South Africa Wine Research Projects 2026

2. VITICULTURE RESEARCH COMPLETED PROJECT

AUTOMATED VISUAL ASSESSMENT OF GRAPEVINE LEAFROLL SYMPTOMS USING ARTIFICIAL INTELLIGENCE

SU-PP LM 23-01 Prof Lizel Mostert, Department of Plant Pathology, Stellenbosch University Project completed: 2025

AIM AND INDUSTRY RELEVANCE: This study aimed to develop an artificial intelligence system capable of identifying Grapevine Leafroll-associated Virus 3 (GLRaV-3) symptoms in white and red wine cultivars using smart phone images. Early detection of infected vines is important for effective vineyard management, particularly for removing infected plants (rogueing) to limit disease spread. A dataset of 28,771 smartphone images was collected from vineyards containing seven red and three white wine cultivars. Images included healthy vines, GLRaV-3 infected vines, and other abnormalities such as nutrient deficiencies, mechanical damage, aster yellows, and esca. A pre-trained InceptionV3 Deep Learning Convolutional Neural Network was trained to classify images into healthy, diseased, or other symptom categories. The optimised multiclass model achieved 96% overall accuracy in identifying GLRaV-3 symptoms. Accuracy exceeded 96% for red cultivars and 83% for white culti vars, with the highest accuracy for Pinot Noir (99.53%) and the lowest for Chardonnay (83.41%). The results demonstrate that artificial intelligence-based image recognition using smartphone photos can effectively identify leafroll symptoms, providing a foundation for a future smartphone application to assist growers in diagnosing and managing leafroll disease in vineyards.

SOUTH AFRICA WINE RESEARCH 2026

37