Towards automated grape vine pruning: Learning by example using recurrent graph neural networks

Vine pruning in vineyards is an important canopy management activity. It requires skill and experience to perform well and poor pruning results in low yield and can have a long‐term effect on the productivity of the vine. Automated systems usually rely on simplified pruning rules that need to be specified before operation and largely prune all vines in the same way. This is not a realistic approach and limits the pruning quality. We propose a step toward an automated system that can learn pruning behavior from expert examples without the need for explicit pruning rules. By using a novel neural network architecture we train a classifier that can choose which canes should be pruned and which should be kept. We show that the classifier can learn simple pruning rules without prior knowledge and show how this can be used as part of an automated system. The system is robust to vines that are more complex and diverse than those it was trained on and shows potential for future models to be pretrained using synthetic data and fine tuned on a minimal set of expertly pruned vines.

[1]  Enhong Chen,et al.  Word Embedding Revisited: A New Representation Learning and Explicit Matrix Factorization Perspective , 2015, IJCAI.

[2]  Tinkara Toš,et al.  Graph Algorithms in the Language of Linear Algebra , 2012, Software, environments, tools.

[3]  Tom Botterill,et al.  An expert system for automatically pruning vines , 2012, IVCNZ '12.

[4]  Samuel Williams,et al.  A Robot System for Pruning Grape Vines , 2017, J. Field Robotics.

[5]  Ah Chung Tsoi,et al.  The Graph Neural Network Model , 2009, IEEE Transactions on Neural Networks.

[6]  Sebastian Scherer,et al.  VoxNet: A 3D Convolutional Neural Network for real-time object recognition , 2015, 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).

[7]  Lukasz Kaiser,et al.  Attention is All you Need , 2017, NIPS.

[8]  XiaoQi Chen,et al.  A specialised collision detector for grape vines , 2015 .

[9]  Geoffrey Zweig,et al.  Spoken language understanding using long short-term memory neural networks , 2014, 2014 IEEE Spoken Language Technology Workshop (SLT).

[10]  Zachary Chase Lipton.  A Critical Review of Recurrent Neural Networks for Sequence Learning , 2015, ArXiv.

[11]  M. Greven,et al.  Influence of retained node number on Sauvignon Blanc grapevine vegetative growth and yield , 2014 .

[12]  R. Bramley,et al.  Vineyard variability in Marlborough, New Zealand: characterising variation in vineyard performance and options for the implementation of Precision Viticulture , 2011 .

[13]  Franco Scarselli,et al.  Recursive neural networks for processing graphs with labelled edges: theory and applications , 2005, Neural Networks.

[14]  Del Moral HernandezEmilio.  2005 Special Issue , 2005 .

[15]  Ming-Wei Chang,et al.  BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , 2019, NAACL.

[16]  Alex Sherstinsky,et al.  Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) Network , 2018, Physica D: Nonlinear Phenomena.

[17]  Pietro Liò,et al.  Graph Attention Networks , 2017, ICLR.

[18]  Jürgen Schmidhuber,et al.  Long Short-Term Memory , 1997, Neural Computation.

[19]  Yoshua Bengio,et al.  Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation , 2014, EMNLP.

[20]  E. Archer,et al.  Effect of Bud Load and Rootstock Cultivar on the Performance of V. vinifera L. cv. Red Muscadel (Muscat noir) , 2017 .

[21]  George Kurian,et al.  Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation , 2016, ArXiv.