Articles
Definition of virtual profiles of banana texture as an alternative to feature extraction from XTA curves
Article number
1382_8
Pages
59 – 68
Language
English
Abstract
Magness-Taylor (MT) penetration of fruit flesh is a standard test for the evaluation of firmness, that provides a complex texture profile when operated with a universal testing machine; to extract the main features in this test (penetration force and slope) the intervention of a trained operator is required.
The aim of this study, is to generate virtual texture-type banana based on 78 MT texture-profiles of samples, considering a significant variance in the texture of the banana tissue produced by both the storage time and stress conditions that lead to the development of stay-green ripe banana.
Principal component analysis (PCA) is used to reduce the dimensionality of interpolated texture profiles, and the quality of automated non-supervised texture reconstruction is assessed for increasing number of components: 99.8% with 8PCs.
The analysis of the relationship between the conventional MT values (force and slope) and the PCs, shows a correlation coefficient of r=0.94 and 0.91, respectively, with PC_1 while being neglectable with the rest of PCs, concluding that a significant amount of information from the texture profiles is lost when using the conventional MT values.
On the other hand, non-supervised clustering based on the 8PCs allows the reconstruction of fingerprints representative of abnormal and sound banana ripening; 98% of texture profiles are well identified when using automated feature extraction compared to 86% using conventional MT parameters (force and slope).
The aim of this study, is to generate virtual texture-type banana based on 78 MT texture-profiles of samples, considering a significant variance in the texture of the banana tissue produced by both the storage time and stress conditions that lead to the development of stay-green ripe banana.
Principal component analysis (PCA) is used to reduce the dimensionality of interpolated texture profiles, and the quality of automated non-supervised texture reconstruction is assessed for increasing number of components: 99.8% with 8PCs.
The analysis of the relationship between the conventional MT values (force and slope) and the PCs, shows a correlation coefficient of r=0.94 and 0.91, respectively, with PC_1 while being neglectable with the rest of PCs, concluding that a significant amount of information from the texture profiles is lost when using the conventional MT values.
On the other hand, non-supervised clustering based on the 8PCs allows the reconstruction of fingerprints representative of abnormal and sound banana ripening; 98% of texture profiles are well identified when using automated feature extraction compared to 86% using conventional MT parameters (force and slope).
Authors
E.C. Correa, N. Benito, M. Bustelo, B. Diezma, P. Barreiro
Keywords
fruit quality, stay-green bananas, flow-pack, shelf life
Online Articles (30)