Abstract for HONS 03/20
An AI System for Automatic Grapevine Pruning Decision Making
Marvin Goesmann
Department of Computer Science and Software Engineering
University of Canterbury
Abstract
Vine pruning is a commonly manual process in vineyards requiring costs and time and requiring skill and expertise from pruners. Over the years, advancements made in the related fields of AI has provided the potential to automate this process and aim to increase grape yield while reducing costs. This paper focuses on one specific step of the process, in which an AI classifies vines to be either pruned or laid down. We introduce a neural network architecture involving a graph structure to encode the vine sequence to overcome limitations in past research relating to lack of encoding structural information. By training with a simple and fairly short dataset, the system is able to learn simple pruning rules through training examples and managed to achieve a 97.4% classification accuracy when evaluated by a dataset from the same source as the training data, and 93.5% when evaluated by a more complex and unknown dataset, proving it is able to learn unknown features and can generalise to structures more complex than what is was trained on.