Development of imaging devices and algorithms to quantify black rot symptoms in grapevine under semi-controlled conditions: detached leaves in laboratory and growing cuttings in greenhouse

Abstract: Concerning one of the most threatful cryptogamic diseases in viticulture worldwide,
i. e., black rot “BR” caused by Phyllosticta ampelicida (Engelm.) syn. Guignardia bidwellii, an acceleration of research is required for high-throughput phenotyping in order to promote
research in the near future. This is crucial for developing new biocontrol tools and further
studying key epidemiological issues. Two original image capture devices (photographic
benches) have been developed in laboratory and greenhouse to be adapted specifically for
detached BR symptomatic leaves and grapevine cuttings, respectively. They will aim at
assessing more precisely and rapidly, respectively: i) foliar BR severity (lesion size, number,
density) and also by counting pycnidia conceptacles (macrophotography), and ii) BR severity
in the growing cuttings, notably those used in biocontrol bioassays according to the published
methodology (Raveau et al., 2024).
As for detached leaves and standard BR lesion analysis – not based on macro-photographs
– a database of approx. 1,000 images has been compiled and specifically annotated using the
CVAT software. Then, an off-the-shelf deep learning algorithm, namely the YOLO V11s,
proved suitable for, satisfactorily, detecting and quantifying the BR foliar symptoms. Another
more specific neural network architecture, dedicated to particle counting, is under development to detect and count individually occurrences of pycnidia. Concerning growing cuttings in greenhouse used in various biocontrol assays, an image database based on 32 photographs per plant is currently in progress. A specific annotation protocol has been elaborated specifically using the CVAT software. The algorithms for detached leaves will be extended, if possible, to also be applied to detect and count BR leaf lesion (excluding pycnidia) at the level of the whole young plant (grapevine cuttings). The underlying methodologies, both from the epidemiological and biological as from the deep-learning/AI points of view will be discussed. As a current prospect, near-future research developments are going to address the vineyard issue with foliar BR detection and severity quantification based on field-images. As a more long-term prospect, such methods would support BR biocontrol studies and decision-making in managing BR in vineyards.