In viticulture, timely identification and management of grapevine diseases are essential to prevent significant deterioration in grape quality. Early disease detection systems play a crucial role in ensuring effective management and mitigating potential losses. This work proposes a novel deep learning-based method that leverages multimodal data for the estimation of the risk level of developing grey mold, powdery mildew and downy mildew grapevine diseases. More specifically, the proposed multimodal method processes weather data for grapevine disease risk level estimation, as well as visual information from grapes and leaves for the classification of grape maturity and leaf health status, respectively. The ultimate goal is to fuse these complementary data sources to estimate expert-defined disease risk levels and support vineyard decision-making. Experimental results indicate that the proposed multimodal method can achieve promising performance in grapevine disease risk level assessment and highlight the potential benefits of incorporating multiple data modalities in the model’s predictions. Finally, the proposed method is validated in vitro , demonstrating its applicability for processing real-world data and supporting disease risk level assessment under real-life scenarios.
Kotaidou et al. (Sun,) studied this question.