Automatic Detection of Flavescence Dorée Symptoms Across White Grapevine Varieties Using Deep Learning

Abstract

Flavescence dorée (FD) is a grapevine disease caused by phytoplasmas and transmitted by leafhoppers that has been spreading in European vineyards despite significant efforts to control it. In this study, we aim to develop a model for the automatic detection of FD-like symptoms (which encompass other grapevine yellows symptoms). The concept is to detect likely FD-affected grapevines so that samples can be removed for FD laboratory identification, followed by uprooting if they test positive, all to be conducted quickly and without omission, thus avoiding further contamination in the fields. Developing FD-like symptoms detection models is not simple, as it requires dealing with the complexity of field conditions and FD symptoms' expression. To address these challenges, we use deep learning, which has already been proven effective in similar contexts. More specifically, we train a Convolutional Neural Network on image patches, and convert it into a Fully Convolutional Network to perform inference. As a result, we obtain a coarse segmentation of the likely FD-affected areas while having only trained a classifier, which is less demanding in terms of annotations. We evaluate the performance of our model trained on a white grape variety, Chardonnay, across five other grape varieties with varying FD symptoms expressions. Of the two largest test datasets, the true positive rate for Chardonnay reaches 98.48% whereas for Ugni-Blanc it drops to 8.3%, underlining the need for a multi-varietal training dataset to capture the diversity of FD symptoms. To obtain more transparent results and to better understand the model's sensitivity, we investigate its behavior using two visualization techniques, Guided Gradient-weighted Class Activation Mapping and the Uniform Manifold Approximation and Projection. Such techniques lead to a more comprehensive analysis with greater reliability, which is essential for in-field applications, and more broadly, for all applications impacting humans and the environment.

Keywords: Flavescence dorée; convolutional neural networks; explainable artificial intelligence; fully convolutional networks; grapevine yellows; plant diseases detection; precision viticulture; smart farming.

FIGURE 1

Processing setup overview.

FIGURE 2

Illustration of Flavescence dorée symptoms for the different grape varieties found in the Ext. test dataset. Source: D. Vergnes, FREDON Aquitaine.

FIGURE 3

Examples of samples from the CNN training dataset for FD and Non-FD classes.

FIGURE 4

Summary of accuracy values obtained on the validation dataset for the 10 runs.

FIGURE 5

Segmentation maps with minimum and maximum Intersection over Union (IoU) values on Chardonnay datasets. On the Pers. Chardonnay dataset, Images (A) with the lower IoU value, (B) with the higher IoU value. On the Ext. Chardonnay dataset, (C) with the lower IoU value, (D) with the higher IoU value.

FIGURE 6

Examples of false predictions from the Pers. Chardonnay Test dataset. (A,B) Undetected Flavescence dorée (FD) symptoms, (C,D) Over-detection close to FD symptomatic areas, (E): Visually unexplainable false FD detection.

FIGURE 7

Examples of false predictions from Ext. datasets. (A) Original image of Exalta leaf with Flavescence dorée (FD) symptoms; (B) Prediction associated to image A, only the rolled areas of the leaf are identified as FD; (C) Undetected FD symptoms on Ugni-Blanc grapevine variety; (D) Over-detection close to FD symptomatic areas; (E) False detection on non-grapevine elements; and (F) Undetected FD symptoms on Exalta grapevine variety.

FIGURE 8

Segmentation maps with minimum and maximum Intersection over Union (IoU) values on two Ext. datasets. On the Ext. Semillon dataset: Images (A) with the lower IoU value, and (B) with the higher IoU value. On the Ext. Merlot Blanc dataset: (C) with the lower IoU value, and (D) with the higher IoU value.

FIGURE 9

Illustration of representative features from Guided Grad-CAM that are used by the model to predict the Flavescence dorée (FD) and Non-FD classes. (A–C): Non-FD images, (D–H): FD images. The predicted label and the source dataset are given next to each image.

FIGURE 10

Grape varieties in the model space: UMAP visualization of the embeddings based on an arbitrarily-set seed.

FIGURE 11

Illustration of irrelevant features from GG-CAM that are used by the model to predict the Flavescence dorée (FD) class. (A) Trellising wire, petioles, canes, soil and rolled shapes, (B) Canes, (C) Canes and grapes, (D) Trellising wire and triangular shadow shape, (E) Trellising wire and triangular shadow shape, and (F) Trellising wire and soil.