Computer vision-based mapping of grapevine vigor variability for enhanced fertilization strategies through intelligent pruning estimation

The objective of this study is to develop an affordable and non-invasive method using computer vision to estimate pruning weight in commercial vineyards. The study aims to enable controlled fertilization by leveraging pruning data as an indicator of plant vigor [1]. The methodology entails the analysis of RGB and DEPTH images acquired through an embedded platform (Figure 1) in a vineyard cultivating corvina grapes using the guyot method [2]. Initially, pruning weight was evaluated by processing pictures taken manually with a controlled background. Then, we developed an algorithm to estimate pruned wood weight based on these images. Subsequently, a mobile sensor platform was utilized to automatically capture grapevine images without a controlled background. Collected data will then be used to deploy a convolutional neural network (CNN) for intelligent pruning estimation capable of extracting meaningful data from real-world environments. Additionally, we integrated and validated a visual-odometry sensor (Intel Realsense T265) to map the spatial variability of pruning estimation results.