XU Jing，HOU Pei-guo，WANG Yu-tian，PAN Zhao

Measurement Technology and Instrumentation Key Lab of Hebei Province, Yanshan University, Qinhuangdao 066004, China

Determination of Component Contents of Blend Oil Based on Characteristics Peak Value Integration

XU Jing，HOU Pei-guo*，WANG Yu-tian，PAN Zhao

Measurement Technology and Instrumentation Key Lab of Hebei Province, Yanshan University, Qinhuangdao 066004, China

Edible blend oil market is confused at present. It has some problems such as confusing concepts, randomly named, shoddy and especially the fuzzy standard of compositions and ratios in blend oil. The national standard fails to come on time after eight years. The basic reason is the lack of qualitative and quantitative detection of vegetable oils in blend oil. Edible blend oil is mixed by different vegetable oils according to a certain proportion. Its nutrition is rich. Blend oil is eaten frequently in daily life. Different vegetable oil contains a certain components. The mixed vegetable oil can make full use of their nutrients and make the nutrients more balanced in blend oil. It is conducive to people’s health. It is an effectively way to monitor blend oil market by the accurate determination of single vegetable oil content in blend oil. The types of blend oil are known, so we only need for accurate determination of its content. Three dimensional fluorescence spectra are used for the contents in blend oil. A new method of data processing is proposed with calculation of characteristics peak value integration in chosen characteristic area based on Quasi-Monte Carlo method, combined with Neural network method to solve nonlinear equations to obtain single vegetable oil content in blend oil. Peanut oil, soybean oil and sunflower oil are used as research object to reconcile into edible blend oil, with single oil regarded whole, not considered each oil’s components. Recovery rates of 10 configurations of edible harmonic oil is measured to verify the validity of the method of characteristics peak value integration. An effective method is provided to detect components content of complex mixture in high sensitivity. Accuracy of recovery rats is increased, compared the common method of solution of linear equations used to detect components content of mixture. It can be used in the testing of kinds and content of edible vegetable oil in blend oil for the food quality detection, and provide an effective reference for the creation of national standards.

Spectroscopy；Characteristics peak value integration；Quasi-Monte Carlo；Edible blend oil

Biography：XU Jing, (1989-), female, PhD, Measurement Technology and Instrumentation Key Lab of Hebei Province, Yanshan University e-mail: ysuxujing@163.com *Corresponding author e-mail: pghou@ysu.edu.cn

Introduction

With the improvement of residents’ quality of life, the demand for edible oil nutritional value is also higher. Blend oil contained a variety of vegetable oil has a rich nutrition structure. It is more and more popular with people. However, blend oil market is confused at present. It has the phenomenon of randomly named and raw material proportioning is unknown.

The drafting of corresponding national standards started from 2005, but it still fail to come out. Its basic reason lies in the lack of effective measuring method of vegetable oil content in the blend oil. Qualitative check is easier and it can be implemented at present. However quantitative detection method is more difficult to achieve, which is the main factor that affect the national standards to come on schedule.

Even requirements of contents of vegetable oil in the blend oil marked in the national standard, but due to lack of related detection method, they can’t be measured accurately, so it is not feasible. Therefore, a fast effective mean for vegetable oils in blend oil is necessary. Fluorescence technique is often used to study the properties of the material characteristics[1，2].Because of a variety of the different concentration fluorescent ingredients in edible vegetable oils, such as: fat hydrocarbon, vitamin[3，4]and so on, so it is feasible for the measurement of the type and content of vegetable oils by fluorescence technique. A method with higher measuring sensitivity of data analysis is proposed for three dimensional spectra based on Quasi-Monte Carlo method, combined with the neural network method. It improves measuring sensitivity in numeric and is sensitive to changes of the content.

1 Theory

When the fluorescent substance molecules are excitated by ultraviolet, they will produce electron transition. The excited electron is unstable. When they returned to the ground state with the release of energy, they will produce fluorescence. According to the Lambert-Beer’s law, the intensity of light through the solution has a quantitative relationship with the concentration of test solution, and the material fluorescence intensity has a proportional relationship with the excitation intensity which can be absorbed. The combination of above two can deduce the relation between the intensity of incident light and fluorescence intensity. The result can be seen from Eq.(1). This is the theoretical foundation of the measurement of solute concentration by fluorescence intensity.

If=KI0(1-10-εcl)

(1)

Where，Ifis fluorescence intensity;Kis proportional coefficient about spectrometer;I0is the intensity of incident light.εis molar absorption coefficient；cis solution concentration;lis optical path length from the incidence to the exit.

Whenεcl≤0.05, high order terms are omit through the Lagrange method. The result can be seen from Eq.(2).

If=2.3KI0εcl=K′c

(2)

Integralexpressionofhigh-dimensionalspaceisshownasEq.(3).Thecalculationisdifficulty.TheprincipleofQuasi-Monte-Carlointegrationistoturnintegralvaluesinthehigh-dimensionalspaceintosumcalculation.Thespecificimplementationprocessis:firstofallproduceuniformpointseriesinthespaceofDm(volumeisSD), and then calculate the sum of function on discrete points [Eq.(4)], and use this sum as a result of integral approximation.

(3)

(4)

The three dimensional spectral representation is commonly used as the excitation wavelength, emission wavelength and fluorescence intensity data matrix. Due to different kinds and content of different fluorescent material contained in the vegetable oil, its corresponding fluorescence intensity matrix is also different. Three dimensional spectra contain large amount of information. The commonly used treatment method is for dimension reduction, which to a certain extent, will lose information. Uniform distribution point series are produced used for Quasi-Monte-Carlo integral on the selected excitation wavelength-emission wavelength planar regions. Take cumulative sum of third power of the corresponding intensity values at point series as the corresponding peak area integration-the characteristic peaks of integral. Because the fluorescence intensity slope is expanded to third power, it improves the detection sensitivity for density change. At the same time, because the sum of the third power of fluorescence intensity at produced super uniform point series, to a certain extent it can suppress random noise influence on measurement results. In order to obtain high accuracy of measurement, integral areas are selected in each constituent with strong fluorescence peak and different fluorescent peak from the rest of the components. Finally contents of each single component in the mixture are obtained by using the neural network method to solve nonlinear equations.

Specific analysis steps are shown as follows:

Preparation step: The algorithm is based on the linear relationship of the fluorescence intensity and concentration as the theoretical part mentioned above, so if some interference factors, such as scattered light and inner filter effect, contained in analysed data set makes the data set not meet tri-linear relatioship, then some preparation steps are needed to make the data set accord with tri-linear features.

Step 1: The three dimensional fluorescence spectra of the single plant oils have been measured. By observing selected representative characteristic peak of integral areaSD. Generate ultra evenly distributed point series in the area of the selected, and divide the selected area into small grids. Calculate the exponential integral value of unit concentration and crossing concentration fluorescence intensity on each grid.

Step 2: Determine the three dimensional fluorescence spectra of the blend oil under test. Calculate characteristic peak integral value in each grid selected above.

Step 3: List the nonlinear equations of concentration by the fluorescence intensity additivity and integral linear additivity.

Step 4: Simultaneous nonlinear equations in each of the selected area, based on the type and degree of spectral overlap between mixture different weights are given to different grid, using the method of neural network solving nonlinear equations. The solutions of equations systems are the concentrations of each component in the mixture. For a certain composition of blend oil, good weights can be obtained through neural network training. In order to reduce the number of nonlinear equations, and improve the operation speed, the different characteristic peaks of the same vegetable oil can be given weights to compose a combination characteristic peak represented for one vegetable oil.

2 Experimental and result

Buy pure vegetable oils (peanut oil, soybean oil, sunflower oil) from the market, and mix them according to a certain proportion to verify the algorithm is effective to the blend oil of different components.

A portable optical fiber edible blend oil fluorometer is used here, with filter wheel as spectral components, with 49 optical fiber bundle as the light components, including 36 roots of transmission optical fiber of excitation light, 4 roots of reference transmission optical fiber and 9 roots transmission optical fiber of fluorescence. Excitation light source is short arc pulse xenon lamp. Excitation light after collimating lens group focuses on the excitation light filter roulette. The filter wheel turns by drive motor and make the needed wavelength of exciting light get to fiber bundle. Part of light is taken as the reference light beam though optical fiber after photoelectric conversion for weak signal detection, while other part of the light into the optical fiber probe for fluorescence measurement. Test fluorescence though optical fiber probe to launch filter roulette, uses for photoelectric conversion. After the weak signal detection, fluorescence light together with the reference light goes into CPU for subsequent analysis. CPU controls light source driver circuit. Instrument has keys on it to the display and data storage, to output and to display the results of the analysis at the same time. The communication interface uses for data transmission. Portable fiber edible blend oil fluorometer system structure diagram is shown in figure 1.

Three-dimensional fluorescence spectra and fingerprint spectra of sunflower oil, soybean oil, peanut oil, diagram are shown in Fig.2 to Fig.4.

Fig.1 Measurement system structure

Fig.2 3-dimensional fluorescence spectra and fluorescence contour spectrum of sunflower oil

Dreparation steps are applied to make the data set accord with tri-linear features.

Integral areas are selected as 350～353/430～433 nm，355～364/391～400 nm，377～380/467～470 nm. The second area in three integral areas is divided into small grids of 3 nm×3 nm. The area 1 and 3 are taken as two small grids. Make characteristics peak value integration in each grid. List the nonlinear equations of the concentration. Select suitable weights for each nonlinear equation. Solve the equations using neural network method. Calculate concentration using the commonly used method of solving linear equations in the same small grid. The corresponding recovery rates are shown in table 1.

Fig.3 3-dimensional fluorescence spectra and fluorescence contour spectrum of soybean oil

Fig.4 3-dimensional fluorescence spectra and fluorescence contour spectrum of peanut oil

Table 1 Concentration and recovery rate of each vegetable oil in blend oil

It can be seen from the table that forecast effect of characteristics peak integral method is good. The recovery rates are ideal. Calculation accuracy is higher. The characteristic peak integral method can be used for determination of single kind of oil in blend oil with a simple calculation.

3 Conclusion

Characteristic peak integral method is a kind of treatment method of three dimensional spectra based on Quasi-Monte-Carlo method, combined with the neural network method to solve the nonlinear equations. It can accurately measure the contents of each component in the mixture. Different meshing in selected characteristics regions and different weights given to each grid will directly affect the accuracy of analysis. Ideal weight for specific components of the blend oil can be got through a lot of training. Take sunflower oil, soybean oil and peanut oil as the research object, as a whole regardless of the composition of fat hydrocarbon. With their mixture as simulated blend oil, determine the contents of mixture components. The recovery rates are good for using to test in the complex components in the mixture. It has higher recovery rates than the method of solving linear equations often used to measure a variety of mixture components. The optimal weights for a particular form of blend oil can be determined through neural network training, for testing blend oil in the market in the future. It provides an effective means for measurement of vegetable oil contents in the blend oil.

[1] Li Ren, Chen Guoqing, Zhu Chun, et al. Spectroscopy and Spectral Analysis，2014, 34(1): 111.

[2] Mahato P, Saha S, Suresh E, et al. Inorganic Chemistry, 2011, 51(3): 1769.

[3] Zhao Shoujing, Chen Bin, Lu Daoli. Journal of the Chinese Cereals and Oils Association, 2012, 27(3): 104.

[4] Zhang Yong, Cheng Yao, Yan Yu-tong, et al. Laser Technology，2013, 37(1): 109.

*通訊聯(lián)系人

O657.3

A

基于特征峰值積分法的調(diào)和油組分含量的測量

徐 婧，侯培國*，王玉田，潘 釗

燕山大學電氣工程學院，河北省測試計量技術(shù)與儀器重點實驗室，河北 秦皇島 066004

目前食用調(diào)和油市場混亂，存在混淆概念、隨意冠名、以次充好等問題，特別是調(diào)和油成分和配比標準模糊不清。國家食用調(diào)和油標準歷經(jīng)八年仍未能如期出臺，其根本原因在于缺乏對調(diào)和油中植物油定性及定量檢測的有效方法。食用調(diào)和油是由不同的植物油按一定比例混合而成，含有豐富的營養(yǎng)成分，在日常生活中經(jīng)常使用。不同的植物油含有特定的組成成分，將各種植物油進行混合可以充分利用其中的營養(yǎng)物質(zhì)，使調(diào)和油中營養(yǎng)成分更加均衡，有利于人的身體健康。因此準確測定調(diào)和油中單一植物油的含量可以有效對調(diào)和油市場進行監(jiān)管。同時由于調(diào)和油中植物油種類是確定的，僅需對其含量進行準確測定。利用三維熒光光譜對調(diào)和油中植物油含量進行測定，提出一種新的數(shù)據(jù)處理方法，采用擬蒙特卡洛原理對選定的特征區(qū)域進行特征峰積分，結(jié)合神經(jīng)網(wǎng)絡(luò)方法求解非線性方程組，得出調(diào)和油中各單一植物油含量。選用花生油，大豆油，葵花油為研究對象，用不同比例單一植物油調(diào)和成食用調(diào)和油，不考慮每種單一植物油的具體組分，僅將其作為整體研究。通過測定10組不同調(diào)和比例的調(diào)和油的回收率，驗證特征峰積分法的有效性，為高靈敏度檢測混合物復(fù)雜組分含量提供一種有效方法,與常用的解線性方程組測定混合物組分濃度的方法進行比較，回收率的準確度提高，可以用于食品質(zhì)量檢測人員對食用調(diào)和油中所用植物油種類及含量進行檢測，為國家標準的出臺提供一種有效參考。

光譜學；特征峰峰值積分；擬蒙特卡洛；食用調(diào)和油

2014-09-25，

2015-01-20)

2014-09-25; accepted：2015-01-20

Project supported by the National Natural Science Foundation of China (61471312) and the National Natural Science Foundation of Hebei (F2012203189，F(xiàn)2015203072, C2014203212)

10.3964/j.issn.1000-0593(2016)01-0298-05