An expertized grapevine disease image database focused on Flavescence dorée and its confounding diseases

Grapevine is a plant subject to many diseases, deficiencies and is vulnerable to many pests. The current disease controls consist in monitoring the vineyard and the spraying of phytosanitary products at the block scale. The automatic detection of disease symptoms would allow a reduction of these products for the treatment of these diseases before they spread. Flavescence dorée (FD) is one of the most monitored vine diseases in Europe as it decreases vine productivity while being very infectious and leading to significant yield losses. The diagnosis of FD is usually done by the association of symptoms on 3 organs (leaf, shoot and bunch). It is carried out by experts, as many others diseases and stresses, either biotic or abiotic, imply similar symptoms. These experts require a decision support tool to better plan their survey to improve the efficiency of their scouting. To work towards this goal, a dataset composed of 1483 RGB images showing vines suffering from various diseases and stresses, was acquired. In particular, it is focused on the grapevine disease called Flavescence dorée (FD) and its confounding factors and diseases. Images of 5 grape varieties (Cabernet Sauvignon, Cabernet Franc, Merlot, Ugni Blanc and Sauvignon Blanc) were acquired during 2 years (2020 and 2021) in the field by proximal sensing. Images were taken through vine rows identified by scouting experts as containing many cases of FD. They pointed out vines affected by FD, ESCA or vines showing symptoms similar to those of FD. In order to see entire vines in the images, images were acquired at a distance between 1 and 2 meters depending on the row width. An industrial flash ensured a constant luminance on the images regardless of the environmental circumstances. Two types of annotations have been made. First, an annotation at the image scale: an annotation file containing the diagnosis of the experts at the vine scale was filled directly in the field for each image. Secondly, an annotation at the symptom scale: the same scouting experts were asked to annotate on a computer the symptoms present on the images. They annotated 744 images by creating bounding boxes around the interesting leaves and separated them into 3 classes: ‘FD symptomatic leaves’, ’Esca symptomatic leaves’ and ‘Confounding leaves’, this last class containing all the leaves different from healthy leaves. In addition, on 110 of these images, symptomatic bunches and shoots were respectively annotated by bounding boxes a