Detecting vineyards using multispectral UAV imagery and artificial intelligence: A case study from Northern Greece

The recent technological advancements in the field of Unmanned Aerial Vehicles (UAVs) and Artificial Ιintelligence (ΑΙ) have led to their widespread adoption across different sectors of agriculture. In particular, there has been a growing interest in the application of these technologies in viticulture, but their operational implementation and validation under diverse environmental and management conditions remain limited. To this end, the evaluation of different AI methodological frameworks for vine detection across different settings constitutes an important step for the sufficient and cost-effective deployment in real-world vineyards.

Our study aims at contributing towards this direction by evaluating different segmentation approaches for vines detection tested in a real-world vineyard of Ktima Lazaridi vinery located in the prefecture of Drama, Macedonia, Northern Greece.  The vineyard acting as the study’s experimental site spanned across approximately 4.94 hectares and consisted of Sauvignon Blanc vines. In this site, multispectral imagery was acquired at 40 meters Above Ground Level (AGL) on 30 July 2025 from a UAV equipped with a high-definition RGB camera, a red-edge and Near infrared bands. Experiments were performed using state-of-the-art segmentation methods such as Segment Anything Model (SAM) and object-based image analysis frameworks using multimodal UAV imagery (RGB, NIR, Red Edge bands, Vegetation Indices). Standard statistical metrics were employed to quantitatively assess the modelling results using as reference ground truth masks generated with direct photointerpretation. ArcGIS Pro was used for the implementation of the AI algorithms as well as for the evaluation of the experimental analysis.

Our study findings suggest that UAV-based multimodal imagery combined with advanced AI algorithms, can serve as a cost-effective and scalable solution for vineyard monitoring, management, and decision-making. Future work will focus on evaluating these methods under different grape varieties, phenological stages, and environmental conditions to further generalize their applicability and optimize vineyard management strategies.