Please use this identifier to cite or link to this item:
Title: Fast olive quality assessment through RGB images and advanced convolutional neural network modeling
Authors: Salvucci, Giorgia
Pallottino, Federico
De Laurentiis, Leonardo
Del Frate, Fabio
Manganiello, Rossella
Tocci, Francesco
Vasta, Simone
Figorilli, Simone
Bassotti, Beatrice
Violino, Simona
Ortenzi, Luciano 
Antonucci, Francesca
Issue Date: 2022
The presence of olive external damages influences consumers perception in the case of table olive, lowering consumers acceptance and willingness to purchase. Defects cause a decrease of extra virgin olive oil quality and its shelf-life. Indeed, fruit external quality represents an important factor for marketing and oil quality characteristics. In this context, RGB image processing systems can potentially support the production of high-quality products through the automatic and rapid classification into different qualitative classes of both lots for oil production and table olives. The neural networks known as Deep Neural Networks represent a kind of artificial intelligence that demonstrated very high levels of accuracy in different application fields. The aim of the present study regards the rapid classification through RGB images and advanced Convolutional Neural Network modeling (YOLO, You Only Look Once) for olives selection on the base of defects and color. The model was trained, tested and evaluated for the future realization of an optomechanical RGB sorting system for real-time olive classification into different classes (e.g., ripening, defects, etc.) through the simultaneous extraction of parameters and dedicated features. The images acquisition was carried out with a high-resolution RGB camera equipped on a laboratory conveyor belt. The algorithm was trained using two datasets: the first made of 1500 oil olive images (i.e., Carboncella, Frantoio and Leccino cultivars), the second one of 930 table olive images (i.e., Bella di Cerignola cultivar). The classification accuracy resulted to be above 95% for both datasets, as required by the high-efficiency standards of a selection prototype.
ISSN: 1438-2377
DOI: 10.1007/s00217-022-03971-7
Appears in Collections:A1. Articolo in rivista

Files in This Item:
File Description SizeFormat Existing users please
s00217-022-03971-7.pdf1.54 MBAdobe PDF    Request a copy
Show full item record


checked on Jun 3, 2023

Page view(s)

checked on Jun 3, 2023


checked on Jun 3, 2023

Google ScholarTM



All documents in the "Unitus Open Access" community are published as open access.
All documents in the community "Prodotti della Ricerca" are restricted access unless otherwise indicated for specific documents