Investigating 2-D and 3-D Proximal Remote Sensing Techniques for Vineyard Yield Estimation.

Area of Science:

• Viticulture
• Agricultural Engineering
• Computer Vision

Background:

• Accurate vineyard yield estimation is crucial for winegrower management and winery logistics.
• Proximal remote sensing techniques have shown limited success in viticulture for yield estimation.

Purpose of the Study:

• To investigate the effectiveness of 2-D RGB and 3-D RGB-D (Kinect sensor) imagery for yield estimation in vertical shoot positioned (VSP) vineyards.
• To compare the performance of RGB and RGB-D imagery under different experimental conditions (laboratory vs. in-situ, bunch-level vs. plant-level).

Main Methods:

• Utilized 2-D RGB and 3-D RGB-D imagery for vineyard yield estimation.
• Applied image segmentation for fruit area estimation (RGB) and 3-D reconstruction for fruit volume estimation (RGB-D).
• Employed cross-validation to determine fruit mass and subsequently estimate yield.

Main Results:

• RGB-D imagery achieved high yield estimation agreement in laboratory bunch-level experiments (r² = 0.950), outperforming RGB imagery (r² = 0.889).
• RGB and RGB-D performance was similar at the bunch-level in experiment two.
• RGB imagery outperformed RGB-D at the plant-level in experiment three.

Conclusions:

• The RGB-D sensor (Kinect) is best suited for controlled laboratory conditions.
• The robust RGB methodology demonstrates suitability for both laboratory and in-situ vineyard yield estimation.