The feasibility of quantifying the soluble solids content of intact apples was investigated by visible and near infrared (vis–NIR)
transmittance spectroscopy combined with the least squares support vector machines (LS-SVM) method. The spectra were pretreated
by Savitzky-Golay smoothing, first and second derivatives, standard normal variate transformation, and multiplicative scatter
correction. The regression models were developed by LS-SVM and partial least squares (PLS). The accuracy of the LS-SVM and
PLS models was compared.
Apples are a very popular fruit in people's daily lives because of their delicious taste and nutritional value. Soluble solids
content (SSC) is one of the most important evaluation criteria affecting the consumers' appreciation for selection. However,
the traditional measurement methods have many disadvantages, such as long analysis time, high costs, and complications.
Visible–near infrared (vis–NIR) spectroscopy is a rapid, reliable, and nondestructive approach for the measurement of SSC
in several fruits. Many researchers have reported using nondestructive techniques to assess the SSC value of fruits. In 2004,
Chauchard and colleagues (1) showed that the least squares support vector machines (LS-SVM) method was more accurate in prediction
than partial least square (PLS) and multiple linear regression (MLR) for predicting the total acidity in fresh grapes. In
2006, Zude and colleagues (2) applied acoustic impulse resonance frequency sensors and miniaturized vis–NIR spectrometers
to predict fruit flesh firmness and SSC when the fruit was still on the tree as well as its shelf life. As a result, SSC prediction
of freshly harvested apples had a standard error of cross validation (SECV) of 1.29 °Bx. In 2009, Fan and colleagues (3) investigated
fruit orientation in the examination of SSC in red Fuji apples by vis–NIR transmittance spectroscopy and concluded that the
best fruit orientation was when the stem–calyx axis was vertical and the fruit surface was illuminated from the upper side.
Also in 2009, Paz and colleagues (4) performed a comparative study which was made of the performance of different spectrophotometers
as part of some research into the potential of NIR reflectance spectroscopy as a nondestructive method for predicting soluble
solids content. Many publications have proven that multivariate calibration methods in NIR spectroscopy for estimating varieties
of fruit properties are a good alternative (5,6). Although linear models such as multiple linear regression (MLR), principal
component regression (PCR), and PLS are widely used in the prediction of fruit quality, nonlinear calibration methods often
have better performance, especially in improving the robustness of NIR spectroscopy. Support vector machines (SVM) is a powerful
methodology for solving problems in nonlinear classification, function estimation, and density estimation. LS-SVM is an improved
method of standard SVM put forward by Suykens and Vande in 1999 and Suykens and colleagues in 2002. It transformed the quadratic
programing problem of a standard SVM demand solution to a linear problem by using the least square value function and equality
constraints and increased the training speed and restraining precision. Some researchers have applied LS-SVM to regression
models. Kovalenko and colleagues (7) determined amino acid composition of soybeans and concluded that the performance of LS-SVM
was better than that of artificial neural networks (ANN). Sun and colleagues (8) reported that LS-SVM models were better than
PLS models with correlation coefficient (R) and root mean square error of prediction (RMSEP) of (0.88, 0.80 °Bx) and (0.82, 1.01 °Bx) for portable and on-line measurement
mode, respectively. Liu and colleagues (9) developed an NIR spectrometry regression model by LS-SVM, and the results showed
that portable NIR combined with LS-SVM was a feasible method to predict Brix values of intact pears nondestructively. Also,
Liu and colleagues (10) investigated the performance of NIR spectrometers with LS-SVM in determining acetic, tartaric, and
formic acids, and the pH of fruit vinegars. The results indicated that NIR spectroscopy (7800–4000 cm-1) combined with LS-SVM could be utilized as a precision method for the determination of organic acids and pH of fruit vinegars.
In addition, Liu and colleagues (11) applied principal component analysis (PCA) combined with partial least-squares discriminant
(DPLS) and LS-SVM to realize the rapid identification of different varieties of pears. Both of those two models had preferable
results, that the rate of identification is 100%. Nie and colleagues (12) investigated the performance of vis–NIR spectroscopy
as a rapid and nondestructive technique to determine the boiling time of yardlong beans. Pissard and colleagues (13) determined
the vitamin C, polyphenol, and sugar contents in apples by NIR spectroscopy combined with the LS-SVM method.
The objectives of this study were to investigate the feasibility of using vis–NIR spectroscopy combined with the LS-SVM method
to predict the SSC of intact apples nondestructively and to compare the accuracy of the LS-SVM and PLS models.