Hyperspectral Imaging and Machine Learning for Plant Health Assessment

Voiceover

Presenter(s)(s)

Emma Hoopes

Abstract or Description

Crop diseases present a major threat to crop health and productivity in agricultural systems. Often, these diseases cause no visible symptoms in the plant until late in the infection. Thus, early detection and management can be difficult, especially at large scales, leading to high yield loss and plant mortality. When infected, plants respond to the impact by reallocating their resources toward defensive secondary metabolites. It is difficult and time-consuming to manually characterize the changes. However, hyperspectral imaging presents can overcome this problem, as the changes in plant foliar chemistry produce distinct spectral reflectance patterns. Integrating this technology with machine learning (ML) provides a powerful tool for detecting, phenotyping, and characterizing diseases, disease resistance, and assessing plant trait variation.

In this project, we apply these methods to screen and characterize Rhizoctonia infections in greenhouse-grown sugar beets. Hyperspectral images were captured on 122 plants before and after disease inoculation using a Specim IQ sensor with 204 spectral bands between 400-1000 nm. We evaluate the performance of well-established ML algorithms on full spectra and spectral indices in disease detection. Once a baseline is established, we can use hyperspectral drones to gather data from crop fields and analyze the infection in real-time. The results of these flights promote a better understanding of crop-disease interactions, allowing for improved decision-making in crop protection and disease-resistant crop breeding.

Mentor

Phuong Dao