• Keine Ergebnisse gefunden

Detection of Bruises on Apples by Near Infrared Reflectance Spectroscopy P. Guillermin and C. Camps D. Bertrand Institut National d’Horticulture Institut National de la Recherche Agronomique Angers Nantes France France

N/A
N/A
Protected

Academic year: 2022

Aktie "Detection of Bruises on Apples by Near Infrared Reflectance Spectroscopy P. Guillermin and C. Camps D. Bertrand Institut National d’Horticulture Institut National de la Recherche Agronomique Angers Nantes France France"

Copied!
8
0
0

Wird geladen.... (Jetzt Volltext ansehen)

Volltext

(1)

Detection of Bruises on Apples by Near Infrared Reflectance Spectroscopy

P. Guillermin and C. Camps D. Bertrand

Institut National d’Horticulture Institut National de la Recherche Agronomique

Angers Nantes

France France

Keywords: NIR, compression, fruit sorting Abstract

The objective of the study was to evaluate the ability of near infrared reflectance spectroscopy (NIR) to detect the bruises caused on apples by a definite compression test. Two cultivars (‘Elstar’ and ‘Gala’) and six stages of fruit development (from mid May to mid September) were compared. Spectra were first acquired on intact fruit. Two successive compressions at constant speed were then applied on each apple to reach a strain equal to 10% (first samplings) or 5%

(samplings at harvest) of the fruit diameter. Two hours after compression, a second spectrum was recorded on the same face. The spectra were acquired on a NIR spectrometer at wavelength ranging from 800 nm to 2200 nm. Discriminative analyses (FDA) with cross validation procedure were applied on the spectral collections. In most of the cases, these analyses showed good discrimination between apples before and after compression. Nevertheless, the capacity of NIR spectroscopy to detect bruised apples and the more discriminative spectral absorption bands varied according to the cultivar and the stage of development. The relationship between the spectral response and the characteristics of the bruising event or the properties of fruit were studied.

INTRODUCTION

The harvest, packaging and transport of apples involve mechanical damage due to dynamic or static load applied on fruit. These bruises can be visible on the skin with formation of brown blemishes but, more often, they affect the inner tissue without external defects (Rodriguez et al., 1990). The texture of flesh is modified after impact: the fruit becomes soft and loses its juiciness and crunchiness, leading loss in market value.

Therefore the detection of bruised fruit is an important goal for the apple packinghouse managers. Besides, a better understanding of the physical and biological factors involved in bruising is also necessary both to improve the classification of the fruit and to reduce the irreversible mechanical damage occurring throughout the life of the fruit. Previous work has described damage susceptibility by measuring the surface of the damaged tissues but this approach is destructive and time consuming and cannot be automated.

Visible/near infrared spectroscopy has been largely developed for many years as a non-destructive method to evaluate internal quality of fruits and this technology is now present on grading machines, especially to sort fruits according to their sugar concentration. This method may also appear as a relevant tool to study bruises. Geola et al. (1994), Upchurch et al. (1997), Kleynen et al. (2003) and Xing et al. (2003) succeeded in segregating intact or bruised apples according to their spectral features. Near infrared spectrometry has also been used to detect some physiological disorders showing visible symptoms rather similar to bruises (Clark et al., 2003).

Whatever the method employed (destructive or non destructive), all the authors concluded that the bruise susceptibility depends both on the conditions of bruising event and on the internal fruit properties. Canonical discriminant analysis performed on VIS/NIR spectra made it possible to clearly distinguish two types of bruising event:

impact and compression (Xing et al., 2003). Khan and Vincent (1991) have studied the role of the skin for preventing definitive damages in the underlying flesh during compression. When the core of the apple is perpendicular to the direction of the 1355

(2)

compression force, the elastic lateral expansion and the energy absorbed is immediately dissipated as bruising. Bruising appearance and susceptibility are also influenced by fruit turgidity (Garcia et al., 1995), mechanical properties of the flesh or the skin (Grotte et al., 2000), cellular arrangement and cell geometry (Wenian et al., 1991) or cultivars and fruit maturity (Bollen et al., 2001).

The aim of this work was to use near infrared spectroscopy to study the variability of apple bruising susceptibility inside very different sets of apples (from small fruit sampled in May to mature fruit at harvest). For each set of fruit, the differences between bruised and non-bruised spectral data were first compared. Different factors supposed to influence the bruising susceptibility were then studied in order to better understand their respective roles.

MATERIALS AND METHODS

Fruit of ‘Gala’ and ‘Elstar’ trees in an experimental orchard in Val de Loire (France) were sampled at 6 dates: in May (51 and 64 days after full bloom, DAFB), in June (78 and 92 DAFB) and at the end of August, near maturity for commercial harvest (147 and 161 DAFB). Twelve sets of fruit were collected with 25 fruits/set.

Immediately after sampling, each fruit was analysed in the following way: (1) A first spectrum was acquired on the whole intact fruit using a Near Infrared spectrometer at wavelengths ranging from 800 to 2200 nm (direct contact analysis). (2) The fruit was submitted to a mechanical stress by compression using a texture analyser (TA.XT.Plus) fitted with a circular flat probe (5 cm diameter), at a constant loading and unloading rate.

Two successive cycles were applied with a reversion of work fixed at 10% of the initial diameter for dates P1 to P4 and at 5% for dates P5 and P6. (3) Two hours after compression, a second spectrum was acquired. The two spectral measures were carried out on the same zone, on the less coloured face, the face in contact with the plate. During the compression, the force-time curves were recorded. The intensity of the mechanical stress was estimated from the energy applied during the first loading. This value (Work) was computed as 0.5 (F1 x T1) where F1 is the maximum force for the first compression and T1 is the time needed to reach the strain limit. Two other parameters measured physical flesh properties: F1-F2 = difference between the maximum force of the first and the second compression cycle; T2-T3 = difference between the time corresponding to F >

0 during the second loading and the time corresponding to F = 0 during the first unloading. In order to complete these measures, a puncture test (depth = 1 cm) was performed at constant speed on non-peeled fruit using the same texture analyser fitted with a hemispheric probe. The force-time curves were also acquired during this puncture test and three parameters were extracted, namely: Ff = flesh firmness (mean force required to perforate flesh), Fmax (maximum force before skin rupture) and St = Stiffness (slope of the force-time curve from zero to the point of rupture). The two last parameters characterised the skin properties. Eventually, the fruit were also characterised by their percent of dry matter (DW).

In order to reduce the effect of uncontrolled variations, a SNV correction was applied on spectral data before analysis. Models for data analysis will be presented before each result.

RESULTS AND DISCUSSION

Preliminary Analysis of the Variability of Spectra

A principal components analysis on the two spectra of all fruit was performed in order to study the main factor of variability. The factorial map of the centroïd values obtained for each set on the two first components (Fig. 1) showed that the main source of variability between spectra was the dates of sampling, followed by the nature of the cultivar and the bruises. In order to emphasise the bruising effect, the variability of the spectra has been studied independently on each set of fruit (one cultivar x one date) and then compared.

(3)

Capability of NIR to Detect Bruises

Twelve factorial discriminative analysis were performed as described by Bertrand et al. (1990), with the treatment ‘bruised’ or ‘not bruised’ as qualitative groups. Apples were randomly split into two subsets: a calibration and a validation set. Each fruit of the validation set was classified as bruised or non bruised according to the predictive model established from the calibration set. The procedure was repeated 10 times and the 10 results, expressed as the percentage of correctly-classified fruit, were averaged. Two predictive models were compared. The first was optimised: 10 components were chosen among the 30 first principal components. In the second model 5 components were chosen among 5. This last model made it possible to study if the discriminative variables for bruises were representative of the main variability in the set.

With the optimised model (model one), the results (Fig. 2) showed that NIR spectroscopy was able to distinguish bruised and not bruised apples with a very good accuracy: whatever the set, 95 to 100 percent of apples had correct classification.

Nevertheless, the accuracy was largely weaker with the second model. We can conclude that the bruises were not the main cause of variability between spectra, even inside a single cultivar and a single date.

Main Wavelengths Implied on the Detection of Bruises

One-way ANOVA’s with absorbency at each wavelength as variables and treatment (bruised or non-bruised) as unique factor were performed for each cultivar and each date. Fisher F values were plotted according wavelengths (Fig. 3). A great F value signifies that the absorbency on the corresponding wavelength was strongly different from bruised to non-bruised fruit.

The results showed that the differences between the absorbencies of bruised and non-bruised fruit were observed inside a wide range of wavelengths, but the number and the importance of these significant wavelengths remarkably varied according to the dates of sampling. These differences became even larger for the fruit at maturity (P6). The absorbencies around 2000 nanometers seemed to be appropriate for the detection of the bruises for all dates of sampling (P2 to P6), except for P1. At a given date, the comparative results between ‘Elstar’ and ‘Gala’ showed a good similarity between the two cultivars, except for the first part of the spectra (from 800 to 1100 nm).

Relations between the Spectral Response and Bruises or Fruit Characterisation The spectral response to bruising events was measured by computing the differences between the spectra before and after compression. The curves of differences were then analysed like new spectra. The links between these spectral responses and the different factors of variability (features measured on the compression curves: Work, F1- F2 and T1-T3, features measured on the penetration curves: Fmax, St and Ff, and estimation of fruit turgidity by DW) were then studied. Regressions between the factorial scores of an ACP of spectral responses and these parameters were carried out using the Principal Component Regression method (PCR). The results were expressed as the correlation coefficients. Two models were applied: in model 1 (optimised model), 5 latent variables were chosen among the 15 first principal components; in model 2, 2 latent variables were chosen among the 5 first principal components that explained most of 95

% of variability. The comparison between the two models allowed to study in what extend the correlation coefficients were established on the main source of variability of spectral responses.

Fig. 4 showed the regressions obtained with the parameter chosen to characterise the bruising event (Work). With the first model, the coefficients of correlation were always comprised between 0.6 and 0.8. So we can conclude that the spectral response was partly related to the work of compression. Nevertheless, except for ‘Elstar’ at P1, the coefficients obtained with the second model were always less than 0.4. The main source of variability between the spectral responses was therefore not related with the parameter Work.

(4)

In order to study the relations between the spectra and the characteristics of the fruit, we have gathered the factors of variability into three groups: the physical features of flesh (F1-F2, T2-T3 and Ff), the physical features of skin (Fmax and St) and the percent of dry matter (DW). For ‘Gala’ (Fig. 5a), the spectral response was better related to the flesh than to the skin properties, and partly related to the percent of dry matter. For

‘Elstar’ (Fig. 5b), the flesh and skin properties were implied in the same way in spectral response, whereas percent of dry matter was poorly related. For some parameters, the correlation coefficients were rather different at the two dates of sampling, which is for two different physiological states of the fruit. Finally, the coefficients obtained with the second model (data not shown) suggested that the different factors of variability retained in this study were not related to the factorial scores on the first components of ACP, which was the main source of variability.

CONCLUSIONS

This work has confirmed the capability of near infrared spectroscopy to detect bruises. Nevertheless the response is not very specific: many wavelengths are related to bruising. As previously observed by Kleynen et al. (2003), the bruises seem to be one of the most difficult defects to detect by NIR. This can be explained by the fact that the bruise is a complex phenomenon and that most of the biological events occurring in a bruised apple tissue are possibly detected by NIR spectroscopy. A larger sample set would be necessary to validate a model making it possible to correctly define the type and the intensity of bruises from the NIR spectra of the fruit. Studies of the relationship between the spectral features and the parameters describing fruit damage, as those recently proposed by Ragni and Berardinelli (2001), could also improve the interpretation of NIR data.

The calculation of the difference spectra can be useful to explore the main factors of variability of bruising susceptibility. Many factors are able to influence the capacity of NIR to detect bruises: the energy displayed during the initial bruising event, (work of compression in this case), some flesh and skin physical properties (parameters extracted from the force-time curves in this case) and the percent of dry matter. Nevertheless, the importance of these factors varies according to the cultivar and to the physiological stages of the fruit. Besides, these factors do not explain the main differences between spectral responses. A more detailed characterisation of bruising event and fruits properties must be continued in order to explain the differences of bruising susceptibility among fruit.

ACKNOWLEDGEMENTS

The authors thank the Region des Pays de la Loire for their financial support and C. LeMorvan for his technical assistance.

Literature Cited

Bertrand, D., Courcoux, P., Autran, J.C., Méritan, R. and Robert, P. 1990. Stepwise canonical discriminant analysis of continuous digitised signals: application to chromatograms of wheat proteins. J. Chemometrics 4:413-427.

Bollen, A.F., Cox, N.R., Dela Rue, B.T. and Painter, D.J. 2001. A descriptor for damage susceptibility of a population of produce. J. Agri. Engineering Res. 78:391-395.

Clark, C.J., McGlone, V.A. and Jordan, R.B. 2003. Detection of brownheart in 'Braeburn' apple by transmission NIR spectroscopy. Postharvest Biol. Technol. 28:87-96.

Garcia, J.L., Ruiz-Altisent, M. and Barreiro, P. 1995. Factors influencing mechanical properties and bruise susceptibility of apples and pears. J. Agri. Engineering Res.

61:11-18.

Geoola, F. and Peiper, U.M. 1994. A spectrophotometric method for detecting surface bruises on ‘Golden delicious’ apples. J. Agri. Engineering Res. 58:47-51.

Grotte, M., Duprat, F., Loonis, D. and Piétri, E. 2000. Bruising appearance of apples:

involved parameters. Sciences des aliments 20:575-590.

Khan, A.A. and Vincent, J.F. 1991. Bruising and splitting of apple fruit under uni-axial

(5)

compression and the role of skin in preventing damage. J. Texture Studies 22:251- 263.

Kleynen, O., Leemans, V. and Destain, M.F. 2003. Selection of the most efficient wavelength bands for ‘Jonagold’ apple sorting. Postharvest Biol. Technol. 30:221- 232.

Ragni, L. and Berardinelli, A. 2001. Mechanical behaviour of apples, and damage during sorting and packaging. J. Agri. Engineering Res. 78:273-279.

Rodriguez, L., Ruiz, M. and De Felipe, M.R. 1990. Differences in the structural response of 'Granny-Smith' apples under mechanical impact and compression. J. Texture Studies 21:155-164.

Upchurch, B.L.,Throop, J.A.and Aneshansley, D.J. 1997. Detecting internal breakdown in apples using interactance measurements. Postharvest Biol. Technol. 10:15-19.

Wenian, C., Duprat, F. and Roudot, A.C. 1991. Evaluation of the importance of the cellular tissue geometry on the strains observed on apples after a compression or an impact. Sciences des aliments 11:99-110.

Xing, J., Landahl, S., Lammertyn, J., Vrindts, E. and De Baerdemaeker, J. 2003. Effect of bruise type on discrimination of bruised and non-bruised 'Golden delicious' apples by VIS/NIR spectroscopy. Postharvest Biol. Technol. 30:249-258.

Figures

a b a b

Gala non bruised Gala bruised Elstar non bruised Elstar bruised

Fig. 1. Factorial map (components 1 and 2) of a principal components analysis performed on the two spectra of all fruit. Each ellipse represents the confidence interval at 1% around the centroïd value of each set defined by the cultivar (‘Elstar’ or

‘Gala’), the date of sampling (1 to 6) and the treatment (bruised or non-bruised).

(6)

50 60 70 80 90 100

P1 P2 P3 P4 P5 P6 Dates

Discrimination accuracy (%)

ELSTAR

50 60 70 80 90 100

P1 P2 P3 P4 P5 P6 Dates

Discrimination accuracy (%)

Model 1 GALA Model 2

Fig. 2. Classification accuracy (expressed in percentage of correct classification) for the two cultivars, the six dates and the two models. Model 1: 10 components chosen among the 30 first principal components, Model 2: 5 components chosen among 5.

Fig. 3. F values resulting from variance analysis of absorbency at each wavelength, with treatment (bruised or not bruised) as factor, for two dates P3 and P6 and for the two cultivars.

(7)

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9

P1 P2 P3 P4 P5 P6

Correlation coefficients

Elstar - mod1 Gala - mod1 Elstar - mod2 Gala - mod2

Fig. 4. Principal Components Regression between the factorial scores of an ACP of the spectral responses and the parameter Work, for the two cultivars and the six dates (P1 to P6). Model 1 (mod1): 5 latent variables chosen among the 15 first principal components. Model 2 (mod 2): 2 latent variables chosen among 5.

Fig. 5. Principal Components Regression between the factorial scores of an ACP of the spectral responses and flesh properties (Ff, F1-F2 and T2-T3), skin properties (Fmax and St) and the percent of dry matter (DW), for the two cultivars and the two dates (P5 and P6). Results with the first model: 5 latent variables chosen among the 15 first principal components.

DW Fmax St Ff F1-F2 T3-T2

ELSTAR

0,4 0,5 0,6 0,7 0,8 0,9

P5 P6

GALA

0,4 0,5 0,6 0,7 0,8 0,9

P5 P6

Correlation coefficients

(8)

Referenzen

ÄHNLICHE DOKUMENTE