Drone hyperspectral sensing for wheat scab disease phenotyping and breeding scale-up

Overview

This solution applies drone-mounted hyperspectral imaging to detect wheat scab (Fusarium head blight, FHB) symptoms at field scale, enabling breeders and crop researchers to phenotype disease response across 1-40 acres per flight. By combining narrow-band spectral feature selection with deep learning models, the platform targets specific wavebands that carry the most diagnostic information, supporting earlier and more accurate disease detection than visual scouting or conventional RGB imagery. The approach is designed to accelerate wheat breeding for scab resistance by delivering high-throughput, objective, and repeatable disease severity measurements across large breeding nurseries.

Technical specifications

• Drone hyperspectral imaging: Aerial sensor platform capturing both high spectral and high spatial resolution data across field-scale breeding plots.
• Ensemble feature selection: Algorithm that distills hundreds of spectral bands down to a handful of the most informative wavebands, reducing data complexity while preserving diagnostic power.
• Deep learning detection models: Includes Yolo-family object detection (Yolo, Yolo 9000, Yolo v3, Yolo v4), Deep Neural Networks, Hybrid Task Cascade, and Mask-RCNN for instance segmentation of disease symptoms.
• Validated algorithm base: Prior lab and field work has demonstrated the approach for wheat yield prediction, early wheat scab detection under controlled conditions, and nitrogen, potassium, and sulfur stress detection in field corn.
• Scalable coverage: Designed to image 1-40 acres per flight, suitable for breeding nurseries and large experimental plots.

Technology readiness level

The underlying spectral feature selection and deep learning algorithms have been validated in prior lab-based and field-based studies for related crop sensing tasks, including controlled-environment early wheat scab detection. The drone hyperspectral deployment for large-scale FHB severity assessment is at the field validation stage, with planned trials across diverse wheat genotypes at St. Paul and Crookston, Minnesota, using over 600 plots per location and imaging at multiple time points after inoculation to train and benchmark detection models.