Occluded Grape Cluster Detection and Vine Canopy Visualisation Using an Ultrasonic Phased Array

Abstract

Grape yield estimation has traditionally been performed using manual techniques. However, these tend to be labour intensive and can be inaccurate. Computer vision techniques have therefore been developed for automated grape yield estimation. However, errors occur when grapes are occluded by leaves, other bunches, etc. Synthetic aperture radar has been investigated to allow imaging through leaves to detect occluded grapes. However, such equipment can be expensive. This paper investigates the potential for using ultrasound to image through leaves and identify occluded grapes. A highly directional low frequency ultrasonic array composed of ultrasonic air-coupled transducers and microphones is used to image grapes through leaves. A fan is used to help differentiate between ultrasonic reflections from grapes and leaves. Improved resolution and detail are achieved with chirp excitation waveforms and near-field focusing of the array. The overestimation in grape volume estimation using ultrasound reduced from 222% to 112% compared to the 3D scan obtained using photogrammetry or from 56% to 2.5% compared to a convex hull of this 3D scan. This also has the added benefit of producing more accurate canopy volume estimations which are important for common precision viticulture management processes such as variable rate applications.

Plant phenotyping relevance

超音波アレイによる葉に隠れたブドウ房の画像化と、果房体積・樹冠体積の推定手法を開発・評価しており、植物形質取得が研究の中心である。

Code and data availability

The article describes ultrasonic array grape imaging experiments, photogrammetry scans, and MATLAB processing, but contains no public dataset, image, code, or model deposit. The only URL present is the standard CC BY license link, which is not a paper-specific asset.

No evidence-backed public reproduction asset is currently recorded.