Digital ampelography: deep learning (CNN) using Keras to identify grapevine cultivars

S. Vélez, J.A. Rubio, R. Vacas, E. Barajas
Ampelography is the science that studies the identification and classification of grapevines (Vitis). It is a laborious science, and it is carried out manually through visual surveys usually performed by agronomists, which requires a huge amount of time. Image processing and computer vision based on machine learning methods can enable agronomists to minimise the time spent on cultivar identification. Convolutional neural networks (CNN) could be employed for this task since they can efficiently learn increasingly complex visual concepts by identifying spatial hierarchies of patterns and reprocessing the results of convolution layers. This work aims to use digital ampelography to classify grapevine cultivars using deep learning. For this purpose, a convolutional neural network to classify grapevine cultivars was configured and trained using a deep learning framework Keras/TensorFlow written in Python and R, and a data set of 1011 leaf images. The results showed that CNNs can efficiently extract significant features from the input leaf images and the model reached an excellent accuracy rate, successfully predicting 95,8% of Vitis vinifera ‘Verdejo’ leaves and 100% of Vitis vinifera ‘Tempranillo’ leaves.
Vélez, S., Rubio, J.A., Vacas, R. and Barajas, E. (2024). Digital ampelography: deep learning (CNN) using Keras to identify grapevine cultivars. Acta Hortic. 1390, 311-320
DOI: 10.17660/ActaHortic.2024.1390.38
deep learning, Keras, CNN, convolutional neural network, Tensor, leaf recognition, digital ampelography

Acta Horticulturae