Ripening stage classification of Coffea arabica L. var. Castillo using a Machine learning approach with the electromechanical impedance measurements of a contact device

0560331 - ÚFM 2023 RIV NL eng C - Conference Paper (international conference)
Pineda, Maria F. - Tinoco Navaro, Hector Andres - Lopez-Guzman, J. - Perdomo-Hurtado, L. - Cardona, Carlos I. - Rincon-Jimenez, A. - Betancur-Herrera, N.
Ripening stage classification of Coffea arabica L. var. Castillo using a Machine learning approach with the electromechanical impedance measurements of a contact device.
MATERIALS TODAY-PROCEEDINGS. Vol. 62. Amsterdam: Elsevier, 2022, s. 6671-6678. ISSN 2214-7853.
[IC4M - International Conference on Advances in Materials, Mechanics, Mechatronics and Manufacturing. Indie (IN), 09.04.2022-10.04.2022]
Institutional support: RVO:68081723
Keywords : Coffee fruits * coffee arabica L. var. Castillo * Electromechanical impedance * Non-destructive testing * Machine learning * Selective harvesting
OECD category: Materials engineering

The new agricultural prototypes or devices based on deep physical insights as the frequency and vibrational analysis must aid the gap between the plants and their fruit mechanical properties and time dependence. The present study describes a non-destructive method to classify coffee fruits (Coffea arabica L. var. Castillo) according to their ripening stage using high-frequency vibrations, Ripetech. This device's main advantages of this novel proposal are the physical insights of its electro-mechanical response, design functionality to hold fruits, and operability. For this purpose, a vibration technique was developed through electromechanical impedance evaluation of a piezo device that stimulates coffee fruits by holding tweezers. This methodology was planned to conduct electrical impedance measurements and correlate these with the ripening stage. Then, experimental vibration tests were directed between the frequency spectrum 5 and 50 kHz to obtain a spectral vibration database for a total sample of 45 fruits, 15 per each proposed ripening stage. Statistical indexes based on the root mean square (RMS) enabled the implementation of a classifier based on Machine Learning (the Naive-Bayes algorithm). The method proposed in this study tested the effectiveness of classifying fruits in three stages of ripening: unripe, semi-ripe, and ripe/over-ripe. This work evidences an alternative for classifying coffee fruit differently from the traditional operations. As a relevant result, each fruit has exposed its characteristic response signal, which is correlated with the ripening stage. Furthermore, this technology could help select ripe fruits more efficiently, leading to a feasible complementing selective harvesting technology development process. Copyright (C) 2022 Elsevier Ltd. All rights reserved.
Permanent Link: https://hdl.handle.net/11104/0333290