Evaluation of Vineyard Cropping Systems Using On-Board RGB-Depth Perception

Area of Science:

• Agricultural Engineering
• Computer Vision
• Plant Phenomics

Background:

• Advancements in 3D optical sensors like RGB-D cameras offer new possibilities for crop phenomics.
• Non-destructive measurement techniques are crucial for improving phenotyping in agriculture.

Purpose of the Study:

• To assess the utility of a Microsoft Kinect v2 sensor for non-destructively measuring vine geometric traits.
• To evaluate the correlation between 3D vine volume and pruning weight (dry biomass) for yield prediction in vineyard crops.

Main Methods:

• An adaptable mobile platform equipped with a Kinect v2 sensor collected depth images of grapevines.
• 3D point clouds of vine rows were generated under six different management cropping systems.
• Vine branch volume was calculated and correlated with pruning weight and yield data.

Main Results:

• Kinect-derived branch volume showed strong consistency with physical vine parameters (R² = 0.80 for pruning weight).
• A good power law relationship was observed between measured volume and vineyard yield (R² = 0.87).
• Inconsistent results for small details highlight limitations in current depth camera technology.

Conclusions:

• The Kinect v2 system demonstrates significant potential as a low-cost, robust 3D sensor for proximal sensing in agricultural applications.
• This non-destructive method aids in vineyard yield prediction, supporting strategic decision-making for growers.
• Further development is needed to overcome limitations in capturing fine details for more precise individual treatment analysis.