Elsevier

Talanta

Volume 155, 1 August 2016, Pages 347-357
Talanta

Facilitated wavelength selection and model development for rapid determination of the purity of organic spelt (Triticum spelta L.) flour using spectral imaging

https://doi.org/10.1016/j.talanta.2016.04.041Get rights and content

Highlights

  • A new method for feature wavelength selection was devised.

  • Simplified PLSDA model was exploited to qualitatively discriminate OSF.

  • Quantitative determination of adulterants in OSF was achieved by MLR and PLSR.

  • Visualization maps were drawn based on an identification function.

Abstract

Based on a new approach for wavelength selection, a multispectral real-time imaging system was proposed for the staple food industry to determine the fidelity of organic spelt flour (OSF) from three categories of adulterants including rye flour (RF), organic wheat flour (OWF) and spelt flour (SF). Calibration models were first built by partial least squares discriminant analysis (PLSDA) and partial least squares regression (PLSR) with spectral pretreatment for multivariate analysis of hyperspectral image in the spectral range of 900–1700 nm. Instead of qualifying certain groups of characteristic wavelengths for RF, OWF, SF and OSF separately, a set of mutual wavelengths (1145, 1192, 1222, 1349, 1359, 1396, 1541, and 1567 nm) was chosen by first-derivative and mean centring iteration algorithm (FMCIA) for all investigated flour samples. Then these selected feature wavelengths were utilized in PLSDA, PLSR and multiple linear regression (MLR) models to devise multispectral imaging system. Better performances for both qualitative discrimination of OSF and quantitative measure of adulterants were emerged in simplified PLSDA and PLSR models, with mean determination coefficients in cross validation (R2CV) of 0.958 and in prediction (R2P) of 0.957, respectively. To visualize the adulterants in OSF samples, the distribution maps were drawn by computing the spectral response of each pixel on corresponding spectral images at specific frequencies using a quantitative identification function. The results reveal that spectral imaging integrated with multivariate analysis has good potential for rapidly evaluating the purity of organic spelt flour.

Introduction

The hexaploid spelt (Triticum spelta L.) was the predominant cereal food cultivated in Europe from the 5th century and has been substituted by wheat (Triticum aestivum L.) since the 20th century [1]. The spelt yield is much lower, because the husks take up loss of about 30% and the milling procedure requires an extra step for husk separation. With an excellent source of dietary fiber, vitamins and minerals besides carbohydrate and protein, spelt is suggested to have higher nutritional potential and better taste than common wheat, and can be widely used for brewing, baking and production of pasta [2], [3]. In recent years, spelt whose price often achieves twice that of comparable wheat products is undergoing a renaissance in Europe and North America. The advantage of spelt is that it is more resistant to adverse climate and poor soil conditions like wet, cold soils and at high altitudes [4]. Moreover, spelt carries high-level resistance to several fungal pathogens in terms of Pythium aristoporum Vanterpool, Fusarium spp and yellow rust [5]. Therefore, spelt can survive in low-input farming systems without using much fertilizer and pesticide, which makes it more suitable for organic agriculture. It is known that the aim of organic agriculture is to minimise the use of external inputs and avoid using any chemical additive [6]. In organic agriculture, the lower yield and cumbersome qualification process to some extent result in much higher prices of organic spelt products. Even so, the global market for organic food is still huge and increasing significantly, especially in Europe where the largest organic food market exists [7].

Food adulteration is the presence of an extraneous and cheaper constituent. Driven by high profit, food adulteration has become a global challenge during the past years [8,9]. In order to protect consumers from food adulteration, it is necessary to explore effective methods of food quality authentication [10]. By detecting the relative proportions of wheat and spelt in DNA mixtures, the determination of adulteration of spelt flour (SF) mixed with wheat flour was realized based on a sensitive wheat-specific polymerase chain reaction (PCR) system [11]. The result showed that this technique could be employed to measure the wheat content of a self-produced bread from SF. Then, using the typical soft wheat γ-gliadin sequence, another two DNA-based analytical methods including restriction fragment length polymorphism (RFLP) analysis and lab-on-a-chip capillary gel electrophoresis (LOC-CE) were successfully established for the identification and quantification of adulteration of SF mixed with soft wheat [12]. The authentication of conventional wheat from organic wheat was also carried out by measuring δ15N and δ13C of amino acids via gas chromatography–combustion–isotope ratio mass spectrometry (GC-C-IRMS) [13]. Relying on these destructive methods, the detection of food adulteration can be realized. However, these approaches cannot be used in the food industry for real-time online monitoring since it takes too much time to analyze chemical elements and related data. Moreover, as organic spelt flour (OSF) is a species of wheat, the discrimination between OSF and organic wheat flour (OWF) should also be effectively evaluated. In addition, OSF is nearly same in color to SF and similar to rye flour (RF) as well, which makes them more difficult to be identified after adulteration. With the increasing requirement on the subtle adulterations in food, it is of critical importance to develop a more sensitive, accurate, and rapid approach for non-destructive identification and visualization of OSF.

Spectroscopic techniques including visible/infrared spectroscopy [14,15], Raman spectroscopy [16,17], nuclear magnetic resonance spectroscopy [18,19] and hyperspectral imaging [20,21] can provide non-invasive and fast measurement of radiation intensity as a function of wavelength and have been effectively explored for quality evaluation of staple foods on sensory, adulteration, chemical, mycotoxin, parasitic infection, and internal physiological aspects [22]. Although these techniques are widely used in spectral background inspection, only hyperspectral imaging is emerged for simultaneous detection of both spectral and spatial information of an object [23,24], as hyperspectral imaging combines the techniques of spectroscopy and computer vision [25], [26], [27], [28]. Hyperspectral imaging has therefore been applied not only to the determination of their locations based spatial distribution but also the authentication and quantification of spectral properties based chemical component in many applications in the agricultural and food industry [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41]. In order to realize the on-line application, high-efficiency and cost-effective multispectral imaging systems can be eventually established using qualified algorithms and feature wavelengths based on hyperspectral imaging [42], [43], [44]. Thus, it is essential to identify the most characteristic wavelengths from hundreds of contiguous spectral bands to develop effective models. Several classical methods for wavelength selection have already been extensively used in spectral imaging. By applying successive projection algorithm (SPA), seven (466, 525, 590, 620, 715, 850 and 955 nm) and eight characteristic wavelengths (924, 931, 964, 1068, 1262, 1373, 1628 and 1668 nm) were selected respectively for the color distribution of grass carp fillets [45] and the Enterobacteriaceae contamination of salmon flesh [46]. Another group of nine wavelengths (445, 485, 533, 566, 578, 600, 636, 759 and 959 nm) identified by principal components analysis (PCA) was used for predicting pH values of salted pork [47]. Other methods including genetic algorithm (GA) [48], [49], second derivative (SD) [50], [51], regression coefficients from the partial least squares regression (PLSR) [52], [53] and competitive adaptive reweighted sampling (CARS) [54], [55] were also successfully applied to selecting characteristic wavelengths for assessing food quality. After removing all redundancy variables, the remaining informative wavelengths can enhance model robustness and generate better prediction results [56].

The criteria for the selection of wavelength is based on the information content of spectral bands to enhance the prediction accuracy of the calibration model maximally. In this study, we present an approach termed first-derivative and mean centring iteration algorithm (FMCIA), which has the latent capacity to select an optimum wavelength combination in full spectral range. In order to demonstrate the effectiveness of the proposed algorithm, partial least squares discriminant analysis (PLSDA), partial least squares regression (PLSR) and multiple linear regression (MLR) models are constructed to evaluate the inveracity of RF, OWF and SF in OSF, respectively. The accuracy and predictability of these simplified models are verified and compared. To visualize the result, spatial distribution maps of adulteration are generated by image processing.

Section snippets

Sample preparation and hyperspectral image acquisition

The organic spelt and organic common wheat were produced from organic system where the organic food standards such as non-using synthetic fertiliser and pesticide were strictly implemented. To make wholegrain flour, the inedible outer husk of spelt was removed, leaving the inner bran and grain intact. The whole OSF samples (Doves Farm Foods Ltd., Berkshire, UK) micronized by grinding were legally qualified organic (EU Organic - GB-ORG-05, EU/non-EU Agriculture). The OWF samples were certified

Spectra of flour samples

The average spectra of each adulteration level from 3–75% including rye flour adulteration (RFA), organic wheat flour adulteration (OWFA) and spelt flour adulteration (SFA) are depicted in Fig. 1(a, b and c), respectively. The analogous spectral trend throughout the wavelength range was noticed, along with the variances in the magnitude of spectral absorption for different adulteration proportions. The comparison of the mean spectra of RFA, OWFA, and SFA is demonstrated in Fig. 1(d). The

Conclusions

The most vital challenge for this current study is to devise a new wavelength selection approach for real-time determination of three categories of adulterants in OSF using hyperspectral imaging. The results demonstrate that the FMCIA is an effective method of selecting characteristic wavelengths. The efficient multispectral imaging system can be established to detect OSF adulteration for a real-time practical application based on hyperspectral imaging in tandem with multivariate analyses.

Acknowledgments

The authors would like to acknowledge the support of the UCD-CSC programme jointly funded by University College Dublin and Chinese Scholarship Council.

References (67)

  • Di Wu et al.

    Application of long-wave near infrared hyperspectral imaging for measurement of color distribution in salmon fillet

    Innovative Food Sci. Emerg. Technol.

    (2012)
  • Douglas Barbin et al.

    Allen Paul Near-infrared hyperspectral imaging for grading and classification of pork

    Meat Sci.

    (2012)
  • Di Wu et al.

    Potential of time series-hyperspectral imaging (TS-HSI) for non-invasive determination of microbial spoilage of salmon flesh

    Talanta

    (2013)
  • Mohammed Kamruzzaman et al.

    Paul Non-destructive prediction and visualization of chemical composition in lamb meat using NIR hyperspectral imaging and multivariate regression

    Innovative Food Sci. Emerg. Technol.

    (2012)
  • Douglas F. Barbin et al.

    Non-destructive determination of chemical composition in intact and minced pork using near-infrared hyperspectral imaging

    Food Chem.

    (2013)
  • Yao-Ze Feng et al.

    Near-infrared hyperspectral imaging in tandem with partial least squares regression and genetic algorithm for non-destructive determination and visualization of Pseudomonas loads in chicken fillets

    Talanta

    (2013)
  • Yao-Ze Feng et al.

    Near-infrared hyperspectral imaging and partial least squares regression for rapid and reagentless determination of Enterobacteriaceae on chicken fillets

    Food Chem.

    (2013)
  • H. Pu et al.

    Selection of feature wavelengths for developing multispectral imaging systems for quality, safety and authenticity of muscle foods-a review

    Trends Food Sci. Technol.

    (2015)
  • H.-J. He et al.

    Inspection of harmful microbial contamination occurred in edible salmon flesh using imaging technology

    J. Food Eng.

    (2015)
  • D. Liu et al.

    Combination of spectra and texture data of hyperspectral imaging for prediction of pH in salted meat

    Food Chem.

    (2014)
  • P. Jackman et al.

    Identification of important image features for pork and turkey ham classification using colour and wavelet texture features and genetic selection

    Meat Sci.

    (2010)
  • Y.-Z. Feng et al.

    Near-infrared hyperspectral imaging in tandem with partial least squares regression and genetic algorithm for non-destructive determination and visualization of Pseudomonas loads in chicken fillets

    Talanta

    (2013)
  • D.F. Barbin et al.

    Non-destructive assessment of microbial contamination in porcine meat using NIR hyperspectral imaging

    Innov. Food Sci. Emerg. Technol.

    (2013)
  • J.M. Hernández-Hierro et al.

    Preliminary study on the use of near infrared hyperspectral imaging for quantitation and localisation of total glucosinolates in freeze-dried broccoli

    J. Food Eng.

    (2014)
  • H. Li et al.

    Key wavelengths screening using competitive adaptive reweighted sampling method for multivariate calibration

    Anal. Chim. Acta

    (2009)
  • D. Wu et al.

    Potential of time series-hyperspectral imaging (TS-HSI) for non-invasive determination of microbial spoilage of salmon flesh

    Talanta

    (2013)
  • M. Kamruzzaman et al.

    Non-destructive prediction and visualization of chemical composition in lamb meat using NIR hyperspectral imaging and multivariate regression

    Innov. Food Sci. Emerg. Technol.

    (2012)
  • D. Wu et al.

    Advanced applications of hyperspectral imaging technology for food quality and safety analysis and assessment: a review—Part I: fundamentals

    Innov. Food Sci. Emerg. Technol.

    (2013)
  • S. Wold et al.

    PLS-regression: a basic tool of chemometrics

    Chemom. Intell. Lab. Syst.

    (2001)
  • R.C. Barthus et al.

    Determination of the total unsaturation in vegetable oils by Fourier transform Raman spectroscopy and multivariate calibration

    Vib. Spectrosc.

    (2001)
  • S.J. Mazivila et al.

    Discrimination of the type of biodiesel/diesel blend (B5) using mid-infrared spectroscopy and PLS-DA

    Fuel

    (2015)
  • H. Siedler et al.

    Genetic diversity in European wheat and spelt breeding material based on RFLP data

    Theor. Appl. Genet.

    (1994)
  • K.G. Campbell

    Spelt: agronomy, genetics, and breeding

    Plant Breed. Rev.

    (1997)
  • Cited by (49)

    • Near infrared techniques applied to analysis of wheat-based products: Recent advances and future trends

      2022, Food Control
      Citation Excerpt :

      Therefore, aiming for high profits, the adulteration of organic products has become recurrent, and the authentication of these products, consequently, became more necessary. For this objective, Su and Sun (2016) investigated the adulteration of organic spelt flour with a hyperspectral imaging system. The authors proposed an algorithm to select optimum wavelengths in order to evaluate the addition of rye flour, organic wheat flour and spelt flour in organic spelt flour.

    View all citing articles on Scopus
    View full text