1-s2.0-s0958694601000395-main

Upload: yolby-milena-rodriguez-ariza

Post on 04-Apr-2018

221 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/31/2019 1-s2.0-S0958694601000395-main

    1/6

    International Dairy Journal 11 (2001) 9398

    Moisture, solids-non-fat and fat analysis in butter

    by near infrared spectroscopy

    Mar!a Hermidaa, Jose M. Gonzaleza, Mar Sanchezb, Jose L. Rodriguez-Oterob,*aLaboratorio de Mouriscade, Diputaci!on Provincial de Pontevedra, 36500 Lal n (Galicia), Spain

    bFacultad de Veterinaria, Instituto de Investigaci!on y An !alisis de Alimentos, Universidad de Santiago, 27002 Lugo (Galicia), Spain

    Received 10 July 2000; accepted 25 February 2001

    Abstract

    Moisture, solids-non-fat and fat were analysed in butter by near infrared (NIR) spectroscopy without any previous sample

    treatment. A set of 102 samples was used to calibrate the instrument by modified partial least-squares regression. The following

    statistical results were achieved: standard error of calibration (SEC)=0.192 and square correlation coefficient R2 0:90 formoisture, SEC=0.086 and R2 0:93 for solids-non-fat, and SEC=0.168 and R2 0:94 for fat. To validate the calibrationperformed, a set of 24 butter samples was used. Standard errors of validation were 0.255, 0.071 and 0.377 for moisture, solids-non-

    fat and fat, respectively, and R2 for the regression of measurements by reference method versus NIR analysis were 0.83, 0.94 and

    0.72 for moisture, solids-non-fat and fat, respectively. To compare, for all three components of the validation set, the results

    obtained by NIR spectroscopy with those obtained by the reference methods, linear regression and paired t-test were applied

    and the methods did not give significantly different results for p 0:05. When mean square prediction error analysis was applied, itcould be concluded that a strong calibration model was obtained and that for all three components the main error was unexplained.

    # 2001 Elsevier Science Ltd. All rights reserved.

    Keywords: NIR spectroscopy; Butter analysis

    1. Introduction

    At present, with the great increase in food manufac-

    ture and trade, a large number of analyses of both raw

    materials and finished products are in demand. The

    classical analytical methods are slow, expensive and

    need highly qualified staff; therefore these methods are

    not sufficiently effective in satisfying the current needs of

    the food industry.

    The main advantages of near infrared (NIR) spectro-

    scopy for food analysis lies in its speed, no or little

    sample pre-treatment and the avoidance of the use of

    chemicals (Osborne, Fearn, & Hindle, 1993). NIR

    spectroscopic analysis of milk (Albanell, C !aceres, Caja,

    Molina, & Gargouri, 1999; Laporte & Paquin, 1999),

    cheese (Rodr!guez-Otero, Hermida, & Cepeda, 1995;

    Sorensen & Jepsen, 1998a, b; Wittrup & Norgaard,

    1998) and other dairy products (Rodr!guez-Otero &

    Hermida, 1996; Rodr!guez-Otero, Hermida, & Centeno,

    1997; Laporte & Paquin, 1998) is a very useful technique

    for the determination of major components in milk and

    dairy products. However, little work has been done in

    butter analyses; in the literature reviewed only one

    article was found (Weaver, 1984). Measurements at 20

    different wavelengths with equipment fitted with optical

    filters were described in this article, and calibration was

    performed by multiple linear regression (MLR) for

    moisture. Evaluation of data on fat and solids-non-fat

    was not, however, included.

    Moisture determination is the most important control

    in butter production, especially in modern factories

    where hundreds of tons may be produced daily. The

    economy of manufacture demands that the final

    products be as near as possible to the target value. But

    in addition to moisture, solids-non-fat and fat limits are

    specified in butter regulations; therefore these two

    parameters must be analysed routinely.

    Since studies on butter analysis with modern full

    scanning instruments have not been found in the

    literature, the aim of this work is the application of

    NIR spectroscopy to the analysis of all three major*Corresponding author. Tel.: +34-982-252231.

    E-mail address: [email protected] (J.L. Rodriguez-Otero).

    0958-6946/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved.

    PII: S 0 9 5 8 - 6 9 4 6 ( 0 1 ) 0 0 0 3 9 - 5

  • 7/31/2019 1-s2.0-S0958694601000395-main

    2/6

    components of butter: moisture, solids-non-fat and fat,

    without any previous sample treatment. We also study

    the effectiveness of different multivariate analysis

    techniques (Martens & Naes, 1993).

    2. Materials and methods

    2.1. Samples

    A total of 126 butter samples were divided into two

    groups: 102 samples for calibration and 24 samples for

    validation of the calibration performed. Samples of

    salted as well as unsalted butter were included in both

    groups. Samples were supplied by the principal factories

    in, and purchased from the markets of Galicia (NW

    Spain) and Portugal.

    2.2. Reference analysis

    Water, solids-non-fat and fat contents were deter-

    mined in accordance with International Dairy Federa-

    tion (IDF) Standard 80 (1977). The water content was

    determined by drying a known mass of butter at

    102 28C and weighing it afterwards to determine the

    loss of mass. The solids-non-fat content was determined

    after extracting the fat from the dried butter with light

    petroleum (boiling range between 308C and 608C) and

    weighing the residue. The fat content was calculated as a

    percentage by subtracting the percentages of water

    content and solids-non-fat content from 100.

    2.3. NIR analysis

    A wavelength scanning instrument, NIRSystems 6500

    (NIRSystems, Silver Spring, MD), with a scanning

    range from 400 to 2500 nm and wavelength increments

    of 2 nm was used. Instrument checks recommended by

    the manufacturer were performed daily prior to use.

    Samples were analysed at room temperature (approxi-

    mately 208C) in a closed 30 mm thick cuvette, using the

    transport module. Reflectance measurements of mono-

    chromatic light were made from 1108 to 2492 nm. The

    means of 25 spectra scans were taken for each sample;data was recorded as log 1=R, where R is the reflectanceenergy.

    2.4. NIR calibration

    ISI software was used (ISI software, 1992). Scatter

    correction was performed by standard normal variate

    transformation (SNV) and detrend method (Barnes,

    Dhanoa, & Lister, 1989) and by multiplicative scatter

    correction (MSC) (Geladi, McDougall, & Martens,

    1985).

    A general Mahalanobis distance (H statistic) was

    calculated from principal components analysis (PCA)

    scores. The H values were standardized by dividing

    them by the average H value for the calibration file. If

    a new spectrum sample differed by more than 3.0

    standardized units from the mean spectrum of the

    calibration file, the sample was defined as a global Houtlier, liable to give inaccurate predictions.

    The calibrations were performed by MLR, principal

    components regression (PCR) and modified partial

    leasts-square (MPLS) regression (Martens & Naes,

    1993) using first and second derivatives of the spectra

    (Meuret, Dardenne, Biston, & Poty, 1993).

    * The first derivative of the spectra was calculated

    using a subtraction gap and smoothing segment of 4

    data points (1, 4, 4).* The second derivative of the spectra was calculated

    using a subtraction gap and smoothing segment of 6

    data points (2, 6, 6).

    The optimal number of terms for the calibration

    minimising overfitting was based on the standard error

    of cross-validation (SECV). The approach used was as

    follows: 80% of the samples from the calibration set

    were used for calibration, and for the remaining 20%

    the standard error of prediction (SEP) was calculated.

    This operation was carried out a total of 5 times, each

    time using a different group for calibration and

    prediction. The SECV was calculated as the square root

    of the average of the squares of the 5 SEP. The final

    calibration equation was developed with the totalsamples of the calibration set using the number of

    factors with the lowest SECV.

    The standard error of calibration (SEC) was calcu-

    lated, and the critical T value for eliminating outliers

    was fixed at 2.5 (T=residual/SEC).

    To check the calibration performed the validation set

    was used. The standard error of validation (SEV) and

    square correlation coefficient (R2) of reference analyses

    versus NIR values were calculated. Statistical errors

    were calculated in accordance with Workman (1992).

    3. Results and discussion

    3.1. Spectral information and butter samples composition

    Fig. 1 shows the mean spectra of butter in the

    calibration set. The following bands can be observed:

    at 1210 nm the second overtone of the CH stretch of

    CH2 groups; at 1450 nm the first overtone of the OH

    stretch; at 1728 and 1764 nm the first overtone of the

    CH stretch of CH2 groups; at 1940 nm a combination

    of the second overtone of the OH bend and stretching

    band of water; at 2310 nm the second overtone of the

    M. Hermida et al. / International Dairy Journal 11 (2001) 939894

  • 7/31/2019 1-s2.0-S0958694601000395-main

    3/6

    CH bend of CH2 groups; at 2350 nm a stretch and bend

    combination band of CO (Osborne et al., 1993).

    Table 1 shows the range of chemical composition and

    standard deviation (Sd) of butter samples in the

    calibration and validation sets.

    3.2. Repeatability

    Repeatability standard deviation (Sr) was calculatedfor reference methods and for NIR spectroscopy in

    accordance with IDF Standard 128 (1985) by analysing

    in duplicate 12 butter samples. Table 2 shows that for

    the repeatability standard deviation better values were

    achieved in the NIR spectroscopy analysis than by using

    the reference methods. This is certainly due to the

    simplicity of the instrumental method in comparison

    with the reference methods.

    3.3. Calibration

    The calibration set was selected with the aim of

    achieving a strong calibration for butter with and

    without added salt by maximising variability among

    samples composition and obtaining a wide range of

    spectra to avoid H outliers in the validation set.

    Therefore, the mean value for H statistic for the

    validation set of samples was 1.15 with respect to the

    mean value of the calibration set and no H outliers

    were found. Consequently, no statistical differences were

    found between the spectra of the validation and

    calibration sets, indicating that the calibration set was

    wide enough.

    In order to evaluate the different calibrations, the

    SEC, SECV and R2 of the calibration set, and the bias,

    SEV, R2 and mean square prediction error (MSPE) of

    the validation set were evaluated. Although 102 samples

    were used for calibration, T outliers were eliminated

    and the final number of samples used to calibrate each

    parameter is shown in Table 3.

    MPLS regression achieved the best calibration equa-

    tions in all cases. Relatively small differences were found

    for the statistical results of calibrations, when compar-

    ing the use of SNV and detrend method with MSC for

    scatter correction of radiation and when using first or

    second derivative of the spectra. Therefore SNV and

    detrend and first derivative were chosen for all three

    calibrations, with the aim of simplifying the discussion.

    Fig. 1. Mean near infrared spectrum of butter samples.

    Table 1

    Range of chemical composition (% w/w) and Sd of butter in the

    calibration and validation sets

    Set Moisture Solids-non-fat Fat

    Range Sd Range Sd Range Sd

    Calibration 13.0117.05 0.675 1.143.12 0.321 81.0985.52 0.761

    Validation 13.8616.44 0.604 1.292.60 0.282 81.3184.34 0.617

    Table 2

    Repeatability standard deviation (% w/w) of determination of

    moisture, solids-non-fat and fat by reference and NIR methods

    Method Moisture Solids-non-fat Fat

    Reference 0.052 0.060 0.085

    NIR 0.028 0.012 0.023

    M. Hermida et al. / International Dairy Journal 11 (2001) 9398 95

  • 7/31/2019 1-s2.0-S0958694601000395-main

    4/6

    Table 3 shows the number of PLS terms, SEC, SECV

    and R2 values for moisture, solids-non-fat and fat for

    the calibration set. The number of PLS terms is not too

    high, less than 1 per 10 samples of the calibration set.

    Therefore, no overfitting should be expected, the SEC

    values are only about 20% lower than those of the

    SECV for moisture and solids-non-fat and 30% for fatR2 values were between 0.90 and 0.94. These figures can

    be considered acceptable, bearing in mind that butter is

    a product with a very narrow range of composition

    variation, owing to the high level of standardisation in

    butter factories.

    The best values for the SEV and R2 of the validation

    set (Table 4) were those obtained for solids-non-fat,

    perhaps because of its higher variation range due to the

    inclusion of samples of salted butter, which helps to

    broaden the range of this parameter. On the other hand,

    the worst SEV and R2 values were those of fat, in our

    opinion, due to the fact that in the fat determination by

    the reference method two errors are accumulated, as a

    consequence of subtracting the percentages of water and

    solids-non-fat from 100. The bias value is low for

    moisture and zero for solids-non-fat and fat.

    To compare the results obtained by NIR spectroscopyfor all three components of the validation set with those

    obtained by the reference methods, linear regression and

    paired t-test were applied (IDF, Standard 128, 1985):

    (i) When calculating the slope and intercept of NIR

    values versus reference values, no statistical differ-

    ences (p 0:05) were found from the theoreticalvalues 1.00 and 0.00, respectively.

    (ii) The calculated t values were lower than the

    theoretical t values (p 0:05).

    Therefore the null hypothesis was retained: the twomethods did not give significantly different results.

    Graphic comparisons between reference values and

    NIR predicted values of the global validation set are

    shown in Figs. 24.

    The MSPE is the sum of three types of errors

    (Dhanoa, Lister, France, & Barnes, 1999): errors in

    central tendency, errors due to regression and errors

    due to uncontrolled disturbance or unexplained errors

    (Table 5). Errors in central tendency are also known

    as mean bias; this error is very small in percentage

    terms for all three components analysed; the highest

    value was for moisture, where the bias value was also the

    highest.

    The errors linked to regression will be equal to zero

    when the slope of regression is unity. This error also

    accounts for a small proportion of the MSPE and in the

    Table 3

    Statistical data for the calibration set (% w/w)

    Component N PLS terms SEC R2 SECV

    Moisture 96 6 0.192 0.90 0.242

    Solids-non-fat 99 8 0.086 0.93 0.103

    Fat 93 8 0.168 0.94 0.240

    Table 4

    Statistical data for the validation set (% w/w)

    Component N Bias SEV R2

    Moisture 24 0.04 0.255 0.83

    Solids-non-fat 24 0.00 0.071 0.94

    Fat 24 0.00 0.377 0.72

    Fig. 2. Validation set: moisture content. Reference method versus NIR.

    M. Hermida et al. / International Dairy Journal 11 (2001) 939896

  • 7/31/2019 1-s2.0-S0958694601000395-main

    5/6

    case of fat, where the slope is not far from unity, for

    only 0.40%. Therefore it can be stated that a strong

    calibration model was obtained.

    The unexplained error accounts for a high percentage

    of the MSPE, more than 87% for all three components

    and for fat even nearly 100%.

    Fig. 3. Validation set: solids-non-fat content. Reference method versus NIR.

    Fig. 4. Validation set: fat content. Reference method versus NIR.

    Table 5

    Mean square prediction error and its components (percentage in parenthesis) calculated in accordance with Dhanoa et al. (1999)

    Component MSPE Errors in central tendency Errors due to regression Unexplained error

    Moisture 0.0680 1.610-4 0.0070 0.0600

    (2.35%) (10.31%) (87.34%)

    Solids-non-fat 0.0053 6.2510-6 0.0007 0.0046

    (0.12%) (12.71%) (87.17%)

    Fat 0.1300 2.110-5 0.0005 0.1300

    (0.02%) (0.40%) (99.58%)

    M. Hermida et al. / International Dairy Journal 11 (2001) 9398 97

  • 7/31/2019 1-s2.0-S0958694601000395-main

    6/6

    4. Conclusions

    NIR spectroscopy is an adequate technique for the

    analysis of moisture, solids-non-fat and fat in butter

    without any previous sample treatment. Owing to the

    simplicity of the instrumental method, better values of

    repeatability were achieved by NIR spectroscopy thanby the reference methods. The best calibrations were

    performed through MPLS regression. After MSPE

    analysis it can be concluded that a strong calibration

    model was obtained and that the main source of error

    was an unexplained error, accounting for more than

    87% for all three components.

    References

    Albanell, E., C!aceres, P., Caja, G., Molina, E., & Gargouri, A. (1999).

    Determination of fat, protein and total solids in ovine milk bynear-infrared spectroscopy. Journal of AOAC International, 82,

    753758.

    Barnes, R. J., Dhanoa, M. S., & Lister, S. J. (1989). Standard normal

    variate transformation and detrending of near infrared diffuse

    reflectance spectra. Applied Spectroscopy, 43, 772777.

    Dhanoa, M. S., Lister, S. J., France, J., & Barnes, R. J. (1999). Use of

    mean square prediction error analysis and reproducibility measures

    to study near infrared calibration equation performance. Journal of

    Near Infrared Spectroscopy, 7, 133143.

    Geladi, P., McDougall, D., & Martens, H. (1985). Linearization and

    scatter correction for near infrared reflectance spectra of meat.

    Applied Spectroscopy, 39, 491500.

    IDF Standard 80. (1977). Butter}determination of water, solids-non-

    fat and fat contents on the same test portion. Brussels, Belgium.

    IDF Standard 128. (1985). Definition and evaluation of the overallaccuracy of indirect methods of milk analysis}application to

    calibration procedure and quality control in dairy laboratory.

    Brussels, Belgium.

    ISI Software. (1992). ISI NIRS 2: Software for near infrared

    instruments. Intrasoft International, Port Matilda, PA, USA.

    Laporte, M. F., & Paquin, P. (1998). Near-infrared technology and

    dairy food products analysis: A review. Seminars in Food Analysis,

    3, 173190.

    Laporte, M. F., & Paquin, P. (1999). Near-infrared analysis of fat,

    protein and casein in cows milk. Journal of Agricultural and Food

    Chemistry, 47, 26002605.Martens, H., & Naes, T. (1993). Multivariate calibration. Chichester,

    UK: Wiley.

    Meuret, M., Dardenne, P., Biston, R., & Poty, O. (1993). The use of

    NIR in predicting nutritive value of Mediterranean tree and shrub

    foliage. Journal of Near Infrared Spectroscopy, 1, 4554.

    Osborne, B. G., Fearn, T., & Hindle, P. H. (1993). Practical near

    infrared spectroscopy with applications in food and beverage analysis

    (pp. 2833). Harlow, UK: Longman Scientific & Technical.

    Rodr!guez-Otero, J. L., & Hermida, M. (1996). Analysis of fermented

    milk products by near-infrared reflectance spectroscopy. Journal of

    AOAC International, 79, 817821.

    Rodr!guez-Otero, J. L., Hermida, M., & Centeno, J. (1997). Analysis

    of dairy products by near infrared spectroscopy: A review. Journal

    of Agricultural and Food Chemistry, 45, 28152820.

    Rodr!

    guez-Otero, J. L., Hermida, M., & Cepeda, C. (1995).Determination of fat, protein, and total solids in cheese by near-

    infrared spectroscopy. Journal of AOAC International, 78, 802806.

    Sorensen, L. K., & Jepsen, R. (1998a). Comparison of near infrared

    spectroscopy techniques for determination of semi-hard cheese

    constituents. Milchwissenschaft, 53, 263267.

    Sorensen, L. K., & Jepsen, R. (1998b). Assessment of sensory

    properties of cheese by near-infrared spectroscopy. International

    Dairy Journal, 8, 863871.

    Weaver, R. W. V. (1984). Near infrared reflectance analysis applied to

    dairy products. In Abstract of papers; Challenge to contemporary

    dairy analytical technique. Reading, UK: Royal Society of

    Chemistry.

    Wittrup, Ch., & Norgaard, L. (1998). Rapid near infrared spectro-

    scopy screening of chemical parameters in semi-hard cheese using

    chemometry. Journal of Dairy Science, 81, 18031809.Workman (1992). NIR spectroscopy calibration basis. In D. A. Burns,

    E. W. Gurczak (Eds.), Handbook of near-infrared analysis

    (pp. 270278). New York: Marcel Dekker Inc.

    M. Hermida et al. / International Dairy Journal 11 (2001) 939898