Search

Button mushroom (Agaricus bisporus) is widely used on the world market. Browning of mushroom surface is the most intuitive factor affecting consumer purchase. Developments of browning during postharvest are related to the flush number, mechanical damage, polyphenol oxidase, phenolic compounds, pathogens and postharvest environment. Many techniques such as cooling, modified atmosphere packaging (MAP) and coating are developed to control button mushroom browning. In this review, the main factors involved in mushroom browning including mushroom tyrosinase, postharvest factors and pathogen are discussed, and some detection methods for mushroom surface color are introduced. Furthermore, different methods for controlling browning including MAP, polyphenol oxidase (PPO) inhibitors, edible coating, irradiation, cooling and other emerging techniques are reviewed and compared. Internal factors of button mushroom browning include subtracts, enzyme and mushroom cellular integrity, which are affected by postharvest factors such as mechanical damage, temperature, relative humidity, and atmosphere gas. Imaging or computer vision and hyperspectral imaging are emerging methods for the detection of mushroom color and can be used to analyze each pixel of the entire surface of mushrooms. In order to control mushroom browning, preservation methods including MAP, coating, PPO inhibitors and irradiation are extensively studied. In addition, thermally buffered package, commercial film materials and gas composition of MAP, natural coating and PPO inhibitor materials should be investigated in future. • Button mushroom browning is affected by postharvest factors. • Modified atmosphere packaging is often used among all the control methods. • Tendency of inhibitory rate for L* during storage varies with control methods. • Computer vision is a rapid and reliable method for detecting mushroom color. [ABSTRACT FROM AUTHOR]

Cheese is produced around the world in a wide range of flavours, varieties, textures, and shapes, which can be used as a final product in human diet, and as an important ingredient in various foods. With consumer's continuing demand for quality cheese as well as increasing challenge from production facing the industry, nondestructive techniques are increasingly used to evaluate cheese quality. Considering the rapid development of novel nondestructive techniques, relevant literatures in the past 15 years (2004–2018) are reviewed in this paper. The main quality attributes of cheese and the importance of evaluating these attributes are discussed. The principles, developments and applications of computer vision (CV), computed tomography (CT), X-ray system, magnetic resonance imaging (MRI), fluorescence spectroscopy, near-infrared (NIR) spectroscopy, Fourier transform infrared (FTIR) spectroscopy, hyperspectral imaging (HSI), Raman imaging, ultrasonic and acoustic sensing and other nondestructive techniques are discussed in this review. Among all nondestructive techniques used for cheese quality evaluation, fluorescence spectroscopy is the most used method for classification, NIR spectroscopy is mainly used for predicting chemical components, FTIR spectroscopy shows the greatest scope of applications, while CV, X-ray and MRI are only used for monitoring eye growth. HSI and Raman imaging begins to attract research interest, and synchronous fluorescence spectroscopy gradually replaces front-face fluorescence spectroscopy, whereas CV is no longer applied to cheese evaluation and literatures about NIR spectroscopy also becomes less in recent years. It is hoped that this review should provide information on current research gaps and set some directions for future studies. • Nondestructive techniques for cheese quality evaluation are reviewed. • Cheese attributes and their importance are presented. • Applicability and the pros and cons of nondestructive techniques are discussed. • Future research directions are suggested. [ABSTRACT FROM AUTHOR]

This study explored the potential of computer vision system (CVS) and hyperspectral imaging (HSI) technique covering spectral range of 900–1700 nm for identifying freezer burnt salmon fillets after frozen storage. Local binary pattern (LBP) descriptor was applied for the RGB image classification. Reflectance spectra were obtained from various positions surface and pretreated using the standard normal variate (SNV) transformation. TreeBagger classifier was used to build classification models for recognition and authentication of the freezer burnt flesh. The results suggested that hyperspectral discrimination performed much better than CVS with the correct classification rate (CCR) of 0.905 in validation and CCR of 0.945 in cross-validation. The effective wavelengths were selected based upon the feature importance in the TreeBagger model and the corresponding optimized model yielded CCR of 0.914 in validation and 0.978 in cross-validation. Overall, the outcome suggested the capability of HSI for rapid categorization of damaged regions on frozen salmon. [ABSTRACT FROM AUTHOR]

Colour is the first quality attribute of food evaluated by consumers, and is therefore an important component of food quality relevant to market acceptance. Rapid and objective measurement of food colour is required in quality control for the commercial grading of products. Computer vision is a promising technique currently investigated for food colour measurement, especially with the ability of providing a detailed characterization of colour uniformity at pixel-based level. This paper reviews the fundamentals and applications of computer vision for food colour measurement. Introduction of colour space and traditional colour measurements is also given. At last, advantages and disadvantages of computer vision for colour measurement are analyzed and its future trends are proposed. [Copyright &y& Elsevier]

Abstract: A new algorithm for the conversion of device dependent RGB colour data into device independent L*a*b* colour data without introducing noticeable error has been developed. By combining a linear colour space transform and advanced multiple regression methodologies it was possible to predict L*a*b* colour data with less than 2.2 colour units of error (CIE 1976). By transforming the red, green and blue colour components into new variables that better reflect the structure of the L*a*b* colour space, a low colour calibration error was immediately achieved (ΔECAL =14.1). Application of a range of regression models on the data further reduced the colour calibration error substantially (multilinear regression ΔECAL =5.4; response surface ΔECAL =2.9; PLSR ΔECAL =2.6; LASSO regression ΔECAL =2.1). Only the PLSR models deteriorated substantially under cross validation. The algorithm is adaptable and can be easily recalibrated to any working computer vision system. The algorithm was tested on a typical working laboratory computer vision system and delivered only a very marginal loss of colour information ΔECAL =2.35. Colour features derived on this system were able to safely discriminate between three classes of ham with 100% correct classification whereas colour features measured on a conventional colourimeter were not. [Copyright &y& Elsevier]

Computer vision has emerged as a useful alternative to manual expert grading of meat in recent years. Manual grading by experts has a number of essential flaws that can be effectively mitigated with computer vision technology. Computer vision technology is a simple and affordable alternative that, while requiring some expertise, is not excessively technologically demanding. Computer vision technology has shown key attributes of flexibility and ease of compatibility allowing a wide range of meat quality assessment challenges to be successfully tackled. The simplest approaches involving visible light imaging and explicit statistical modelling have proven adequate on the vast majority of occasions. On the other hand, for difficult tasks, more expensive non-visible wavelength imaging and implicit statistical modelling are required. In addition, fully automatic image segmentation still remains a difficult problem in many instances, although image processing is becoming more powerful particularly as computationally demanding texture characterisation algorithms become more viable as computation speeds increase. Finally training and testing a computer vision system will require considerable groundwork as a substantial amount of image and independent meat quality data will be required for good model calibration and validation. [Copyright &y& Elsevier]

Abstract: A method to discriminate between various grades of pork and turkey ham was developed using colour and wavelet texture features. Image analysis methods originally developed for predicting the palatability of beef were applied to rapidly identify the ham grade. With high quality digital images of 50–94 slices per ham it was possible to identify the greyscale that best expressed the differences between the various ham grades. The best 10 discriminating image features were then found with a genetic algorithm. Using the best 10 image features, simple linear discriminant analysis models produced 100% correct classifications for both pork and turkey on both calibration and validation sets. [Copyright &y& Elsevier]

Abstract: A new study was conducted to apply computer vision methods successfully developed using trained sensory panel palatability data to new samples with consumer panel palatability data. The computer vision methodology utilized the traditional approach of using beef muscle colour, marbling and surface texture as palatability indicators. These features were linked to corresponding consumer panel palatability data with the traditional approach of partial least squares regression (PLSR). Best subsets were selected by genetic algorithms. Results indicate that accurate modelling of likeability with regression models was possible (r 2 =0.86). Modelling of other important palatability attributes proved encouraging (tenderness r 2 =0.76, juiciness r 2 =0.69, flavour r 2 =0.78). Therefore, the current study provides a basis for further expanding computer vision methodology to correlate with consumer panel palatability data. [Copyright &y& Elsevier]

Abstract: Textural patterns are often complex, exhibit scale-dependent changes in structure and are difficult to identify and describe. The lacunarity morphometric uses multiscale windowing to measure the scale dependency of spatial heterogeneity. In the current work, the objectives are to investigate the usefulness of lacunarity, using different colour scales (R,G,B, L ∗, a ∗, b ∗, and Gray), as a quantitative descriptor of visual texture in sliced ham surfaces, and compare these results with the binary approach developed by . Images were acquired from three qualities of sliced pork ham, typically consumed in Ireland (200slices/quality). Lacunarity plots reveal important textural content information that corresponds to degree of spatial heterogeneity of intensities and level of self-similar behaviour. The results of intensity lacunarity suggest that window sizes up to 10pixels may be adequate enough to cover textural features and produce meaningful results. Once the box size is larger than 10, lacunarity curves either converge or display atypical behaviour and then decay. Investigation confirmed that both binary and intensity lacunarity approaches are useful quantitative descriptors of visual texture in sliced ham images. Moreover, potential future research directions were suggested for computing lacunarity in colour images directly. [Copyright &y& Elsevier]

Abstract: The palatability of beef has been investigated with digital imaging systems on numerous previous occasions. In the current study, a novel approach was applied using high magnification imaging to develop surface texture features and an alternative colour space greyscale to express muscle surface texture. An automatic segmentation method was applied to develop colour and marbling features and best regression model subsets were selected automatically with genetic algorithms. Results indicated that accurate modelling of beef acceptability with regression models was possible with r 2 up to 0.95. Modelling of acceptability using high magnification images proved more successful than modelling with low magnification images. Linear models performed well compared to non-linear models. Other sensory measurements particularly TPA hardness were more difficult to model, although an accurate model of juiciness was developed. Addition of non-linear terms did not give large improvements except for juiciness. [Copyright &y& Elsevier]

Abstract: An algorithm for automatic segmentation of beef longissimus dorsi (LD) muscle and marbling has been developed. The algorithm used simple thresholding to remove the background and then used clustering and thresholding with contrast enhancement via a customised greyscale to remove marbling. It was possible to attain lean muscle free of obvious marbling or background pixels where specular reflection could be effectively mitigated. Features of the automatically derived LD muscle and marbling images were compared to corresponding features of LD muscle and marbling images derived with a segmentation method requiring manual completion. Very strong correlations (up to r =1) were found between the colour features of both sets of LD muscle images. Strong correlations (up to r =0.96) were found between the features of both sets of marbling images. The automatic segmentation method has shown its good ability to approximate colour and marbling features. The algorithm has adaptable parameters and can be retailored to suit different image acquisition environments. [Copyright &y& Elsevier]

Abstract: The wavelet transform can be used to characterise the surface texture of beef images in a more efficient manner than classical algorithms such as co-occurrence and run lengths. Features extracted from wavelet decompositions have been used to develop predictive models of important palatability attributes. A variety of common wavelet transforms were considered (biorthogonal, reverse biorthogonal, discrete Meyer, Daubechie, symmetric modified Daubechie and Coifman modified Daubechie) to search for the most useful texture features. A classic run length and co-occurrence algorithm was used for comparison. Using the same data analysis methods for each wavelet type, predictive models of beef acceptability, tenderness, juiciness, flavour and hardness were developed. Genetic algorithms succeeded in finding more accurate models than stepwise and manual elimination except for hardness. An accurate model of flavour (r 2 =0.84) was computed. A good model of overall acceptability (r 2 =0.79) was computed that fell just short of an important benchmark of accuracy. An encouraging model of juiciness (r 2 =0.71) was computed showing that with additional palatability information juiciness might be accurately modelled. Tenderness proved difficult to model with only the classic model satisfying stability criteria and a poorer result (r 2 =0.64) meaning substantial additional palatability information is required for accurate modelling. Hardness was particularly difficult to model. The biorthogonal wavelet produced the best model for three palatability measurements but the symmetric modified Daubechie wavelet produced the best model of overall acceptability and thus must be viewed as the most useful wavelet type. [Copyright &y& Elsevier]

Abstract: Beef longissimus dorsi surface texture is an indicator used in predicting beef palatability by expert graders. Computer vision systems have previously used imaging at normal view to develop surface texture features with some success. Good models of beef overall acceptability using imaging at high magnification have been recently developed. As a comparison the same surface texture features were computed from the corresponding images at normal view and used to model overall acceptability. Both sets of texture features were also combined with muscle colour and marbling features and used to model overall acceptability. Models using texture features alone were more successful at normal modality. However colour and marbling features combined much better with texture features at high modality to yield the most accurate model of overall acceptability (r 2 =0.93). Accurate Partial Least Squares Regression (PLSR) models were computed at both modalities with and without inclusion of colour and marbling features. Addition of squared terms to the models failed to improve accuracy. [Copyright &y& Elsevier]

Abstract: Colour, marbling and surface texture properties of beef longissimus dorsi muscle are used in some countries to grade carcasses according to their expected eating quality. Handheld VIA systems are being used to augment the grader assessments, however attempts have been made to develop higher resolution image systems to give consistent and objective predictions of quality based on these properties. Previous efforts have been unable to model sufficiently the variation in eating quality. A new approach has been applied whereby beef carcasses were subjected to homogenous post-slaughter treatment to minimize variation in eating quality related to other factors such as chilling temperature and hanging method. Furthermore a wider range of features were used to better characterize colour and marbling and the wavelet transform was used to characterize texture. Objective and sensory panel tests were performed to evaluate the beef eating qualities. Classical statistical methods of multilinear regression (MLR) and partial least squares regression (PLSR) were used to develop predictive models. It was possible to explain a greater portion of variation in eating quality than before (up to r 2=0.83). Carcasses were classified as high or low quality with a high rate of correct classifications (90%). Genetic algorithms were used to select the model subsets. [Copyright &y& Elsevier]

Abstract: 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 (r 2 =0.88 for sensory overall acceptability and r 2 =0.85 for 7-day WBS). Addition of non-linear texture terms to the models gave some improvements. [Copyright &y& Elsevier]

Abstract: An automatic method for estimating the content of intramuscular fat (IMF) in beef M. longissimus dorsi (LD) was developed using a sequence of image processing algorithm. To extract IMF particles within the LD muscle from structural features of intermuscular fat surrounding the muscle, three steps of image processing algorithm were developed, i.e. bilateral filter for noise removal, kernel fuzzy c-means clustering (KFCM) for segmentation, and vector confidence connected and flood fill for IMF extraction. The technique of bilateral filtering was firstly applied to reduce the noise and enhance the contrast of the beef image. KFCM was then used to segment the filtered beef image into lean, fat, and background. The IMF was finally extracted from the original beef image by using the techniques of vector confidence connected and flood filling. The performance of the algorithm developed was verified by correlation analysis between the IMF characteristics and the percentage of chemically extractable IMF content (P <0.05). Five IMF features are very significantly correlated with the fat content (P <0.001), including count densities of middle (CDMiddle) and large (CDLarge) fat particles, area densities of middle and large fat particles, and total fat area per unit LD area. The highest coefficient is 0.852 for CDLarge. [Copyright &y& Elsevier]

Abstract: The classification of pizza base, sauce spread and topping is highly sensitive to human error for its subjective and inconsistent nature. Image processing techniques combined with machine learning provide an objective and consistent way to accomplish this task. By using a combination of several binary classifiers, support vector machine (SVM) is a state-of-the-art learning algorithm for multi-classification of pizza base, sauce spread, and topping. With the selected features as input, the one-versus-one and directed acyclic graph (DAG) methods achieved 89.17% and 88.33% multi-classification accuracy respectively for pizza base, both 87.5% for pizza sauce spread, and 80.83% and 80.00%, respectively for pizza topping. The results showed that the computer vision systems developed had a great potential to assist in the automatic multi-classification of pizza base, sauce spread, and topping. [Copyright &y& Elsevier]

Abstract: Images were acquired from 25 beef samples before and after cooking and shrinkages were measured from the images in four different ways: volume, surface area, major axis, and minor axis. A total of 15 factors relevant to the samples, which included weight, moisture content, volume, surface area, major and minor axes, cooking temperature, projected area and perimeter, as well as the mean, standard deviation, and Fourier descriptors of both radiuses and lengths of power curves, were trained in an error backpropagation network in order to correlate them to the shrinkages. The correlation coefficients (r 2) were 0.684, 0.674, and 0.745 for the shrinkage of volume, surface area, and major axis, respectively, indicating that the method worked with sufficient confidence in predicting the shrinkage of these three parameters. However, the correlation coefficient for minor axis was only 0.42, showing the limitation of the method in predicting the shrinkage of minor axis. The difference of shrinkage between major and minor axis was possibly caused by different heat transfer behaviour along the axes. Sensitivity analyses were conducted to further explore the ability of the above 15 factors in predicting the shrinkage. Results showed that among these 15 factors, mean length of power curves, projected perimeter, and cooking temperature played the most important role in determining all the four kinds of shrinkage. [Copyright &y& Elsevier]

Abstract: Thirty-six tumbled samples from triceps brachii (beef) were water-immersion cooked and cooled by vacuum, air blast, and cold room cooling. A computer vision system was set up to obtain images of the samples. Colour features including the mean and the standard deviation in two colour spaces, Red, Green, Blue (RGB) and Hue, Saturation, Intensity (HSI) were extracted. Moisture content of the samples was determined by chemical analysis. A partial least square regression (PLSR) model and a neural network (NN) model were proposed for correlating the colour to the moisture content of the beef joints. Correlation coefficients (r 2) of the models were 0.56 (PLSR) and 0.75 (NN). A stepwise selection together with the analysis of the regression coefficients of the PLSR model both showed that among the 12 colour features analysed, saturation was the one that had the largest contribution to the results of the prediction model. However, only saturation itself was not sufficient for establishing the correlation between meat colour and its moisture content. [Copyright &y& Elsevier]

Abstract: An automatic method for estimating the surface area and volume of ham was developed using a sequence of image processing algorithm. To extract the shape of ham, three steps of image processing algorithm were firstly developed, i.e. image segmentation, noise reduction, and edge detection. Then the protrusion of ham formed by elastic netting was deleted with four steps, namely obtaining radii, wavelet transform, locating protrusion, and spline interpolation. Finally, based on the new shape of ham, the surface area and volume were estimated by two methods, i.e. the derived method and the partitioned method. The former was based on the three principal dimensions, i.e. length (L), width (W), and thickness (T), and the surface area and volume were derived mathematically, while the latter firstly divided ham into a number of sections and then summed the surface area and volume of each section to obtain the entire ones. The results obtained have demonstrated that the approaches proposed have the ability to estimate the surface area and volume of ham. [Copyright &y& Elsevier]

Abstract: Pores formed in pork ham have a significant effect on its quality. However, they are mostly characterised using manual methods with special devices. In this paper, an automatic method for pore characterisation of pork ham was developed using computer vision. To segment pores from images of pork ham, three stages of image processing algorithm were developed, i.e., ham extraction, image enhancement, and pore segmentation. From the segmented pores, the porosity, number of pores, pore size, and size distribution were measured. The statistical analysis showed that 79.81% of pores have area sizes between 6.73×10−3 and 2.02×10−1 mm2. Furthermore, it was found that the total number of pore (TNP) and porosity highly negatively related to the water content of pork ham (P <0.05), and had negative correlations with the cooking and cooling time. However, for texture analysis, positive correlations were found between the pore characterisations and WBS, hardness, cohesion, and chewiness, respectively, while springiness and gumminess were negatively related to TNP and porosity. [Copyright &y& Elsevier]

Abstract: In this work, computer vision techniques were applied to estimate the shrinkages of cooked meat joints during the air-blast cooling process. Three kinds of shrinkage were evaluated in the current research, i.e. shrinkages of major and minor axes, volume, and surface area. Since beef joints are normally in irregular ellipse shape, a region-based ellipse fitting method was employed to measure their major and minor axes before and after cooling. To calculate their volume and surface area, the images of beef samples were divided into numerous cross sections and each cross section was considered as a cylindrical disc. Based on the obtained values of beef joints using image processing techniques, linear models (r 2 >0.990) were developed for predicting the shrinkages of major and minor axes, volume, and surface area, respectively. [Copyright &y& Elsevier]

Abstract: Learning techniques have been applied increasingly for food quality evaluation using computer vision in recent years. This paper reviews recent advances in learning techniques for food quality evaluation using computer vision, which include artificial neural network, statistical learning, fuzzy logic, genetic algorithm, and decision tree. Artificial neural network (ANN) and statistical learning (SL) remain the primary learning methods in the field of computer vision for food quality evaluation. Among the applications of learning algorithms in computer vision for food quality evaluation, most of them are for classification and prediction, however, there are also some for image segmentation and feature selection. In this paper, the promise of learning techniques for food quality evaluation using computer vision is demonstrated, and some issues which need to be resolved or investigated further to expedite the application of learning algorithms are also discussed. [Copyright &y& Elsevier]

Abstract: Five transformations of RGB (red, green, and blue) colour space were evaluated for their performance in classifying pizza toppings, i.e., NRGB (normalised RGB), HSV (hue, saturation, and value), I1I2I3, L*a*b*, and YCbCr. Using these five colour space transformations, the performance of three SVM (support vector machine) classifiers (linear, polynomial, and RBF) on pizza topping classification was compared with two classical classification approaches, i.e., C4.5 classifier and an RBF_NN (radial basis function neural network) classifier. The C4.5 classifier obtained the best classification accuracy of 93.3% with L*a*b* or I1I2I3 colour space transformation, and the RBF_NN classifier achieved the best classification accuracy of 86.7% with YCbCr, HSV or L*a*b* colour space transformation. For the SVM classifiers, the polynomial SVM classifier had the best classification accuracy of 96.7% with HSV colour space transformation, while the radial basis function (RBF) SVM classifier obtained the best classification accuracy of 90.0% with YCbCr, L*a*b* or HSV colour space transformation. Among the SVM classifiers, the polynomial SVM classifier combined with HSV colour space transformation proved to be a good approach for the classification of pizza toppings using computer vision. [Copyright &y& Elsevier]

An automated classification system of pizza sauce spread using colour vision and support vector machine (SVM) was developed. To characterise pizza sauce spread with low dimensional colour features, a sequence of image processing algorithms was developed. After image segmentation from the background, the segmented image was transformed from red, green, and blue (RGB) colour space to hue, saturation, and value (HSV) colour space. Then a vector quantifier was designed to quantify the HS (hue and saturation) space to 256-dimension, and the quantified colour features of pizza sauce spread were represented by colour histogram. Finally, principal component analysis (PCA) was applied to reduce the 256-dimensional vectors to 30-dimensional vectors. With the 30-dimensional vectors as the input, SVM classifiers were used for classification of pizza sauce spread. It was found that the polynomial SVM classifiers resulted in the best classification accuracy with 96.67% on the test experiments. [Copyright &y& Elsevier]

The quality of pizza base should be determined before further processing. However such classification is highly sensitive to human error. Image processing techniques can be used to extract shape features from digital images of pizza bases. The support vector machine (SVM) is the state-of-the-art classification technique and is capable of learning in high-dimensional feature space with less training data. In this paper, an algorithm for the shape extraction of pizza base and a new classification algorithm based on Fourier transform and SVM for shape grading of pizza bases are presented. The experiments showed that 86.7% classification accuracy was achieved with a linear SVM classifier, 95.0% with a polynomial SVM classifier, and 98.3% with a Gaussian radial basis function SVM classifier. The computer vision system developed has a great potential to assist in the automation of pizza base classification. [Copyright &y& Elsevier]

An algorithm for segmenting images using stick growing and merging is described, which consists of four major steps: stick initialisation, stick merging, subregion merging, and boundary modification. It is started from an initial decomposition of the image into small sticks and non-sticks. The small sticks are merged to obtain the initial subregions on the basis of homogeneity criteria. Then smaller subregions with only one stick are merged into larger subregions and subsequently all subregions are merged into regions according to the criteria. Finally, non-sticks and separate small sticks are merged and the degree of boundary roughness is reduced by boundary modification. The algorithm was successfully used to segment many types of complex food images including pizza, apple, pork and potato. [Copyright &y& Elsevier]

Image-related parameters including the difference/ratio of luminance features and percentage oil area (POA) were examined under different cooking conditions to investigate if they could be used as oiling off indexes of cheese in a computer vision method. POA was most suitable because, in addition to its high correlation with the fat leakage obtained from the traditional oil ring test, the information regarding the oiling off property of cheese provided by POA was in good accordance with observation. Free oil formation of both Cheddar and Mozzarella cheeses occurred rapidly during the initial stage of cooking, followed by a notable decrease afterwards. For the Cheddar cheese, obvious oiling off was observed at all temperatures from 70 to 200 °C, while the maximum oiling off occurred at 160 °C. Little free oil was formed for the Mozzarella cheese below 100 °C and the highest value of POA was observed at 130 °C. The oiling off property of cheese was significantly influenced by sample dimensions. Regardless of sample dimensions, the maximum oiling off of Cheddar cheese occurred at 1 min of cooking and little free oil was observed after cooking for 4 min. As the side length of cheese slices increased from 25 to 70 mm, the increase in POA after 1 min cooking can be described by a square polynomial. Free oil formation was limited for Mozzarella cheese in spite of a wide range of heating temperature and sample dimensions applied. [Copyright &y& Elsevier]

A wide range of image features including luminance features, histogram related features and percentage oil area were extracted from images of cheese before and after cooking. Parameters calculated from these image attributes were correlated to the fat leakage determined with the traditional oil ring test using filter paper to investigate if the oiling off property of cheese can be evaluated with computer vision methods. Among the parameters derived from the luminance features, the difference of the number of pixels and the ratio of integrated grey value after and before cooking were highly correlated with fat leakage. However, none of the difference/ratio after and before cooking obtained from the histogram related features was significantly correlated with fat leakage. Percentage oil area was correlated to fat leakage with an R value of 0.852 and was a promising parameter for evaluating the oiling off property of cheese with computer vision. The correlation between image luminance features and fat leakage was temperature dependent. Stronger correlation with fat leakage was found for some image parameters extracted at 130 than at 110 °C. [Copyright &y& Elsevier]

With increased expectations for food products of high quality and safety standards, the need for accurate, fast and objective quality determination of these characteristics in food products continues to grow. Computer vision provides one alternative for an automated, non-destructive and cost-effective technique to accomplish these requirements. This inspection approach based on image analysis and processing has found a variety of different applications in the food industry. Considerable research has highlighted its potential for the inspection and grading of fruits and vegetables. Computer vision has been successfully adopted for the quality analysis of meat and fish, pizza, cheese, and bread. Likewise grain quality and characteristics have been examined by this technique. This paper presents the significant elements of a computer vision system and emphasises the important aspects of the image processing technique coupled with a review of the most recent developments throughout the food industry. [Copyright &y& Elsevier]

The growth in the popularity and consumption of pizzas has led to the need for automated quality inspection, so as to maintain confidence and increase production efficiency. In the current research the use of computer vision for inspection of pizza base and tomato sauce spread quality was investigated. Twenty pizza bases were analysed for base area, spatial ratio I (SRI), spatial ratio II (SRII), and circularity. For the measurement of the sauce spread characteristics based on 25 samples, the indexes taken were sauce area and heavy area percentage; a fuzzy logic system was then developed to classify the sauce spread samples into classes of acceptable and defective quality. The base area analysis gave a classification error of 13% when compared to human assessment. The experimental results for the sauce spread analysis show that by using computer vision in conjunction with fuzzy logic a classification accuracy of 92% can be achieved. [Copyright &y& Elsevier]

The ever increasing consumer needs and quality requirements have led to the necessity for more objective and accurate assessment of individual pizza topping quantity and distribution. In the current study the use of computer vision for the evaluation of different features of pizza topping quality were assessed. The indexes used for the analysis of the twenty-five samples examined included ham area percentage, mushroom area percentage and topping area percentage (TAP). A fuzzy logic system was then developed and used to classify the pizza topping quality, in comparison with quality personnel assessment. The TAP index gave the lowest ambiguous degree value hence indicating that it displays the least fuzzyness. A classification error of 24% was determined for the five linguistic personnel classes. When only two-classification levels (i.e. acceptable quality and defective quality) were considered an accuracy of 100% was achieved. [Copyright &y& Elsevier]

A computer vision method was developed to determine the colour change of cheese during heating. The browning properties of both Cheddar and Mozzarella cheeses were analysed and compared. The influence of baking temperature (70–200 °C) and time (0–20 min) on the browning property of cheese was significant for both cheeses. However, the time and temperature dependencies of a specially defined browning factor (BF) were different for Cheddar and Mozzarella cheeses. Cooking for 2–4 min, the BF of Mozzarella cheese increased almost linearly with baking temperature. As for the Cheddar cheese, after heating for 8–12 min, a linear relationship between the BF and baking temperature was observed from 70 to 130 °C. Further browning hardly occurred when cooking temperature increased from 130 to 160 °C, probably because of depletion of reducing sugars and amino groups in cheese that are necessary for the browning reaction. When heating at temperatures above 160 °C, the colour of Cheddar cheese darkened dramatically due to scorching. The results also show that the computer vision method developed provides an objective and efficient approach for assessing the browning of cheese. [Copyright &y& Elsevier]

Computer vision is a rapid, economic, consistent and objective inspection technique, which has expanded into many diverse industries. Its speed and accuracy satisfy ever-increasing production and quality requirements, hence aiding in the development of totally automated processes. This non-destructive method of inspection has found applications in the agricultural and food industry, including the inspection and grading of fruit and vegetable. It has also been used successfully in the analysis of grain characteristics and in the evaluation of foods such as meats, cheese and pizza. This paper reviews the progress of computer vision in the agricultural and food industry, then identifies areas for further research and wider application the technique. [Copyright &y& Elsevier]

The influence of dimensions of cheese slices on the melting property of cheese was investigated with a computer vision method using image analysis techniques. Tests were carried out with Cheddar and Mozzarella cheeses. The results show that the meltability of cheese increased linearly when the side length of square cheese slices increased from 25 to 70 mm. The meltability of cheese also increased proportionally as the thickness of cheese slices was increased from 2 to 4 mm. The influences of the area/thickness of cheese slices on cheese meltability were similar for both the Cheddar and Mozzarella cheeses. The opposite changes may occur for different meltability indices when the initial area of cheese slices varied. [Copyright &y& Elsevier]

A computer vision system was used to evaluate the melting property of cheese. The method is suitable for both cheddar and mozzarella cheese. The effect of cooking time (0–20 min) and temperature (70–200°C) on the melting degree and melting rate of cheddar and mozzarella cheese was investigated and the melting property of both cheeses were compared. Both cheddar and mozzarella cheese melted rapidly at the initial stage of cooking and melting slowed down in about 3–4 min. Maximum melting degree and melting rate occurred at about 130–160°C. Extended cooking and cooking at high temperature may cause reduced melting. Cheddar cheese melted more rapidly and considerably than mozzarella cheese at all cooking temperatures. [Copyright &y& Elsevier]

The development of techniques and methods for rapidly and reliably detecting and analysing food quality and safety products is of significance for the food industry. Traditional machine learning algorithms based on handcrafted features normally have poor performance due to their limited representation capacity for complex food characteristics. Recently, the convolutional neural network (CNN) emerges as an effective and potential tool for feature extraction, which is considered the most popular architecture of deep learning and has been increasingly applied for the detection and analysis of complex food matrices. In the current review, the structure of CNN, the method of feature extraction based on 1-D, 2-D and 3-D CNN models, and multi-feature aggregation methods are introduced. Applications of CNN as a depth feature extractor for detecting and analyzing complex food matrices are discussed, including meat and aquatic products, cereals and cereal products, fruits and vegetables, and others. In addition, data sources, model architecture and overall performance of CNN with other existing methods are compared, and trends of future studies on applying CNN for food detection and analysis are also highlighted. CNN combined with nondestructive detection techniques and computer vision system show great potential for effectively and efficiently detecting and analysing complex food matrices, and the features based on CNN show better performance and outperform the features handcrafted or those extracted by machine learning algorithms. Although there still remains some challenges in using CNN, it is expected that CNN models will be deployed on mobile devices for real-time detection and analysis of food matrices in future. • The principle and architectures of convolutional neural network are presented. • Feature extraction methods based on convolutional neural network are investigated. • Recent applications of convolutional neural network in food are summarized. • Challenges and future work of convolutional neural network are discussed. [ABSTRACT FROM AUTHOR]

Abstract Background Fishery plays an important role in providing abundant nutrition for consumers as well as in international trade and foreign exchange earner. Due to the abundant endogenous enzymes and psychrophilic bacteria in fish bodies and the fragile organization of fish, fish and fish products are easily perishable and are affected by pre-harvest and postmortem factors, thus decreasing their value. It is thus significant to evaluate fish freshness in an effective and timely manner. Scope and approach In the current review, the principles and applications of novel techniques including enzyme biosensor, electrochemical biosensor, electronic nose, colorimetric sensor, electronic tongue, computer vision techniques, Vis/NIR spectroscopy, HSI spectroscopy and fluorescence spectroscopy as reliable and rapid evaluation tools complementing traditional time-consuming, expensive and cumbersome techniques are highlighted and the advantages and disadvantages of these methods are discussed. Moreover, challenges and future research trends are also proposed. Key findings and conclusions Although recent development of novel technologies has demonstrated their superiority for time-saving, rapid, economic, and even nondestructive and on-line detection of fish freshness, some challenges still remain in optimizing and simplifying bionics techniques, developing algorithms and chemometrics in spectroscopic technologies and improving selectivity and sensitivity in biosensor methods. Further studies are still needed to facilitate the applications of these new evaluation technologies in the fish industry. Highlights • Quality attributes for fish freshness evaluation are provided. • Biosensor, bionics, spectroscopy techniques for freshness evaluation are reviewed. • The advantages and disadvantages of these novel technologies are discussed. • Future works of novel technologies for fish freshness evaluation are suggested. [ABSTRACT FROM AUTHOR]

Background For decades, fruit assessment without off-line interruption and sample destruction was a challenge for producers, researchers and food safety agencies. Efforts have therefore been exerted towards the introduction and advancements of innovative imaging technologies for fast, non-invasive and non-destructive monitoring of ripening and maturity stages of fresh produce. Scope and approach Owing to their non-destructive nature, environmental friendliness, reliability accuracy and simplicity of imaging techniques have been successfully exploited for rapid evaluation of ripening and maturity stages of fruits. The current review provides an overview of recent applications in the last few years. The imaging techniques presented include spectral techniques (Raman imaging, hyperspectral imaging, fluorescence imaging and laser light backscattering imaging), nuclear magnetic techniques (magnetic resonance imaging and soft x-ray) and other techniques including thermal imaging, infrared thermography and microwave imaging. The principles of these imaging techniques are also briefly highlighted, and their applicability is summarized. Key findings and conclusions The contemporary imaging techniques presented in this review portray continuous productiveness as excellent quality assessment, particularly for ripening and maturity analysis tools for fruits. Thus, these novel techniques hold great potentiality to replace conventional procedures, because such traditional approaches lack ability to examine fruit ripening and maturity indicators such as lycopene content, flesh freshness, soluble solid content, and other internal quality parameter rapidly and efficiently, furthermore such techniques are time consuming, laborious, destructive and tedious. [ABSTRACT FROM AUTHOR]

This study was carried out to explore the potential of computer vision system (CVS) and two hyperspectral imaging (HSI) systems covering the visible and short-wave near infrared range (400–1000 nm) and the long-wave near infrared range (897–1753 nm), respectively, for differentiation of organic and conventional farm-raised salmon fillets in fresh and chill-stored conditions. Partial least squares-discriminant analysis (PLS-DA), support vector machine (SVM) and random forests (RF) classifiers were used to build classification models for recognition and authentication of the tested samples. The results suggested that hyperspectral discrimination performed much better than CVS. For the same validation set, the highest correct classification rate (CCR) equivalent to 98.2% was presented in two SVM models, one was built from full spectral variables in the 400–1000 nm and the other from four optimal wavelengths in the same spectral region. The best prediction for CVS was obtained by using PLS-DA with CCR of 83.6% for validation, while the most satisfying results for HSI in 897–1753 nm were achieved by applying SVM algorithm on full spectral region as well as 10 important wavelengths with both CCRs of 92.7% for validation. In a nutshell, hyperspectral imaging in the 400–1000 nm has presented the best predictive ability to differentiate between organic and conventional salmon fillets, while SVM classifier has been confirmed to be very powerful for multivariable analysis in our case. Overall, the outcome achieved from this research strongly suggested the capability of the hyperspectral imaging for objective and rapid categorization of the two salmon varieties under circumstances of fresh and chill-stored conditions. [ABSTRACT FROM AUTHOR]

Abstract: Tenderness is an important attribute influencing consumer opinion about the eating quality of fresh meat. Manual assessment of tenderness requires lengthy procedures with tedious sample preparations. An objective, non-destructive, and rapid technique for assessing meat tenderness is required by the meat industry. In this study, the development of partial least squares (PLS) regression models to relate near-infrared (NIR) reflectance spectra and statistical features (mean, standard deviation, norm-1 energy, norm-2 energy, average residual, and entropy) from discrete wavelet transforms (DWT) of raw porcine longissimus dorsi muscle images, to slice shear force (SSF) instrumental measurements, was investigated. The coefficient of determination (R2) of the PLS regression model was 0.63 when only spectral information from hyperspectral (HS) images was analyzed, while PLS models using DWT features extracted from computer vision (CV) images yielded coefficient of determination of 0.48. By combining them, the R2 increased to 0.75. The study has shown the potential for NIR measurements combined with wavelet features from CV images to provide better correlations with muscle tenderness. Industrial relevance: The study has shown the potential for NIR measurements combined with wavelet features from CV images to provide better correlations with muscle tenderness for the meat industry. [Copyright &y& Elsevier]

Abstract: In this study, a hyperspectral imaging technique was developed to achieve fast, accurate, and objective determination of pork quality grades. Hyperspectral images were acquired in the near-infrared (NIR) range from 900 to 1700nm for 75 pork cuts of longissimus dorsi muscle from three quality grades (PSE, RFN and DFD). Spectral information was extracted from each sample and six significant wavelengths that explain most of the variation among pork classes were identified from 2nd derivative spectra. There were obvious reflectance differences among the three quality grades mainly at wavelengths 960, 1074, 1124, 1147, 1207 and 1341nm. Principal component analysis (PCA) was carried out using these particular wavelengths and the results indicated that pork classes could be precisely discriminated with overall accuracy of 96%. Algorithm was developed to produce classification maps of the tested samples based on score images resulting from PCA and the results were compared with the ordinary classification method. Investigation of the misclassified samples was performed and showed that hyperspectral based classification can aid in class determination by showing spatial location of classes within the samples. [Copyright &y& Elsevier]

Abstract: Lacunarity is about quantifying the degree of spatial heterogeneity in the visual texture of imagery through the identification of the relationships between patterns and their spatial configurations in a two-dimensional setting. The computed lacunarity data can designate a mathematical index of spatial heterogeneity, therefore the corresponding feature vectors should possess the necessary inter-class statistical properties that would enable them to be used for pattern recognition purposes. The objectives of this study is to construct a supervised parsimonious classification model of binary lacunarity data–computed by Valous et al. (2009)–from pork ham slice surface images, with the aid of kernel principal component analysis (KPCA) and artificial neural networks (ANNs), using a portion of informative salient features. At first, the dimension of the initial space (510 features) was reduced by 90% in order to avoid any noise effects in the subsequent classification. Then, using KPCA, the first nineteen kernel principal components (99.04% of total variance) were extracted from the reduced feature space, and were used as input in the ANN. An adaptive feedforward multilayer perceptron (MLP) classifier was employed to obtain a suitable mapping from the input dataset. The correct classification percentages for the training, test and validation sets were 86.7%, 86.7%, and 85.0%, respectively. The results confirm that the classification performance was satisfactory. The binary lacunarity spatial metric captured relevant information that provided a good level of differentiation among pork ham slice images. [ABSTRACT FROM AUTHOR]

Abstract: The visual texture of pork ham slices reveals information about the different qualities and perceived image heterogeneity, which is encapsulated as spatial variations in geometry and spectral characteristics. Detrended Fluctuation Analysis (DFA) detects long-range correlations in nonstationary spatial sequences, by a self-similarity scaling exponent α. In the current work, the aim is to investigate the usefulness of α, using different colour channels (R, G, B, L*, a*, b*, H, S, V, and Grey), as a quantitative descriptor of visual texture in sliced ham surface patterns for the detection of long-range correlations in unidimensional spatial series of greyscale intensity pixel values at 0°, 30°, 45°, 60°, and 90° rotations. Images were acquired from three qualities of pre-sliced pork ham, typically consumed in Ireland (200 slices per quality). Results indicated that the DFA approach can be used to characterize and quantify the textural appearance of the three ham qualities, for different image orientations, with a global scaling exponent. The spatial series extracted from the ham images display long-range dependence, indicating an average behaviour around 1/f-noise. Results indicate that α has a universal character in quantifying the visual texture of ham surface intensity patterns, with no considerable crossovers that alter the behaviour of the fluctuations. Fractal correlation properties can thus be a useful metric for capturing information embedded in the visual texture of hams. [Copyright &y& Elsevier]

Abstract: The quaternionic singular value decomposition is a technique to decompose a quaternion matrix (representation of a colour image) into quaternion singular vector and singular value component matrices exposing useful properties. The objective of this study was to use a small portion of uncorrelated singular values, as robust features for the classification of sliced pork ham images, using a supervised artificial neural network classifier. Images were acquired from four qualities of sliced cooked pork ham typically consumed in Ireland (90 slices per quality), having similar appearances. Mahalanobis distances and Pearson product moment correlations were used for feature selection. Six highly discriminating features were used as input to train the neural network. An adaptive feedforward multilayer perceptron classifier was employed to obtain a suitable mapping from the input dataset. The overall correct classification performance for the training, validation and test set were 90.3%, 94.4%, and 86.1%, respectively. The results confirm that the classification performance was satisfactory. Extracting the most informative features led to the recognition of a set of different but visually quite similar textural patterns based on quaternionic singular values. [Copyright &y& Elsevier]

Abstract: An important goal for characterization of texture appearance is the quantification of spatial patterns. The objective was to investigate the potential usefulness of two fractal metrics based on fast Fourier transform and gliding box lacunarity as descriptors of visual texture in ham slices. Images were acquired from three qualities of sliced pork ham, typically consumed in Ireland (200slices/quality). Unexpected characteristics in textural pattern were revealed; the values of fractal dimension were larger for the smoothest surface. Alternatively, the decreasing trend of the power spectrum intercept towards the smoother premium quality ham showed that it correlates well with the overall magnitude of visual roughness. The results of lacunarity suggest that it has a discriminating power among the three ham qualities and its behaviour resembles the one of an exponential decay function. Results showed that Fourier analysis dimension, power spectrum intercept and lacunarity are important fractal parameters and useful quantitative descriptors that capture information embedded in the spatial structure of the underlying image texture of hams. [Copyright &y& Elsevier]

Abstract: Due to the high variability and complex colour distribution in meats and meat products, the colour signal calibration of any computer vision system used for colour quality evaluations, represents an essential condition for objective and consistent analyses. This paper compares two methods for CIE colour characterization using a computer vision system (CVS) based on digital photography; namely the polynomial transform procedure and the transform proposed by the sRGB standard. Also, it presents a procedure for evaluating the colour appearance and presence of pores and fat-connective tissue on pre-sliced hams made from pork, turkey and chicken. Our results showed high precision, in colour matching, for device characterization when the polynomial transform was used to match the CIE tristimulus values in comparison with the sRGB standard approach as indicated by their values. The [3×20] polynomial transfer matrix yielded a modelling accuracy averaging below 2.2 units. Using the sRGB transform, high variability was appreciated among the computed (8.8±4.2). The calibrated laboratory CVS, implemented with a low-cost digital camera, exhibited reproducible colour signals in a wide range of colours capable of pinpointing regions-of-interest and allowed the extraction of quantitative information from the overall ham slice surface with high accuracy. The extracted colour and morphological features showed potential for characterizing the appearance of ham slice surfaces. CVS is a tool that can objectively specify colour and appearance properties of non-uniformly coloured commercial ham slices. [Copyright &y& Elsevier]

Abstract: As a consequence of market competition, the production and manufacture of cheese products are at the stage of innovative dynamics. Pasteurized cheese with vegetable ingredients is one of the new products that may be added to sandwiches, salads, sauce, toast and pizza. Since vegetable ingredients can improve the nutritive value and flavour of cheese, it will probably become more and more popular in the future. Such new products require new techniques for monitoring and evaluating their quality in order to satisfy the increased awareness and expectations of consumers. Computer vision methods have been used increasingly in the food industry for inspection and evaluation purposes as they provide a rapid, economical, consistent and objective assessment. The aim of this study was to develop a computer vision method for inspecting two major quality attributes of pasteurized cheese, i.e. the distribution and amount of ingredients. An image pre-processing algorithm was first developed to delete the border area of cheese. Next a three-step method for ingredient extraction was developed, comprising colour quantification, ingredient location, and mask operation. Finally, the distribution and amount of each ingredient was calculated automatically. Two kinds of pasteurized cheese were evaluated using the above method, i.e. (a) cheese with garlic and parsley and (b) cheese with a mixture of vegetables composed mainly of pepper and parsley. It was found that the distribution and amount of ingredients in the first set of samples were determined within an accuracy of over 88%, compared with the results of a sensory method. As for the second set of samples, accuracies of over 81% and over 71% were achieved for measuring the distribution and amount of ingredients, respectively. [Copyright &y& Elsevier]

Abstract: Images of three qualities of pre-sliced pork and Turkey hams were evaluated for colour and textural features to characterize and classify them, and to model the ham appearance grading and preference responses of a group of consumers. A total of 26 colour features and 40 textural features were extracted for analysis. Using Mahalanobis distance and feature inter-correlation analyses, two best colour [mean of S (saturation in HSV colour space), std. deviation of b∗, which indicates blue to yellow in L∗a∗b∗ colour space] and three textural features [entropy of b∗, contrast of H (hue of HSV colour space), entropy of R (red of RGB colour space)] for pork, and three colour (mean of R, mean of H, std. deviation of a∗, which indicates green to red in L∗a∗b∗ colour space) and two textural features [contrast of B, contrast of L∗ (luminance or lightness in L∗a∗b∗ colour space)] for Turkey hams were selected as features with the highest discriminant power. High classification performances were reached for both types of hams (>99.5% for pork and >90.5% for Turkey) using the best selected features or combinations of them. In spite of the poor/fair agreement among ham consumers as determined by Kappa analysis (Kappa-value<0.4) for sensory grading (surface colour, colour uniformity, bitonality, texture appearance and acceptability), a dichotomous logistic regression model using the best image features was able to explain the variability of consumers’ responses for all sensorial attributes with accuracies higher than 74.1% for pork hams and 83.3% for Turkey hams. [Copyright &y& Elsevier]