Principal component analysis of chewing sounds to detect differences in apple crispness

Citation
N. De Belie et al., Principal component analysis of chewing sounds to detect differences in apple crispness, POSTH BIOL, 18(2), 2000, pp. 109-119
Citations number
34
Categorie Soggetti
Agriculture/Agronomy
Journal title
POSTHARVEST BIOLOGY AND TECHNOLOGY
ISSN journal
09255214 → ACNP
Volume
18
Issue
2
Year of publication
2000
Pages
109 - 119
Database
ISI
SICI code
0925-5214(200003)18:2<109:PCAOCS>2.0.ZU;2-K
Abstract
An investigation was made to establish the basic relationship between the c rispness of 'Cox's Orange Pippin' apples (Malus domestica Borkh.) and recor ded chewing sounds. Crispness groups were created by submitting apples to v arying storage conditions. After carrying out a fast Fourier transformation on the time signal of the generated sound, principal component analysis (P CA) was carried out on the power spectra of a training set, and a calibrati on matrix for group prediction was created. The PC values were compared wit h mechanical parameters, including apple firmness measured with the acousti c impulse response technique, maximum force and slope of the force-deformat ion curve during a penetrometer measurement, and tensile strength in a ring tensile test. PCA on Fourier-transformed chewing sounds appeared to be a p romising technique to separate apple crispness groups. By further developme nt this technique has potential as an objective measure for crispness evalu ation. Mealy and crisp apples could be distinguished by PCA. The frequencie s between 100 and 500 Hz and between 800 and 1100 Hz contributed most to th e PCs calculated from the original power spectra, corresponding to peaks in power spectra of crisp apples. When PCA was carried out on the logarithm o f the original power spectra, all frequencies contributed to some degree to the PCs. In further experiments on a variety of stored fruit, the position of the power spectra in the PC1-PC2 space was correlated with apple tensil e strength. Prediction of group belonging, using a calibration matrix based on the first 15 PCs, gave much better results for the logarithm of the pow er spectra than for the original signals. Chewing sounds from apples stored under normal air composition or under ULO conditions could be distinguishe d relatively well, while two RH groups could not be separated. This corresp onded to results from sensory analysis. (C) 2000 Elsevier Science B.V. All rights reserved.