Use of FT-NIR spectrometry in non-invasive measurements of internal quality of ‘Fuji’ apples

Postharvest Biology and Technology 37 (2005) 65–71

Use of FT-NIR spectrometry in non-invasive measurements of internal quality of ‘Fuji’ apples Yande Liu a,b , Yibin Ying a,∗ a

College of Biosystems Engineering and Food Science, Zhejiang University, 268 Kaixuan Road, Hangzhou 310029, PR China b College of Engineering, Jiangxi Agriculture University, Nanchang 330045, PR China Received 29 August 2004; accepted 22 February 2005

Abstract This research studied the feasibility of making rapid measurements of the soluble solids contents (SSC) and acidity of ‘Fuji’ apple (Malus domestica Borkh. cv. Fuji) fruit. FT-NIR spectra were recorded in the interactance mode, using fiber optics and a special sample holder. Calibration models related the FT-NIR spectra to SSC, titratable acidity (TA) and available acidity (pH) and were developed based on partial least square (PLS) regression with respect to the logarithms of the reflectance reciprocal and its first and second derivative. The prediction performance of calibration models in different wavelength regions was also investigated. The best models gave a standard errors of prediction (SEP) of 0.455, 0.044 and 0.068, and correlation coefficients of 0.968, 0.728 and 0.831 for SSC, TA and pH, respectively, in the wavelength range of 812–2357 nm. Based on the results, it was concluded that FT-NIR spectrometry could be easy to facilitate, reliable, accurate and a fast method for non-invasive measurements of apple SSC and acidity. © 2005 Elsevier B.V. All rights reserved. Keywords: FT-NIR spectrometry; Non-invasive measurements; Apples; Fruit quality; PLS

1. Introduction Consumer acceptance of fresh or processed apples is the ultimate goal of apple breeders, food scientists and supermarket managers. Soluble solids contents (SSC) and acidity are the properties most likely to match the consumer’s perception of internal quality, but measurements of these properties are still largely destructive. ∗

Corresponding author. E-mail address: [email protected] (Y. Ying).

0925-5214/$ – see front matter © 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.postharvbio.2005.02.013

Therefore, the development of a reliable, non-invasive method for the quality measurement of apples, before harvest and at the packing site, is critical to the success of the apple industry. Near-infrared spectroscopy (NIRS) has been used as a rapid and non-invasive technique for measuring SSC and acidity of fruit. In this technique, information about the internal quality of the product by measuring the absorption of near-infrared light by functional groups and scattering at specific wavelength is obtained (Banwell, 1983). Most research concerning

66

Y. Liu, Y. Ying / Postharvest Biology and Technology 37 (2005) 65–71

the quality determination of apples has been achieved with dispersive NIR spectroscopy. These instruments use a grating to separate the individual frequencies of the radiation leaving the sample. Lammertyn et al. (1998) examined the prediction capacity of the quality characteristics such as acidity, firmness and SSC of ‘Jonagold’ apples with a wavelength range between 380 and 1650 nm. The SEP of the partial least squares (PLS) models were 0.07, 0.61 and 0.26, respectively. Peiris et al. (1999) established a relationship between the reflectance spectra (800–1650 nm) and the soluble solids content of ‘Jonagold’ apples by means of the PLS technique and obtained the correlation coefficients of 0.79–0.91. Peirs et al. (2002) performed a statistical analysis on a large spectral data set to analyze the effect of orchard, season and cultivar and found that the accuracy of the models for apple quality increased considerably when including more variability in the calibration set. McGlone et al. (2003) reported the accuracy of dry-matter (DM: r2 > 0.95; RMSEP < 0.32%) and the post-storage SSC (r2 = 0.93) predictions, based on 800–1000 nm near infrared spectrometric measurements for ‘Royal Gala’ apples. They concluded that harvest-time DM predictions were strongly correlated with the post-storage SSC. However, there is little reported about the FTNIR method used for measuring internal quality of intact fruit in the literature. Peirs et al. (2002) compared Fourier transform (FT-NIR) spectroscopy with dispersive near-infrared spectroscopy, and the instrument stability, depth of light penetration and the predictive capacity of some quality characteristics between both instruments were compared. Based on the results, it was concluded that FT-NIR reflectance spectroscopy is an interesting alternative for standard dispersive instruments for non-destructive quality measurement. Some other techniques (e.g. Defernez et al., 1995; Belton et al., 1995; Rodriguez-Saona et al., 2001) have also used FT-NIR transmission to measure the main components of fruit juices. In our research, we studied the feasibility of rapidly measuring intact apple fruit SSC, TA and pH with a fiber optic FT-NIR spectrometer. The specific objectives of our research were (1) to establish relationships between the FT-NIR measurements and the major physiological quality indices which relate to the internal quality of the ‘Fuji’ apple and (2) to compare the prediction performance of calibration models with dif-

ferent spectra treatment and wavelength ranges and to find out the optimal wavelength range and develop the best calibration models.

2. Materials and methods 2.1. Apple fruit A total of 333 ‘Fuji’ apples which came from Shandong and Shanxi orchards were purchased at a local auction and stored for 2 days at 25 ◦ C and 68% RH to equilibrate before being examined by the FT-NIR technique, 235 apples were used for the calibration models, and 98 samples were used for prediction models. All measurements including spectral collection and quality analysis were carried out on the same day. 2.2. FT-NIR measurements FT-NIR spectra were recorded on a Nexus FTNIR spectrometer (Thermo Nicolet Corporation, USA) equipped with a NIR fiber-optic probe (Type 847072200), an interferometer, an InGaAs detector, a wide band light source (50 W), and quartz halogen to provide interactance measurements. Apples were placed steadily upon the fruit holder, with the stem-calyx axis horizontal. On each apple, an interactance spectrum was measured on three opposite, equatorial positions and the averaged spectrum per apple was used for analysis. In the head of the bifurcated cable, the source and detector fibers were situated randomly, the light was guided to the sample by source fibers, and from the sample with the detector fibers to a Nexus FT-NIR spectrometer, which has a spectral range of 800–2500 nm. The mirror velocity was 0.9494 cm s−1 and the resolution was 16 cm−1 in this experiment. In order to avoid surface reflectance and guarantee subsurface penetration of the light into the apple flesh, the bifurcated optical probe was placed at a 75◦ angle to the level. 2.3. Determination of quality parameters During ripening, there is an increase in the SSC and a concurrent reduction in TA (which can be expressed as the % malic acid in the flesh). To minimize physiological changes in the fruit, SSC, TA and pH measurements were performed in the afternoon on all fruit.

Y. Liu, Y. Ying / Postharvest Biology and Technology 37 (2005) 65–71

Tissue samples of 30–40 g were then cut from each fruit separately, from the marked area close to the region in which the FT-NIR readings had been taken, and were macerated with a manual fruit squeezer (Model: HL56, Shanghai, China). Samples of the filtered juice were then taken for SSC measurement by digital refractometer (Model: WYT-4, Quanzhou, China), and TA was measured on approximately 10 g of cortical tissue and reported as percentages, by fruit fresh weight, of malic acid equivalent. The tissue was macerated and juice, filtered from the homogenate, titrated to an endpoint at pH of 8.05 using 0.05N NaOH and an autotitrator (Model: ZD-2 Titrino, Shanghai, China). A pH meter (SJ-4A, Instrument Co. Ltd., Shanghai, China) was used to measure the pH. 2.4. Data analysis The spectral data were analyzed using PLS regression with pre-processing techniques includes the use of multiplicative scatter correction, first, second derivatives and logarithmic transformations. 2.4.1. Model calibration and validation The Nicolet TQ Analyst v6.0 (Madison, WI, USA) was used for processing the data and FT-NIR models were developed for the calibration set. Processed spectral data such as the log(1/R), first and second derivative were analyzed for PLS calibration techniques. Models were formulated which related the FT-NIR spectra and the SSC, TA and pH in each tested apple. The calibration models include two validation procedures. Cross-validation was performed on the calibration samples based on excluding a certain number of observations for the calibration model. The number of latent variables in the PLS models is determined by the predictive residual error sum of squares (PRESS). By means of full cross-validation with one sample omitted, the number of latent variables is obtained according to the smallest PRESS. These excluded samples are subsequently used as the validation set. The process is repeated until all observations are left out of the calibration set once. Since the validation samples originate from the same sample set, a comparable kind of spectral variability can be expected. On the other hand, the spectral measurements used for external validation might deviate from the calibration samples as they originate from another sample

67

set. In this way, the ability of the calibration model to withstand unknown variability is assessed. The accuracy of the calibration models is defined as RMSECV for cross-validation and RMSEP for external validation (Peirs et al., 2003).
 Ip 1   RMSEP or RMSECV = (1) (ˆyi − yi )2 Ip i=1

with yˆ i = predicted value of the ith observation, yi = measured value of the ith observation, and Ip = number of observations in the prediction set. 2.4.2. Wavelength range selection The selection of the portion of the spectra to be used in calibration is crucial and determines the quality of the process. Wavelengths above 2400 nm were discarded due to the insensitivity of the InGaAs detector above this value, where the intensity spectrum drops drastically and the absorbance spectrum becomes noisy (see Fig. 1). In this work, three wavelength ranges (812–2357, 812–1100 and 1100–2357 nm) were used in the calibration process according to the FT-NIR spectrometer. These ranges include the full NIR spectral range, shortand long-wave NIR range.

3. Results and discussion 3.1. Quality parameter distributions The mean SSC, TA and pH measurements were 12.49% (S.D. = 1.82), 0.30% (S.D. = 0.07) and 3.65% (S.D. = 0.12), respectively. The SSC measurements (n = 333) were normally distributed around the mean (max = 17.60%, min = 7.60%), but the TA distribution (n = 333) was skewed with the high-end tail stretching further than the corresponding low-end tail (max = 0.50%, min = 0.13%). Among the samples analyzed for the different parameters, some of them were selected for calibration by Nicolet TQ Analyst v6.0 (Madison, WI, USA) software models and the remaining samples were used for the validation set. Some details of apple quality parameters used for the calibration and validation sets are summarized in Table 1.

68

Y. Liu, Y. Ying / Postharvest Biology and Technology 37 (2005) 65–71

Fig. 1. Typical original spectra of log(1/R) for 20 apple fruit samples.

3.2. Wavelength selection for calibration models Developing the calibration models involved deciding on how many wavelengths to use and selecting an optimal format. Wavelengths where the data were noisy and provided little predictive ability were eliminated prior to selection of regressions. In our work, three ranges were used in the calibration process, one being a subset of the other. The three ranges included were: 812–1100 nm, which was covered by Silicon sensors; 1100–2357 nm, which was covered by PbS sensors; and the third was the full NIR region 812–2357 nm, which was covered by InGaAs sensors. The calibration set was selected with the aim of providing strong calibration for SSC, TA or pH by maximizing the variability among sample composition

and obtaining a wide range of spectra to avoid outliers in the validation set. Selection of the optimum wavelength range for the best prediction model was done by PLS analysis in TQ Analyst v6.0 (Nicolet Co., USA). The SSC, TA and pH prediction results of different wavelengths ranges are presented in Table 2. The optimum wavelength range for the best prediction models of SSC, TA and pH were in the 812–2357 nm range with a correlation coefficient of determination r2 = 0.937, 0.530 and 0.691; a RMSEC of 0.475, 0.041 and 0.068; a RMSEP of 0.452, 0.043 and 0.068, respectively, and was therefore selected as the optimum wavelength range for SSC, TA and pH predictions (Table 2). In order to explain the result from PLS regression, the regression coefficient plots of the FT-NIR spectra using the wavelength region from

Table 1 Means, ranges, and S.D.s of apple samples of calibration and validation sets Parameter

Data set

Samples

Mean

S.D.

CV (%)

Range

SSC (%)

Calibration Validation

235 98

12.453 12.563

1.809 1.741

14.527 13.858

7.60–17.60 8.20–15.20

TA (g/100 g)

Calibration Validation

235 98

0.030 0.030

0.007 0.006

21.976 20.340

0.013–0.050 0.013–0.048

pH

Calibration Validation

235 98

3.651 3.639

0.121 0.118

3.320 3.235

3.400–4.030 3.420–3.990

Y. Liu, Y. Ying / Postharvest Biology and Technology 37 (2005) 65–71

69

Table 2 The results for calibration models of PLS regression methods for the log(1/R), mean values of RMSEC, RMSEP and RMSECV, the correlation coefficients of determination, r2 with 812–2357, 1100–2357 and 812–1100 nm wavelength ranges Parameter

Wavelength range (nm)

No. of factors

r2

RMSECV

RMSEC

RMSEP

SSC

812–2357 1100–2357 812–1100

10 10 8

0.910 0.910 0.861

0.541 0.540 0.673

0.475 0.490 0.485

0.452 0.444 0.615

TA

812–2357 1100–2357 812–1100

10 6 6

0.510 0.494 0.394

0.0047 0.0047 0.0052

0.0041 0.0043 0.0046

0.0043 0.0042 0.0044

pH

812–2357 1100–2357 812–1100

10 10 5

0.582 0.460 0.554

0.079 0.089 0.081

0.068 0.084 0.064

0.068 0.076 0.073

812 to 2357 nm is shown in Fig. 2. The wavelength of 2306 nm had an important role in the calibration model from Fig. 2. On the other hand, the FT-NIR system often has more than a 20 000:1 ratio of signal-to-noise which helps to detect constituents at quite low concentrations. The InGaAs detector used has sensitivity in the wavelength range of 812–2357 nm. However, FTIR systems become much noisier at the lower spectral wavelengths because of a combination of source noise, detector noise and increasing bandwidth due to uncertainties as the wavelength approaches that of the reference laser used for measuring mirror movement. By comparison, the precise nature of the relevant spectral information is unclear because the 812–1100 nm range used for predictions contains a number of different carbohydrate and water absorption bands (McGlone and Kawano, 1998). Therefore, the results showed that the calibration models in the wavelength range from 813 to 2357 nm would have higher accuracy than the other wavelength ranges (Table 2).

Fig. 2. Regression coefficient plots for PLS calibration of apple spectra measured the wavelength region of 814–2357 nm.

3.3. Calibration and prediction models PLS calculations were carried out using the log(1/R), its first derivative D1 log(1/R), and second derivative D2 log(1/R) spectra for the FT-NIR system. The calibration and prediction results are presented in Table 3. The best wavelength range was from 812 to 2357 nm, which was wider than the other wavelength ranges (810–1100 and 1100–2357 nm). Analysis of the spectral data from the intact ‘Fuji’ apple material showed that 8–10 factors were required for predicting SSC, 5–11 for TA and 5–10 for pH, depending on the wavelength range used respectively. The average inter-correlations among SSC–TA, TA–pH, and SSC–pH were 0.47, −0.57 and 0.64, respectively. The relatively low inter-correlations indicated that the prediction by the FT-NIR method is applicable.

Fig. 3. Predictions of PLS by the FT-NIR system vs. laboratory measurements of SSC of ‘Fuji’ apple fruit.

70

Y. Liu, Y. Ying / Postharvest Biology and Technology 37 (2005) 65–71

Table 3 The results for calibration models of PLS methods for the log(1/R), first derivative D1 log(1/R), and its second derivative D2 log(1/R), mean values of RMSEC, RMSEP and the coefficients of determination, r2 Parameter

Sample sets

Statistical parameter

Spectrum D1 log(1/R)

D2 log(1/R)

SSC

Calibration

No. of factors rc2 RMSEC rp2 RMSEP

10 0.931 0.476 0.937 0.452

5 0.922 0.505 0.913 0.540

7 0.923 0.500 0.925 0.512

No. of factors rc2 RMSEC rp2 RMSEP

10 0.623 0.0041 0.530 0.0043

3 0.538 0.0045 0.537 0.0043

4 0.441 0.0050 0.520 0.0044

No. of factors rc2 RMSEC rp2 RMSEP

10 0.685 0.068 0.691 0.0678

4 0.658 0.071 0.6114 0.0739

8 0.693 0.067 0.5993 0.0701

log(1/R)

Prediction TA

Calibration

Prediction pH

Calibration

Prediction

It is desirable that a model has a low error of calibration, with as few factors as possible. The selection of calibration models based on this criterion, with the total described variance considered to be appropriate (about 90%), was conducted by means of the Nicolet TQ Analyst v6.0 software. PLS prediction results for SSC, TA and pH are presented in scatter plots shown in Figs. 3–5. In all figures, the ordinate and abscissa axes represent the measured and the fitted values of the appropriate parameters, respectively. The correlation relationship between

Fig. 5. Predictions of PLS by the FT-NIR system vs. laboratory measurements of pH of ‘Fuji’ apple fruit.

the measured and predicted values of the parameters is higher for SSC than the results reported by previous research (Slaughter, 1995; Lu, 2001).

4. Conclusions

Fig. 4. Predictions of PLS by the FT-NIR system vs. laboratory measurements of TA of ‘Fuji’ apple fruit.

Predictive models based on interactance mode measurements, and using the spectral range 812–2357 nm, appear to be optimal for measuring ‘Fuji’ apple SSC, TA and pH. The log(1/R) spectra with PLS method was found to provide the best prediction of the physical

Y. Liu, Y. Ying / Postharvest Biology and Technology 37 (2005) 65–71

properties of the apple with an SEP of 0.455, 0.044 and 0.068, respectively. This is still similar to NIR prediction accuracies achieved on many other fruit and so might be good enough for some applications. The non-invasive FT-NIR measurements provided good estimates of the internal quality indices of the apple fruit, especially the SSC and the predicted values were correlated with destructively measured values for SSC and acidity.

Acknowledgement The authors wish to express their sincere thanks to the financial support from the Natural Science Foundation of China (Projects 30270763 and 60468002).

References Banwell, C.N., 1983. Fundamentals of Molecular Spectroscopy, 3rd ed. McGraw-Hill, London, UK, 323. Belton, P.S., Kemsley, E.K., McCann, M.C., Ttofis, S.R., Wilson, H., Delgadillo, I., 1995. The identification of vegetable and matter using Fourier transform infrared spectroscopy. J. Food Chem. 54, 437–441. Defernez, M., Kemsley, E.K., Wilson, R., 1995. The use of infrared spectroscopy and chemometrics for the authentication of fruit purees. J. Agric. Food Chem. 43, 109–113.

71

Lammertyn, J., Nicolai, B.M., Ooms, K., De Smedt, V., De Baerdemaeker, J., 1998. Non-destructive measurement of acidity, soluble solids and firmness of ‘Jonagold’ apples using NIRspectroscopy. Trans. ASAE 41, 1089–1094. Lu, R., 2001. Predicting firmness and sugar content of sweet cherries using near-infrared diffuse reflectance spectroscopy. Trans. ASAE 44, 1265–1271. McGlone, V.A., Jordan, R.B., Seelye, R., Clark, C.J., 2003. Drymatter—a better predictor of the post-storage soluble solids in apples. Postharvest Biol. Technol. 28, 431–435. McGlone, V.A., Kawano, S., 1998. Firmness, dry matter and soluble solids assessment of postharvest kiwifruit. Postharvest Biol. Technol. 13, 131–141. Peiris, K.H.S., Dull, G.G., Leffler, R.G., Kays, S.J., 1999. Spatial variability of soluble solids or dry-matter content within individual fruits, bulbs, or tubers: implications for the development and use of NIR spectrometric techniques. HortScience 34, 114– 118. Peirs, A., Scheerlinck, N., Touchant, K., Nicolai, B.M., 2002. Comparision of Fourier transform and dispersive near-infrared reflectance spectroscopy for apple measurements. Biosyst. Eng. 81, 305–311. Peirs, A., Tirry, J., Verlinden, B., Darius, P., Nicola¨ı, B.M., 2003. Effect of biological variability on the robustness of NIR models for soluble solids content of apples. Postharvest Biol. Technol. 28, 269–280. Rodriguez-Saona, L.E., McLaughlin, M.A., Fry, F.S., Calvey, E.M., 2001. Rapid analysis of sugars in fruit juices by FT-NIR spectroscopy. J. Carb. Res. 336, 63–74. Slaughter, D.C., 1995. Non-destructive determination of internal quality in peaches and nectarines. Trans. ASAE 38, 617– 623.