ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

Precision Clusters and Peduncle Cutting Points Detection for Automated Table Grape Harvesting Using Deep Learning

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2601312.(doi:10.13031/aim.202601312)
Authors:   Auster Zhang, Bo Liu
Keywords:   Deep Learning, Computer Vision, YOLOv8, YOLO26, Table Grapes, Automation

Abstract. Grapes are extremely high valued commodities in the United States, especially in California. The California table grape industry has been facing challenges regarding shortage of available workers, unfavorable labor laws, rising production costs, and overall lack of interest towards the workforce, forcing the industry to consider a solution through automation. This study explores the development of a deep learning-based computer vision system with the ability to accurately detect grape clusters and the corresponding peduncle cutting point. Pictures were gathered from the Cal Poly vineyard, taken with a RealSense D415 camera. Images were manually annotated using CVAT.ai, defining a skeleton around the cluster to label grapes and cutting point. The dataset was trained on a modified YOLO (You Only Look Once) architecture to provide predictions on grape clusters and peduncle cutting, producing a model with a mean average precision (mAP) of 0.861 for grape cluster detection and 0.738 for peduncle point, indicating reliable cluster identification, with frailer cutting point prediction. This was done by adding a P2 detection head, 4-scale feature fusion, ghost bottleneck backbone, and adding new SEBlock and CBAM attention blocks. The model prediction is incorporated into a multi-layered module that will provide assistance in cutting prediction, such as geometry refinement, depth validation, temporal smoothing, and much more. The future scope of this project leads into the development of full integration with the UR5 robotic arm, shear end effector, and a mobile platform with self-navigation with the aim for a fully autonomous grape harvesting system. The implication of this work is to incorporate automation in the industry to not replace workers but substitute manual labor with abstract complex problem solving.

(Download PDF)    (Export to EndNotes)