Eve Laroche-Pinel, Benjamin Corrales, Khushwinder Singh, Erica Sawyer, Monica Cooper, Marc Fuchs, and Luca Brillante*
*Department of Viticulture and Enology, California State University Fresno, 2360 E Barstow Ave, Fresno, CA, 93740 (lucabrillante@csufresno.edu)

North American vineyards are highly affected by two groups of viruses that cause major economic losses: grapevine leafroll-associated viruses (GLRaVs) and red blotch virus (GRBV). Unfortunately, no curative solution has been found to eradicate these viruses in diseased vineyards. Therefore, the only way to limit their spread is to quickly identify and remove infected vines (i.e., roguing). For this purpose, remote sensing, especially hyperspectral imagery, is an encouraging tool to identify infected vines autonomously and on a large scale. We used >2000 canopy images acquired in the vineyard with a stationary VIS-NIR hyperspectral camera (from 510 nm to 900 nm). Images were obtained at six times per season, from the onset of veraison to harvest, in two consecutive growing seasons. Pre-trained machine-learning models were used to extract the canopy signal from the images and predict plant infection status previously assessed by molecular analyses. Binary (healthy, infected) and four classification categories (healthy, infected by GLRaVs, infected by GRBV, or infected by both viruses) were tested. Prediction accuracy across phenological stages was determined and compared. Additional analysis was conducted to highlight the most relevant wavelengths among the 234 acquired by the camera to identify these viral diseases. This work showed that VIS/NIR imagery combined with machine learning is a promising tool to identify infected vines in the vineyard from static images acquired on the ground.

Funding Support: CDFA-SCBGP, CSU-ARI System Grants