Automatic diagnosis of a multi-symptom grape vine disease using computer vision

“Flavescence dorée” (FD) is a highly transmissible disease very closely monitored in Europe as it reduces vine productivity and causes vine death. Currently, this disease is controlled by a two-pronged approach: spray insecticide on a regular basis to kill the vector and by experts surveying each row in a vineyard. Unfortunately, these experts are not able to carry out such a task every year on every vineyard and need an aid for planning their surveys. In this study, we propose and evaluate an original automatic method for the detection of FD, based on computer vision and artificial intelligence applied to images acquired by proximal sensing. A two-step approach is used, mimicking expert’s scouting in the vine rows: i) the three known isolated symptoms are detected, ii) isolated detections are combined to make a diagnosis at vine scale. To achieve this, a detection deep neural network is used to detect and classify non-healthy leaves into three classes – ‘FD symptomatic leaf', ‘Esca leaf’ and ‘Confounding leaf’ – while a segmentation network retrieves FD symptomatic shoots and bunches. Finally, the association of the detected symptoms is performed by a RandomForest classifier allowing a diagnosis at the image scale. The experimental evaluation is conducted on images collected on 14 blocks planted with 5 grape cultivars, allowing the study of the impact of acquisition conditions and variability of symptom expressions among grape cultivars.