Automated yield prediction in vineyard using RGB images acquired by a UAV prototype platform

In viticulture, accurate yield estimation is crucial for enhancing vineyard management and improving the quality of commercial grapes. However, traditional monitoring methods, which rely on visual assessments or destructive sampling, face significant limitations, including subjectivity and lengthy processing times. Additionally, these methods often fail to capture the spatial variability within vineyards, potentially leading to unrepresentative observations. This article describes the development of a fast and automated method based on aerial digital image analysis for yield quantification in vineyards with different slopes and scenarios. In this work, digital images of vines were acquired by unmanned aerial vehicles (UAVs) and coloured tags of a known size were placed in the vineyards to automate image processing. To properly extract quantitative data from the images, the geometric distortion due to deformation between the aerial image and the real world was first corrected using a control points tool. For grape detection, colour thresholds and image filtering were applied and tested in different scenarios. Then, the number of grape pixels was converted in yield (kg/vine) using a linear regression model calculated between the bunch surface derived from image analysis and the bunch weight observed by ground measurements on representative vines. The yield estimated from UAV images was validated against ground truth measurements and the R2 in two seasons was equal to 0.85 and 0.89, respectively. Future prospects are to improve image processing by exploiting the identification of objects present in the environment to avoid the installation of the tags in the field, extend the developed approach to other agricultural contexts, and create easy-to-use tools that do not require any specific skills for the agronomist.