ReConEsca

DiVine

ESSENSE

Raising Starts

X2.0 AgriTech

BioSens Accelerator

The goal of the ReConEsca (Revealing Concealed Esca) project is to create a data set containing multispectral drone images for training machine learning algorithms (i.e. neural networks) for early detection of vine diseases.

Creating a data set is crucial when working with unbalanced data (we have more images containing healthy plants than images containing infected plants). In order to achieve this goal, within the ReConEsca project we will create a special protocol for drone mapping and a data augmentation method based on innovative neural networks.

The STARTECH program is implemented by NALED and supported by Philip Morris Operations a.d.

Vineyard monitoring is a critical component of effective vineyard management. Traditionally conducted on foot, this process has recently seen advancements through innovative (imaging) technologies, offering more efficient means of surveillance. The use of multispectral cameras mounted on drones is particularly gaining traction. However, the majority of commercial drone-based solutions primarily rely on vegetation indices. While these indices measure the overall stress level in a vineyard, they fall short in differentiating between various stress causes such as disease, nutrient deficiencies, or water scarcity. This distinction is even more vital in organic vineyards, where pinpointing the exact stress source is crucial. The limitation of vegetation indices lies in their inability to differentiate between visually observable issues (like disease symptoms) and those detectable only beyond the visual spectrum (such as mineral deficiencies). Thus, our aim is Discovering Invisible VIne Nutrition Elemens (DiVine) by developing detailed digital vineyard maps that can distinctly identify areas likely affected by disease from those suffering from nutritional deficiencies.

This sub-project presents a meticulously designed pilot for Esca Surveillance and Sensing in Serbian Vineyards (ESSENSE), structured into three distinct stages. Initially, the planning stage involves the strategic selection of testing vineyards across Serbia. Despite its small size, Serbia boasts 22 diverse wine regions, necessitating a comprehensive selection process to account for this variability. The implementation stage begins with data acquisition and preprocessing, focusing on data collected during the 2023 and 2024 growing seasons. Multispectral and RGB images captured via a drone owned by Veles Sense will be scrutinized to curate the appropriate datasets. If required, these images will undergo preprocessing using advanced image restoration techniques specifically tailored for multispectral data. Subsequently, the processed data will be uploaded to the CHAMELEON platform for further analysis – vegetation indices calculation and automatic detection of vines with problems. The resulting outputs will undergo rigorous evaluation through seven distinct validation approaches, including expert validation in situ, standard computer vision validation techniques on the resulting images/maps, and comparative analysis with other disease detection methodologies. Finally, the dissemination and exploitation phase encompasses drafting an open-access journal paper to disseminate the findings and integrate the results to enhance Veles Sense activities. By adhering to this structured approach, the ESSENSE sub-project aims to advance the methodologies for disease surveillance and sensing, ensuring robust and reliable disease detection in Serbian vineyards.