Discrimination of Thai melon seeds using near-infrared spectroscopy and adaptive self-organizing maps

Makmuang, Sureerat, Vilaivan, Tirayut, Maher, Simon, Ekgasit, Sanong and Wongravee, Kanet (2024) Discrimination of Thai melon seeds using near-infrared spectroscopy and adaptive self-organizing maps. Chemometrics and Intelligent Laboratory Systems, 245. p. 105060.

Text
Revised-Manuscript_20231206_SM1.pdf - Author Accepted Manuscript
Available under License Creative Commons Attribution.
Download (2MB) | Preview

Official URL: http://dx.doi.org/10.1016/j.chemolab.2023.105060

Abstract

Melon (Cucumis melo L.) is a popular fruit consumed around the world. It has significant economic value as a crop, export product, and source of essential nutrients. Thus, using high-quality, authentic seed varieties is the first step toward achieving impactful agricultural production. Unfortunately, distinguishing between seed varieties using only human perception can be difficult because of their similar traits. Thus, dishonest distributors may trade low-quality seeds for high-quality seeds. In this study, seeds from five Thai melon varieties, Singapore Thai melon (ST), Nan Thai melon (NT), Round Thai melon (RT), Striped Singapore Thai melon (SST), and Golden and Long Thai melon (GLT), were classified using a distinctive discrimination method that combines modified self-organizing maps (SOMs) with near-infrared (NIR) spectroscopy. The physical characteristics, morphology, and thermal behavior of the seeds were also examined through optical microscopy, scanning electron microscopy, and thermogravimetric analysis, respectively. Attenuated total reflection–Fourier transform infrared, and NIR spectroscopy revealed that different varieties of melon seeds possess significant variations in lignin content and carbohydrate composition. Seed samples from the five Thai melon varieties were further classified using a modified SOM map created with optimized scaling value, map size, and a number of iteration parameters. Binary classification with the One vs Rest strategy and multiclass classification was performed to verify the constructed classifier model. The supervised SOMs developed herein can achieve the multiclassification of seed types effectively and efficiently, with a high accuracy of 95.52 % for the training set and 91.59 % for the test set, which were significantly superior to those of well-established discrimination models.

Item Type:	Article
Divisions:	Faculty of Science and Engineering > School of Electrical Engineering, Electronics and Computer Science
Depositing User:	Symplectic Admin
Date Deposited:	09 Jan 2024 10:40
Last Modified:	26 Jan 2024 14:08
DOI:	10.1016/j.chemolab.2023.105060
Related URLs:	Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3177769