AI pipeline pinpoints single-rice-panicle anthesis start and peak in the field

A new field-ready computer-vision pipeline can pinpoint when individual rice panicles begin and peak flowering, a brief and highly heat-sensitive window that strongly influences grain set and yield. The system, called OPAL-Flow, combines orientation-aware panicle detection and tracking, pose normalization, lightweight super-resolution, and temporal event decoding to produce start and peak anthesis timestamps for the same tracked panicle. It also links these panicle-level times to same-day weather, enabling variety comparisons and tests of how photosynthetically active radiation (PAR), temperature, relative humidity (RH), and rainfall shift within-day flowering rhythms. The work aims to replace labor-intensive visual scoring with scalable, minute-scale phenotyping for breeding and crop management.

Rice anthesis occurs within a short daily window, and its start and peak can shift rapidly with microclimate, so minute-scale timing is essential for linking flowering events to heat or humidity exposure. Yet large-scale field breeding still relies mainly on manual visual inspection, which is labor-intensive, subjective, and difficult to synchronize across plots. Existing image-based methods often classify frames or patches and then aggregate predictions at plot or image level, rather than producing explicit start and peak timestamps for the same panicle. Dense, tilted canopies, occlusion, variable illumination, and pose changes further destabilize detection, tracking, and annotation. Given these challenges, there is a need for in-depth research on robust, geometry-aware, temporally precise single-panicle anthesis phenotyping under real field conditions.

Researchers at Nanjing Agricultural University and colleagues developed OPAL-Flow to address this gap. The study was published (DOI: 10.1016/j.plaphe.2026.100216) in Plant Phenomics (2026, article 100216). The pipeline uses YOLO (You Only Look Once)-SnakePanNet for oriented panicle detection and tracking, Real-ESRGAN (enhanced super-resolution generative adversarial network) for super-resolution, and PanicleTimeMAE (a panicle temporal masked-autoencoder model) for temporal event pinpointing. Across two rice-growing seasons, it estimated single-panicle anthesis start and peak times and related them to same-day meteorological conditions across 100 varieties.

The technical core has three parts. YOLO-SnakePanNet replaces standard convolutions with Dynamic Snake Convolution (DySnakeConv) to follow slender, curved panicles and adds LSCD_LQE (lightweight shared convolutional detection head with localization quality estimation) to regress oriented bounding boxes (OBBs). In ablation tests, it reached 94.4% mean average precision at intersection-over-union 0.5 (mAP@50), 3.6 percentage points above the YOLOv11 baseline, while reducing computation by about 0.7 GFLOPs (giga floating-point operations). BoT-SORT (Bag of Tricks for SORT) tracking achieved 0.854 multi-object tracking accuracy (MOTA), 0.932 identity F1 (IDF1), and 202 identity switches across 17,672 ground-truth panicle instances. PanicleTimeMAE uses a VideoMAEv2 (video masked autoencoder version 2) encoder, a Pyramid Dilated Temporal Convolutional Network (TCN), and a Fast Adaptive Savitzky-Golay (FASG) smoothing gate to produce start and peak probability curves. On 5-minute 2024 images, it achieved an average mean absolute error (MAE) of 1.35 frames (6.75 minutes), with 0.85 and 0.90 accuracy within ±1 and ±2 frames; on independent 10-minute 2023 images, average MAE was 1.31 frames. It outperformed TimeSformer and ResNet18+BiLSTM baselines. Environmentally, higher photosynthetically active radiation (PAR) advanced anthesis, while higher relative humidity (RH) and rainfall delayed it; meteorological blocks explained about 64% of variance, variety about 33.5%, and year about 3%.

The authors said the main advance is decoupling panicle geometry from temporal event decoding, so the same tracked panicle receives explicit start and peak timestamps instead of coarse plot-level labels. They said this makes it possible to test how minute-scale weather shifts flowering rhythms, while requiring only sparse sequence-level annotations rather than dense pixel labels. They said performance remained stable across 5-minute and 10-minute image cadences, although extreme sunlight can still cause larger timing errors. They added that the approach should transfer to other rapid organ-level phenological events.

OPAL-Flow could help breeders screen varieties with earlier or more heat- and humidity-resilient anthesis windows, and help agronomists time irrigation, shading, or other interventions to the hours when a variety typically flowers. By providing comparable start and peak times across varieties, dates, and environments, it also supports ecophysiological studies linking microclimate to flowering dynamics. Future work will validate the pipeline across more sites, years, and canopy structures, and integrate thermal or in-canopy sensors to measure panicle-surface temperature and boundary-layer humidity.

###

References

DOI

10.1016/j.plaphe.2026.100216

Original Source URL

https://doi.org/10.1016/j.plaphe.2026.100216

Funding information

The research was supported by the “JBGS” Project of Seed Industry Revitalization in Jiangsu Province (JBGS [2021] 007).

About Plant Phenomics

Plant Phenomics publishes breakthrough research that integrates genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics and computer science to advance plant phenotyping—from the cellular level to entire populations. The journal tackles key scientific challenges in phenomics by showcasing cutting‑edge technologies for data acquisition, management, interpretation and modelling. Its scope spans practical applications across all plant sciences, aiming to translate novel phenotyping tools into real‑world solutions for breeding, crop management and fundamental plant biology. By fostering interdisciplinary innovation, Plant Phenomics serves as a hub for researchers developing high‑throughput, accurate and scalable methods to decipher complex plant traits, ultimately accelerating progress in agriculture and plant science.