Quantitative Analysis of Fruit Firmness by Near Infrared Spectroscopy and Least-Squares Support Vector Machine

Xiaping Fu,Yibin Ying,Huirong Xu
DOI: https://doi.org/10.13031/2013.27183
2009-01-01
Abstract:Near infrared (NIR) spectroscopy has been proved to be a powerful analytical tool for fruit internal quality determination. This paper proposes the use of least-squares support vector machine (LS-SVM) as an alternative multivariate calibration method for the quantification of pear firmness, using near infrared spectroscopy with direct measurements by diffuse reflectance in the spectral range of 800-2500 nm. Spectrum outliers were diagnosed and removed based on Chauvenet test before modeling. LS-SVM models were established using principal component (PC) scores based on principal component analysis (PCA) and spectral information from factor loadings based on partial least square regression (PLSR). The performances of LS-SVM models using PC scores were not so good. The calibration results of LS-SVM model using spectral information were almost the same as those of PLSR model with 6 factors used; while the prediction results of LS-SVM model using spectral information were much better, with much smaller RMSEP. rc, rp, RMSEC and RMSEP were 0.89, 0.88, 2.40 N and 2.57 N, respectively. From the results, it can be concluded that LS-SVM model using spectral information from factor loadings of PLSR is a feasible and promising method for predicting pear firmness based on NIR spectra.
What problem does this paper attempt to address?