The Feasibility and Stability of Distinguishing the Kiwi Fruit Geographical Origin Based on Electronic Nose Analysis

Abstract

This study focused on the feasibility and stability of electronic nose in differentiating kiwi fruit according to geographical origin. Ninety kiwi fruit samples with protected designation of origin were collected from Hunan, Shaanxi, and Sichuan Provinces of China respectively. The freeze-dried kiwi fruit samples were analyzed by a PEN3 portable electronic nose coupled with pattern recognition techniques. The stability of traceability based on this method was also investigated. The difference analysis results indicated that kiwi fruit samples from different regions had their unique aroma compounds fingerprint characteristics. Principal component analysis (PCA) showed a good visualization of clusters between regions, and freeze-dried samples (before or after storage) from the three areas were successfully identified by stepwise discriminant analysis (SLDA) and soft independent modeling of class analogy (SIMCA). Therefore, it was acceptable that the electronic nose is cost-effective, practical and could be used as a potential technology for protecting or authenticating specialty agricultural products and geographical indication. But its stability for tracing needs to be further enhanced.

Introduction

Nowadays, vegetables and fruits are the main food consumption around the world on account of their health benefits for prevention and treatment of various diseases (Park et al., 2011). Kiwi fruit, which originated in China, has significant amounts of vitamin C and also contains essential amino acids and mineral elements necessary for human, and its quality is closely related to geographical origin (Ma et al., 2013). At present, there are more than ten kinds of kiwi fruit geographical indication products that have already been registered in China, such as “Zhouzhi kiwi fruit”, “Muchuan kiwi fruit”, “Meixian kiwi fruit”, and so on (Xiong et al., 2010). Products with geographical indication are popular in the market and have good and special characteristics, but they often are replaced with low-quality products by illegal operators, which brings unfair competitions and break the normal market order seriously. For this reason, the geographical origin identification of kiwi fruit is urgently needed in order to protect featured products and prompt the sustainable development of local brand industry.

Some food components such as proteins, vitamins, sugars, acids, elements as well as aroma compounds are influenced by special geographical climate conditions of producing area, which are responsible for the quality difference in agricultural products from different regions. Most of the previous research on geographical origin traceability focused on detecting a set of specific indexes, for example stable isotopes or mineral elements, and obtained correct discrimination rates of 60%∼100% (Fabani et al., 2010; Zhao et al., 2011; Pilgrim et al., 2010; Bontempo et al., 2011). However, analysis of stable isotopes or mineral elements is expensive and need professional operators. Therefore, it is practical to seek some of the new variables as traceable markers to make geographical classification easy to conduct.

Volatile compounds as important quality characteristics of food were confirmed to be origin-related (Longobardi et al., 2011). Feudo et al. (2011) determined the volatile fraction of tomatoes from Italian four different regions and classified them successfully based on geographical origin. Electronic nose (E-nose), known as artificial olfactory system, is an instrument which imitates the sense of smell. There are many types of e-nose on basis of metal oxide semiconductors sensors, piezoelectric sensors, thermal sensors or mass spectrometer available in the laboratory. The e-nose analysis is rapid, low-cost, and recently has been successfully applied to identify the geographical origin of wine (Berna et al., 2009), virgin olive oils (Casale et al., 2010), hard cheeses (Gursoy et al., 2009) and propoli (Cheng et al., 2013).

Until now, researchers have only discussed the applicability of electronic nose for classification according to the region from the perspective of differences. However, volatile compounds in fresh fruit change quickly during storage, so the practicability and stability of traceability based on this method is not clear. This research focused on the use of electronic nose PEN in distinguishing the geographical origin of freeze-dried kiwi fruit and the stability of traceability based on this method.

Materials and Methods

Kiwi fruit samples and pre-treatment    Ninety kiwi fruit samples with protected designation of origin were collected in 2012 from Shaanxi, Hunan, and Sichuan Provinces of China, respectively (Fig.1). In Shaanxi Province, fifteen samples were collected from Zhouzhi county (34°04 – 34°11N, 108°03 – 108°19E) in August, and fifteen samples from Mei county (34°09 – 34°16N, 107°43 – 107°54E) in September. The average annual temperature of this Province is 12.9°C, the altitude is 410 – 760 m, and the mean annual rainfall is 631 mm. Thirty samples were collected from Muchuan county (28°58 – 29°03N, 103°49 – 104°01E) in Sichuan Province in September. Muchuan′s average annual temperature is 17.0°C, the altitude is 412 – 895 m, and the mean annual rainfall is 1350 mm. The remaining thirty samples were collected from Yongshun county (28°47 – 29°00N, 109°58 – 110°06E) in Hunan Province in September. Its average annual temperature is 17.5°C, the altitude is 404 – 842 m, and the mean annual rainfall is 970 mm. To obtain a representative sampling, the main kiwifruit-producing counties and towns and the most common varieties (“Qinmei—A. deliciosa C.F.Liang et A.R.Ferguson” and “Hongyang—A. Chinensis Planch” from Shaanxi Province; “Hongyang —A. Chinensis Planch” from Sichuan Province; “Miliang No.1—A. deliciosa C.F.Liang et A.R.Ferguson” from Hunan Province) were chosen in each province. 1 kg kiwi fruits samples that reached commercial maturity (firmness was about 10 – 15 kg·cm⁻², soluble solids content was about 6.0% – 12.0%.) were picked.

Fig. 1.

Geographical origin of kiwi fruit samples collected.

In order to slow the loss rate of fruit aromas during ripening process, before the test, fruits were quickly peeled and cut by a stainless steel paring knife, frozen at −20°C and then freeze-dried for 72 hours using a Christ Alpha 1-2LD Plus freeze-dryer (Marin Christ, Germany). Freeze-dried samples were powdered and stored at 0 – 4°C in sealed containers in desiccators. Each freeze-dried sample was divided into two parts, one part as a member of sample set 1 and the other as a member of sample set 2. All samples from sample set 1 were immediately analyzed by e-nose after pre-treatment, and samples in sample set 2 were stored at 0 – 4°C for 3 months and then sent to the e-nose for analysis.

Portable electronic nose    The aroma compounds analysis was conducted with a PEN3 portable electronic nose (Win Muster Airsense, Germary), which basically consists of an array of 10 different metal oxide sensors (MOS) positioned into a 1.8 mL chamber and pattern recognition software for data processing. The sensors have good selectivity for sulfur organic compounds, methane, hydrogen, alcohol and hydrocarbons (Table 1). The limit of detection (LOD) of the instrument ranged from 0.1 to 5 ppm. The electronic nose has external calibration procedure, which can immediately notices deviations from the “standard smell”.

Table 1. The sensors array of electronic nose PEN 3

Number in array	sensor	Performance characteristics	sensitivity
1	W1C	Aromatic-aliphatic compounds	10 ppb
2	W5S	Oxynitride	1 ppb
3	W3C	Ammonia, Aroma compounds	10 ppb
4	W6S	Hydrogen	10 ppb
5	W5C	Hydrocarbons, Aroma compounds	10 ppb
6	W1S	Methane	10 ppb
7	W1W	Sulfide	1 ppb
8	W2S	Alcohol	10 ppb
9	W2W	Organic sulfur compounds, Aroma compounds	1 ppb
10	W3S	Hydrocarbons	10 ppb

1 g of fruit powder was placed in 20 mL airtight glass vial, 4 mL of distilled water was added and then sealed with a screw cap having PTFE/silicone septum. Vials were equilibrated at 20 ± 1°C for one hour and analyzed at ambient temperature (Benedetti et al., 2008). The volatile compounds were sampled by a manual dynamic headspace technique. One luer-lock needle connected to a tube was used to absorb the aroma accumulated inside the vial. While a second needle connected to a clean air source (charcoal filter) was used to enrich aroma concentration. The sample headspace was pumped over the sensors array with a flow of 300 mL/min and lasted for 60 s for response record. After that, clean air was used to rinse the system for 180 s in order to keep the gas sensor signal back to the baseline. In total, 180 freeze-dried samples were analyzed (90 before storage + 90 after storage).

Data handling and pattern recognition techniques    Sensor data was analyzed by analysis of variance (ANOVA), principal component analysis (PCA) and stepwise discriminant analysis (SLDA) with SPSS 18.0 (SPSS Inc., Chicago), soft independent modeling of class analogy (SIMCA) with Unscrambler 9.7 (CAMO ASA, Norway).

ANOVA was performed to examine whether there were significant differences of samples from different provinces. PCA is frequently used to reduce the dimensionality of original data by calculating several components that best describe the differences between objects and acquire visualization of cluters (Cynkar et al., 2010). As an unsupervised technique, PCA is able to offer important characteristic information about samples with a small number of factors. But if the similarities are very remarkable, PCA may fail to obtain precise conclusion, while SLDA can give a successful model. SLDA is a supervised classification technique that maximizes the variance between-class and minimizes the variance within-class based on a number of linear discriminant functions (Kovács et al., 2010). The principle of SLDA for variables selection is the maximum contribution for group differentiation.

SIMCA, focusing on modeling the similarities between members of the same class (Tominaga, 1999), has been widely used in origin classification (Stanimirova et al., 2010; Bevilacqua et al., 2012; Vitale et al., 2012). This method is based on making a PCA model for each class in the training set. Unknown samples are compared with the class models and then assigned to classes according to their analogy to the training samples. The actual classification stage uses significance tests, where the decisions are based on statistical tests performed on the orthogonal projection (object-to-model) distances. The distance is computed as the square root of the sample residual variance. If a sample belongs to a class, it should have a small distance to the class model (the ideal situation being “distance=0”); Otherwise, it is rejected. For this method, the kiwi fruit samples were randomly divided into two categories, with 2/3 of the samples as a training set and the remaining 1/3 as a test set. Before we classify new samples by SIMCA, each class must be described by a PCA model. A cross-validation procedure was adopted to determine the optimal number of principal components. In general, non-error classification and prediction rate as the results of all classification methods were used to represent the method's performances (Casale et al., 2010; Zhao et al., 2012). Apart from non-error classification and prediction, sensitivity and specificity are two additional parameters used to evaluate the prediction quality of models (Stanimirova et al., 2010; Marini et al., 2006). Sensitivity is defined as the percentage of samples from the modeled class that are accepted by the class model, while specificity is the percentage of samples belonging to other classes that are rejected by the class model (Marini et al., 2006). A model with sensitivity and specificity of 100% is considered ideal for geographical origin differentiation.

Results

Spectral response characteristics and difference analysis    Fig.2 shows the electronic nose spectra of one freeze-dried kiwi fruit sample, which was collected by 10 sensors that defined as variables. It was observed that the signals of W5S and W1W sensors were stronger than others. Almost all of the sensors had reached the peak between 0 and 30 s and were kept stable till the end of measurement. In addition, we can see that the spectral response of freeze-dried samples was significantly reduced after storage especially the W5S and W1W sensors, in comparison with those of freeze-dried samples before storage.

Fig. 2.

(a)&(b) Electronic nose spectra of kiwi fruit sample before and after storage.

The signal values of sensors were the ratio of detected resistance value (unit: ohm) and base value (unit: ohm). As a result of ANOVA, the signals of W1C, W3C, W5C, W1S, W2S, W3S, W5S, W1W and W2W sensors from freeze-dried samples before storage were significantly different among regions, and those of all sensors from freeze-dried samples after storage were significantly different (Table 2, Table 3). Furthermore, the values of W1S, W2S, W3S, W5S, W6S, W1W and W2W sensors decreased significantly after storage while those of W1C, W3C and W5C sensors increased significantly. In addition, the result of differences analysis of the same kiwi fruit variety planted in two different regions showed that the sensor values of W1C, W1W and W2W were significantly different (Table 4), which demonstrated that geographical origin had certain influence on the aroma components of kiwi fruit.

Table 2. Sensor values of samples before storage from different provinces

Sensor	Hunan Province	Shaanxi Province	Sichuan Province
W1C	0.50 b ± 0.04	0.57 a ± 0.06	0.45 c ± 0.04
W5S	28.29 a ± 3.51	20.16 b ± 3.40	31.63 a ± 8.63
W3C	0.67 b ± 0.04	0.73 a ± 0.04	0.63 c ± 0.05
W6S	1.48 a ± 0.04	1.50 a ± 0.03	1.48 a ± 0.02
W5C	0.82 b ± 0.02	0.85 a ± 0.02	0.79 c ± 0.03
W1S	4.91 b ± 0.54	4.60 c ± 0.45	5.21 a ± 0.68
W1W	13.64 a ± 1.40	7.57 b ± 1.59	14.80 a ± 5.15
W2S	3.98 b ± 0.72	3.65 c ± 0.42	4.67 a ± 0.65
W2W	4.63 b ± 0.54	3.68 c ± 0.61	6.21 a ± 1.89
W3S	2.08 c ± 0.05	2.40 a ± 0.07	2.28 b ± 0.05

Note: The different letters in this table represent significant difference (P < 0.05); Unit: dimensionless

Table 3. Sensor values of samples after storage from different provinces

Sensor	Hunan Province	Shaanxi Province	Sichuan Province
W1C	1.09 a ± 0.07	0.81 c ± 0.03	0.98 b ± 0.07
W5S	6.33 a ± 1.84	3.61 b ± 0.33	7.51 a ± 3.87
W3C	1.05 a ± 0.04	0.88 c ± 0.02	0.98 b ± 0.05
W6S	0.99 b ± 0.01	1.02 a ± 0.02	0.99 b ± 0.01
W5C	1.01 a ± 0.02	0.94 c ± 0.01	0.98 b ± 0.02
W1S	1.00 c ± 0.10	1.42 a ± 0.07	1.14 b ± 0.09
W1W	3.90 a ± 0.73	2.54 b ± 0.26	4.25 a ± 1.56
W2S	0.98 c ± 0.07	1.27 a ± 0.06	1.12 b ± 0.08
W2W	1.85 b ± 0.25	1.44 c ± 0.15	2.13 a ± 0.33
W3S	1.04 c ± 0.01	1.09 a ± 0.02	1.06 b ± 0.02

Note: The different letters in this table represent significant difference (P < 0.05); Unit: dimensionless

Table 4. Sensor values of samples of the same variety in the different regions

Sensor	Shaanxi Province	Sichuan Province
W1C	0.52 a ± 0.04	0.45 b ± 0.04
W5S	24.69 a ± 1.72	31.63 a ± 8.63
W3C	0.68 a ± 0.04	0.63 a ± 0.05
W6S	1.46 a ± 0.01	1.48 a ± 0.02
W5C	0.82 a ± 0.02	0.79 a ± 0.03
W1S	4.51 a ± 0.38	5.21 a ± 0.68
W1W	9.29 b ± 0.84	14.80 a ± 5.15
W2S	4.56 a ± 0.48	4.67 a ± 0.65
W2W	4.11 b ± 0.32	6.21 a ± 1.89
W3S	2.26 a ± 0.06	2.28 a ± 0.05

Note: The different letters in this table represent significant difference (P < 0.05); Unit: dimensionless

Principal component analysis (PCA)    Sensor data of electronic nose were submitted to PCA in order to preliminarily study differences of kiwi fruit from three provinces. The score plot of the first three principal components of freeze-dried samples before storage is shown in Fig.3, where we can see a good trend of separation among samples produced in different regions. The first principal component (PC1) accounted for about 59.1% of the total variance, the second one (PC2) 21.3% and the third one (PC3) 13.8%. PC1 mainly described the signals from W1C, W3C, W5C, W1W, W2W and W2S sensors. PC2 represented the information about W6S, W1S, and W3S sensors. PC3 was mainly correlated with W5S sensor. The good visualization of clusters among the three regions indicated the feasibility of aroma compounds fingerprinting based on electronic nose in distinguishing geographical origin of kiwi fruit.

Fig. 3.

Three dimensional PCA plot of kiwi fruit samples before storage.

Stepwise discriminant analysis (SLDA)    The SLDA based on Wilks' lambda was performed to assess the validity of electronic nose spectra for kiwi fruit origin discrimination. The value of Wilks' lambda ranges from 0 to 1. A variable (sensor) with small Wilks' lambda value is to be reserved for its significant impact on discrimination model. Thus sensors of electronic nose possessing relevant information for geographical origin classification were selected during analysis procedure, while other irrelevant sensors were eliminated. The selection criteria was based on significance of difference among three regions.

The results of canonical discriminant analysis of freeze-dried kiwi fruit samples (before and after storage) are shown in Fig.4. From Fig.4a, it can be observed that freeze-dried samples before storage from the different regions were classified into three groups. Function 1 described 83% of the total variance, while Function 2 described 17%. The classification functions (Model 1) for sample set 1 based on W1C, W3C, W5C, W1S, W2S, W3S and W2W were as follow:

•    Y₁ (Hunan) = 4097.651(W1C) − 13310.924(W3C) + 21219.845(W5C) + 96.157(W1S) − 81.798(W2S) + 908.817(W3S) + 98.732(W2W) − 6571.793
•    Y₂ (Shaanxi) =4352.893(W1C) − 13745.425(W3C) + 21463.535(W5C) + 98.166(W1S) − 92.885(W2S) + 968.315(W3S) + 100.251(W2W) − 6716.718
•    Y₃ (Sichuan) = 4136.113(W1C) − 13317.609(W3C) + 21158.610(W5C) + 102.244(W1S) − 86.816(W2S) + 904.272(W3S) + 100.549(W2W) − 6543.254

Fig. 4.

(a)&(b) Scatter plot of canonical discriminant analysis of samples before and after storage.

The overall correct discrimination rates of original and cross-validation were 84.4% and 76.7%, respectively. Samples from Shaanxi were separated well with those from Hunan and Sichuan, which acquired 100% classification. However, there were several samples from Hunan that were misclassified into Sichuan, which led to the lower average classification. Although some overlaps existed, most of the samples from Hunan (83.3% for original, 76.7% for cross-validation) and Sichuan (70.0% for original, 63.3% for cross-validation) were able to be correctly distinguished according to their geographical origin.

Fig.4b shows the result of canonical discriminant analysis for the freeze-dried samples after storage. Function 1 described 91% of the total variance while Function 2 described 9%. The correct classifications of 93.3% and 92.2% for original and cross-validation were acquired. The new discriminant functions (Model 2) for each region were as follow:

•    Y₁ (Hunan) = 27294.831(W1C) − 142194.032 (W3C) + 251573.662(W5C) + 9170.460(W1S) − 10102.401(W2S) + 14165.508(W3S) + 971.332(W2W) − 75540.178
•    Y₂ (Shaanxi) = 26485.255(W1C) − 140923.288 (W3C) + 251577.418(W5C) + 9467.417(W1S) − 10350.613(W2S) + 14221.018(W3S) + 931.364(W2W) − 76076.189
•    Y₃ (Sichuan) = 27117.442(W1C) − 141539.514(W3C) + 250576.822(W5C) + 9047.947(W1S) − 9929.153(W2S) + 14005.252(W3S) + 975.571(W2W) − 74917.941

According to our results, part of freeze-dried kiwi fruit aroma compounds had escaped during cool-storage with the phenomenon of spectral response decline from Fig.1a to Fig.1b. However, a better classification among three provinces can be observed from Fig.4b. When the samples from sample set 2 were analyzed using the discriminant model 1 built with sample set 1, the obtained classification results were not good (only few samples were correctly differentiated). This indicated the considerable influence of storage on aroma compound fingerprints of freeze-dried kiwi fruit thus giving rise to a poor prediction ability of the models. Therefore, the detection time and storage condition are crucial factors for geographical differentiation based on aroma fingerprints analysis, and should be fully regarded.

SIMCA models    In our study, the first three principal components of training set were selected to optimise SIMCA class models at 95% confidence level. Table 5 shows the result of SIMCA modeling referring to sample set 1 and sample set 2. The percentage of correct classification and prediction for sample set 1 were 100% and 93.3%, respectively. Only two samples from Shaanxi were out of the model, which resulted in a high sensitivity of 93.3%. However, the models showed a poor specificity (71.3%) mainly due to the numerous similarities between the samples from Sichuan and Hunan. In particular, the model built for Shaanxi was the only one with an acceptable specificity (100%), while models for Sichuan (45.5%) and Hunan (68.4%) seemed to be difficult to differentiate between samples from the two regions. Moreover, it can be observed that the models obtained by sample set 2 showed a better sensitivity (97.0%), but they presented a lower specificity (59.6%). Though with poor specificity, good prediction ability (96.7%) could be acquired.

Table 5. Results of sample set 1 and 2 by SIMCA

Sample	Classification, %	Prediction, %	Mean sensitivity, %	Mean specificity, %
Sample set 1	100	93.3	93.9	71.3
Sample set 2	100	96.7	97.0	59.6

Discussion

A number of studies concerning chemical compounds and sensory evaluation of kiwi fruit have been reported (Soufleros et al., 2001; Esti et al., 1998). From our study, kiwi fruit samples from different regions had their unique aroma compounds fingerprint characteristics, which was possibly dependent upon the geographical location and climate factors. Castro-Vázquez et al. (2010) investigated the effect of the geographical origin on the volatile composition of chestnut honeys produced in different areas, and found that samples from the north-east presented higher concentrations of aldehydes, alcohols, lactones and volatile phenols. Samples from the north-west area showed superior levels of terpenes, esters and some benzene derivatives, and those produced in the south-east area were rich in norisoprenoids content. The different climatic conditions influence the honeys volatile composition since they possess a characteristic aroma profiles intimately related with the geographical area of provenance. Zhouzhi and Meixian counties of Shaanxi Province are located in China's inland heartland, where the climate is temperate continental monsoon climate, the rainfall is relatively low, and kiwi fruit are grown in the plain areas. Muchuan county of Sichuan Province is located in southwest hinterland of China, where features a subtropical monsoon climate, and the weather is milder with more rainfall compared to Shaanxi Province. Kiwi fruit here are mainly grown in the high altitudes areas. Yongshun county of Hunan Province is located in middle southern China. The climate here is identical to Sichuan province while the rainfall is moderate. The differences in the climate, terrain, and cultivation pattern of fruit geographical origin affected the quality and sensory characteristics, which resulted in different electronic nose spectral response in the samples we used and made it possible for geographical identification.

Results of ANOVA showed the significant difference of sensor values among regions and indicated the crucial roles of W1C, W3C, W5C, W1S, W2S, W3S and W2W sensors in differentiating kiwi fruit provenance. Steine et al. (2001) analyzed the volatile organic fractions (hydrocarbons, methane, fluoride, etc.) of Valencia orange juices from five different regions with Fox 3000 electronic nose and obtained good discrimination (98%) of their provenance. Casale et al. (2010) successfully authenticated Ligurian e.v. olive oil geographical origin using electronic nose with chemometric package. In this work, the discriminant results of SLDA and SIMCA indicated that electronic nose analysis can successfully distinguish the geographical origin of freeze-dried kiwi fruit whether before or after storage.

The feasibility of electronic nose in food authentication according to geographical origin has been confirmed by researchers, but some factors (storage, variety, etc.) may affect the stability of classification model. Previous studies reported that the flavour volatiles in fresh fruit would change during ripening process (Wan et al., 1999; Talens et al., 2003; Torres et al., 2012). This study found that the sensor values of W1S, W2S, W3S, W5S, W6S, W1W and W2W in freeze-dried kiwi fruit were significantly reduced, while those of W1C, W3C and W5C were significantly rised after storage at 0 − 4°C (Table 3). Besides, the responses of W5S and W1W declined greatly after storage for 3 months, which might imply that the most heavily affected sensors were also those with the highest response values. Due to the adaptability of kiwi fruit to climates, each region has their unique varieties. The varietal differences may have several contributions to the regional difference. From Table 4, it can be observed that the aroma fingerprint characteristics of the same kiwi fruit variety from different regions were significantly different. This demonstrated that geographical origin had an effect on the aroma fingerprint information of kiwi fruit. Exploring the contributions of geographical origin and variety to the aroma compounds in kiwi fruit samples is necessary for providing more reliable classification functions from further study.

The fruit aroma compounds changed during cold storage thereby presenting different classified models for freeze-dried samples before and after storage. E-nose identification of origins was better with samples after storage (93.3%) than those before storage (84.4%). This result possibly because the aroma profiles of freeze-dried fruit from each region were more concentrated and stable with storage time, thus the better separation tendency of samples after storage was observed (Fig.4). However, the results of using the discriminant model 1 (built with freeze-dried samples before storage) to identify the set 2 samples (freeze-dried samples after storage) were not satisfying. This suggested that the storage could affect the discriminant validity of classified model, and the stability of traceability based on this method remains to be improved and perfected.

This study showed the fact that geographical origin had an effect on the aroma compounds composition of kiwi fruit, and electronic nose analysis coupled with chemometrics is fast and economical as a potential method for identifying the provenance of kiwi fruit and protecting geographical indication products from being fabricated, although storage affected the spectral characteristics of kiwi fruit.

Conclusion

Kiwi fruit samples from different regions have their unique aroma compounds fingerprint characteristics. Electronic nose analysis is a potential technique for authenticating the geographical origin of kiwi fruit, but the storage maybe affect its spectral characteristics, and the stability of classified model needs to be further improved.

Acknowledgment    This study was funded by National Science & Technology Pillar Program (No.2012BAK17B06). The authors would like to thank local personnels for the sampling work assistance and Anthony Okholaye Ojokoh, working in the Department of Microbiology, Federal University of Technology, Nigeria, for his careful corrections of this manuscript.

References

•  Benedetti,  S.,  Buratti,  S.,  Spinardi,  A.,  Mannino,  S., and  Mignani,  I. (2008). Electronic nose as a non-destructive tool to characterise peach cultivars and to monitor their ripening stage during shelf-life. Postharvest Biol. Technol., 47, 181-188.
•  Berna,  A.Z.,  Trowell,  S.,  Clifford,  D.,  Cynkar,  W., and  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.
•  Bevilacqua,  M.,  Bucci,  R.,  Magrì,  A.D.,  Magrì,  A.L., and  Marini,  F. (2012). Tracing the origin of extra virgin olive oils by infrared spectroscopy and chemometrics: A case study. Anal. Chim. Acta, 717, 39-51.
•  Bontempo,  L.,  Camin,  F.,  Manzocco,  L.,  Nicolini,  G.,  Wehrens,  R.,  Ziller,  L., and  Larcher,  R. (2011). Traceability along the production chain of Italian tomato products on the basis of stable isotopes and mineral composition. Rapid Commun. Mass Spectrom., 25, 899-909.
•  Casale,  M.,  Casolino,  C.,  Oliveri,  P., and  Forina,  M. (2010). The potential of coupling information using three analytical techniques for identifying the geographical origin of Liguria extra virgin olive oil. Food Chem., 118, 163-170.
•  Castro-Vázquez,  L.,  Díaz-Maroto,  M.,  De Torres,  C., and  Pérez-Coello,  M. (2010). Effect of geographical origin on the chemical and sensory characteristics of chestnut honeys. Food Res. Int., 43, 2335-2340.
•  Cheng,  H.,  Qin,  Z.H.,  Guo,  X.F.,  Hu,  X.S., and  Wu,  J.H. (2013). Geographical origin identification of propolis using GC-MS and electronic nose combined with principal components analysis. Food Res. Int., 51, 813-822.
•  Cynkar,  W.,  Dambergs,  R.,  Smith,  P., and  Cozzolino  D. (2010). Classification of Tempranillo wines according to geographic origin: Combination of mass spectrometry based electronic nose and chemometrics. Anal. Chim. Acta, 660, 227-231.
•  Esti,  M.,  Messia,  M.C.,  Bertocchi,  P.,  Sinesio,  F.,  Moneta,  E.,  Nicotra,  A.,  Fantechi,  P., and  Palleschi,  G. (1998). Chemical compounds and sensory assessment of kiwifruit (Actinidia chinensis (Planch.) var. chinensis): electrochemical and multivariate analyses. Food Chem., 61, 293-300.
•  Fabani,  M.P.,  Arrúa,  R.C.,  Vázquez,  F.,  Diaz,  M.P.,  Baroni,  M.V., and  Wunderlin,  D.A. (2010). Evaluation of elemental profile coupled to chemometrics to assess the geographical origin of Argentinean wines. Food Chem., 119, 372-379.
•  Feudo,  G.L.,  Macchione,  B.,  Naccarato,  A.,  Sindona,  G., and  Tagarelli,  A. (2011). The volatile fraction profiling of fresh tomatoes and triple concentrate tomato pastes as parameter for the determination of geographical origin. Food Res. Int., 44, 781-788.
•  Gursoy,  O.,  Somervuo,  P., and  Alatossava,  T. (2009). Preliminary study of ion mobility based electronic nose MGD-1 for discrimination of hard cheeses. J. Food Eng., 92, 202-207.
•  Kovács,  Z.,  Dalmadi,  I.,  Lukács,  L.,  Sipos,  L.,  Szántai-Kőhegyi,  K.,  Kókai,  Z., and  Fekete,  A. (2010). Geographical origin identification of pure Sri Lanka tea infusions with electronic nose, electronic tongue and sensory profile analysis. J. Chemometrics, 24, 121-130.
•  Longobardi,  F.,  Casiello,  G.,  Sacco,  D.,  Tedone,  L., and  Sacco,  A. (2011). Characterisation of the geographical origin of Italian potatoes, based on stable isotope and volatile compound analyses. Food Chem., 124, 1708-1713.
•  Ma,  Y.Y.,  Guo,  B.L.,  Wei,  Y.M.,  Zhao,  H.Y., and  Wei,  S. (2013). Discrimination of kiwi fruit according to geographic origin based on organic compound fingerprint analysis. Agr. Sci. China, 46, 3864-3870. (in Chinese)
•  Marini,  F.,  Magrì,  A.L.,  Bucci,  R.,  Balestrieri,  F., and  Marini,  D. (2006). Class-modeling techniques in the authentication of Italian oils from Sicily with a Protected Denomination of Origin (PDO). Chemometr. Intell. Lab., 80, 140-149.
•  Park,  Y.S.,  Leontowicz,  H.,  Leontowicz,  M.,  Namiesnik,  J.,  Suhaj,  M.,  Cvikrová,  M.,  Martincová,  O.,  Weisz,  M., and  Gorinstein,  S. (2011). Comparison of the contents of bioactive compounds and the level of antioxidant activity in different kiwifruit cultivars. J. Food Comp. Anal., 24, 963-970.
•  Pilgrim,  T.S.,  Watling,  R.J., and  Grice,  K. (2010). Application of trace element and stable isotope signatures to determine the provenance of tea (Camellia sinensis) samples. Food Chem., 118, 921-926.
•  Soufleros,  E.,  Pissa,  I.,  Petridis,  D.,  Lygerakis,  M.,  Mermelas,  K.,  Boukouvalas,  G., and  Tsimitakis,  E. (2001). Instrumental analysis of volatile and other compounds of Greek kiwi wine; sensory evaluation and optimisation of its composition. Food Chem., 75, 487-500.
•  Stanimirova,  I.,  Üstün,  B.,  Cajka,  T.,  Riddelova,  K.,  Hajslova,  J.,  Buydens,  L., and  Walczak,  B. (2010). Tracing the geographical origin of honeys based on volatile compounds profiles assessment using pattern recognition techniques. Food Chem., 118, 171-176.
•  Steine,  C.,  Beaucousin,  F.,  Siv,  C., and  Peiffer,  G. (2001). Potential of semiconductor sensor arrays for the origin authentication of pure Valencia orange juices. J. Agric. Food Chem., 49, 3151-3160.
•  Tominaga,  Y. (1999). Comparative study of class data analysis with PCA-LDA, SIMCA, PLS, ANNs, and k-NN. Chemometr. Intell. Lab., 49, 105-115.
•  Torres,  J.D.,  Chiralt,  A., and  Escriche,  I. (2012). Development of volatile fraction of fresh cut osmotically treated mango during cold storage. Food Chem., 130, 921-927.
•  Vitale,  R.,  Bevilacqua,  M.,  Bucci,  R.,  Magrì,  A.D.,  Magrì,  A.L., and  Marini,  F. (2012). A rapid and non-invasive method for authenticating the origin of pistachio samples by nir spectroscopy and chemometrics. Chemometr. Intell. Lab., 121, 90-99.
•  Wan,  X.M.,  Stevenson,  R.J.,  Chen,  X.D., and  Melton,  L.D. (1999). Application of headspace solid-phase microextraction to volatile flavour profile development during storage and ripening of kiwifruit. Food Res. Int., 32, 175-183.
•  Xiong,  W.Z.,  Zhang,  M., and  Sun,  Z.G. (2010). Protection of the Geographical Indication Intellectual Property Rights of Xiangxi Kiwifruit in Hunan Province. Hunan Agric. Sci, 21, 150-153. (in Chinese)
•  Zhao,  H.Y.,  Guo,  B.L.,  Wei,  Y.M., and  Zhang,  B. (2012). Near infrared reflectance spectroscopy for determination of the geographical origin of wheat. Food Chem., 138, 1902-1907.
•  Zhao,  H.Y.,  Guo,  B.L.,  Wei,  Y.M.,  Zhang,  B.,  Sun,  S.M.,  Zhang,  L., and  Yan,  J.H. (2011). Determining the geographic origin of wheat using multielement analysis and multivariate statistics. J. Agric. Food Chem., 59, 4397-4402.