Hyperspectral imaging for high-throughput vitality monitoring in ornamental plant production

Abstract

Ornamental heather (Calluna vulgaris) production is characterized by high risks such as occurrence of fungal diseases and plant losses. Given the general absence of formal research on this economically important production system, farmers depend on their own approaches to assess plant vitality. We provide a reproducible, affordable and transparent workflow for assessing ornamental plant vitality with spectroscopy data. We use hyperspectral imaging as a non-invasive alternative for monitoring plant performance by combining the long-term experience of experts with hyperspectral images taken with a portable hyperspectral camera. We tested a custom-made setup deployed in a horticultural production facility and screened thousands of heather plants over a period of 14 weeks during their development from cuttings to young plants under production conditions. The vitality of shoots and roots was classified by experts for comparison with spectral signatures of shoot tips of healthy and stressed plants. To identify wavelengths that allow distinguishing between healthy and stressed heather plants, we evaluated the datasets using Partial Least Squares regression. Reflectance in the green (519–575 nm) and red-edge (712–718 nm) region of the spectrum was identified as most important for classifying plants as healthy or stressed. We transferred the trained Partial Least Squares regression model to independent test data obtained on a different date, correctly classifying 98.1% of the heather plants. The setup we describe here is adjustable and can be used to measure different plant species. We identify challenges in data evaluation, point out promising evaluation approaches, and make our dataset available to facilitate further studies on plant vitality in horticultural production systems.