(Hellenic Society of Agricultural Engineers, 2025) López Maestresalas, Ainara; Ruiz de Gauna González, Jon; Jarén Ceballos, Carmen; León Ecay, Sara; Arazuri Garín, Silvia; Agronomía, Biotecnología y Alimentación; Agronomia, Bioteknologia eta Elikadura; Ingeniería; Ingeniaritza; Institute on Innovation and Sustainable Development in Food Chain - ISFOOD
Esca is one of the most destructive vine diseases in the world. It causes significant economic losses,
mainly due to reduced grape yield and quality. Currently, the approved methods of controlling esca include
preventive methods such as the use of fungicides on plant wounds or the use of planting systems that do not
require intensive pruning, among others. It is therefore advisable to monitor the crop to identify those vines
that are susceptible to the disease. For this reason, in this study a proximal hyperspectral camera was used
for early detection of esca presence in asymptomatic grapevine leaves. Images of 11 vines of the Tempranillo
variety grown in Etxauri (Navarre, Spain) were analysed. Hyperspectral images were acquired using a
Specim IQ snapshot camera, mounted on a tripod, working in the range of 400¿1000 nm with a spectral
resolution of 7 nm (204 bands), and an image resolution of 512 × 512 pixel including an RGB camera (5
Mpix). The images were taken under natural ambient light conditions on August 21, 2023. From the 11 vines
selected, 9 showed visual symptoms of esca and the remaining 2 were asymptomatic to the naked eye. A total
of 200 pixels were randomly selected from the dataset, 100 from asymptomatic leaves of asymptomatic vines
(class 1) and 100 from asymptomatic leaves of symptomatic vines (class 2). Partial Least Square Discriminant
Analysis (PLS-DA) was performed to classify the leaves into the two classes. Classification rates of 97%
were achieved in the cross-validation dataset. Models were externally validated at pixel-level using one image
of an asymptomatic vine and another of a symptomatic vine. The visualisation of the images confirmed the
correct classification of the pixels into the two classes, indicating that by using proximal hyperspectral
sensing an early identification of the disease is possible.
