Preliminary study to determine the phenolic maturity stage of grape seeds by computer vision

Depósito de investigación de la Universidad de Sevilla https://idus.us.es/ “This is an Accepted Manuscript of an article published by Elsevier in Analytica Chimica Acta on 30 June 2012, available at: https://doi.org/10.1016/j.aca.2012.01.005 .” 1 *Manuscript Click here to view linked References 1 Preliminary study to determine the phenolic maturity stage of grape seeds 2 by computer vision 3 4 Francisco J. Rodríguez-Pulido 1, Raúl Ferrer-Gallego 2, M. Lourdes 5 González-Miret 1, Julián Carlos Rivas-Gonzalo 2, María Teresa Escribano- 6 Bailón 2, Francisco J. Heredia 1*, 7 8 1 Food Colour & Quality Lab., Dept. Nutrition & Food Science. Facultad de 9 Farmacia. Universidad de Sevilla. 41012-Sevilla, Spain. 10 2 Grupo de Investigación en Polifenoles. Unidad de Nutrición y Bromatología, 11 Facultad de Farmacia, Universidad de Salamanca, Campus Miguel de 12 Unamuno, E 37007 Salamanca, Spain. 13 14 15 Corresponding author: 16 Heredia, Francisco J.; 17 Food Colour & Quality Lab., Dept. Nutrition & Food Science. Facultad de 18 Farmacia. Universidad de Sevilla. 41012-Sevilla, Spain 19 Tel.: +34 954556495 Fax: +34 954557017 20 e-mail: [email protected] 21 2 22 ABSTRACT 23 The applications of computer vision technology for acquiring and analysing 24 images have been extended to the quality evaluation in food industry. This 25 technique involves great advantages for the objective, rapid, non-contact and 26 automated quality inspection and control. The aim of this work was to evaluate 27 the potential of the computer vision to determine the phenolic maturity stage of 28 grape seeds. Up to 21 phenolic compounds were determined by HPLC-DAD- 29 MS in order to obtain reference values to develop the model. The CIELAB 30 parameters, area, aspect, roundness, length, width and heterogeneity of seeds 31 were analysed using a DigiEye® system. The technique reported in this work 32 can be a good and rapid tool for taking decisions at harvest time. 33 Notwithstanding, a comprehensive study should be made in order to develop 34 more robust models. 35 36 KEYWORDS 37 Image analysis, color, Vitis vinifera, grape seeds, phenolics. 38 3 39 1. INTRODUCTION 40 41 The phenolic composition of grape seeds plays an important role on the red 42 wine quality. In this sense, the contribution of seed tannins to several sensorial 43 attributes of wines, such as body, structure, bitterness, astringency and colour 44 are well known [1,2]. Also, the presence of phenolic compounds in wine has 45 been the subject of great interest due to their potentially beneficial effects on 46 human health [3]. 47 Many physical and biochemical changes take place during grape seeds maturity 48 which affect their characteristics like astringency, tannic intensity, colour and 49 morphology [4,5]. Quantity, structure and degree of polymerization of seed 50 tannins are modified during ripening development [6,7]. Monitoring these 51 changes could be an important tool for deciding the optimal moment for 52 harvesting, but this would require to have rapid analytical methods involving 53 minimal or no sample preparation. 54 Colour and appearance are closely related to the chemical composition and 55 sensory properties of food. Usually, colour is determined by tristimulus 56 colorimetry and expressed in terms of L*a*b* variables, corresponding to the 57 uniform colour space CIELAB. The CIELAB colour space (1976 L*a*b* colour 58 space) is an international standard for colour measurements, recommended by 59 (CIE) in 1976. Within CIELAB, a 60 psychometric index of lightness (L*) and two colour coordinates (a* and b*) are 61 defined. L* is a qualitative attribute of relative luminosity, which is the property 62 according to which each colour can be considered as equivalent to a member of 63 the grey scale, ranging between black (L*=0) and white (L*=100). Coordinate a* 4 64 takes positive values for reddish colours and negative values for greenish ones, 65 whereas b* takes positive values for yellowish colours and negative values for 66 bluish ones [8]. From these Cartesian coordinates, two angular parameters can 67 be defined: chroma (C*ab) and hue (hab) [9]. Hue angle (hab) is the qualitative 68 attribute of colour (blue, purple, green...), and chroma (C*ab) is considered the 69 quantitative attribute of colourfulness, allowing assessing the degree of 70 difference of any given hue relative to a grey colour with the same lightness. 71 The CIELAB colour space is considered as uniform by the CIE, i.e., the 72 Euclidean distance between two different colour points corresponds 73 approximately to the colour difference perceived by the human eye. 74 For colour measurement, different kinds of instruments such as colorimeters, 75 spectrophotometers and spectroradiometers can be used. However, in the case 76 of non-homogeneous colour, these instruments require homogenizing the 77 sample to achieve uniform colour. This fact makes complicated and tedious 78 measuring colour of heterogeneous stuff or small objects, such as grape seeds. 79 Computer vision provides fast, accurate, direct and non-destructive analysis 80 that this case needs. Moreover, not only colour but also other characteristics 81 such as shape, texture and homogeneity can be determined [10,11]. 82 A computer vision system includes an illumination system, a charge-coupled 83 device (CCD), a frame grabber which converts the analogue image from the 84 camera into a digitized one, and a computer with the suitable software for image 85 processing and interpretation of results [12]. The digital camera receives 86 images onto a CCD that registers them in gradations of three basic colours: red, 87 green and blue (RGB colour space). This colour space is widely used in 88 electronics devices but it depends on the device where it is visualized. So, we 5 89 can not use this space for absolute measures; it is necessary to transform the 90 data from RGB to CIELAB colour spaces. This transformation requires 91 calibration and depends on the illumination when images are taken. 92 In the image analysis perform, it is indispensable to recognize objects into the 93 image before analyse them. For this purpose, the representation of the image is 94 changed into something more meaningful and easier to analyse [13]. This 95 process is known as segmentation and refers to the partition of a digital image 96 into multiple sets of pixels. Once this recognition is done, the morphology of the 97 objects are analysed and colour coordinates are calculated from the pixels that 98 they are composed of. Frequently, HSI (hue, saturation and intensity) space is 99 used for the segmentation process [14]. HSI colour space is easier to 100 understand than RGB due to its likeness with the L*, C*ab and hab stimuli, which 101 are more related to the visual appreciation of colour. 102 The aim of this study was to find relationships between chemical (phenolic 103 composition) and appearance (colour and morphological) data of grape seeds 104 which can yield to the assessment of the phenolic maturation stage based on 105 105 106 106 parameters obtained by image analysis. 107 107 108 108 2. MATERIALS AND METHODS 109 2.1 Samples 110 Vitis vinifera L. cv. Graciano red grape berries grown on Rioja Designation of 111 Origin (Spain) were collected at two vineyards in 2010, at six different stages 112 during the berry maturation (from veraison until over-ripeness). In order to have 113 representative samples, berries were collected from the top, middle and bottom 6 114 of the clusters from different vines along the vineyard. The samples (whole 115 grapes) were immediately frozen and stored at -20 ºC until the analyses were 116 116 117 117 performed. 118 2.2 Phenolic extraction and analysis 119 Phenolic extraction was carried out as described in García-Marino et al. [15]. 120 For extraction, grape seeds were separated manually and freeze-dryed. After 121 this, samples were ground to obtain homogeneous powder. The grape seed 122 powder was extracted with a solution of methanol:water (75:25), and the 123 methanolic extracts were concentrated at low pressure until aqueous extract 124 was obtained. The extracts were filtered through a 0.45 m filter and injected 125 directly into the chromatographic system. All analyses were performed in 126 triplicate. 127 Liquid chromatography-diode array detection-mass spectrometry (LC-DAD-MS) 128 analysis was performed using an Agilent Technologies 1200 series liquid 129 chromatograph (Palo Alto, CA, USA). A Spherisorb S3 ODS-2 reverse phase 3 130 m particle size C18 column 150 4.6 mm i.d. (Waters, Ireland) thermostatted 131 at 25ºC was used. The solvents used were (A) 2.5% acetic acid, (B) 2.5% 132 acetic acid/acetonitrile (90:10, v:v), and (C) HPLC-grade acetonitrile, 133 establishing the following gradient: from 0 to 100% B for 5 min, from 0 to 15% C 134 for 25 min, from 15 to 50% C for 5 min, and isocratic 50% C for 5 min, at a flow 135 rate of 0.5 ml min 1. Detection was carried out at 280 nm as preferred 136 wavelength. The mass spectrometer (MS) was connected to HPLC system via 137 the DAD cell outlet. MS detection was performed with API 3200 Qtrap (Applied 138 Biosystems, Darmstadt, Germany) equipped with an electrospray ionization 7 139 (ESI) source and a triple quadrupole-ion trap mass analyzer, controlled by the 140 Analyst 5.1 software. 141 Quantification was performed by HPLC-DAD using calibration curves of (+)- 142 catechin, purchased from Sigma Aldrich (St. Louis, USA), and of procyanidin 143 143 144 144 standards previously obtained in our laboratory [16,17]. 145 2.3. Image Acquisition 146 The DigiEye® imaging system based upon the calibrated digital camera was 147 used [18]. This device consists on a closed illumination box, specially designed 148 by VeriVide Ltd. (Leicester, UK), and a digital camera (10.2-megapixel Nikon® 149 D80 with Nikkor® 35 mm f/2D objective) connected to a computer (Pentium IV 150 3.00 GHz processor) via USB. The cabinet is equipped with two CIE D65 151 standard illuminant emulators which allow illuminating the samples consistently 152 under stable lighting conditions. For obtaining morphological and appearance 153 parameters, as well as the CIELAB coordinates from RGB colour space data, 154 the software DigiFood® [19] was used. 155 155 156 3. RESULTS AND DISCUSSION 157 3.1. Chemical analysis 158 Based on the chromatographic and spectral features, twenty one phenolic 159 compounds were identified. These were quantified and grouped into 160 hydroxybenzoic acids, flavanol monomers (including catechin and epicatechin), 161 dimers of procyanidins, trimers of procyanidins and galloylated flavanols 162 (epicatechin 3-O-gallate and galloylated procyanidins) (Figure 1). These 8 163 compounds correspond to the main phenolic compounds present in the 164 chromatograms of the seed extracts. 165 Total phenolic content, defined as the sum of the content of the phenolic 166 compounds identified, decrease during ripening, in agreement with other works 167 [4,20,21,22,6]. Monomers and galloylated flavanols undergo the sharpest drop. 168 Since these last compounds could be related to astringency and roughness 169 [23], their decrease throughout the ripening process, would contribute to 170 improve the mouth-feel of grape seeds at maturity. Near maturity, the content of 171 galloylated compounds is close to that found by other authors for this variety 172 (24% of total) [4,24]. The pronounced decrease of the flavanol monomers 173 provokes that the ratio of the concentration of monomers to oligomers (dimers 174 and trimers) tends to diminish during ripening. Thus, values close to 1 can be 175 calculated at veraison while ca. 0.6 is obtained at harvest. This is consistent 176 with possible polymerization reactions occurring during maturation. 177 Hydroxybenzoic acids are present in seeds in low amounts (from 0.4 to 0.5 mg 178 g-1) and their concentration do not suffer as noticeable changes during ripening 179 179 180 180 as flavanols. 181 3.2. Image Analysis 182 Images of the seeds corresponding to 100 berries per sample were acquired by 183 using the DigiEye system. For seed recognition within the images, the following 184 selection criteria were established: a) regions having values of saturation (S) 185 lower than 180 units, and intensity (I) lower than 140 units, b) objects having 186 area values between 6 and 35 mm2, and length values less than 15 mm. By 15 327 FIGURE CAPTIONS: 328 Figure 1: Contents of the different polyphenols divided in groups at each stage 329 of maturation. 330 Figure 2: Scatterplot of observed vs. predicted values of stage of maturity. 331 331 332 332 Dashed lines shows confidence bands at (95%). Figure1 Click here to download Figure: Rodriguez-Pulido_FIGURE1.eps Figure2 Click here to download Figure: Rodriguez-Pulido_FIGURE2.eps