Deep transfer learning to verify quality and safety of ground coffee. (April 2021)

• Record Type: Journal Article
• Authors: Pradana-López, Sandra; Pérez-Calabuig, Ana M.; Cancilla, John C.; Lozano, Miguel Ángel; Rodrigo, Carlos; Mena, María Luz; Torrecilla, José S.
• Is Part Of: Food control. Volume 122(2021)
• Publication Date: 2021-04
• Subjects: Coffee classification -- Food quality -- Adulteration -- Photographic camera -- Convolutional neural networks -- Transfer learning -- ResNet34; Food -- Quality -- Periodicals; Food -- Analysis -- Periodicals; Food handling -- Periodicals; Food industry and trade -- Quality control -- Periodicals; Aliments -- Industrie et commerce -- Qualité -- Contrôle -- Périodiques; Aliments -- Qualité -- Périodiques; Aliments -- Analyse -- Périodiques; Hygiène alimentaire -- Périodiques; Food -- Analysis; Food handling; Food -- Quality; Periodicals; Electronic journals; 664.07
• DOI: 10.1016/j.foodcont.2020.107801
• Languages: English
• ISSNs: 0956-7135

Abstract

Abstract: In this work, convolutional neural networks were trained with images of ground coffee captured with a camera. The objective is the quality control and detection of adulterations of Arabica and Robusta coffee with other foods such as chicory and barley. The convolutional algorithms are based on the previously trained ResNet34 convolutional system combined with transfer learning to reduce the images required, thus reducing the design costs of the final mathematical models. The models presented in this paper are capable of classifying different types of ground coffee, chicory, and barley with errors below 1.0%. They are also able to detect adulterations comprising from 5.0% to 0.5% in weight with errors below 1.4%. These results have led to a prototype capable of detecting adulterations of coffee in a straightforward, practically immediate, and accurate manner, intended for producers, distributors, and consumers. Highlights: Arabica and Robusta coffee images classified via convolutional neural networks. Deep and transfer learning to locate low amounts of coffee adulterations. Up to 60 image classes distinguished with an overall >98% accuracy. Cost-effective and real-time approach to fight fraud in coffee sector.