Near-infrared spectroscopy and multivariate analysis techniques were employed to nondestructively evaluate the rancidity of perilla seed oil by developing prediction models for the acid and peroxide values. The acid, peroxide value, and transmittance spectra of perilla seed oil stored in two different environments for 96 and 144 h were obtained and used to develop prediction models for different storage conditions and time periods. Preprocessing methods were applied to the transmittance spectra of perilla seed oil, and multivariate analysis techniques, such as principal component regression (PCR), partial least squares regression (PLSR), and artificial neural network (ANN) modeling, were employed to develop the models. Titration analysis shows that the free fatty acids in an oil oxidation process were more affected by relative humidity than temperature, whereas peroxides in an oil oxidation process were more significantly affected by temperature than relative humidity for the two different environments in this study. Also, the prediction results of ANN models for both acid and peroxide values were the highest among the developed models. These results suggest that the proposed near-infrared spectroscopy technique with multivariate analysis can be used for the nondestructive evaluation of the rancidity of perilla seed oil, especially the acid and peroxide values.
Recently, westernization and gentrification of food has increased the variety of processed food, and the expanding consumption pattern aimed at pursuing wellbeing and health is increasing the interest and demand in healthy high-functioning vegetable oils. Also, with the advent of problems, such as obesity and adult diseases, caused by trans fats and cholesterol from vegetable oils like soybean oil and corn oil, which were previously extensively used, the demand for high-quality vegetable oils, such as olive oil and grape seed oil, is increasing; recently, the interest in premium vegetable oils, for example, canola oil, green tea oil, and brown rice oil, has rapidly increased. Oils are known to be not only a high-energy source, as one of the three major nutrient groups of carbohydrates, proteins, and oils, but also an important and useful component of the human body; it is present in cell membranes as a fat-soluble carrier and also protects hypodermic tissues and organs [
Note that the smell, fragrance, and taste of vegetable oils used in most homes, restaurants, and the food processing industry tends to change because of the various chemical and microbial factors during storage and processing. This decreases its nutritive value, which in turn deteriorates quality as well as that of the processed foods, and sometimes, toxic agents that are harmful to humans are also generated. This oil deterioration is called rancidity [
As an existing method to measure oil rancidity, physicochemical titration is widely used to measure the acid value, peroxide value, and so forth. However, physicochemical titration analysis may generate errors resulting from the experimenter’s skills and is expensive and time-consuming and hence is not suitable for repetitive experiments. Therefore, the development of a technique to nondestructively analyze vegetable oil rancidity in real time is attracting much interest. Recently, a study has been conducted into spectroscopy to nondestructively analyze and predict variation in the components and quality of agricultural products or food. Near-infrared (NIR) spectroscopy can obtain a signal of relatively high energy, relative to far-infrared radiation and microwaves, and can detect a particular spectrum of an intrinsic component of a test object during wavelength bandwidth measurement. Its relatively simple device composition of sensor and light source makes it easy to implement; therefore, NIR spectroscopy is being widely used in researches into the nondestructive quality analysis of agricultural products and food [
Generally, near-infrared rays have a high energy level relative to mid- and far-infrared rays, and hence, they have low optical absorbance on test samples and excellent penetrability; this means that the technique is little influenced by test sample thickness and there is no need for the preprocessing of the test sample. Also, the components of near-infrared ray light splitters, such as fiber-optic cables, monochromators, and detectors, are reported to be easy to set up and operate relative to mid-infrared ray light splitters and have excellent measurement repeatability. In particular, fiber-optic communications have been developed based on near-infrared rays; consequently, their use makes it possible to measure and analyze spectra remotely in real time and allows the development of telemetering device. Therefore, near-infrared rays can be used for the composition of real-time analysis devices for material process control and quality analysis. In spite of the many merits of NIR spectroscopy, the measurement environment and instruments may cause noise in the NIR spectra, which may lead to the degradation of spectrum analysis. Therefore, it is very important to perform appropriate preprocessing methods to remove these noise components and to develop a suitable optimal model to analyze and predict sample components by using spectra data. Until now, NIR spectroscopy research on agricultural products has been applied to the internal quality analysis of fruits and vegetables by using multivariate analysis methods, such as PLSR (partial least square regression), PCR (principal component regression), and MLR (multiple linear regression) [
This study employed NIR spectroscopy and multivariate analysis techniques to nondestructively discern the rancidity of perilla seed oil, in conjunction with the storage conditions, and to evaluate the effectiveness of the analysis methods. For this purpose, perilla seed oil was stored in specific environments for a certain period of time, and the acid value, peroxide value, and transmittance spectra of the same perilla seed oil were measured by physicochemical titration analysis and NIR spectroscopy. We used PCR, PLSR, and ANN analysis methods to quantitatively predict the rancidity extent of perilla seed oil from the collected data, and then we evaluated the performance of each model in conformity with the applied preprocessing methods.
Perilla (
The acid value (AV) of the samples was determined by the method of the American Oil Chemists’ Society (1977) with some modifications [
The peroxide value (PV) is a measure of the peroxides contained in the oil during storage. The PV of the oil was measured by iodine released from potassium iodide (KI) according to the method of the Association of Analytical Communities (1995) with slight modifications [
As illustrated in Figure
Schematic representation of the NIR spectra measuring instrument.
Perilla seed oil samples were separated into two groups and stored in two different environments. The first sample group was stored at a controlled temperature of 60°C with a relative humidity of 40% for 4 d (96 h) to accelerate rancidity. Samples were taken at time zero, and then 10 g of 12 oil samples were collected every 24 h in order to obtain transmittance spectra after different periods. Therefore, we acquired 60 spectra for the first perilla seed oil sample group in 4 d (96 h). The environmental conditions for the second sample group were similar but employed higher temperature and lower relative humidity, that is, the temperature and relative humidity of the storage chamber were controlled as 80°C, 10% for 6 d (144 h), respectively. Samples were again taken at time zero, and then in this case, 10 g of nine perilla seed oil samples were collected every 24 h to obtain transmittance spectra after different periods. Therefore, we acquired 63 spectra for the second perilla seed oil sample group in 6 d (144 h). At the end, we had prepared 123 spectra for the two different groups of perilla seed oil samples. All spectra were prepared with ten replicates at each measuring point with 50 ms of measuring exposure time.
The transmittance of measured spectra was calibrated by using both white-referenced and dark-referenced spectra as shown in (
Light scattering, optical path changes, and noise can be induced in the obtained transmittance spectra from the measuring environment and instruments; therefore, preprocessing methods should be applied to the measured spectra in order to remove noise components and correct the spectra. In this study, we employed scatter correction techniques, such as MSC (multiplicative scatter correction) and SNV (standard normal variate), and normalization techniques, including maximum normalization, mean normalization, and range normalization, as preprocessing methods.
We used the PCR and PLSR methods that are defined as (
In the development process of both PCR and PLSR models, the selection of the number of PCs (principal components) should be significantly considered. If the number of PCs is too small, the regression model becomes inaccurate because the complete measurement information cannot be sufficiently reflected in the developing model. By contrast, overfitting of the regression model, which may reduce its performance, can be induced if a large number of PCs are selected. The performance of the PCR and PLSR models that were developed to estimate the extent of rancidity was assessed with the RMSE (root-mean-square error), as defined in (
We also developed ANN-based regression models that can predict the acid and peroxide values by using NIR transmittance spectra obtained from perilla seed oil samples. As shown in Figure
Schematic structure of the ANN model.
Oil oxidation, the main cause of rancidity, is the process by which the double bonds of unsaturated fatty acids in the oil combine with oxygen in the air to form oxidation products, with accompanying off-flavors and smells. The AVs and PVs measured in this study are widely used as key indicators of the extent to which rancidity reactions have occurred during the initial and intermediate stages of oil oxidation. They could be used as an indication of the quality and stability of oils. Generally, the more the oil rancidity has progressed, the more these values have increased [
Acid values and peroxide values of perilla seed oil for various storage temperatures, relative humidity, and time periods. (a) Acid value. (b) Peroxide value.
Statistics of acid values and peroxide values of perilla seed oil according to storage time.
Measurements | Storage conditions | Time (h) | ||||||
---|---|---|---|---|---|---|---|---|
0 | 24 | 48 | 72 | 96 | 120 | 144 | ||
Acid value (mg/g) | 60°C, 40% RH | 3.32 ± 0.02 | 3.81 ± 0.05 | 3.92 ± 0.04 | 3.9 ± 0.14 | 4.19 ± 0.23 | — | — |
80°C, 10% RH | 3.57 ± 0.04 | 3.63 ± 0.01 | 3.65 ± 0.01 | 3.71 ± 0.19 | 3.78 ± 0.13 | 3.80 ± 0.06 | 3.82 ± 0.08 | |
Peroxide value (meq/kg) | 60°C, 40% RH | 1.48 ± 0.68 | 2.18 ± 0.50 | 2.99 ± 0.30 | 3.65 ± 0.45 | 4.22 ± 0.45 | — | — |
80°C, 10% RH | 2.91 ± 0.10 | 2.98 ± 0.38 | 5.64 ± 0.54 | 8.45 ± 0.66 | 10.33 ± 0.50 | 12.03 ± 0.29 | 14.00 ± 0.11 |
From the incremental results of AVs and PVs that were obtained during the initial and intermediate processes of oil oxidation under the two storage conditions, it can be observed that the AV of the oil sample stored under condition 1 (60°C, 40% RH) was higher than that of the oil sample stored under condition 2 (80°C, 10% RH). That is, the AV of the oil sample stored at the lower relative humidity of 10% was generally stabilized during the six days at under 3.82 mg/g, whereas the observed AV at the higher relative humidity of 40% increased continuously during the four days until it reached 4.19 mg/g. In contrast to the AV, the PV of the oil sample stored under condition 2 (80°C, 10% RH) increased rapidly above the PV of the oil sample stored under condition 1 (60°C, 40% RH). That is, the PV was generally stabilized during the four days at less than 4.2 meq/kg at a 60°C, whereas the observed PV of the oil sample stored at the higher temperature of 80°C increased rapidly until it reached 10.3 meq/kg on the same fourth day. Moreover, the PV of the oil sample stored at the higher temperature of 80°C condition was measured as 14 meq/kg on the sixth day. Based on a previous study, which reported that rancid tastes and odors are clearly noticeable when the PV exceeds 20 meq/kg [
These results indicate that the free fatty acids that were produced by the oil oxidation process can be more easily affected by relative humidity than temperature, whereas the peroxides that were formed by iodine released from potassium iodide by an oil oxidation process were more significantly affected by temperature than relative humidity during the initial and intermediate processes of the oil oxidation under the storage environments in this study. Similarly to our study results, previous studies showed that the oxidation process of oil stored at around 60°C was significantly affected by both the moisture content of the oil itself and the relative humidity of the storage environment [
Figure
Transmittance spectra of perilla seed oil samples according to storage time.
Locations of function groups and corresponding vibration modes in the near-infrared region.
Wavelength (nm) | Vibration mode | Structure |
---|---|---|
900 | C–H str. third overtone | CH3 |
1195 | C–H str. second overtone | CH3 |
1215 | C–H str. second overtone | CH3 |
1395 | 2C–H str. + 2C–H def. | CH3 |
1410 | O–H str. first overtone | ROH |
The performances of PCR models for determining the AVs and PVs of the oil samples are summarized in Table
Results of PCR calibration and validation for acid and peroxide values of perilla seed oil according to preprocessing methods.
Preprocessing methods | Acid value | Peroxide value | ||||||
---|---|---|---|---|---|---|---|---|
Calibration | Validation | Calibration | Validation | |||||
|
RMSE |
|
RMSE |
|
RMSE |
|
RMSE | |
Raw data | 0.8979 | 0.0699 | 0.6501 | 0.1293 | 0.8240 | 1.6833 | 0.7748 | 1.9136 |
MSC | 0.8645 | 0.0805 | 0.7091 | 0.1179 | 0.8787 | 1.4042 | 0.7328 | 2.0841 |
SNV | 0.8373 | 0.0882 | 0.6859 | 0.1225 | 0.8120 | 1.7484 | 0.7552 | 1.9951 |
Normalization | ||||||||
Max | 0.8390 | 0.0877 | 0.5344 | 0.1492 | 0.8403 | 1.6113 | 0.7654 | 1.9529 |
Mean | 0.8785 | 0.0762 | 0.6985 | 0.1201 | 0.8575 | 1.5156 | 0.7833 | 1.8771 |
Range | 0.8105 | 0.0952 | 0.6454 | 0.1302 | 0.8157 | 1.7308 | 0.7474 | 2.0265 |
Prediction results of PCR models for acid and peroxide values from the highest
Table
Results of PLSR calibration and validation for acid and peroxide values of perilla seed oil according to preprocessing methods.
Preprocessing methods | Acid value | Peroxide value | ||||||
---|---|---|---|---|---|---|---|---|
Calibration | Validation | Calibration | Validation | |||||
|
RMSE |
|
RMSE |
|
RMSE |
|
RMSE | |
Raw data | 0.8257 | 0.0913 | 0.6887 | 0.1220 | 0.8209 | 1.7061 | 0.8036 | 1.7869 |
MSC | 0.8643 | 0.0806 | 0.7369 | 0.1122 | 0.8255 | 1.6842 | 0.7858 | 1.8660 |
SNV | 0.8703 | 0.0787 | 0.7272 | 0.1142 | 0.8268 | 1.6781 | 0.7905 | 1.8455 |
Normalization | ||||||||
Max | 0.8741 | 0.0776 | 0.7225 | 0.1152 | 0.8340 | 1.6428 | 0.7914 | 1.8416 |
Mean | 0.8328 | 0.0894 | 0.6681 | 0.1260 | 0.8037 | 1.7863 | 0.7816 | 1.8844 |
Range | 0.9163 | 0.0633 | 0.7413 | 0.1112 | 0.8460 | 1.5821 | 0.7829 | 1.8787 |
Prediction results of PLSR models for acid and peroxide values from the highest
Tables
Performance of ANN models for acid value predictions.
Nodes | Training | Validation | Test | ||||||
---|---|---|---|---|---|---|---|---|---|
|
RMSE | Elapsed time (sec) |
|
RMSE | Elapsed time (sec) |
|
RMSE | Elapsed time (sec) | |
50 | 0.8440 | 0.0847 | 1 | 0.6001 | 0.1586 | 1 | 0.6764 | 0.1177 | 1 |
70 | 0.8942 | 0.0745 | 1 | 0.7728 | 0.1008 | 1 | 0.6572 | 0.1181 | 1 |
90 | 0.9037 | 0.0694 | 1 | 0.8175 | 0.0963 | 1 | 0.8555 | 0.1112 | 1 |
110 | 0.9374 | 0.0611 | 2 | 0.8393 | 0.0779 | 2 | 0.6777 | 0.0935 | 2 |
Performance of ANN models for peroxide value predictions.
Nodes | Training | Validation | Test | ||||||
---|---|---|---|---|---|---|---|---|---|
|
RMSE | Elapsed time (sec) |
|
RMSE | Elapsed time (sec) |
|
RMSE | Elapsed time (sec) | |
50 | 0.8439 | 1.5760 | 1 | 0.8266 | 1.9282 | 1 | 0.6458 | 2.1710 | 1 |
70 | 0.9388 | 0.9862 | 1 | 0.8304 | 1.6262 | 1 | 0.8140 | 2.0883 | 1 |
90 | 0.9210 | 1.1986 | 1 | 0.9341 | 0.9806 | 1 | 0.8286 | 1.5867 | 1 |
110 | 0.9195 | 1.1220 | 2 | 0.8224 | 1.8588 | 2 | 0.8159 | 1.7379 | 2 |
Graphs of the best results for the prediction of acid and peroxide values (90 nodes). (a) Acid value. (b) Peroxide value.
In this study, we developed prediction models and estimated the acid and peroxide values of perilla seed oil to nondestructively estimate the rancidity in conjunction with the storage conditions by using NIR spectroscopy and multivariate analysis methods. These methods have some merits in reducing time-consuming repetitive experiments and supporting reliable results by minimizing the errors that may originated from the operator’s skills. It is generally known that the NIR spectroscopy technique we used in the study can precisely and quantitatively estimate the optical response of functional groups for a target molecular structure through multivariate analysis methods. We employed several multivariate analysis methods, such as PCR, PLSR, and ANN, and it was determined that the prediction results for ANN models for both the acid value (
The NIR spectroscopy technique used in this study has been commonly used in both qualitative and quantitative analyses of target samples by considering the vibrational energies of molecules, and the low optical absorbance characteristics of near-infrared rays facilitate deeper penetration than those of mid-infrared rays. In spite of its merits, NIR spectroscopy has limited ability to resolve noise, which is induced from measurement environments and the overlap tendencies of NIR spectra. A great deal of research is being performed in order to employ possible preprocessing methods and develop proper multivariate analysis models. We expect that our application of the PCR, PLSR, and ANN multivariate analysis methods for the nondestructive evaluation of the rancidity of perilla seed oil by using NIR spectroscopy techniques has great potential for use in the component analysis of agricultural products and foods, as well as many other scientific areas, such as life sciences and biomaterial researches.
The authors declare that they have no competing interests.
This work was supported by the National Research Foundation of Korea (Grant 2017R1D1A1A02019090).