Vineyards are affected by pathogens globally. Some of the most damaging pathogens are Uncinula necator, Plasmopara viticola, and thrips, which affect the plant entirely and threaten the health and productivity of vineyards. To control the emergence and spread of pathogens, early detection is essential. Studies to date focus on visual or molecular detection of pathogens but are limited in terms of scalability, labor intensity, need for equipment and expertise. To tackle these limitations, we propose the early detection of grapevine virus infections using Convolutional Neural Networks on both RGB and thermal infrared imagery captured via a smartphone and an FLIR sensor. To do so, we employ a four-step workflow where we first acquire nearly 500 images detecting symptoms of pathogens, which we then crop in smaller tiles. Then, we use the ArcGIS Train Deep Learning Model tool trained with Single Shot Detector and RetinaNet frameworks to detect image areas showing pathogen presence. Finally, we calculate the IoU score to compare precisions between different tile sizes and frameworks. The results demonstrate that pathogen detection using these models is highly effective, with most images having a IoU score above 0.7. Moreover, 30% of images score a precision of 1.0. The consequences of these findings highlight the importance of early detection of pathogens to better understand their spread and effects on vineyards, which finally contribute to proposing effective management measures.
Petre et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: