Monitoring and grading of tea by computer vision - A review

Tea being a high value crop throughout the world, its quality plays a significant role in its marketability. Currently, organoleptic methods such as inspection by human sensory panel, and instrument based approaches such as gas chromatography and colorimetric method have been reported as the quality monitoring tools in various stages of tea processing. These methods are time consuming, laborious, expensive and sometimes inaccurate. Therefore, to overcome the inaccuracy and inconsistency, computer vision techniques can be explored as an alternative to conventional techniques. This paper presents an overview of various computer vision based algorithms for colour and texture analysis with a special orientation towards monitoring and grading of made tea. Computer vision and image analysis are non-destructive procedures for sorting tea on the basis of its physical parameters viz. granule colour, shape, size and texture. Although diverse methods for estimation of above parameters were developed by researchers independently, all these can be related to each other.     

Quality of fresh‐cut purple‐fleshed sweet potatoes after X‐ray irradiation treatment and refrigerated storage

The effect of X‐ray irradiation on the quality of fresh‐cut, refrigerated purple‐fleshed sweet potato (PFSP) cubes was investigated. Packaged sweet potato cubes were treated with 0, 250, 500, 750 or 1000 Gy X‐ray irradiation and stored at 4 ± 1 °C for 14 days. After 14 days, total aerobic bacteria counts were 4.1 and 3.2 log₁₀ CFU g^?1, and mould–yeast counts were 3.3 and 3.0 log₁₀ CFU g^?1 in 750 and 1000 Gy treated samples, respectively. Doses up to 1000 Gy did not affect the firmness, moisture content and anthocyanin content of PFSP cubes throughout storage. PFSP cubes' flesh colour did not change during the first week of storage, but lightness (L*) increased after 14 days. Also, irradiation doses at 750 and 1000 Gy decreased saturation (C*) significantly, producing duller flesh colour than controls. Results indicate that X‐ray irradiation treatment at doses up to 1000 Gy can reduce microbial populations while maintaining the physical quality and anthocyanin content of PFSP cubes up to 14 days of storage.     

Escherichia coli biotransformation of daidzein fermentation products from soy‐based foods—relevance to food oestrogenicity‐based functionality

Oestrogenic isoflavones (e.g. daidzein) present in soy‐based foods are likely to be important in food functionality. Biotransformation of daidzein during fermentation‐based soy‐containing food manufacture forms three trihydroxy metabolites, 6‐ortho‐hydroxydaidzein (6‐OHD), 8‐ortho‐hydroxydaidzein (8‐OHD) and 3′‐ortho‐hydroxydaidzein (3′‐OHD) which alters the oestrogenicity of the final food product. We report that Escherichia coli (a key component of the gut microbiome) metabolism converts 8‐OHD to 6‐OHD with a likely concomitant increase in oestrogenicity. This means that the functionality of 8‐OHD‐containing soy‐based foods is altered by gut microbiome metabolism. This change in oestrogenicity‐based food functionality might have benefits for postmenopausal women, while being a feminising health risk for males and could add to the risk of oestrogen‐mediated precocious puberty in girls.     

Prediction of beef eating quality from colour, marbling and wavelet texture features

Beef longissimus dorsi colour, marbling fat and surface texture are long established properties that are used in some countries by expert graders to classify beef carcasses, with subjective and inconsistent decision. As a computer vision system can deliver objective and consistent decisions rapidly and is capable of handling a greater variety of image features, attempts have been made to develop computerised predictions of eating quality based on these and other properties but have failed to adequately model the variation in eating quality. Therefore, in this study, examination of the ribeye at high magnification and consideration of a broad range of colour and marbling fat features was used to attempt to provide better information on beef eating quality. Wavelets were used to describe the image texture of the beef surface at high magnification rather than classical methods such as run lengths, difference histograms and co-occurrence matrices. Sensory panel and Instron analyses were performed on duplicate steaks to measure the quality of the beef. Using the classical statistical method of partial least squares regression (PLSR) it was possible to model a very high proportion of the variation in eating quality (r² = 0.88 for sensory overall acceptability and r² = 0.85 for 7-day WBS). Addition of non-linear texture terms to the models gave some improvements.     

Gluten‐free bread: multiple time–intensity analysis,physical characterisation and acceptance test

The formulation of gluten‐free breads is challenging, once the gluten network is responsible for the physical and sensory properties of bread. This study presents a novel concept for making gluten‐free breads using sugar substitutes. The quality of gluten‐free breads was analysed by means of physical (specific volume and colour), textural (firmness, elasticity and chewiness) and sensory properties (time‐intensity analysis and acceptance test). Time‐intensity analysis showed that the sample developed with raw sugar had a higher intensity of sweetness, and the sample with stevia had a higher intensity of yeast flavour. The sample with frutooligosaccharides (FOS) presented a higher acceptance by consumers. Sample with stevia presented the higher intensity of yeast flavour (8.10) and the lesser mean of specific volume (2.83 cm³.g^?1); whereas sample with sugar who presented higher intensity of sweetness stimulus (5.20) also presented the higher mean of specific volume (3.80 cm³.g^?1). The addition of prebiotic and sweetener opens up new opportunities to develop gluten‐free breads that may present similar properties to those of wheat‐based breads.     

Prediction of Mechanical Properties of Corn and Tortilla Chips by Using Computer Vision

Deep-fat frying is a unit operation which develops unique sensorial attributes in foods. For instance, texture is the principal quality parameter of tortilla and corn chips. On the other hand, computer vision is a useful tool for quality evaluation and prediction of some physical properties in different either raw or processed foods. The objective of this research was to characterize corn and tortilla chips by using computer vision, and to build proper mathematical models which permit to predict mechanical properties of these chips (maximum force, such as hardness, and distance to maximum force, such as toughness) by using chromatic features extracted from their corresponding digital images. Corn and tortilla chips (thickness of 2 mm; diameter of 37 mm) were made from masa of maize and fried at constant oil temperatures of 160, 175, and 190 °C. A high linear correlation (R ² > 0.9400) was obtained between mechanical properties and some image features (Hu, Fourier, and Haralick moments). Cross-validation technique demonstrated the repeatability and good performance (>90%) of the models tested, indicating that can be used to predict the textural properties of the tortilla and corn chips by using selected features extracted from their digital images, without the necessity of measuring them in a texture analyzer.     

A study on chemical estimation of pu‐erh tea quality

Chemical compositions and infusion colour differences of seven pu‐erh tea samples and their correlation to sensory quality were investigated. The results showed that the pu‐erh tea contained 37.1 mg g^?1 caffeine, 15.7 mg g^?1 amino acids, 67.0 mg g^?1 polyphenols and 8.41 mg g^?1 total catechins, on average. Among the 17 tested volatile compounds, n‐valeraldehyde was not detected. The most abundant volatile was β‐ionone and the next was linalool oxide II. Infusion colour analysis showed that the pu‐erh tea had deep hue with ΔE ranging from 66.8 to 79.2. Spearman's linear correlation analysis showed that total quality score (TQS) of the pu‐erh tea was significantly correlated to concentration of amino acids, linalool oxide II and infusion colour indicator ΔE. Five components were extracted from the 34 tested indicators by principal component analysis and were regressed on the TQS to produce six Pearson's linear regression equations for estimating sensory quality of pu‐erh tea, among which two were statistically significant, ie TQS = 57.47 ? 0.18geraniol + 0.33polyphenols ? 1.14n‐caproaldehyde ? 1.38linalool oxide I + 0.21caffeine (p < 0.01) and TQS = 57.42‐0.03Citral + 0.33polyphenols ? 1.14n ? caproaldehyde ? 1.40linalool oxide I + 0.20caffeine (p < 0.01). Copyright © 2004 Society of Chemical Industry     

Consumer exposures to anthocyanins from colour additives,colouring foodstuffs and from natural occurrence in foods

Anthocyanins are responsible for the red/blue colour of grapes, currants, and other fruits and vegetables. They may also be extracted for use as colour additives (E163) or concentrated for use as colouring foods. Consumer exposures have been assessed using data on natural occurrence, use levels and frequencies from food manufacturers and European food consumption data. Intakes from natural occurrence can be up to 4 mg kg bw^?1 day^?1 at the mean and up to 17 mg kg bw^?1 day^?1 for children who are high level consumers of red/black berries and small fruits. High-level intakes for children from food colour and colouring food applications lie in the range 0.3–6.3 mg kg bw^?1 day^?1 and for adults at 0.6–2.8 mg kg bw^?1 day^?1. Exposures from food colour use and colouring foods separately or combined are therefore lower than those from natural occurrence in foods.     

Formation of protein adducts of 2‐amino‐1‐methyl‐6‐phenylimidazo[4,5‐b]pyridine in cooked foods

Heterocyclic amines (HCAs) are mutagenic and carcinogenic compounds found in cooked meat and fish. Although HCAs are known to form adducts with protein after metabolic activation, adduct formation during cooking has not been elucidated. In this study, we showed that 2‐amino‐1‐methyl‐6‐phenylimidazo[4,5‐b]pyridine (PhIP) is released from high molecular weight compounds by acid or enzymatic hydrolysis of cooked foods. Formation of free and protein adduct forms of PhIP was dependent on cooking temperature and time, and PhIP–protein adducts were estimated to form after formation of free PhIP. We also demonstrated that PhIP–protein adduct is formed by heating of PhIP and albumin as a model protein. A new adduct peak including [M+H]⁺ (m/z=225) of PhIP as a fragment ion was detected in the high molecular weight fraction of heat‐treated protein by LC–MS analysis. From model experiments by heating of PhIP and amino acids, the adduct was estimated to be produced by condensation of the amino group of PhIP and the carboxyl group of protein. PhIP–protein adducts were detected in several cooked meat and fish at ng/g food level as PhIP content. These results suggest that food‐borne protein adducts of HCAs may influence human HCA exposure and carcinogenic risk.     

Reducing the glycaemic index and increasing the slowly digestible starch content in gluten‐free cereal‐based foods: a review

Commercially available gluten‐free (GF) cereal‐based foods are generally characterised by a lower nutritional quality than their gluten‐containing counterparts, relatively lower resistant starch (RS) content, along with higher glycaemic index (GI) often being reported. To overcome this nutritional imbalance, extensive research has been conducted to investigate the preparation of a new generation of staple GF products. This review reported the main strategies currently adopted in GF cereal‐based food recipes to formulate products with overall slowly digestible starch properties. They are mainly obtained by the utilisation of alternative ingredients to be incorporated into standard food formulation (including native starch and GF flours) or by technological treatments that may contribute to impact starch digestibility. Considering data from in vitro digestion trials, indications suggested that, aiming to obtain a RS content of about 5% (dry weight) and lowering the in vitro GI values, the minimum high amylose starch (amylose >60%) replacement level in GF bread and cookie formulations should be in the order of 20% by weight of total flours. Overall, with respect to un‐substituted GF foods, two‐ to three‐times higher RS contents, along with a parallel in vitro GI decrease (up to ?50%), were obtained for GF pasta and ready‐to‐eat snacks enriched with legume flours at inclusion levels from 40% to 100% by weight of total flours. The use of flours from pseudocereals (e.g., quinoa, amaranth and buckwheat) with and without sourdough did not always guarantee favourably slowly digestible starch GF foods.     

Design of an Optimum Computer Vision‐Based Automatic Abalone (Haliotis discus hannai) Grading Algorithm

An automatic abalone grading algorithm that estimates abalone weights on the basis of computer vision using 2D images is developed and tested. The algorithm overcomes the problems experienced by conventional abalone grading methods that utilize manual sorting and mechanical automatic grading. To design an optimal algorithm, a regression formula and R² value were investigated by performing a regression analysis for each of total length, body width, thickness, view area, and actual volume against abalone weights. The R² value between the actual volume and abalone weight was 0.999, showing a relatively high correlation. As a result, to easily estimate the actual volumes of abalones based on computer vision, the volumes were calculated under the assumption that abalone shapes are half‐oblate ellipsoids, and a regression formula was derived to estimate the volumes of abalones through linear regression analysis between the calculated and actual volumes. The final automatic abalone grading algorithm is designed using the abalone volume estimation regression formula derived from test results, and the actual volumes and abalone weights regression formula. In the range of abalones weighting from 16.51 to 128.01 g, the results of evaluation of the performance of algorithm via cross‐validation indicate root mean square and worst‐case prediction errors of are 2.8 and ±8 g, respectively.     

Influence of particle characteristics and E/Z‐isomer ratio on the colour of concentrated β‐carotene dispersions

The influence of particle concentrations (0.02–0.40%), mean particle sizes (0.45–1.5 μm) and all‐E‐isomer ratios (95–30%) on the CIELAB colour coordinates (L*, a*, b*) of concentrated β‐carotene dispersions was investigated. Particle concentration between 0.06% and 0.40% had a slight impact on the colour parameters. However, with decreasing the mean particle size from 1.50 to 0.45 μm, there were increases in ΔL* = 8.0, in Δa* = 2.26 and in Δb* = 13.1. And with decreasing the all‐E‐isomers ratio from 95% to 30%, there were increases in ΔL* = 8.66, in Δb* = 17.51 but decrease in Δa* = 7.42. The experimental results were explained in terms of the scattering and absorption of light by dispersions. These findings have important implications for food industry as they offer a means to control and optimise the colour of food dispersions.     

Non-destructive evaluation of quality and ammonia content in whole and fresh-cut lettuce by computer vision system

The paper describes the developed hardware and software components of a computer vision system that extracts colour parameters from calibrated colour images and identifies non-destructively the different quality levels exhibited by lettuce (either whole or fresh-cut) during storage. Several colour parameters extracted by computer vision system have been evaluated to characterize the product quality levels. Among these, brown on total and brown on white proved to achieve a good identification of the different quality levels on whole and fresh-cut lettuce (P-value < 0.0001). In particular, these two parameters were able to discriminate three levels: very good or good products (quality levels from 5 to 4), samples at the limit of marketability (quality level of 3) and waste items (quality levels from 2 to 1). Quality levels were also chemically and physically characterized. Among the parameters analysed, ammonia content proved to discriminate the marketable samples from the waste in both product's typologies (either fresh-cut or whole); even the two classes of waste were well discriminated by ammonia content (P-value < 0.0001).A function that infers quality levels from the extracted colour parameters has been identified using a multi-regression model (R² = 0.77). Multi-regression also identified a function that predicts the level of ammonia (an indicator of senescence) in the iceberg lettuce from a colour parameter provided by the computer vision system (R² = 0.73), allowing a non-destructive evaluation of a chemical parameter that is particularly useful for the objective assessment of lettuce quality.The developed computer vision system offers flexible and simple non-destructive tool that can be employed in the food processing industry to monitor the quality and shelf life of whole and fresh-cut lettuce in a reliable, objective and quantitative way.     

Simple determination of formaldehyde in fermented foods by HS‐SPME‐GC/MS

We describe an automatic method to detect formaldehyde (FA) in some fermented foods. This method is based on derivatisation with 2,2,2‐trifluoroethylhydrazine (TFEH) and consecutive headspace solid‐phase micro‐extraction and gas chromatography‐mass spectrometry. FA in food reacted for 30 min at 40 °C with 2,2,2‐TFEH in a headspace vial, and the formed FA‐TFEH derivative was simultaneously vaporised and adsorbed on 85‐μm carboxen–polydimethylsiloxane. Under the established condition, the limit of detection was 0.1 μg kg^?1 by using 1.0‐g solid food and 1.0‐mL liquid food, and the relative standard deviation was less than 10% at FA concentrations between 0.050 and 0.500 mg kg^?1. The concentrations of FA in several traditional Korean foods including gimchi, watery radish gimchi, soybean paste, red pepper paste, soy sauce and bean‐paste soup were measured. All food samples had detectable levels of FA (0.104–13.05 mg kg^?1).     

The effect of probiotic‐fermented soy milk on enhancing the NO‐mediated vascular relaxation factors

BACKGROUND: Soy milk is one of the common soy‐based foods in Asia. In this study the effects of soy milk fermented with selected probiotics on nitric oxide (NO)‐mediated vascular relaxation factors in cell model systems were investigated. RESULTS: Soy milk fermented with Lactobacillus plantarum TWK10 or Streptococcus thermophilus BCRC 14085 for 48 h showed a greater transformation of glucoside isoflavones to aglycone isoflavones (P < 0.05). An increase in aglycone isoflavones in ethanol extracts from fermented soy milk stimulated NO production and endothelial NO synthase (eNOS) activity in human umbilical vein endothelial cells. It also had a stimulating effect on superoxide anion scavenging and prostaglandin E₂ production. In addition, it enhanced mRNA expression of the E‐prostanoid 4 receptor in rat thoracic aorta smooth muscle cells. Moreover, a small amount of O induced by water extracts from fermented soy milk at low concentration (1 mg mL^?1) increased the content of calcium ions and activated eNOS, thereby promoting NO production and the coupling state of eNOS. CONCLUSION: Soy milk fermented with selected probiotics promotes the relaxation factors of vascular endothelial cells and can be applied in the development of functional foods. © 2012 Society of Chemical Industry     

Comparison of Minolta colorimeter and machine vision system in measuring colour of irradiated Atlantic salmon

BACKGROUND: Minolta and machine vision are two different instrumental techniques used for measuring the colour of muscle food products. Between these two techniques, machine vision has many advantages, such as its ability to determine L*, a*, b* values for each pixel of a sample's image and to analyse the entire surface of a food regardless of surface uniformity and colour variation. The objective of this study was to measure the colour of irradiated Atlantic salmon fillets using a hand‐held Minolta colorimeter and a machine vision system and to compare their performance. RESULTS: The L*, a*, b* values of Atlantic salmon fillets subjected to different electron beam doses (0, 1, 1.5, 2 and 3 kGy) were measured using a Minolta CR‐200 Chroma Meter and a machine vision system. For both Minolta and machine vision the L* value increased and the a* and b* values decreased with increasing irradiation dose. However, the machine vision system showed significantly higher readings for L*, a*, b* values than the Minolta colorimeter. Because of this difference, colours that were actually measured by the two instruments were illustrated for visual comparison. Minolta readings resulted in a purplish colour based on average L*, a*, b* values, while machine vision readings resulted in an orange colour, which was expected for Atlantic salmon fillets. CONCLUSION: The Minolta colorimeter and the machine vision system were very close in reading the standard red plate with known L*, a*, b* values. Hence some caution is recommended in reporting colour values measured by Minolta, even when the ‘reference’ tiles are measured correctly. The reason for this discrepancy in colour readings for salmon is not known and needs further investigation. Copyright © 2009 Society of Chemical Industry     

Green Pea and Garlic Puree Model Food Development for Thermal Pasteurization Process Quality Evaluation

Development and selection of model foods is a critical part of microwave thermal process development, simulation validation, and optimization. Previously developed model foods for pasteurization process evaluation utilized Maillard reaction products as the time–temperature integrators, which resulted in similar temperature sensitivity among the models. The aim of this research was to develop additional model foods based on different time–temperature integrators, determine their dielectric properties and color change kinetics, and validate the optimal model food in hot water and microwave‐assisted pasteurization processes. Color, quantified using a^* value, was selected as the time–temperature indicator for green pea and garlic puree model foods. Results showed 915 MHz microwaves had a greater penetration depth into the green pea model food than the garlic. a^* value reaction rates for the green pea model were approximately 4 times slower than in the garlic model food; slower reaction rates were preferred for the application of model food in this study, that is quality evaluation for a target process of 90 °C for 10 min at the cold spot. Pasteurization validation used the green pea model food and results showed that there were quantifiable differences between the color of the unheated control, hot water pasteurization, and microwave‐assisted thermal pasteurization system. Both model foods developed in this research could be utilized for quality assessment and optimization of various thermal pasteurization processes.     

Modelling the consensus and differences between assessors inspecting the colour quality of apples by ‘tree‐based modelling’

This research investigated the consensus and differences between different quality inspectors assessing the colour quality classification of a two‐coloured apple variety (Malus domestica Borkh cultivar ‘Jonagold’). An image analysis system measured the colour characteristics objectively. These objective measures formed the basic reference to compare the grading behaviour. Different quality inspectors were asked to assess the colour quality of different sets of apples applying the commercial quality standards. Agreement and association measures of the intra‐inspector contingency tables indicated a moderate ability of the inspectors to reconstruct their own quality classification. If the intra‐agreement and intra‐association were lower than a threshold, the repeatability of the inspector's assessments was considered as too poor and his classification results were omitted from the model development that simulated the apple colour classification behaviour. The statistical method ‘tree‐based modelling’ was applied to connect the individual quality assignments with the objective apple colour measurements. These models indicated that the blush colour was more important than the blush area for the quality assessment.The individual grading decision models were compared by correspondence analysis. Four different grading archetypes were detected and simulated. The predictive power of the four archetypal models was much higher than the predictive power of a consensus model including all inspectors. These results underlined the fact that poorly defined commercial standards lead to different quality interpretation. © 2000 Society of Chemical Industry