Feasibility in multispectral imaging for predicting the content of bioactive compounds in intact tomato fruit

Highlights

• BPNN improved the performance of multispectral imaging for predicting lycopene.

• BPNN improved the performance of multispectral imaging for predicting phenolics.

• Multispectral imaging together with chemometric methods is a promising technique.

Abstract

Tomato is an important health-stimulating fruit because of the antioxidant properties of its main bioactive compounds, dominantly lycopene and phenolic compounds. Nowadays, product differentiation in the fruit market requires an accurate evaluation of these value-added compounds. An experiment was conducted to simultaneously and non-destructively measure lycopene and phenolic compounds content in intact tomatoes using multispectral imaging combined with chemometric methods. Partial least squares (PLS), least squares-support vector machines (LS-SVM) and back propagation neural network (BPNN) were applied to develop quantitative models. Compared with PLS and LS-SVM, BPNN model considerably improved the performance with coefficient of determination in prediction ($R_P^2$) = 0.938 and 0.965, residual predictive deviation (RPD) = 4.590 and 9.335 for lycopene and total phenolics content prediction, respectively. It is concluded that multispectral imaging is an attractive alternative to the standard methods for determination of bioactive compounds content in intact tomatoes, providing a useful platform for infield fruit sorting/grading.

Introduction

Tomato fruit (Solanum lycopersicum) is the second most important vegetable crop worldwide and consumed either fresh or in the form of processed products (Viuda-Martos et al., 2013). It is well known as a health-stimulating fruit because of the antioxidant properties of its main bioactive compounds. The most important bioactive compounds in tomato fruit are lycopene and phenolic compounds (Canene-Adams et al., 2005, Lenucci et al., 2006). Lycopene, exhibiting the highest antioxidant activity and singlet oxygen quenching ability of all dietary carotenoids (Gärtner, Stahl, & Sies, 1997), is the most abundant carotenoid in ripe tomato fruit, comprising approximately 80–90% of the total carotenoids content (Shi & Le Maguer, 2000). Phenolic compounds are famous group of secondary metabolites in plants, which are vital determinants in the nutrition and sensory quality in fruits. Vinson et al. (1998) studied the amount of phenolics in commonly consumed vegetables and, on the basis of their average consumption, tomatoes were identified as the most important suppliers of phenolics in human diet. The main phenolic compounds in tomato fruit are namely the flavonoids rutin (quercetin-3-rutinoside), kaempferol-3-O-rutinoside, and naringenin chalcone (Long et al., 2006) and the phenolic acids caffeic acid, p-coumaric acid and ferulic acid (Luthria, Mukhopadhyay, & Krizek, 2006). Differentiation of the final product in the market requires an accurate evaluation of these value-added compounds. However, the conventional methods to determine these bioactive components are generally based on either colorimetric or chromatographic techniques (liquid chromatography coupled with tandem mass spectrometry, LC-MS; high-performance liquid chromatography, HPLC). These techniques are mostly time-consuming, require preparation of sample and the use of chemical products, and they are laborious and expensive. Moreover, since conventional methods are destructive, they only enable quality control of a few samples per batch, rather than each individual fruit. In order to overcome these disadvantages, developing a rapid, non-contaminant and non-destructive determination methods, preferably based on optical properties, is urgently required.

At present, various optical/electromagnetic methods have been developed for rapid, accurate, and non-destructive determination of lycopene content in tomato fruit, which included visible-near infrared (vis–NIR) spectroscopy (Clément, Dorais, & Vernon, 2008a), hyperspectral reflectance imaging (Polder, van der Heijden, van der Voet, & Young, 2004), attenuated total reflection infrared (ATR-IR) spectroscopy (Baranska, Schütze, & Schulz, 2006), surface colour measurement (Arias, Lee, Logendra, & Janes, 2000), Raman chemical imaging (Qin, Chao, & Kim, 2011), and magnetic resonance imaging (Cheng, Wang, Chen, & Lin, 2011). In addition, near infrared (NIR) reflectance spectroscopy has been used for the determination of total phenolic content in many other fruits or vegetables, such as blueberries (Sinelli, Spinardi, Di Egidio, Mignani, & Casiraghi, 2008), and grapes (Ferrer-Gallego, Hernández-Hierro, Rivas-Gonzalo, & Escribano-Bailón, 2011). Regarding the chemometric techniques, most of these methods were developed using partial least square (PLS) analysis. New chemometric analysis such as least square-support vector machine (LS-SVM) and back propagation neural network (BPNN) appear promising in that they enable the non-linearity of data to be modelled using local or specific equations which could improve prediction models.

Multispectral imaging is an emerging non-destructive technology that integrates conventional imaging and spectroscopy to obtain both spatial and spectral information from an object simultaneously. Multispectral imaging analyses are non-destructive, rapid, simple to perform, and require no sample pre-treatment, which makes this technology ideally suited for on-line process monitoring and quality control (Feng and Sun, 2012, Gowen et al., 2007). More importantly, this technique has the great potential to measure the multiple components at the same time for quality assurance. Recently, this technology has been applied as a powerful process analytical tool for rapid, non-destructive inspection of internal and external attributes in various fruits and vegetables such as apple (Lu, 2004, Lunadei et al., 2011, Peng and Lu, 2006, Peng and Lu, 2007), peach (Lleó, Barreiro, Ruiz-Altisent, & Herrero, 2009), fresh-cut spinach leaves (Lunadei et al., 2012), packaged wild rocket (Løkke, Seefeldt, Skov, & Edelenbos, 2013), and strawberry (Liu et al., 2014). In regards of tomato fruit, multispectral imaging was originally applied to prediction of unripe tomatoes and an accuracy of over 85% was achieved (Hahn, 2002). And it has also recently emerged as a powerful approach for identification of cherry-tomato varieties (Yang, Nie, Feng, He, & Chen, 2010). However, to our knowledge, there is no published data on the multispectral imaging for determination of the content of bioactive compounds (lycopene and total phenolics) in tomato fruit. Therefore, the main objective of this study was to assess the application of multispectral imaging for predicting the content of lycopene and total phenolics in tomato fruit, and compare the performance of prediction models obtained using PLS, LS-SVM and BPNN.