Vineyard condition detection method using Google Earth Images and the YOLOv8 model in Northern Italy

Climate change threatens agriculture, altering growing seasons and challenging viticulture in traditionally suitable regions. Grape quality, essential for good wine, depends on proper management and suitable territories. In Northern Italy's Oltrepò Pavese region, enhancing wine production and economic stability for farmers is crucial. Moreover, detection of problematic vineyards is also a key point to enhance wine production. To address them, a PhD project is co-funded by Confagricoltura Pavia, Sezione Vino, and PNRR’s NODES Spoke 6 VINO.

Abandoned vineyards harbour diseases like flavescence dorée: an incurable phytoplasma disease, spread by the vector Scaphoideus titanus that infects nearby fields annually. Control relies on insecticides and uprooting of infected vines, making abandoned vineyards persistent reservoirs of infection. Identifying such vineyards is challenging due to legal complications for owners.

This study employs the YOLOv8 deep learning model to detect active and abandoned vineyards using high-resolution satellite imagery. A custom dataset was created from Google Earth Pro imagery of the Lombardy region in Italy, comprising 188 images of abandoned vineyards and 178 images of active vineyards. Pre-processing techniques, including auto-orientation, static cropping with 25–75% horizontal and vertical regions, resizing to 640×640 pixels, and adaptive equalization for contrast adjustment, were applied to enhance image quality. Augmentation technique was also applied on the dataset to increase the overall dataset size.

Preliminary results show the YOLOv8 model detects accurately (F1 score = 58%) active and abandoned vineyards, providing a reliable, systematic tool for vineyard management. This approach addresses a critical gap in regional viticulture, aiding in the mitigation of disease spread and supporting sustainable agricultural practices. The research results proved that the proposed method has strong generalization and good detection performance for identifying vineyard abandonment using satellite images and machine learning. This work can contribute as part of the economic stability and sustainability of wine grape production in Oltrepò Pavese.  

This work was funded by the European Union - NextGenerationEU, Mission 4 Component 1.5 - ECS00000036 - CUP F17G22000190007.