S18 - Session O2 - In-field detection of a multi-symptom grape vine disease by proximal sensing and artificial intelligence

S18 - Session O2 - In-field detection of a multi-symptom grape vine disease by proximal sensing and artificial intelligence

Thursday, August 18, 2022 11:15 AM to 11:30 AM · 15 min. (Europe/Paris)
Angers Congress Centre
S18 III International symposium on mechanization, precision horticulture, and robotics: precision and digital horticulture in field environments

Information

Authors: Malo Tardif *, Ahmed Amri, Barna Keresztes, Aymeric Deshayes, Marc Greven, Jean-Pierre Da Costa

"Flavescence doree" (FD) is a grape vine disease caused by the bacterial agent Candidatus phytoplasma vitis and spread by a leafhopper (Scaphoideus titanus ). It reduces vine productivity and causes vine death while being highly infectious and hard to control. Classified as a quarantine disease in Europe since 1993, it is subject to mandatory reporting. A grape vine that has contracted FD must be uprooted to avoid the spread of the disease. These days, the control method used against FD is a two-pronged approach: i) regular insecticide spraying to kill the vector, ii) vineyard surveying by experts. Unfortunately, experts are unable to survey the totality of vineyards at regional level on an annual basis and need decision support tools as an aid for planning their survey. In this study, we propose an original automatic detection method of FD, based on computer vision and artificial intelligence applied to images acquired by proximal sensing at vine scale. A two-step approach is used, mimicking experts' prospection in the vine rows: (i) the three known isolated symptoms (red or yellow leaves depending on variety, together with a lack of shoot lignification and the presence of dried off bunches) are detected, (ii) individual detections are combined to make a diagnosis at vine scale. A detection deep neural network is used to detect and classify non-healthy leaves into 3 classes ('FD symptomatic leaf', 'Esca leaf', 'Confounding leaf') while a segmentation network allows to retrieve FD symptomatic shoots and bunches. Finally, the association of the detected symptoms is performed by a RandomForest classifier, which allows a diagnosis at the image scale. Despite reduced learning sets and the presence of confounding factors, this procedure enables precision and recall rates greater than 80% in the diagnosis of FD at vine scale for two grape varieties: Cabernet Sauvignon and Ugni blanc.

Type of sessions
Oral Presentations
Type of broadcast
In Replay (after IHC)In personIn remote
Keywords
artificialintelligencecomputervisiondeeplearningflavescencedoréeImageprocessingproximalsensingvinedisease
Room
Botanical Room - Screen 1

Oral session including this Oral presentation

S18 - Session O2 - Imaging based diagnosis

Angers Congress Centre

Log in