Electronic noses for food quality: A review

Journal of Food Engineering 144 (2015) 103–111

Contents lists available at ScienceDirect

Journal of Food Engineering journal homepage: www.elsevier.com/locate/jfoodeng

Review

Electronic noses for food quality: A review Amy Loutfi a, Silvia Coradeschi a, Ganesh Kumar Mani b, Prabakaran Shankar b, John Bosco Balaguru Rayappan b,⇑ a b

Center for Applied Autonomous Sensors Systems, Fakultetsgatan 1, Orebro, Sweden Centre for Nanotechnology & Advanced Biomaterials (CeNTAB), School of Electrical & Electronics Engineering (SEEE), SASTRA University, Thanjavur 613 401, India

a r t i c l e

i n f o

Article history: Received 21 March 2014 Received in revised form 24 July 2014 Accepted 25 July 2014 Available online 4 August 2014 The authors dedicate this review article to Prof. Silvia Coradeschi who sadly passed away in February 2014. Prof. Silvia was known for her unique leadership quality and dedication with unparalleled energy and enthusiasm. Her absence has made an unmatchable vacuum in our team. It is a great loss to us all and she is greatly missed.

a b s t r a c t This paper provides a review of the most recent works in electronic noses used in the food industry. Focus is placed on the applications within food quality monitoring that is, meat, milk, fish, tea, coffee and wines. This paper demonstrates that there is a strong commonality between the different application area in terms of the sensors used and the data processing algorithms applied. Further, this paper provides a critical outlook on the developments needed in this field for transitioning from research platforms to industrial instruments applied in real contexts. Ó 2014 Elsevier Ltd. All rights reserved.

Keywords: Food quality e-Nose Metal oxide Pattern recognition

Contents 1. 2. 3.

4.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Electronic noses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Application areas in food . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Milk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Wine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Tea and coffee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4. Fish and meat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Common methodologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. Electronic nose comparison with sensor panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2. Hybrid electronic nose technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3. New pattern recognition methods applied to food analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Toward new sensor materials for electronic nose instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

104 104 105 105 105 106 106 106 106 107 108 108

Abbreviations: ANN, artificial neural network; APLSR, analysis of variance partial least squares regression; BP-MLP, back propagation multilayer perceptron; BPNN, back-propagation neural network; CA, cluster analysis; DA, discriminant analysis; DBN, deep belief network; DFA, discriminant factorial analysis; FNN, fuzzy neural network; LDA, linear discriminant analysis; LVQ, learning vector quantization; MGLH, multivariate general linear hypothesis; MLP, multi-layer perceptron; MOS, metal oxide semiconductor; MOSFET, metal oxide semiconductor field effect transistor; m-TDNN, multiple-time-delay neural networks; PCA, principal component analysis; PLS, partial least squares regression; PNN, probabilistic neural network; QDA, quadratic discriminant analysis; RBF, radial basis function; SQC, statistical quality control; SVM, support vector machines; VOCs, volatile organic compounds; KAMINA, Karlsruhe Micro. Nose. ⇑ Corresponding author. Tel.: +91 4362 264 101x2255, mobile: +91 9944468389; fax: +91 4362 264120. E-mail address: [email protected] (J.B.B. Rayappan). http://dx.doi.org/10.1016/j.jfoodeng.2014.07.019 0260-8774/Ó 2014 Elsevier Ltd. All rights reserved.

104

5.

A. Loutfi et al. / Journal of Food Engineering 144 (2015) 103–111

Discussion and future perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

1. Introduction Foodborne illnesses cause about 76 million cases of illnesses, 325,000 hospitalizations, and 5000 deaths in the United States each year (Hedberg, 1999). Common symptoms of foodborne illness include diarrhea, nausea, abdominal cramps, headache, dizziness and fever. In the developed/developing countries, surveillance of foodborne disease is a fundamental component of food safety systems (WHO). According to the estimates of US Centers for Disease Control and Prevention (CDC) in the year 2011, roughly 1 out of 6 Americans or 48 million people get sick, 128,000 are hospitalized and 3000 die of foodborne illnesses (CDC, 2011). Hence researchers started exploring better way for quality discrimination of perishable foods. Highly perishable, muscle foods like fish, meat and poultry have become an integral part of human diet over many decades. However, in the past two decades awareness about the food safety from the point of specific pathogenic bacteria has exemplified the requirement for a rapid and accurate detection system for microbial spoilage in fish and meat (Frost, 2001; Haugen et al., 2006). In general fish and meat quality will be assessed either by examining the structure (texture, tenderness, flavor, juiciness, and color) or by detecting the microorganism and its count or by detecting the gas/VOCs generated by these microorganisms. The practical application of human nose as a smell assessment instrument is severely limited by the fact that our sense of smell is subjective, gets tired easily, and is therefore difficult to use. Consequently, there is considerable need for an instrument that could mimic the human sense of smell and its use in routine industrial applications. To promote this technology to industrial application, metal oxide gas/odour sensors became exemplary candidates in areas like food industry, environment control, automobile industry, indoor air quality check and monitoring, industrial production, medicine and in safety aspects, to name a few – Scientific groups worldwide are investigating them giving due importance to the various aspects of gas/odour sensing properties. Electronic nose instruments are attractive for a number of significant features: the relatively fast assessment of headspace, a quantitative representation or signature of a gas and cheap sensors which can be easily integrated in current production processes. Despite these features, there are still relatively few applications of electronic noses adopted in industry. This could be attributed to difficulties in robustness, selectivity and reproducibility of the sensors and to the need for pattern recognition algorithms which can cope with the complex signal analysis. Nonetheless, the use of electronic noses is rapidly expanding and there have been notable achievements relevant for the food industry, particularly in the past few years. Furthermore, this progress coincides with an increased understanding of the biological mechanisms behind the human olfactory system. Specifically, we now have a greater understanding not only of the genetics behind the olfactory receptors but also of the relationships between an odorant’s molecular property and the quality of an odor. This paper focuses on the latest developments within key areas related to foodstuff where a quantitative approach to quality estimation is important as it regulates the economy of food (e.g. pricing) and quality control (e.g. detection of bacteria and spoilage). Specifically, this paper reviews the progress in the past decade for the following areas: milk, wine, coffee, tea, fish and meat.

A broad list of e-nose reviews can be found in the literature that are structured and focused on mass spectrometry based electronic noses (Peris and Escuder-Gilabert, 2009), biomedical and health care applications (Wilson and Baietto, 2011), agriculture and forestry applications (Wilson, 2013), microbial quality control of food products (Falasconi et al., 2012), pharmaceutical applications (Alam et al., 2012), for developing chemical sensor arrays (James et al., 2005). Our review is distinctive in a way that we focus on the methodologies which are common across the various applications. Therefore the motivation of this review is to make accessible to the various research groups not only the progress within their specific area but also in adjacent areas and most importantly common methodologies that can be used to solve related challenges in different fields. A secondary motivation is to promote the use of electronic noses for food quality monitoring fin an industrial setting by summarizing the extensive work in the past few years that indicates promising results with respect to the applicability of e-noses to a vast number of areas.

2. Electronic noses During the 1980s research on machine olfaction lead to a generally accepted definition of an electronic nose as an instrument that comprises an array of heterogeneous electrochemical gas sensors with partial specificity and a pattern recognition system (Gardner and Bartlett, 1999; Persaud and Dodd, 1982). However, in more recent years, the term electronic nose has been used in a broader sense to refer to gas sensors that measure the ambient gas atmosphere based on the general principle that changes in the gaseous atmosphere alter the sensor properties in a characteristic way. A variety of different sensor types have been developed, to which three types of materials are commonly used: metal oxides, conducting polymers composites and intrinsically conducting polymers. Apart from conductive sensors, gas detection has also been done using optical sensors, surface acoustic wave sensors, gas sensitive field effect transistors and quartz microbalance (QMB) sensors. Micro-electro-mechanical systems (MEMS) plus nanotechnologies are the most promising emerging technologies in the area. The term electronic nose has also been used to characterize systems where ultra-fast gas chromatography or mass spectrometry is employed in the detection process. Once the data from the individual sensors from the array is collected, the electronic nose systems require a suitable post processing procedure to analyze and classify the data. Pre-processing of multivariate signals in sensor arrays represents an essential part of the measuring system. Data processing techniques used in post processing of pattern recognition routines include principal component analysis (PCA), linear discriminate analysis (LDA), partial least squares (PLS), functional discriminate analysis (FDA), cluster analysis (CA), fuzzy logic or artificial neural network (ANN) such as probabilistic neural network (PNN). Among these techniques, PCA, PLS, LDA, FDA and CA are based on a linear approach while fuzzy logic, ANN and PNN are regarded as nonlinear methods (Scott et al., 2006). Particular to the food industry is the sample handling system used for exposing the volatile compounds present in the headspace (HS) to the sensor array in the e-nose. For some applications, specific techniques are used such as purge & trap (P&T), dynamic headspace (DHS), solid-phase microextraction (SPME) used by

105

A. Loutfi et al. / Journal of Food Engineering 144 (2015) 103–111 Table 1 A list of some of the common commercially available electronic nose models. Model

No. of sensors

Technology

Manufacturer

Country

i-Pen, PEN2, PEN3 QCS Artinose FF2 GFD1 FOX 2000, 3000 & 4000 Promethus Air quality module EOS 835, Ambiente Bloodhound ST 214 Aromascan A32S Cyranose 320

6 2 38 6 6 6, 12 & 18 18 2 6 14 32 32

MOS Sensor MOS Sensor MOS Sensor MOS Sensor MOS Sensor MOS Sensor MOS Sensor MOS Sensor MOS Sensor Conducting polymers Conducting polymers Conducting polymers

Airsense Analytics Gerstel GmbH & Co. KG Sysca AG RST Rostock System-Technik GmbH

Germany Germany Germany Germany

Alpha MOS

France

Applied Sensor Sacmi Scensive Technologies Osmetech Plc Intelligent Optical Systems, Inc.

Sweden Italy UK USA USA

Berna et al., 2009 to preprocess wine samples. Comparisons between methods is a further subject of study such as in, where two types of purging, trapping methods and four types solidphase-micro extraction methods were compared (Lozano et al., 2008a,b). Aside from the sample handling, the sample itself may be preprocessed such as in the analysis of the quality of wine and beer, the preprocessing procedure of dehydration and dealcoholization helped the e-nose to classify aromas to a better extent (Ragazzo-Sanchez et al., 2009). In the analysis of tea, a novel sampling system has been devised based on illumination-controlled heating together with physical raking (Bhattacharya et al., 2008c, 2008d). Today, several electronic noses were commercially available in the market. These e-noses are often promoted as generic devices and suitable for a range of applications. Table 1 provides a list of some of the common commercially available electronic nose models. 3. Application areas in food The applications of electronic noses have been numerous and range from environmental monitoring (Kashwan and Bhuyan, 2005) to medical applications (Längkvist and Loutfi, 2011). Since 1993, the amount of publications in the area of electronic olfaction is more than 12,000 articles. The main application areas related to the food industry have been: fish, meat, milk, wine, coffee and tea, and constitute approximately 5000 publications since 1993. This demonstrates that food applications are central to electronic olfaction and nearly half of the publications are in this area. Within each application in the food industry, research contributions have focused on the detection of a variety of aspects such as freshness, adulteration, off-flavors and bacteria detection. We provide in this section a short description of the main application areas for a number of specific food stuff. 3.1. Milk Headspace of milk from a healthy cow typically contains complex mixtures of VOCs (acetone, hexanal, 2-pentanone, 2-butanone, toluene, limonene, heptanal, etc.) at various concentrations (Ampuero and Bosset, 2003; Pardo and Sberveglieri, 2005). The concentration of these VOCs varies due to various factors such as bacterial metabolism, ageing, photo-oxidation and presence of prooxidant metals such as copper, iron and nickel. Milk from a healthy cow contains only few bacteria, which may multiply and the rate of multiplication will increase, as time passes from the time of milking to the time of processing; it also depends on the standard of milking, handling practices and storage. These bacterial growth leads to the spoilage of milk with the production of off-flavor. As the microbial spoilage of starting milk severely affects the industrial

quality, due to undesirable aroma, physical defects and metabolic toxicity. Hence identification of food spoilage, before the product formation became mandatory to avoid consumption of such spoiled food and this can be satisfied with the assistance of electronic noses. Likewise, spoilage of milk due to various other factors should also be identified at the earliest possible stage to avoid complications and complaints in the final product. Microbial milk spoilage can be identified from the presence of acetaldehyde, 3-methyl-1butanol, acetic acid and ethanol in headspace of the milk (Ampuero and Bosset, 2003; Magan et al., 2001). In this work authors determined the specified markers employing 14 conducting polymer sensors and discriminant function analysis to separate the unspoiled and bacterial containing milk. Whereas Marsili et al. studied the off-flavors in milk using solid-phase microextraction, mass spectrometry and multivariate analysis (SPME–MS–MVA) and found the increase in hexanal and dimethyl disulfide concentration in milk when it was exposed to the fluorescent light (typical exposure in supermarkets). Also suggested that pentanal, hexanal, heptanal and isopentanal can indicate the copper induced oxidation in milk (Marsili, 1999). Ageing of milk can be significantly identified through the presence of dimethyl sulfide, ethyl acetate, 2-heptanone, pentanal, etc. Also he observed (Marsili, 2000) that the headspace of milk from the cow bearing genetic defect contains trimethylamine. Authors (Eriksson et al., 2005) have observed higher levels of sulfides, ketones, amines and acids in milk collected from the cow affected by mastic disease. For this experiment, a commercial hybrid sensor array system (Applied Sensor 3320, Linköping, Sweden) consisting of 10 MOSFET sensors, 12 MOS sensors, one sensor for humidity were used to collect the data from milk and PCA was employed to analyze the milk data. Among these, the dominating components were ethanol, trimethylamine, 2, 3-butanedione, 2-pentanone, 2-methylbutanal, methane-thiol, dimethyl sulfide, acetic acid and in general, the infected milk had higher CO2 content than the healthy reference milk. Therefore by detecting the higher concentration of CO2, mastic milk can be identified. 3.2. Wine The main application of electronic noses with respect of wine is quality assurance. Wine is a commercial product, which can vary greatly in aroma and flavor according to the large possible variations in its production. The formation and transformation of organic acids at must fermentation and wine production are of great importance in wine-making. Biochemical processes caused by yeast enzymes are significant to achieve better quality of wine. Organic acids also protect wine against bacterial diseases. However, high content of some acids influences wine flavor badly. The qualitative and quantitative analyses of the most dominant

106

A. Loutfi et al. / Journal of Food Engineering 144 (2015) 103–111

aromas and flavors evolve from the headspace of wine namely Dimethylamine (DMA), Trimethylamine (TMA), Ethyl octanoate, 1-Hexanol, Ethanethiol, Ethylacetate, and 2,4,6-Trichloroanisole play a significant role in determining the quality of wine (Arroyo et al., 2009; Roy and Basu, 2004). Authors (Arroyo et al., 2009) have assessed the quality of wine using an electronic nose comprising of 16 tin oxide thin film based sensors. To enhance the sensitivity of these sensing elements, chromium and indium elements doped thin films with varying thickness ranging from 200 to 800 nm were utilized. PCA, PNN and Leave One Out (LOO) algorithms were applied for sensor data classification. Electronic nose techniques have been used to detect defects or spoilage e.g. caused by high concentrations of 4-ethylphenol and 4-ethylguaiacol (Berna et al., 2008; Cynkar et al., 2007). Other applications include the discrimination between aging techniques (Lozano et al., 2008a, 2008b; Prieto et al., 2012), discrimination between products (based on geographic origin or grape) (Berna et al., 2009) and the prediction of sensorial descriptions (Buratti et al., 2007). 3.3. Tea and coffee The infusions and extract of the leaves Camellia sinensis (L.) is the source for the beverage tea. The extract of leaves has the chemical components such as flavanol, caffeine, phenolic substances, fats, amino acids, theaflavin, thearubigin and volatile components (Chaturvedula and Prakash, 2011). These chemical components are the source to determine the flavors and aromas of the beverage, which are depending on the leaves collected from plants and manufacturing processes like withering, pre-conditioning, cut-tear-curl operation, fermentation and drying (Bhattacharya et al., 2008c). All these processes referred to focus on the enzymatic oxidation of tea leaves. Thus degree of fermentation process dominantly determined by the enzymatic oxidation of tea leaves which produces specific aromas and flavors of tea. The pioneering work reported by Dutta et al. employed Warwick metal oxide electronic nose (WOLF – Enose, UK) with radial basis function (RBF) network pattern recognition system and successfully discriminated the flavors of tea manufactured in different conditions (Dutta et al., 2003). Researchers (Bhattacharya et al., 2008c) have also calibrated an e-nose with an array of eight metal oxide sensors (TGS, Fiagaro Inc., Japan) toward volatile emission signatures which are evolving during the fermentation process. This work has resulted in the e-nose which can be used to predict the optimum fermentation time in real time applications. Similarly, Coffee also provides various flavors depending on species like Arabica, Robusta, Bengal, Congo and etc., and variety of plant, other agricultural factors, harvesting, picking, sorting, processing, hulling, polishing, roasting, and packaging. In addition, coffee is changing in all its forms, from green to roast to brew. Traditionally, human olfaction technique was used to discriminate the quality of coffee beans aroma and flavors. In general, presence of species like Arabica has been considered as one of the markers indicating the high quality coffee bean due to its pleasing flavors, aromatic properties, low in caffeine and low in acidity. Since coffee has gained significant commercial and economic importance, people adulterate it by mixing certain low quality or even other species of coffee beans to yield more profit. But, it is very difficult to discriminate the quality of adulterated coffee using human sensory panels and gas measurement system (Kottawa-Arachchi et al., 2012). In this scenario, Electronic noses have been designed to quantify the concentration levels of the identified aromas in coffee and its quality levels (Bhattacharya et al., 2008a; Dutta et al., 2003). An e-nose comprising of 12 preheated metal oxide gas sensors (Figaro Inc., Japan) were used to recognize the flavors and aromas by detecting the variations of volatile organic components in the mixtures of low quality Robusta beans and high quality Arabica beans.

Differential signal processing with SVM classifier (Brudzewski et al., 2012) schemes were used to classify the data for decision making. The discrimination of aromas has also been fine-tuned by altering the sensor materials (Wang et al., 2012) and data classification systems (Shi et al., 2012) which are employed in the electronic nose. 3.4. Fish and meat In the recent past, inspection of food quality and safety especially to examine the sea foods because they are highly nutritious to provide better health care compounds. Fatty acids, amino acids, proteins, fat content, flavor, color and texture of the fish depends upon the farming systems, control of growth rate, water chemistry, water temperature, feeding behavior, nutrition intake and freshness (before and after the harvest). Among them freshness of the fish is the major issue in the fish market industry. Spectroscopy and imaging techniques were used to test the freshness level of fish flesh. But these techniques are having their own advantages and limitations (Herrero, 2008). Sensory analysis is considered to be the best tool to determine the fish freshness in the industry and consumer market (Warm et al., 2001). Seafood export industry has taken several steps to prevent the decomposition of fish flesh quality by reducing the microbial spoilage factor. Fatty acids profiles, aldehydes (hexanal, 2-methyl-1-butanal, nonanal, 2,4-heptadienal), ketones (2-octanone, 2-decanone, 2-propanone), trimethyl amine (TMA), volatile organic compounds (1-butanol, 1-penten-3-ol) are the major aromatic compounds identified as the biomarkers of spoilage levels. Sensing the concentration levels of these biomarkers (Catarina Bastos and Magan, 2006; Macagnano et al., 2005; Olafsdottir et al., 2004) using Electronic nose is found to be one of the best and promising solutions to analyze the quality of food (Gholamhosseini et al., 2007). The freshness levels of selected varieties of fish species namely Red Snapper, Gurnard, Tarakihi, and Trevally using an array of thirty-two polymer carbon black composite sensors in the portable e-nose (Cyranose 320, USA) with ANN classifiers have been reported (Gholam Hosseini et al., 2008). A fairly accurate assessment confirmed that this e-nose can be employed to estimate the fish freshness, rate of degradation and quantitative time period for quality sustenance. Similarly, the application of electronic nose technology to meats is one of the main application areas in the food industry. A significant number of works have been presented over the past two decades and (Ghasemi-Varnamkhasti et al., 2009) gives an additional review of the field. The main applications of electronic noses with respect to meat are in assessing quality, spoilage identification, detection of off-flavors, taints and classification of bacterial strains. In addition, processing techniques and discrimination of different types of meats have been examined (Blixt and Borch, 1999; Hansen et al., 2002; Olsen et al., 2005). The recognition of meat freshness as well as decomposition of meat food under various situations like temperature and volume were successfully tested using an e-nose (KAMINA e-nose, Germany) with an array of 38 sensors in which LDA algorithm was used to classify the observed data (Musatov et al., 2010). 4. Common methodologies 4.1. Electronic nose comparison with sensor panels Comparative studies between sensory panels and the response from an electronic nose dates back to the early 1990s. In the literature there has been a twofold motivation for developing correlations with human panels. On one hand, the pattern recognition process, whose output is highly dependent on the given labels, is synchronized with current descriptors used in the food industry.

A. Loutfi et al. / Journal of Food Engineering 144 (2015) 103–111

The emerging challenge is to correlate the rich descriptors used by humans to the response of the electronic nose e.g., by using PLS. Other approaches have attempted to provide human like descriptors to an electronic nose response using natural language symbols which have been manipulated to provide new descriptions to unseen samples via relation to known concepts e.g. ’’smells a lot like lavender’’ (Loutfi et al., 2001). A secondary motivation for correlation with human panels is to encourage the adaptation of new standards and eventually exploit the e-nose’s ability to provide finer grading of attributes e.g. freshness. In determination of fish freshness, the most promising works has been done in the project FAIR CT98-4076 (MUSTEC) by Olafsdottir et al. (2004) by considering various combinations of physico-chemical techniques and its integration to attain output that relates fish freshness. The project aims at producing a Artificial Quality Index (AQI) combining the outputs of the instrumental techniques and calibrating them with sensory scores of Quality Index Method (QIM) used by test panel for attributes like appearance, smell and texture. The works which are most prevalent in sensor comparison with panels are related to wine quality testing. This is rather expected due to the rich and varied qualifiers often used in the wine industry. Arroyo et al. investigated a homemade electronic nose with semiconducting thin film based sensing elements and twenty-five human tasters were trained to classify seventeen different aromas which are used to discriminate the quality of wine. It was found that human tasters performed better in identifying certain aromas (Arroyo et al., 2009). However, an electronic nose subsequently developed by Santos et al. was found to be better in detecting the specific thresholds of typical red wine compounds such as ethyl acetate and eugenol and white wine compounds such as hexanol and ethyl octanoate (up to 8 times lower than a human panel) (Santos et al., 2010). In another work, the success rate of sixteen tin oxide sensors based e-nose system was compared to a human sensory panel and to a HP-6890 gas chromatograph with HP Mass Selective 5973 Spectrometer. Twenty-eight wine samples were analyzed. From this comparison it was found that the results provided by e-nose employing PLS regression algorithm corresponded better to the sensory panel results than to the predictions of classification form the GC–MS. Tea, like wine is also characterized by rich descriptors and electronic noses can be trained to achieve a high correlation with tea tasters. For example, researchers (Tudu et al., 2009a) used various methods to correlate with the tester’s scores for assessing the quality of black tea samples. Typically, testers assign scores to tea on a scale of 1 to 10 based on flavor, aroma and taste of the sample. In this work, a set of Figaro gas sensors were first selected for black tea classification by measuring the sensitivity to various compounds that are typical in black tea aroma (Geraniol, Linalool, 2-phenyl-ethanol). In total, an array of five sensors were then used in an instrument whose responses were correlated with two groups of tea testers – thereby deriving a computational model to predict the score for unknown samples. Testing with unknown samples resulted in a classification rate of more than 90%. In certain applications such as meat quality testing, the detection of specific bacteria or chemical compounds is the primary aim, and thus correlation with human panels can be limited. Nevertheless, certain off-odors such as boar taint, a unpleasant cooking odor from non-castrated male pigs, are often characterized by human panels. Five different meat parts (loin, neck, shoulder, outer and inner thigh) were tested with an electronic nose and validated by a trained human sensory panel (Kirsching, 2012). A high determination coefficient (R2 = 0.915) was obtained between reference values of boar taint (obtained by sensory panel) and predicted values calculated from e-nose data. Two compounds deposited in the fat tissue of pig are held responsible for boar taint: androstenone and skatole. Typically in any sample population about 23% are

107

insensitive to androsterone (Bonneau and Squires, 2004), making the selection of human panels particularly challenging. In sum, sensor panels are an essential aspect in the industrial food processes, but the difficulty and cost to maintain such panels is high. Consequently, for day-to-day production having a tool such as an electronic nose which would in effect be synchronized with an eventual panel would be valuable for industrial use. The use of an electronic nose to complement human panels particularly if ‘‘non-odorant’’ gases are to be detected is promising. Nevertheless the general practice is to use the sensor panel to define the targets for quality detection and train the nose on these descriptors. In order for a symbiosis between electronic noses and human panels to emerge in the industry, a high reproducibility remains to be demonstrated in the electronic counterpart. 4.2. Hybrid electronic nose technologies While e-noses present a number of advantages over traditional analysis, the sensors also present a number of shortcomings which have yet to be solved. These include issues such as sensor poisoning, sensor drift and sensitivity. Selectivity and sensor drift have been the focus of investigation. Specific to the food industry is profile masking (e.g. ethanol) which affects the response. Profile masking for example can be found in wines and often requires a pre-processing step as described in Section 3.2. Recent trends to overcome sensor shortcomings include combining semiconductor chemical sensors with other types of gas sensors. While this complicates the sampling system (requiring more bulk and electronics), this hybrid technology is able to compensate for the shortcomings in current chemical sensor technology. In particular, a new generation of electronic noses referred to as mass spectrometry-based electronic noses (MS-E-nose) are increasingly used in the literature where a MS-E-nose consists of a mass spectrometer instrument without prior chromatographic separation. Comparisons of MS-based e-nose and other commercial e-nose systems have been presented in several works. The work by Berna et al. has compared the performance of MOS-e-nose (FOX 3000 E-Nose from Alpha MOS, France) and MS-E-nose in estimating the defects in red wine (Berna et al., 2008). These two techniques were used to identify the presence of the two major components of taint in red wine namely 4-ethylphenol and 4-ethylguaiacol (Cynkar et al., 2007) used MS-e-nose for identifying spoilage factor due to the presence of 4-ethylphenol and a success rate of 67% was achieved with Stepwise Linear Discriminant Analysis (SLDA) classifier. A combination of twelve metal oxide sensors (MOS-e-nose) and a mass spectrometry-based e-nose (MS-E-nose) was used (Ragazzo-Sanchez et al., 2009). The preprocessing procedure of dehydration and dealcoholization helped the e-nose to do better classification of aromas (Cozzolino et al., 2008). This approach has coupled the mass spectrometry and e-nose for better estimation of aroma properties in Australian Riesling wines. After collecting the data using MS-e-nose, it was classified and analyzed using PCA and partial least squares (PLS1) regression adopting leave one out method. Even though this technique did not help for quantitative discrimination of various aroma elements of selected compounds, it helped for better screening of wines prior to sensory analysis. Similarly, Berna (Berna et al., 2008) have compared the performance of MOS-Enose (FOX 3000 E-Nose from Alpha MOS, France) and MS-Enose in estimating the defects in red wine. These two techniques were used to identify the presence of the two major components of taint in red wine namely 4-ethylphenol and 4-ethylguaiacol. MS-E-nose was used prior to ethanol removal from the sample and MOS-Enose later. As expected the selectivity of MOS-Enose was not helpful for rapid discrimination of quality of red wine. In an electronic nose, ion mobility spectrometry was

108

A. Loutfi et al. / Journal of Food Engineering 144 (2015) 103–111

used for boar taint measurements where samples varying in androstenone and skatole levels were tested (Vestergaard et al., 2006). Their findings indicated that sensory perceptible boar taint was found to be more related to androstenone than to skatole. Specifically the e-nose was indentified as a useful tool in for ordering samples with respect to low and high levels of androsterone and skatole. The e-nose based on ion mobility spectrometry was particularly useful for fast analysis and identified as suitable approach for application in production e.g. at the slaughter line. In sum, electronic noses based on chemical gas sensors still present the advantage of ease of use and low costs. Nevertheless, the more complex MS-nose could be used as a complement to provide better verification of results based on an initial screening from a more portable and lightweight device. 4.3. New pattern recognition methods applied to food analysis The pattern recognition component of the electronic nose is non-trivial due to: the non-linearity of the sensor response; the need to compensate for sensor drift; and the selectivity which requires that an electronic nose is trained to recognize specific patterns representing odors. The machine learning community has used electronic nose data as a way to validate new and generic algorithms that cope with the aforementioned challenges. In particular with respect of machine learning applied to food related applications, a number of aspects have been considered that include, feature extraction, sensor selection and incremental learning. In (Bhattacharyya et al., 2007) a selection of specific sensors for a customized e-nose for tea is presented. In this, (Bhattacharya et al., 2008b) an incremental probabilistic neural network is used for black tea grade discrimination. The incremental classifiers show the ability to accommodate new classes and new knowledge within an already trained model, thus promoting the possibility to dynamically augment the training data. This is similar to the works done by Tudu et al., in which a radial basis function (RBF) is combined with incremental learning techniques that preserve the knowledge already learnt by the network on previous training sets while allowing the acquisition of new knowledge as it becomes available (Tudu et al., 2009a). Other combinations of incremental learning have been shown by the same authors using fuzzy techniques (Tudu et al., 2009b). Other learning methods that have been applied also include non-linear support vector machines in the area of spoilage detection in meat. For example, using non-linear support vector machines, a high classification rate of 98.81% and 96.43% were obtained in identifying spoiled/unspoiled samples of beef and sheep meat respectively (El Barbri et al., 2008). A trend in machine learning that can also have relevant application in electronic olfaction is the use of deep learning. A first attempt has been applied to the classification of meat spoilage markers where the combination of novel sensing materials and auto encoders were shown to provide an electronic nose system with high selectivity to TMA as well as fast response when using the transient response for detection (Längkvist et al., 2013). For real industrial deployments, it will be necessary to have systems that can be easily maintained, and do not require an expert in machine learning to manually tune the learning parameters for optimal performance. This is the main challenge for the machine learning community in developing algorithms for food related applications. 4.4. Toward new sensor materials for electronic nose instruments An important new trend is the developing of nanostructured sensors for electronic nose instruments. This new kind of sensors have promising features as they offer controllable grain size (Yamazoe et al., 2003). This can open an entirely new era in development of innovative metal oxide gas sensors. Recently,

nanostructured metal oxide thin films based sensors are playing a key role in the field of perishable food quality assessment particularly meat and fish (Galdikas et al., 2000; Jung and Lee, 2011; Mani and Rayappan, 2013, 2014a, 2014b; Muniyandi et al., 2014; Panigrahi et al., 2006; Perera et al., 2010). Researcher Roy and Basu modified the surface of nanostructured zinc oxide thin film by dispersing palladium and doping aluminum (Roy and Basu, 2004). The authors reported that the surface modified ZnO film with palladium exhibited a better sensitivity to DMA, whereas Al doped ZnO showed better sensitivity toward TMA over DMA. Since TMA and DMA are the very good indicators of fish freshness level, the developed ZnO films can be used as one of the sensors in the sensor array. SnO2 and CuO mixed oxide nanowires exhibited a better sensitivity than that of pure tin oxide nanowires toward acetone, ethanol and ethyl acetate which are all the good markers of meat, fish freshness level (Li et al., 2011). The works by Kumar et al. have reported that the nanostructured cerium oxide thin film is highly selective for one of the markers in perishable food items TMA compared with the ethanol (Kumar et al., 2013) . Clearly for electronic nose applications, the selection of the materials used in the array is important to achieve good discrimination for the intended application. However, solving the problems related to the selectivity in gas sensing applications still remains a major challenge. In this context, the different sensing behavior of F and Mn doped ZnO thin films toward ethanol and TMA is encouraging and will help to solve the problem of selectivity (Sivalingam et al., 2011, 2012). Though F/Mn doped films are n-type semiconductors and ethanol/TMA are reducing gases, the presence of grain boundary scattering during sensing in the first case and the absence of the same in the second case helped to develop sensing elements with inherent selective nature. Adding such sensing elements to the array of sensors, electronic nose will provide new possibilities for detection of specific markers. 5. Discussion and future perspectives In this paper, we have outlined the major contributions of electronic nose technologies relevant to the most published fields within the food industry (Table 2). As each paper addresses just one application area, the field gives the impression of being fragmented. This is in general due to the need to tune either the software and/or hardware to the specific application. However, much overlap exists in hardware and software algorithms used. It is also clear that the utility of using e-noses in an industrial context is high, and that most works have in fact shown cooperation with industrial partners that have made available the samples and conditions of use. The open question is therefore why there is an absence of electronic noses in industrial processes. It is likely that the underlying reason for the reluctance of the uptake of e-noses in an industrial context is multifold. On the materials side, major focus must be given to the design and development of drift free sensors that can be used reliably over long temporal horizons. It is likely that only experiments showing long term use will be a convincing factor for industry when considering the uptake of such a device. Consequently, the internal drift influencing factors like crystalline structure variation, grain size variations, grain boundary effects, uniformity in the dopant concentrations, perfect contact materials, thickness of the sensing elements should be addressed. In addition external factors like humidity, ambient temperature and atmosphere, presence of other gases/aromas during standardization procedure, accuracy in analog to digital data conversion process, etc. must be given special attention for enhancing the performance of the sensor system as a whole. In this context, nanostructured surfaces may be used for better selectivity and sensitivity and Metal organic framework may be incorporated for filtering the atmospheric effects and cross interferences and hence

109

A. Loutfi et al. / Journal of Food Engineering 144 (2015) 103–111 Table 2 Summary of the applications of e-nose in different food matrices. Data Processing Algorithm

Ref.

5

PCA

Capone et al. (2001)

Bloodhound Sensors Ltd. UK (BH-114)

14

DFA

Magan et al. (2001)

Alpha MOS, France (Fox 4000)

18

PCA

Wang et al. (2010)

MOS MOS

Airsense Analytics, Germany (PEN 2) Figaro Engineering Inc., Japan (TGS 2620, 2610 and 2600)

10 5

Yu et al. (2007) Botre et al. (2009)

MOSFET + MOS

Applied Sensor, Sweden (3320)

22

PCA, LDA Neuro Solutions 5.0 from Neuro Dimension Inc., USA PCA

Conducting polymer QCM MOS

OSMETECH PLC, England

28

LDA

Biolatto et al. (2007)

SES Piezo Ltd., England Alpha MOS, France (Fox 4000)

6 18

PCA PCA, DFA

MOS MOS (SnO2)

Airsense Analytics, Germany (PEN 2) Lab Made

10 16

Genetic Algorithms PCA

Ali et al. (2003) Ragazzo-Sanchez et al. (2009) Buratti et al. (2007) Lozano et al. (2008a,b)

MOS

Airsense Analytics, Germany (PEN 2)

10

PCA, LDA and BPNN

Yu et al. (2009)

MOS

Figaro Engineering Inc., Japan (TGS-832, 823, 831, 816, 2600, 2610, 2611 and 2620) Figaro Engineering Inc., Japan (TGS 832, 823, 2600, 2610, and 2611), Figaro Engineering Inc., Japan (TGS 832, 823, 2600, 2610 and 2611) Figaro Engineering Inc., Japan (TGS 816, 823, 831, 832, 2600, 2610, 2611 and 2620) Airsense Analytics, Germany (PEN 2)

8

m-TDNN

5

10

Incremental fuzzy classification Incremental RBF neural network PCA, PNN, BP-MLP, RBF PCA, LDA

Bhattacharya et al. (2008c) Tudu et al. (2009b)

10

PCA, CA

Yang et al. (2009)

4

MLP, LVQ, PNN, RBF

Dutta et al. (2003)

4

PCA, MLP

Pardo et al. (2000)

ANN, FNN

Singh et al. (1996)

8

PCA, MLP

Rodríguez et al. (2009)

6

PCA

Amari et al. (2009)

MGLH, DA

Kirsching (2012)

Application

Sensing Element

E-Nose Model

Rancidity of two different kinds of milk (ultrahigh temperature processed and pasteurized) To detect unspoiled and spoiled milk To discriminate different milk flavorings Identification of adulterated milk Identification of odours (milk, rancid milk and yoghurt)

MOS (SnO2)

Lab Made

Conducting polymer MOS

Differentiate the milk collected from healthy and mastitic disease affected cows Seasonal changes in milk powder Bacterial contamination in milk Detect different off-flavors in wines of different origin Prediction of Italian red wine Classification of four types of red wines Detection of aroma from green tea at different storage times Identification of optimum fermentation time for black tea Quality evaluation of black tea

MOS

Quality evaluation of black tea

MOS

Classification of black tea

MOS

Discriminating different grades of green teas Distinguish the Japanese green teas with different content of coumarin Discrimination of tea with different qualities Classification of different brands of Espresso coffee Classification of different coffee

MOS

Number of sensors

MOS

Fragrance & Flavor Analyzer, Shimadzu, Japan (FF-2A)

MOS

Figaro Engineering Inc., Japan (TGS 880, 826, 825 and 822) Lab Made

MOS (SnO2) MOS

5 8

14

Eriksson et al. (2005)

Tudu et al. (2009a) Bhattacharya et al. (2008a) Yu et al. (2008)

Analysis of Colombian coffee samples

MOS

Shelf-life determination of red meat

MOS

Discrimination of boar tainted samples of different meat parts Quality detection of porcine meat loaf Detection of boar taint

MOS

Figaro Engineering Inc., Japan (TGS 800, 815, 816, 821, 823, 824, 825, 830, 831, 842, 880, 881, 882, 883) Figaro Engineering Inc., Japan and FIS Inc., Japan (SP-12A, SP 31, TGS 813, TGS 842, SP-AQ3, TGS 823, ST-31 and TGS 800) Figaro Engineering Inc., Japan (TGS 815, 821, 822, 824, 825 and 842) Alpha MOS France (Fox 4000)

MOS

Danish Odour Sensor System (DOSS)

6

APLSR

Hansen et al. (2002)

MOS

5

–

Spoilage classification of red meat

MOS

6

PCA, PLS, SVM

Bourrounet et al. (1995) El Barbri et al. (2008)

Meat spoilage markers detection Discrimination of microbial population Quality identification of rapid fermented fish

MOS (ZnO) MOS

Figaro Engineering Inc., Japan (TGS 825, 882, 824, 822, 800) Figaro Engineering Inc., Japan (TGS 823, 825, 826, 831, 832, 882) Lab Made Figaro Engineering Inc., Japan (TGS 812, 822, 880, 2602, 2611 and 2611) Alpha MOS, France (Fox 4000)

3 6

PCA, SVM, DBN LDA, QDA

Längkvist et al. (2013) Panigrahi et al. (2006)

PCA, SQC

Yang et al. (2009)

MOS

improving selectivity and sensitivity. On the software side, researchers apply many of the available linear and nonlinear algorithms (either separately or combined). While complex data processing can achieve good results on a specific dataset it fails to compensate for the core problems of stability and reliability of the sensors. Clearly, selectivity is also a key issue and it is important that the relevant markers for the given application are in fact detected and in certain cases quantifiable. So far, the examination

18

18

of such markers has been done by building relevant models which correspond to GC–MS studies, till today, there is no significant and accurate mathematical modeling proposed. It is also possible to address the problem of selectivity via changing properties in the sensing array, for example, synchronization or modulation of the operating temperatures are viable approaches to improve selectivity. Ultimately, however, such methods still must demonstrate a robustness over time to be considered for industrial use. On the

110

A. Loutfi et al. / Journal of Food Engineering 144 (2015) 103–111

market a selection of generic instruments are available as shown in Table 1, it is quite possible that an indication of a maturing of the field will be demonstrated when specific instruments are available for the specific applications and when such instruments are being used over a long time period in real settings. In addition, the top five needs in this area of research are (1) drift free sensors that can be used reliably (2) investigation of new material for achieving better selectivity (3) better modeling and correlation between presence of chemical markers and the sensor response in the enose array (4) application of specific instruments with carefully selected sensor arrays and sampling system (5) Better understanding of the industrial need related to quality control and monitoring of food processing. Acknowledgements The authors wish to express their sincere thanks to the Department of Science & Technology, New Delhi, India and VINNOVA, Sweden for their financial support (Project ID: INT/SWD/VINN/ P-04/2011). We also thank SASTRA University, Thanjavur, India and Orebro University, Orebro, Sweden for extending the infrastructural support to carry out this work. References Alam, H., Saeed, S.H., Engg, C., 2012. Electronic nose in food and healh applications: a review. Int. J. Comput. Corp. Res. 2, 1–17. Ali, Z., O’Hare, W., Theaker, B., 2003. Detection of bacterial contaminated milk by means of a quartz crystal microbalance based electronic nose. J. Therm. Anal. Calorim. 71, 155–161. Amari, A., El Barbri, N., El Bari, N., Llobet, E., Correig, X., Bouchikhi, B., Pardo, M., Sberveglieri, G., 2009. Potential application of the electronic nose for shelf-life determination of raw milk and red meat. AIP Conf. Proc. 1137, 457–460. Ampuero, S., Bosset, J.O., 2003. The electronic nose applied to dairy products: a review. Sensors Actuators B Chem. 94, 1–12. Arroyo, T., Lozano, J., Cabellos, J.M., Gil-Diaz, M., Santos, J.P., Horrillo, C., 2009. Evaluation of wine aromatic compounds by a sensory human panel and an electronic sose. J. Agric. Food Chem. 57, 11543–11549. Berna, A.Z., Trowell, S., Clifford, D., Cynkar, W., Cozzolino, D., 2009. Geographical origin of Sauvignon Blanc wines predicted by mass spectrometry and metal oxide based electronic nose. Anal. Chim. Acta 648, 146–152. Berna, A.Z., Trowell, S., Cynkar, W., Cozzolino, D., 2008. Comparison of metal oxide based electronic nose and mass spectrometry based electronic nose for the prediction of red wine spoilage. J. Agric. Food Chem. 56, 3238–3244. Bhattacharya, N., Bandyopadhyay, R., Bhuyan, M., Tudu, B., Ghosh, D., Jana, A., 2008a. Electronic nose for black tea classification and correlation of measurements with tea taster marks. IEEE Trans. Instrum. Meas. 57, 1313–1321. Bhattacharya, N., Metla, A., Bandyopadhyay, R., Tudu, B., Jana, A., 2008b. Incremental PNN classifier for a versatile electronic nose. In: 2008 3rd Int. Conf. Sens. Technol. 242–247. Bhattacharya, N., Tudu, B., Jana, A., Ghosh, D., Bandhopadhyaya, R., Bhuyan, M., 2008c. Preemptive identification of optimum fermentation time for black tea using electronic nose. Sensors Actuators B Chem. 131, 110–116. Bhattacharya, N., Tudu, B., Jana, A., Ghosh, D., Bandhopadhyaya, R., Saha, A.B., 2008d. Illumination heating and physical raking for increasing sensitivity of electronic nose measurements with black tea. Sensors Actuators B Chem. 131, 37–42. Bhattacharyya, N., Seth, S., Tudu, B., Tamuly, P., Jana, A., Ghosh, D., Bandyopadhyay, R., Bhuyan, M., Sabhapandit, S., 2007. Detection of optimum fermentation time for black tea manufacturing using electronic nose. Sensors Actuators B Chem. 122, 627–634. Biolatto, A., Grigioni, G., Irurueta, M., Sancho, A.M., Taverna, M., Pensel, N., 2007. Seasonal variation in the odour characteristics of whole milk powder. Food Chem. 103, 960–967. Blixt, Y., Borch, E., 1999. Using an electronic nose for determining the spoilage of vacuum-packaged beef. Int. J. Food Microbiol. 46, 123–134. Bonneau, M., Squires, J., 2004. Boar taint: causes and measurement. Encycl. Meat Sci., 91–96. Botre, B., Gharpure, D., Shaligram, A., Sadistap, S., Pardo, M., Sberveglieri, G., 2009. Semiconductor sensor array based electronic nose for milk, rancid milk and yoghurt odors identification. AIP Conf. Proc. 1137, 587–590. Bourrounet, B., Talou, T., Gaset, A., 1995. Application of a multi-gas-sensor device in the meat industry for boar-taint detection. Sensors Actuators B Chem. 27, 250–254. Brudzewski, K., Osowski, S., Dwulit, A., 2012. Recognition of coffee using differential electronic nose. IEEE Trans. Instrum. Meas. 61, 1803–1810. Buratti, S., Ballabio, D., Benedetti, S., Cosio, M.S., 2007. Prediction of Italian red wine sensorial descriptors from electronic nose, electronic tongue and spectrophotometric measurements by means of Genetic Algorithm regression models. Food Chem. 100, 211–218.

Capone, S., Epifani, M., Quaranta, F., Siciliano, P., Taurino, A., Vasanelli, L., 2001. Monitoring of rancidity of milk by means of an electronic nose and a dynamic PCA analysis. Sensors Actuators B Chem. 78, 174–179. Catarina Bastos, A., Magan, N., 2006. Potential of an electronic nose for the early detection and differentiation between Streptomyces in potable water. Sensors Actuators B Chem. 116, 151–155. Chaturvedula, V., Prakash, I., 2011. The aroma, taste, color and bioactive constituents of tea. J. Med. Plants Res. 5, 2110–2124. Cozzolino, D., Smyth, H.E., Cynkar, W., Janik, L., Dambergs, R.G., Gishen, M., 2008. Use of direct headspace-mass spectrometry coupled with chemometrics to predict aroma properties in Australian Riesling wine. Anal. Chim. Acta 621, 2–7. Cynkar, W., Cozzolino, D., Dambergs, B., Janik, L., Gishen, M., 2007. Feasibility study on the use of a head space mass spectrometry electronic nose to monitor red wine spoilage induced by Brettanomyces yeast. Sensors Actuators B Chem. 124, 167–171. Dutta, R., Hines, E.L., Gardner, J.W., Kashwan, K.R., Bhuyan, M., 2003. Tea quality prediction using a tin oxide-based electronic nose: an artificial intelligence approach. Sensors Actuators B Chem. 94, 228–237. El Barbri, N., Llobet, E., El Bari, N., Correig, X., Bouchikhi, B., 2008. Electronic nose based on metal oxide semiconductor sensors as an alternative technique for the spoilage classification of red meat. Sensors 8, 142–156. Eriksson, Å., Waller, K.P., Svennersten-Sjaunja, K., Haugen, J.-E., Lundby, F., Lind, O., Eriksson, A., 2005. Detection of mastitic milk using a gas-sensor array system (electronic nose). Int. Dairy J. 15, 1193–1201. Falasconi, M., Concina, I., Gobbi, E., Sberveglieri, V., Pulvirenti, A., Sberveglieri, G., 2012. Electronic nose for microbiological quality control of food products. Int. J. Electrochem. 2012, 1–12. Frost, J.A., 2001. Current epidemiological issues in human campylobacteriosis. J. Appl. Microbiol. 90, 85S–95S. Galdikas, A., Mironas, A., Senulien, D., Strazdien, V., Šetkus, A., Zelenin, D., 2000. Response time based output of metal oxide gas sensors applied to evaluation of meat freshness with neural signal analysis. Sensors Actuators B Chem. 69, 258– 265. Gardner, J., Bartlett, P., 1999. Electronic Noses, Principles and Applications. Oxford University Press, New York, NY, USA. Ghasemi-Varnamkhasti, M., Mohtasebi, S.S., Siadat, M., Balasubramanian, S., 2009. Meat quality assessment by electronic nose (machine olfaction technology). Sensors 9, 6058–6083. Gholam Hosseini, H., Luo, D., Xu, G., Liu, H., Benjamin, D., 2008. Intelligent fish freshness assessment. J. Sensors 628585, 8 pages. Gholamhosseini, H., Luo, D., Liu, H., Xu, G., 2007. Intelligent Processing of E-nose Information for Fish Freshness Assessment. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. pp. 173–177. Hansen, T., Boholt, K., Gammelgaard, E., Byrne, D.V., 2002. Sensory and electronic nose analysis of porcine meat loaf in relation to ingredient quality. In: Sensors, 2002. Proceedings of IEEE, vol. 1. pp. 736–740. Haugen, J.E., Chanie, E., Westad, F., Jonsdottir, R., Bazzo, S., Labreche, S., Marcq, P., Lundby, F., Olafsdottir, G., 2006. Rapid control of smoked Atlantic salmon (Salmo salar) quality by electronic nose: correlation with classical evaluation methods. Sensors Actuators B Chem. 116, 72–77. Hedberg, C., 1999. Food-related illness and death in the United States. Emerg. Infect. Dis. 5, 607–625. Herrero, A.M., 2008. Raman spectroscopy a promising technique for quality assessment of meat and fish: a review. Food Chem. 107, 1642–1651. James, D., Scott, S.M., Ali, Z., O’Hare, W.T., 2005. Chemical sensors for electronic nose systems. Microchim. Acta 149, 1–17. Jung, J.-Y., Lee, C.-S., 2011. Characteristics of the TiO2/SnO2 thick film semiconductor gas sensor to determine fish freshness. J. Ind. Eng. Chem. 17, 237–242. Kashwan, K.R., Bhuyan, M., 2005. Robust electronic-nose system with temperature and humidity drift compensation for tea and spice flavour discrimination. In: Sensors Int. Conf. new Tech. Pharm. Biomed. Res. 2005 Asian Conf. 154–158. Kirsching, A., 2012. Classification of meat with boar taint using an electronic nose. Acta Agric. Slov. 3, 99–103. Kottawa-Arachchi, J.D., Gunasekare, M.T.K., Ranatunga, M.A.B., Jayasinghe, L., Karunagoda, R.P., 2012. Analysis of selected biochemical constituents in black tea (Camellia sinensis) for predicting the quality of tea germplasm in sri lanka. Trop. Agric. Res. 23. Kumar, K.L.A., Durgajanani, S., Jeyaprakash, B.G., Rayappan, J.B.B., 2013. Nanostructured ceria thin film for ethanol and trimethylamine sensing. Sensors Actuators B Chem. 177, 19–26. Längkvist, M., Coradeschi, S., Loutfi, A., Rayappan, J.B.B., 2013. Fast classification of meat spoilage markers using nanostructured ZnO thin films and unsupervised feature learning. Sensors (Basel) 13, 1578–1592. Längkvist, M., Loutfi, A., 2011. Unsupervised feature learning for electronic nose data applied to bacteria identification in blood. In: NIPS 2011 Workshop on Deep Learning and Unsupervised Feature Learning. Li, X., Gu, Z., Cho, J., Sun, H., Kurup, P., 2011. Tin–copper mixed metal oxide nanowires: synthesis and sensor response to chemical vapors. Sensors Actuators B Chem. 158, 199–207. Loutfi, A., Coradeschi, S., Duckett, T., Wide, P., 2001. Odor source identification by grounding linguistic descriptions in an artificial nose. Proc. SPIE Conf. Sens. Fusion Archit. Algorithms Appl. V 4385, 273–282.

A. Loutfi et al. / Journal of Food Engineering 144 (2015) 103–111 Lozano, J., Arroyo, T., Santos, J.P., Cabellos, J.M., Horrillo, M.C., 2008a. Electronic nose for wine ageing detection. Sensors Actuators B Chem. 133, 180–186. Lozano, J., Santos, J.P., Horrillo, M.C., 2008b. Enrichment sampling methods for wine discrimination with gas sensors. J. Food Compos. Anal. 21, 716–723. Macagnano, A., Careche, M., Herrero, A., Paolesse, R., Martinelli, E., Pennazza, G., Carmona, P., D’Amico, A., Natale, C.D., 2005. A model to predict fish quality from instrumental features. Sensors Actuators B Chem. 111–112, 293–298. Magan, N., Pavlou, A., Chrysanthakis, I., 2001. Milk-sense: a volatile sensing system recognises spoilage bacteria and yeasts in milk. Sensors Actuators B Chem. 72, 28–34. Mani, G.K., Rayappan, J.B.B., 2013. A highly selective room temperature ammonia sensor using spray deposited zinc oxide thin film. Sensors Actuators B Chem. 183, 459–466. Mani, G.K., Rayappan, J.B.B., 2014a. Novel and facile synthesis of randomly interconnected ZnO nanoplatelets using spray pyrolysis and their room temperature sensing characteristics. Sensors Actuators B Chem. 198, 125–133. Mani, G.K., Rayappan, J.B.B., 2014b. Influence of copper doping on structural, optical and sensing properties of spray deposited zinc oxide thin films.pdf. J. Alloys Compd. 582, 414–419. Marsili, R.T., 1999. SPME-MS-MVA as an electronic nose for the study of Off-flavors in milk. J. Agric. Food Chem. 47, 648–654. Marsili, R.T., 2000. Shelf-life prediction of processed milk by solid-phase microextraction, mass spectrometry, and multivariate analysis. J. Agric. Food Chem. 48, 3470–3475. Muniyandi, I., Mani, G.K., Shankar, P., Rayappan, J.B.B., 2014. Effect of nickel doping on structural, optical, electrical and ethanol sensing properties of spray deposited nanostructured ZnO thin films. Ceram. Int. 40, 7993–8001. Musatov, V.Y., Sysoev, V.V., Sommer, M., Kiselev, I., 2010. Assessment of meat freshness with metal oxide sensor microarray electronic nose: a practical approach. Sensors Actuators B Chem. 144, 99–103. Olafsdottir, G., Nesvadba, P., Di Natale, C., Careche, M., Oehlenschlager, J., Tryggvadottir, S.V., Schubring, R., Kroeger, M., Heia, K., Esaiassen, M., Macagnano, A., Jorgensen, B.M., 2004. Multisensor for fish quality determination. Trends Food Sci. Technol. 15, 86–93. Olsen, E., Vogt, G., Ekeberg, D., Sandbakk, M., Pettersen, J., Nilsson, A., 2005. Analysis of the early stages of lipid oxidation in freeze-stored pork back fat and mechanically recovered poultry meat. J. Agric. Food Chem. 53, 338–348. Panigrahi, S., Balasubramanian, S., Gu, H., Logue, C.M., Marchello, M., 2006. Design and development of a metal oxide based electronic nose for spoilage classification of beef. Sensors Actuators B Chem. 119, 2–14. Pardo, M., Niederjaufner, G., Benussi, G., Comini, E., Faglia, G., Sberveglieri, G., Holmberg, M., Lundstrom, I., 2000. Data preprocessing enhances the classification of different brands of Espresso coffee with an electronic nose. Sensors Actuators B Chem. 69, 397–403. Pardo, M., Sberveglieri, G., 2005. Classification of electronic nose data with support vector machines. Sensors Actuators B Chem. 107, 730–737. Perera, A., Pardo, A., Barrettino, D., Hierlermann, A., Marco, S., 2010. Evaluation of fish spoilage by means of a single metal oxide sensor under temperature modulation. Sensors Actuators B Chem. 146, 477–482. Peris, M., Escuder-Gilabert, L., 2009. A 21st century technique for food control: electronic noses. Anal. Chim. Acta 638, 1–15. Persaud, K., Dodd, G.H., 1982. Analysis of discrimination mechanisms in the mammalian olfactory system using a model nose. Nature 299, 352–355. Prieto, N., Rodriguez-MÃéndez, M.L., Leardi, R., Oliveri, P., Hernando-Esquisabel, D., Iñiguez-Crespo, M., de Saja, J.A., 2012. Application of multi-way analysis to UV– visible spectroscopy, gas chromatography and electronic nose data for wine ageing evaluation. Anal. Chim. Acta 719, 43–51. Ragazzo-Sanchez, J.A., Chalier, P., Chevalier-Lucia, D., Calderon-Santoyo, M., Ghommidh, C., 2009. Off-flavours detection in alcoholic beverages by electronic nose coupled to GC. Sensors Actuators B Chem. 140, 29–34.

111

Rodríguez, J., Durán, C., Reyes, A., 2009. Electronic nose for quality control of colombian coffee through the detection of defects in cup tests. Sensors 10, 36– 46. Roy, S., Basu, S., 2004. ZnO thin film sensors for detecting dimethyl- and trimethylamine vapors. J. Mater. Sci.: Mater. Electron. 15, 321–326. Santos, J.P., Lozano, J., Aleixandre, M., Arroyo, T., Cabellos, J.M., Gil, M., del Carmen Horrillo, M., 2010. Threshold detection of aromatic compounds in wine with an electronic nose and a human sensory panel. Talanta 80, 1899–1906. Scott, S.M., James, D., Ali, Z., 2006. Data analysis for electronic nose systems. Microchim. Acta 156, 183–207. Shi, B., Zhao, L., Zhi, R., Xi, X., 2012. Optimization of electronic nose sensor array by genetic algorithms in Xihu-Longjing Tea quality analysis. Math. Comput. Model. Singh, S., Hines, E.L., Gardner, J.W., 1996. Fuzzy neural computing of coffee and tainted-water data from an electronic nose. Sensors Actuators B Chem. 30, 185– 190. Sivalingam, D., Gopalakrishnan, J.B., Rayappan, J.B.B., 2012. Structural, morphological, electrical and vapour sensing properties of Mn doped nanostructured ZnO thin films. Sensors Actuators B Chem. 166–167, 624–631. Sivalingam, D., Rayappan, J.B.B., Gandhi, S., Madanagurusamy, S., Sekar, R.K., Krishnan, U., 2011. Ethanol and TMA sensing by ZnO based nanostructured thin films. Int. J. Nanosci. 10, 1161–1165. Tudu, B., Jana, A., Metla, A., Ghosh, D., Bhattacharyya, N., Bandyopadhyay, R., 2009a. Electronic nose for black tea quality evaluation by an incremental RBF network. Sensors Actuators B Chem. 138, 90–95. Tudu, B., Metla, A., Das, B., Bhattacharyya, N., Jana, A., Ghosh, D., Bandyopadhyay, R., 2009b. Towards versatile electronic nose pattern classifier for black tea quality evaluation: an incremental fuzzy approach. IEEE Trans. Instrum. Meas. 58, 3069–3078. Vestergaard, J.S., Haugen, J.-E., Byrne, D.V., 2006. Application of an electronic nose for measurements of boar taint in entire male pigs. Meat Sci. 74, 564–577. Wang, B., Xu, S., Sun, D.-W., 2010. Application of the electronic nose to the identification of different milk flavorings. Food Res. Int. 43, 255–262. Wang, G., He, X., Zhou, F., Li, Z., Fang, B., Zhang, X., Wang, L., 2012. Application of gold nanoparticles/TiO2 modified electrode for the electrooxidative determination of catechol in tea samples. Food Chem. 135, 446–451. Warm, K., Martens, M., Nielsen, J., 2001. Sensory quality criteria for five fish species predicted from near-infrared (NIR) reflectance measurement. J. Food Qual. 24, 389–403. Wilson, A.D., 2013. Diverse applications of electronic-nose technologies in agriculture and forestry. Sensors (Basel) 13, 2295–2348. Wilson, A.D., Baietto, M., 2011. Advances in electronic-nose technologies developed for biomedical applications. Sensors (Basel) 11, 1105–1176. Yamazoe, N., Sakai, G., Shimanoe, K., 2003. Oxide semiconductor gas sensors. Catal. Surv. Asia 7, 63–75. Yang, X., Xie, W., Zhang, C., Wu, H., Li, S., Yang, L., 2009. Identification of sensory quality of rapid fermented fish using electronic nose. Int. Conf. Inf. Eng. Comput. Sci. 1–3. Yang, Z., Dong, F., Shimizu, K., Kinoshita, T., Kanamori, M., Morita, A., Watanabe, N., 2009. Identification of coumarin-enriched Japanese green teas and their particular flavor using electronic nose. J. Food Eng. 92, 312–316. Yu, H., Wang, J., Xu, Y., 2007. Identification of adulterated milk using electronic nose. Sensors Mater. 19, 275–285. Yu, H., Wang, J., Zhang, H., Yu, Y., Yao, C., 2008. Identification of green tea grade using different feature of response signal from E-nose sensors. Sensors Actuators B Chem. 128, 455–461. Yu, H., Wang, Y., Wang, J., 2009. Identification of tea storage times by linear discrimination analysis and back-propagation neural network techniques based on the eigenvalues of principal components analysis of e-nose sensor signals. Sensors 9, 8073–8082.