UAS and machine learning platform for peanut yield estimation, disease monitoring, and breeding selection

Overview

This solution combines Unoccupied Aircraft Systems (UAS, commonly known as drones) with machine learning to deliver high-throughput phenotyping and crop monitoring tailored for peanut production. By capturing large volumes of aerial phenotypic data and analyzing it through trained models, the platform can estimate yield, detect diseases and other stress factors not visible to the naked eye, and help breeders identify peanut lines with stronger host plant resistance. The result is faster, more consistent, and less labor-intensive data collection that directly supports selection decisions in peanut breeding programs and management decisions in commercial fields.

Technical specifications

Data collection and processing:

• UAS-based protocols for systematic aerial data capture across field-scale trials
• Processing pipelines that transform raw imagery into standardized phenotypic traits
• Phenotypic outputs including canopy cover, plant height, canopy volume, canopy temperature, and vegetation indices such as Excess Green (ExG)

Machine learning framework:

• Artificial neural network (ANN) models trained on multi-temporal UAS data combined with non-temporal and qualitative inputs
• Validated feature sets for crop monitoring, including canopy volume, ExG, boll count, boll volume, and irrigation status
• Yield prediction demonstrated across cotton, corn, wheat, and grain sorghum, with extension to peanut
• Disease and stress monitoring designed to detect biotic and abiotic stress invisible to the human eye

Field performance (demonstrated in prior crops):

• Plant height accuracy with RMSE under 5 cm
• Canopy temperature accuracy within 1 °C
• Improved reliability and usability compared to manual measurement in breeding programs

Technology readiness level

The UAS data collection and processing protocols have been developed and applied across multiple crops including cotton, grain sorghum, wheat, and corn. Yield prediction using UAS-derived multi-temporal data and machine learning has been demonstrated in published studies on those crops. The peanut-specific pipeline, including data collection, processing, analysis, and modeling, is currently in progress, with field-scale experiments planned across multiple seasons and locations in commercial production settings. Approximately one additional year of data collection is required to complete validation, after which the framework is expected to support management decisions related to monitoring and threshold levels of plant-parasitic nematodes and other stresses in peanut.