Multi-scale sensing system for crop stress detection

On the ground

Our lab is developing a ground-based imaging platform built around a robotic arm integrated with a 3D depth camera for target localization and a high-definition webcam mounted on the arm tip for multi-view RGB imaging of plant stems and leaves.
Results coming soon: we will test this system this summer (2026) to distinguish winter wheat leaf rust and stem rust diseases.

In the air

Unsupervised pre-training enhances deep learning model for crop stress detection
Our proposed unsupervised pre-training strategy mitigated the requirement of intensive training data for deep learning models. Leveraging unlabeled UAV-collected RGB images, this approach significantly enhanced performance in rating the severity of soybean iron deficiency chlorosis, surpassing models without pre-training.

Relevant Publications

Li, J., Oswald, C., Graef, G. L., & Shi, Y. (2020). Improving model robustness for soybean iron deficiency chlorosis rating by unsupervised pre-training on unmanned aircraft system derived images. Computers and electronics in agriculture, 175, 105557. paper link