat
It could mean that people with celiac disease — roughly 1 in 100 worldwide — may one day safely eat bread made from real wheat, without sacrificing the taste...
Rice is a cereal grain and in its domesticated form is the staple food of over half of the world's population, particularly in Asia and Africa. Rice is the seed of the grass species Oryza sativa —or, much less commonly, Oryza glaberrima. Asian rice was domesticated in China some 13,500 to 8,200 y...