Acoustic Identification of Grape Clusters Occluded by Foliage

Abstract

The performance of a vineyard can be influenced by accurate yield estimations prior to harvest. Traditionally, this is a manual process. However, due to the high labour costs and subjective nature of manual assessments, researchers have been working on automated techniques. Utilising 2D computer vision has shown promising results but is inherently limited due to occlusions. The algorithms can only count grapes that are directly visible. Often this shortcoming is accounted for by using coarse occlusion ratio estimates, which themselves need to be manually determined. As a result, researchers have begun looking at alternative methods of grape detection. Synthetic Aperture Radar (SAR) has been demonstrated as a feasible approach to see grape clusters behind leaves. However, this comes at a significant financial cost. This paper introduces an alternative approach that utilises low frequency ultrasound to detect grape clusters in the presence of foliage occlusion. We demonstrate that such low frequency signals have the ability to propagate through foliage and reflect off grapes behind. Additionally, by agitating the leaves we can analyse the variance of consecutive samples and determine which volumes are likely to belong to grape clusters.