SenNet

^Algorithm

SenNet is a dual-branch semantic segmentation network designed for robust wheat senescence evaluation from in-field side-view RGB images. By coupling a semantic stream for global context with a boundary stream guided by edge priors, SenNet delineates key organs and separates green vs. yellow leaves under complex backgrounds. This enables pixel-level quantification of senescence dynamics—such as senescence ratio and curve-derived traits—and supports practical screening of high-yielding varieties across seasons and sites.

SOTA Performance

0.954

mIoU (Overall)

0.934

mIoU (green leaves)

0.934

mIoU (yellow leaves)

0.972

IoU (sky)

0.972

IoU (soil)

0.950

IoU (wheat ear)

Ideal High-yielding Variety Screening

01. Time-series Senescence Segmentation

SenNet generates consistent semantic masks from field side-view RGB time series, separating green/yellow leaves and key organs under complex backgrounds. This enables reliable, pixel-level tracking of senescence progression.

02. GDD-aligned Senescence Curve Fitting

The dynamic traits and parameters of the senescence process include the starting point, ending point, maximum curvature of senescence, minimum curvature, maximum senescence rate, senescence cycle, and cumulative effect (area under the curve).

03. Senescence Dynamics Across Yield Groups

Fitted trajectories are compared across high/mid/low-yield groups to quantify differences in onset, rate, and duration of senescence. These dynamics support yield-related phenotype interpretation and screening.

04. Integrated Screening in Daily Workflows

From segmentation to traits and plot-level summaries, SenNet delivers ready-to-use outputs for routine phenotyping. Dashboards (e.g., heatmaps) help rapidly identify genotypes with favorable senescence patterns.

Citation

If you find this work useful for your research, please consider citing our paper:

@article{yao2025sennet,
  title={SenNet: A dual-branch image semantic segmentation network for wheat senescence evaluation and high-yielding variety screening},
  author={Yao, Jiaqi and Jin, Shichao and Zang, Jingrong and Zhang, Ruinan and Wang, Yu and Su, Yanjun and Guo, Qinghua and Ding, Yanfeng and Jiang, Dong},
  journal={Computers and Electronics in Agriculture},
  volume={237},
  pages={110632},
  year={2025},
  publisher={Elsevier}
}