• Convolutional Autoencoders
• Convolutional Neural Networks
• Deep Learning
• Grapevine
• Leaf Area Index
• Machine Learning
• Self-Supervised Learning
• Vineyards
• Viticulture

The Leaf Area Index (LAI) is an important indicator of grape canopy vigour, which is related to several aspects, such as the health of the plant as well as the quality and quantity of its production. A new approach based on a Deep Learning model, the Convolutional Neural Network (CNN), is proposed to estimate the LAI from a set of RGB images portraying the top surfaces of canopies. Such images are extracted from mosaics of an experimental vineyard, which are collages of images taken by means of high resolution cameras mounted on Unmanned Aerial Vehicles (UAVs). The obtained dataset is composed with a limited amount of labeled data and many unlabeled ones. In this context a baseline CNN, trained on labeled data using Supervised Learning techniques, is compared with other CNNs developed using a Self-Supervised Learning method, the Convolutional Autoencoders (CAEs). The last models are employed to test whether they can exploit unlabeled data for improving the performance of the baseline model.