- Email: [email protected]
Discrimination of different types damage of rice plants by electronic nose
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/issn/15375110
Research Paper
Discrimination of different types damage of rice plants by electronic nose Bo Zhou, Jun Wang* Department of Bio-systems Engineering, Zhejiang University, 268 Kaixuan Road, Hangzhou 310029, PR China
article info
The profiles of volatile compounds emitted by plants varies in response to damage or
Article history:
herbivore attack. The potential of electronic nose technology to monitor such changes,
Received 4 August 2010
with the aim of diagnosing plant health was investigated. An electronic nose (E-nose) was
Received in revised form
used to analyse rice plants that were subjected to different types of treatments causing
29 December 2010
damage, and the results were compared to those of undamaged control plants. Principal
Accepted 8 March 2011
component analysis (PCA), linear discrimination analysis (LDA), cluster analysis (CA), back-
Published online 19 May 2011
propagation neural network (BPNN), and learning vector quantisation (LVQ) network were used to evaluate the E-nose data. The results indicated that the E-nose can successfully discriminate between rice plants with different types of damage. The discrimination was more pronounced after the LDA than after the PCA. The front 5 principal component values of the PCA were extracted and they acted as the input date for the neural network analyses. Good discrimination results were obtained using these front 5 principal component values in LVQ and BPNN. The results demonstrated that it is plausible to use E-nose technology as a method for monitoring rice cultivation practices. ª 2011 IAgrE. Published by Elsevier Ltd. All rights reserved.
1.
Introduction
During evolution, plants have evolved various ways to interact with their environment, including the release into the atmosphere of arrays of volatile compounds from their leaves, flowers, and fruits. The volatile compounds can carry information about the plants’ physiological status and the stress conditions they have been subjected to (Dudareva, Negre, Nagegowda, & Orlova, 2006). In the past 2 decades, it has become increasingly clear that, as a means of self-protection, the vegetative parts of plants produce blends of volatile compounds in response to damage and herbivore attack. There is considerable variation among rice species and varieties in the composition, quantity, and quality of the volatile blends (Lou et al., 2006). Plants respond to insect feeding damage by
releasing a variety of volatiles from the damaged site, and the profile of the emitted volatiles is significantly different from that of undamaged or mechanically damaged plants (DP) (Mccall, Turlings, Loughrin, Proveaux, & Tumlinson, 1994; Pare & Tumlinson, 1999). The profiles of plant volatile compounds produced in response to insect and disease attacks is so specific that they can potentially be used for noninvasive plant insect and disease monitoring in agricultural settings. There has been significant progress in plant volatile compound research as a result of the increased sensitivity of analytical instruments and of improvements in molecular and biochemical approaches in recent years. Sankaran, Mishra, Ehsani, and Davis (2010) reviewed the volatile profiling-based plant disease detection methods that are currently in use for monitoring the disease states of plants. GCeMS is a commonly
* Corresponding author. Tel.: þ86 571 86971881; fax: þ86 571 86971139. E-mail address: [email protected] (J. Wang). 1537-5110/$ e see front matter ª 2011 IAgrE. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.biosystemseng.2011.03.003
251
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
used technique for qualitative as well as quantitative analyses of volatile compounds released by plants (D’Alessandro & Turlings, 2006; Tholl et al., 2006). Kushalappa, Lui, Chen, and Lee (2002) inoculated potato tubers with Erwinia carotovora subsp. carotovora, E. carotovora subsp. atroseptica, Pythium ultimum, Phytophthora infestans, or Fusarium sambucinum, and analysed their volatile compounds profiles using GCeMS. Vuorinen, Nerg, Syrjala, Peltonen, and Holopainen (2007) utilised volatile compound emission pattern of silver birch to distinguish between healthy plants, plants that were damaged by larvae (herbivore arthropod Epirrita autumnata), and plants infected with pathogenic leaf spot (Marssonina betulae). However, the GCeMS method involves several stages, and the trapping of volatile compounds and sample preparation are time-consuming stages. In addition, the technology is very expensive, the equipment bulky, and it requires skilled staff to carry out the analysis. Therefore, GCeMS is unlikely to succeed as an easy-to-operate method for measurements in the field. Therefore, for crop health analysis, there is a need to develop new technologies that focus on monitoring changes in the volatile profiles. The concept of the electronic nose (E-nose) was developed a number of years ago, and it provides an easier and quicker method for measuring odours and flavours than those used so far. The principle is different from that of common methods for chemical analysis, such as gas and liquid chromatography, mass spectrometry, and spectrophotometry. The E-nose does not resolve the volatiles encountered in the sample into individual components. Instead, it responds to the whole set of volatiles with a unique digital pattern (Pathange, Mallikarjunan, Marini, O’Keefe, & Vaughan, 2006). E-nose technology has proven a useful tool in the food industry (Gomez, Wang, Hu, & Pereira, 2007; Parpinello et al., 2007; Wang, Wang, Zhou, & Lu, 2009), medicine (D’Amico et al., 2008), and environmental monitoring (Gomez et al., 2007; Pan, Yang, & DeBruyn, 2007; Parpinello et al., 2007; Sohn, Smith, Yoong, Leis, & Galvin, 2003; Wang et al., 2009). The application of E-nose systems for characterising plant damage is a relatively new domain.
Previous reports on the use of the E-nose for plant damage monitoring. Laothawornkitkul et al. (2008) addressed its potential for use in pest and disease control in cucumber, pepper, and tomato plants. Henderson, Khalilian, Han, Greene, and Degenhardt (2010) used a commercially available E-nose (Cyranose 320) for detecting stink bugs and damage induced by stink bugs in cotton. There is no published information on the use of E-nose technology for measuring damage in rice plants. Therefore, the development of efficient E-nose technology for detecting the presence of insects or differentiating between damaged and undamaged rice plants would be highly valued and is likely to be adopted by crop consultants. The objective of the present study was to determine the feasibility of detecting rice plant damage caused by the rice striped stem borer (SSB) Chilo suppressalis, the rice brown planthopper (BPH) Niaparvata lugens, and mechanical means and to compare damaged and undamaged control plants (CP) utilising an E-nose (PEN2). The aim was to develop discrimination models of E-nose signals for different types of damage. Principal component analysis (PCA), linear discrimination analysis (LDA), cluster analysis (CA), back-propagation neural network (BPNN) and learning vector quantisation (LVQ) network were used to verify the classification capacity of the proposed methods.
2.
Materials and methods
2.1.
Electronic nose
An E-nose device (PEN2), provided by WMA Airsense Analysentechnik GmbH (Schwerin, Germany), was used. The PEN2 system consists of a sampling apparatus, a detector unit containing the array of sensors, and pattern recognition software (WinMuster v.1.6) for data recording. The sensor array is composed of 10 different metal oxide sensors (MOS) positioned in a small chamber. Each sensor has a certain degree of affinity towards specific chemicals or volatile compounds. Table 1 lists
Table 1 e Sensors used and their main applications in PEN 2. Number in array
Sensor-name
General description
S1 S2
W1C W5S
S3 S4 S5 S6
W3C W6S W5C W1S
S7
W1W
S8
W2S
S9 S10
W2W W3S
Aromatic compounds Very sensitive, broad range sensitivity, react on nitrogene oxides, very sensitive with negative signal Ammonia, used as sensor for aromatic compounds Mainly hydrogen, selectively, (breath gases) Alkanes, aromatic compounds, less polar compounds Sensitive to methane (environment) ca. 10 ppm. Broad range, similar to No. 8 Reacts on sulphur compounds, H2S 0.1 ppm. Otherwise sensitive to many terpenes and sulphur organic compounds, which are important for smell, limonene, pyrazine Detects alcohol’s, partially aromatic compounds, broad range Aromatics compounds, sulphur organic compounds Reacts on high concentrations >100 ppm, sometime very selective (methane)
Reference Toluene, 10 ppm NO2, 1 ppm Propane, 1 ppm H2, 100 ppb Propane, 1 ppm CH3, 100 ppm H2S, 1 ppm
CO, 100 ppm H2S, 1 ppm CH3, 10 CH3, 100 ppm
252
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
all the sensors used and their main applications (Gomez et al., 2007). This table contains the currently known or specified reactions. The sensor response is expressed as resistance (U). The MOS sensors rely on changes in conductivity induced by the adsorption of gas phase molecules and on subsequent surface reactions. They consist of a ceramic substrates coated with a metal oxide semiconducting film and are heated by a wire resistor. Because of the high operating temperatures (200e500 C), the organic volatile compounds transferred to the surface of the sensors are totally combusted to carbon dioxide and water, leading to changes in the resistance. The high temperature avoids water interference and results in a rapid response and recovery time. The detection limit of hot sensors is in the range of 1 ppm.
2.2.
Rice plants
The rice variety used was Zhou 903. Pre-germinated seeds were sown in a greenhouse, and after 20 d, 200 seedlings were transplanted singly into common clay pots (80 mm diameter and 100 mm high). Plants were watered daily, and each pot was supplied with 0.10 g of urea 15 d and 25 d after transplanting. All plants were placed in a controlled climate room that was maintained at 28 2 C, 70e80% relative humidity (RH)., and 12-h photophase. After 30e35 days, 60 uniform plants were selected for the experiment.
2.3.
Insects
Overwintered larvae of the SSB, Chilo suppressalis, were collected from rice fields in Hangzhou, China. The larvae were reared on rice plants in 15 cm pots until pupation. After they emerged from the pupae, the moths were fed a 10% honey solution and placed on rice plants grown in a nylon mesh oviposition cage (1 1 1.5 m). The cages were placed in a greenhouse at 27 2 C with a photoperiod of 14 h light/10 h dark (L/D) and 70e80% RH. Egg masses were collected in the greenhouse and kept in glass tubes in a refrigerator. They were allowed to hatch as needed for the tests. The rice BPH, N. lugens, culture was originally obtained from the China National Rice Research Institute, Zhejiang, and maintained on Taichung Native 1 (TN1; susceptible to BPH) rice plants in a greenhouse. Late instar nymphs of BPH were captured in the greenhouse and reared on potted Zhou 903 rice plants, which were placed in cages (110 mm diameter 40 mm high), covered with a 0.1 mm mesh nylon cloth. Newly emerged adults of BPH were collected daily and cultured on potted fresh Zhou 903 rice plants. BPH adults at a uniform age of 2 d after emergence were used in the experiment.
2.4.
Experiment protocols
2.4.1.
Plant treatment
In this study, 4 treatments were tested: (1) damage caused by the rice SSB, (2) damage caused by the rice BPH, (3) mechanically damage, and (4) undamaged. The level of damage inflicted (e.g., the density of pests or number of mechanical pricks) was as described before (Lou et al., 2006; Lu, Wang, Lou, & Cheng, 2006). For the SSB treatment, 1 larva was used per rice plant. For the BPH treatment, 20 BPH adults were used per rice plant. For the
mechanical treatment, a single plant was punched 200 times with a needle. Preliminary experiments showed that after 2 h of infestation, differences in the plant volatile profiles were observed between treated and untreated plants; therefore, the experiments were conducted at this time point after infestation. Sixty plants were selected for the experiment, and there were 15 replicates for each treatment. For the SSB-treated plants (SP), pots with single plants were used. Plants were individually infested using 1 third-instar larva of SSB that had been starved for 2 h. The larvae were left to feed on the stems for 2 h before the E-nose analysis. Each of the BPH-treated plants (BP; 1 per pot) was individually infested with 20 BPH adults. The plant hoppers were contained in parafilm bags (40 mm 80 mm, with 60 small holes made by a needle) that were fixed to the plant stems in the lower position. Two hours after infestation, just before the analysis, the insects were removed from the plants. The mechanically DP (MP; 1 per pot) were individually damaged using a needle on the lower part of the rice stems (about 20 mm long). In preliminary experiments, treatments of 10, 50, 100, and 200 needle pricks per plant were tested. Using 200 pricks, it was found that the E-nose could sufficiently differentiate between damaged and undamaged rice plants. CP were not subjected to any treatment.
2.4.2.
E-nose measurement
Fig. 1 shows a schematic diagram of the PEN2 E-nose measurements during the experiments. Each rice plant was enclosed in a cylindrical stainless steel support (400 mm high 100 mm internal diameter (ID)) covered with a plastic bag over the top, such that the volume of headspace was 3.14 l. The plastic bag was a colourless and odourless food-grade polyethylene bag (0.08 mm thick). The plant was sealed at the lower stem by 2 hermetic boards with a hole in the middle. This excluded the odour coming from the medium. The plant was kept at room temperature (28 2 C) for 20 min before static headspace sampling was begun. Before each measurement, the E-nose system was cleaned with zero-air (air filtered on active carbon). The main purpose of this was to clean the circuit and to return the sensors to their baselines. During the measurement, the headspace gas of a plant was pumped into the sensor chamber at a constant rate of 200 ml min1 through a Teflon tube (3 mm) connected to a needle. When the headspace gas that had accumulated in the bag was pumped into the sensor chamber, the ratio of conductance of each sensor changed. The response of each sensor was expressed as a ratio of conductance (G/G0; G and G0 are the respective conductivities of the sensors when the plant gas and the zero gas pass over). The measurement procedure was controlled by a computer program. The measurement time was 65 s, which was sufficient for the sensors to reach stable values. The interval for data collection was 1 s. A computer recorded the response of the E-nose every second. The flush time was set to 40 s. When the measurement was completed, the acquired data was stored for later use.
2.5.
Data analysis
The number of data points per sample was 650: 1 point per second over 65 s using 10 sensors. However, a strong
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
253
Fig. 1 e Schematic diagram of the electronic nose measurements of PEN2.
correlation was obtained between the data from the same sensor but corresponding to different points. This allowed a reduction in the number of data per sample such that only 1 point per sensor was selected at the time for 60 s. Moreover, the data was subjected to different statistical analyses, including PCA, LDA, CA, BPNN and LVQ. The data were first analysed by PCA and LDA to reduce the dimensionality and visualisation of datasets. CA was employed to examine the sensorial data and to test the relationships between various rice plant groups. Finally, BPNN and LVQ were used to classify the rice plant samples using the first 5 principal components obtained by PCA as inputs. The analyses of the data were carried out using DPS version 3.11 (Data Processing System Statistical Software package), SAS v8 (SAS Institute, Cary, NC, USA), SPSS 11.5 (SPSS Inc., Chicago IL, USA) and MINITAB v14 (Pennsylvania state University, PA, USA).
3.
Results and discussion
3.1.
Electronic nose responses to rice plant volatiles
Fig. 2 shows the typical responses of 10 sensors during measurements on a rice plant. Each curve represents the change in conductivity of each sensor over time. Changes are due to electro-valve action when the volatiles from the rice plant reach the measurement chamber. It is clear that the ratio of conductance (G/G0) of each sensor was close to 1.0 at the initial period, then increased or decreased gradually, and stabilised after about 50 s. In our experiments, we used the sensor signals at 60 s for the analyses.
3.2.
Signal analysis
Fig. 3 shows the changes in the sensor signals in response to rice plants with different types of damage. Each point represents the mean value of each sensor’s response signal to rice plants, linked to the measurements of conductance increases or decreases experienced by the sensor. The E-nose sensor response changed for rice plants during 4 types of treatments. The values of the sensor response signals differed with the different types of damage. The variation in the sensor responses in the CP was small compared to the 3 groups of damaged rice plants (Fig. 3). This result is due to the specific
changes in the chemical composition in rice plants after the damage. Previous studies have shown that BPH attack alters the volatile profiles of rice plants (Lou et al., 2006). Plants damaged by herbivores emit higher levels and a larger variety of odorous hydrocarbons into the atmosphere compared with those emitted by undamaged plants (Pare & Tumlinson, 1999). BPH infestation significantly enhanced the release of ethylene during 2e24 h after infestation (Lu et al., 2006). The sensor responses to CP were more similar to those to mechanically BP than to those to herbivore-damaged plants (BP, SP). These results are in agreement with those obtained by Henderson et al. (2010) when testing cotton plants. There was a strong correlation between the number of stink bugs in a sample and the response of E-nose sensors.
3.3.
PCA and LDA analysis
PCA and LDA are 2 commonly used techniques for data classification and dimensionality reduction. The data from 60 samples (15 samples from each group) obtained by the E-nose for 4 rice groups were used for the PCA. Fig. 4 shows the results of the PCA. The first component contributed 80.8% of the total variance; the second component, 11.1%; and the third component, only 4.5%. This added up to 96.4% in total. In Fig. 4, clusters of data were divided into 3 groups (CP/MP, BP, and SP). Samples from groups BP and SP were easily discriminated, whereas samples from groups CP and MP overlapped partially. PCA did not clearly discriminate between these different types of rice plants. Therefore, LDA was used for further separation of the rice plant samples. Other methods (BPNN and LVQ; Section 3.5) are considered useful for developing classification models from the first 5 component scores of the PCA. The results of the LDA are shown in Fig. 5. Function1 (LD1) contributed 76.2% of the total variance; Function2 (LD2), 22.3%; and Function3 (LD3), only 1.5%, adding up to a total of 100%. In the LDA plot, 4 groups of rice plant samples could be clearly distinguished. The separation of the samples was better using LDA than using PCA.
3.4.
CA analysis
CA is an exploratory statistical technique that aims to identify natural groupings among individuals and to present these
254
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
Fig. 2 e Type response curves of E-nose sensors for rice (control plant) volatiles.
groupings in the form of a hierarchical tree or dendrogram. The dendrogram shows the similarities between individuals. In this paper, the similarities are based on the Euclidean distance. The clustering method used was the ‘minimum distance method’. The result was presented as a dendrogram (Fig. 6). Using a threshold of 0.09, similar results were achieved with PCA, showing that the plant samples from the 4 treatments were grouped into 3 clusters: cluster I (CP 1e15 and MP 1e15), cluster II (BP 1e15), cluster III (SP 1e15). This can be attributed to the marked difference in the profile of volatile compounds between undamaged plants and herbivoredamaged plants. Cluster I could be further separated into 3 sub-clusters at the threshold of 0.07. The first sub-cluster consisted of 11 rice plant samples of CP; the second sub-cluster, 15 rice plant samples of MP; and the third sub-cluster, 4 rice plant samples
of CP. The existence of 2 CP sub-clusters in cluster I indicates that the profile of volatile compounds of undamaged plants was more similar to that of mechanically DP than to that of herbivore-damaged plants. These results also indicate that the mechanically DP can be distinguished from the rest by CA.
3.5.
BPNN and LVQ
In this study, BPNN and LVQ were applied using features derived from the PCA (Section 3.3). Sixty samples (15 duplicates for each of the 4 groups) were divided into 2 groups: 40 samples (10 samples of each group) for the training set and 20 samples for the testing set. For the BPNN, the chosen architecture of the artificial neural network was a 5 11 4 three-layer back-propagation according to Kolmogorov’s theorem. The 5 input neurons
Fig. 3 e Response values of the sensors to rice plants with different types of damage.
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
255
Fig. 4 e PCA plot for rice plants with different types of damage. (CP, control plants; MP, mechanically damaged plants; BP, BPH-treated plants; SP, SSB-treated plants).
correspond to the first 5 component scores from the PCA, while the 4 outputs are the different types of damage. The training function is ‘traingda’ and the training epoch is 2000. The training parameters were chosen with a maximum epoch of 1000 and a goal of 0.01. The result is shown in Table 2. The correct rate of the training set was 100%. The correct rates of the testing set for the CP, MP, BP, and SP groups were 60%, 100%, 100%, and 80%, respectively. The LVQ network also had 5 and 4 neurons in the input and output layers, respectively. The number of neurons in the Kohonen layer was 10. The LVQ analysis showed an overall classification success of 100% for the training sets and 90% for
Fig. 6 e Dendrogram of cluster analysis. (CP, control plants; MP, mechanically damaged plants; BP, BPH-treated plants; SP, SSB-treated plants).
the testing sets. The result is shown in Table 2. The results indicate that it is possible to use E-nose signals to discriminate between rice plants with different types of damage.
3.6. Fig. 5 e LDA plot for rice plants with different types of damage. (CP, control plants; MP, mechanically damaged plants; BP, BPH-treated plants; SP, SSB-treated plants).
Correlation of analysis results
Our data shows that the E-nose can discriminate between different types of damage inflicted on rice plants. The
256
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
Table 2 e Results of BPNN and LVQ analyses of data from plants with different types of damage. Network style
BPNN LVQ
Correct rate of training set
100% 100%
Correct rate of testing set CP
MP
BP
SP
3(60%) 4(80%)
5(100%) 5(100%)
5(100%) 5(100%)
4(80%) 4(80%)
variation in sensor responses in the MP, BP, and SP groups was higher than that in the CP group, and the most obvious variation was in the SP group (Fig. 3). This suggests that the emission of volatiles by rice plants changes in response to different types of damage. Previously, significant differences were detected in volatile emissions from undamaged, mechanically damaged, and infested plants using GC and GCeMS methods (Xu et al., 2002). Plants infested by BPH emitted several volatiles (e.g., linalool, (3E)-4,8-dimethyl-1,3,7nonatriene, and indole) that were not detected in undamaged CP or mechanically DP. N. lugens infestation significantly enhanced the levels of salicylic acid (SA) at 1 and 8 h compared to the corresponding controls (Wang et al., 2008). Although the E-nose technique does not provide information about the identity of the different volatile compounds, the major advantage of this technique over the standard GCeMS measurements is the shorter analysis time (Saevels et al., 2004). The control group was more similar to the mechanically damaged group than to the herbivore-damaged groups in the results of the PCA and LDA (Figs. 4 and 5). Similar results can be obtained when using CA (Fig. 6). The BPNN and LVQ provided corresponding classification results (Table 2).
4.
Conclusions
To sum up, the results prove that the E-nose PEN 2 can successfully distinguish between rice plants with different types of damage. The LDA discriminates more effectively between the different groups than the PCA. The results of the CA show obvious differentiation between the rice plant samples with different types of damage. In particular, the herbivore-damaged plants (BP, SP) are easily distinguished from the CP. The front 5 principal component values of the PCA were extracted and acted as the input of the neural network analysis. Good discrimination results are obtained using these front 5 principal component values in LVQ and BPNN.
Acknowledgements The authors acknowledge the financial support of the Chinese National Foundation of Nature and Science through Project 30771246 and 31071548, the National High Technology Research and Development Program of China through Project 2006AA10Z212, the funded by Zhejiang Provincial Natural Science Foundation Z5100155 and the supported by Science Foundation of Chinese University.
references
D’Alessandro, M., & Turlings, T. C. J. (2006). Advances and challenges in the identification of volatiles that mediate interactions among plants and arthropods. Analyst, 131, 24e32. D’Amico, A., Di Natale, C., Paolesse, R., Macagnano, A., Martinelli, E., Pennazza, G., et al. (2008). Olfactory systems for medical applications. Sensors and Actuators B-Chemical, 130, 458e465. Dudareva, N., Negre, F., Nagegowda, D. A., & Orlova, I. (2006). Plant volatiles: recent advances and future perspectives. Critical Reviews in Plant Science, 25, 417e440. Gomez, A. H., Wang, J., Hu, G. X., & Pereira, A. G. (2007). Discrimination of storage shelf-life for mandarin by electronic nose technique. Lwt-Food Science and Technology, 40, 681e689. Henderson, W. G., Khalilian, A., Han, Y. J., Greene, J. K., & Degenhardt, D. C. (2010). Detecting stink bugs/damage in cotton utilizing a portable electronic nose. Computers and Electronics in Agriculture, 70, 157e162. Kushalappa, A. C., Lui, L. H., Chen, C. R., & Lee, B. (2002). Volatile fingerprinting (SPME-GC-FID) to detect and discriminate diseases of potato tubers. Plant Disease, 86, 131e137. Laothawornkitkul, J., Moore, J. P., Taylor, J. E., Possell, M., Gibson, T. D., Hewitt, C. N., et al. (2008). Discrimination of plant volatile signatures by an electronic nose: a potential technology for plant pest and disease monitoring. Environmental Science & Technology, 42, 8433e8439. Lou, Y. G., Hua, X. Y., Turlings, T. C. J., Cheng, J. A., Chen, X. X., & Ye, G. Y. (2006). Differences in induced volatile emissions among rice varieties result in differential attraction and parasitism of Nilaparvata lugens eggs by the parasitoid Anagrus nilaparvatae in the field. Journal of Chemical Ecology, 32, 2375e2387. Lu, Y. J., Wang, X., Lou, Y. G., & Cheng, J. A. (2006). Role of ethylene signaling in the production of rice volatiles induced by the rice brown planthopper Nilaparvata lugens. Chinese Science Bulletin, 51, 2457e2465. Mccall, P. J., Turlings, T. C. J., Loughrin, J., Proveaux, A. T., & Tumlinson, J. H. (1994). Herbivore-induced volatile emissions from cotton (Gossypium-Hirsutum L) seedlings. Journal of Chemical Ecology, 20, 3039e3050. Pan, L., Yang, S. X., & DeBruyn, J. (2007). Factor analysis of downwind odours from livestock farms. Biosystems Engineering, 96, 387e397. Pare, P. W., & Tumlinson, J. H. (1999). Plant volatiles as a defense against insect herbivores. Plant Physiology, 121, 325e331. Parpinello, G. P., Fabbri, A., Domenichelli, S., Mesisca, V., Cavicchi, L., & Versari, A. (2007). Discrimination of apricot cultivars by gas multisensor array using an artificial neural network. Biosystems Engineering, 97, 371e378. Pathange, L. P., Mallikarjunan, P., Marini, R. P., O’Keefe, S., & Vaughan, D. (2006). Non-destructive evaluation of apple maturity using an electronic nose system. Journal of Food Engineering, 77, 1018e1023. Saevels, S., Lammertyn, J., Berna, A. Z., Veraverbeke, E. A., Di Natale, C., & Nicolai, B. M. (2004). An electronic nose and a mass spectrometry-based electronic nose for assessing apple quality during shelf life. Postharvest Biology and Technology, 31, 9e19. Sankaran, S., Mishra, A., Ehsani, R., & Davis, C. (2010). A review of advanced techniques for detecting plant diseases. Computers and Electronics in Agriculture, 72, 1e13. Sohn, J. H., Smith, R., Yoong, E., Leis, J., & Galvin, G. (2003). Quantification of odours from piggery effluent ponds using an electronic nose and an artificial neural network. Biosystems Engineering, 86, 399e410.
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
Tholl, D., Boland, W., Hansel, A., Loreto, F., Rose, U. S. R., & Schnitzler, J. P. (2006). Practical approaches to plant volatile analysis. Plant Journal, 45, 540e560. Vuorinen, T., Nerg, A. M., Syrjala, L., Peltonen, P., & Holopainen, J. K. (2007). Epirrita autumnata induced VOC emission of silver birch differ from emission induced by leaf fungal pathogen. Arthropod-Plant Interactions, 1, 159e165. Wang, X., Zhou, G. X., Xiang, C. Y., Du, M. H., Cheng, J. A., Liu, S. S., et al. (2008). beta-Glucosidase treatment and infestation by the rice
257
brown planthopper Nilaparvata lugens elicit similar signaling pathways in rice plants. Chinese Science Bulletin, 53, 53e57. Wang, Y. W., Wang, J., Zhou, B., & Lu, Q. J. (2009). Monitoring storage time and quality attribute of egg based on electronic nose. Analytica Chimica Acta, 650, 183e188. Xu, T., Zhou, Q., Xia, Q., Zhang, W. Q., Zhang, G., & Gu, D. X. (2002). Effects of herbivore-induced rice volatiles on the host selection behavior of brown planthopper, Nilaparvata lugens. Chinese Science Bulletin, 47, 1355e1360.
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/issn/15375110
Research Paper
Discrimination of different types damage of rice plants by electronic nose Bo Zhou, Jun Wang* Department of Bio-systems Engineering, Zhejiang University, 268 Kaixuan Road, Hangzhou 310029, PR China
article info
The profiles of volatile compounds emitted by plants varies in response to damage or
Article history:
herbivore attack. The potential of electronic nose technology to monitor such changes,
Received 4 August 2010
with the aim of diagnosing plant health was investigated. An electronic nose (E-nose) was
Received in revised form
used to analyse rice plants that were subjected to different types of treatments causing
29 December 2010
damage, and the results were compared to those of undamaged control plants. Principal
Accepted 8 March 2011
component analysis (PCA), linear discrimination analysis (LDA), cluster analysis (CA), back-
Published online 19 May 2011
propagation neural network (BPNN), and learning vector quantisation (LVQ) network were used to evaluate the E-nose data. The results indicated that the E-nose can successfully discriminate between rice plants with different types of damage. The discrimination was more pronounced after the LDA than after the PCA. The front 5 principal component values of the PCA were extracted and they acted as the input date for the neural network analyses. Good discrimination results were obtained using these front 5 principal component values in LVQ and BPNN. The results demonstrated that it is plausible to use E-nose technology as a method for monitoring rice cultivation practices. ª 2011 IAgrE. Published by Elsevier Ltd. All rights reserved.
1.
Introduction
During evolution, plants have evolved various ways to interact with their environment, including the release into the atmosphere of arrays of volatile compounds from their leaves, flowers, and fruits. The volatile compounds can carry information about the plants’ physiological status and the stress conditions they have been subjected to (Dudareva, Negre, Nagegowda, & Orlova, 2006). In the past 2 decades, it has become increasingly clear that, as a means of self-protection, the vegetative parts of plants produce blends of volatile compounds in response to damage and herbivore attack. There is considerable variation among rice species and varieties in the composition, quantity, and quality of the volatile blends (Lou et al., 2006). Plants respond to insect feeding damage by
releasing a variety of volatiles from the damaged site, and the profile of the emitted volatiles is significantly different from that of undamaged or mechanically damaged plants (DP) (Mccall, Turlings, Loughrin, Proveaux, & Tumlinson, 1994; Pare & Tumlinson, 1999). The profiles of plant volatile compounds produced in response to insect and disease attacks is so specific that they can potentially be used for noninvasive plant insect and disease monitoring in agricultural settings. There has been significant progress in plant volatile compound research as a result of the increased sensitivity of analytical instruments and of improvements in molecular and biochemical approaches in recent years. Sankaran, Mishra, Ehsani, and Davis (2010) reviewed the volatile profiling-based plant disease detection methods that are currently in use for monitoring the disease states of plants. GCeMS is a commonly
* Corresponding author. Tel.: þ86 571 86971881; fax: þ86 571 86971139. E-mail address: [email protected] (J. Wang). 1537-5110/$ e see front matter ª 2011 IAgrE. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.biosystemseng.2011.03.003
251
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
used technique for qualitative as well as quantitative analyses of volatile compounds released by plants (D’Alessandro & Turlings, 2006; Tholl et al., 2006). Kushalappa, Lui, Chen, and Lee (2002) inoculated potato tubers with Erwinia carotovora subsp. carotovora, E. carotovora subsp. atroseptica, Pythium ultimum, Phytophthora infestans, or Fusarium sambucinum, and analysed their volatile compounds profiles using GCeMS. Vuorinen, Nerg, Syrjala, Peltonen, and Holopainen (2007) utilised volatile compound emission pattern of silver birch to distinguish between healthy plants, plants that were damaged by larvae (herbivore arthropod Epirrita autumnata), and plants infected with pathogenic leaf spot (Marssonina betulae). However, the GCeMS method involves several stages, and the trapping of volatile compounds and sample preparation are time-consuming stages. In addition, the technology is very expensive, the equipment bulky, and it requires skilled staff to carry out the analysis. Therefore, GCeMS is unlikely to succeed as an easy-to-operate method for measurements in the field. Therefore, for crop health analysis, there is a need to develop new technologies that focus on monitoring changes in the volatile profiles. The concept of the electronic nose (E-nose) was developed a number of years ago, and it provides an easier and quicker method for measuring odours and flavours than those used so far. The principle is different from that of common methods for chemical analysis, such as gas and liquid chromatography, mass spectrometry, and spectrophotometry. The E-nose does not resolve the volatiles encountered in the sample into individual components. Instead, it responds to the whole set of volatiles with a unique digital pattern (Pathange, Mallikarjunan, Marini, O’Keefe, & Vaughan, 2006). E-nose technology has proven a useful tool in the food industry (Gomez, Wang, Hu, & Pereira, 2007; Parpinello et al., 2007; Wang, Wang, Zhou, & Lu, 2009), medicine (D’Amico et al., 2008), and environmental monitoring (Gomez et al., 2007; Pan, Yang, & DeBruyn, 2007; Parpinello et al., 2007; Sohn, Smith, Yoong, Leis, & Galvin, 2003; Wang et al., 2009). The application of E-nose systems for characterising plant damage is a relatively new domain.
Previous reports on the use of the E-nose for plant damage monitoring. Laothawornkitkul et al. (2008) addressed its potential for use in pest and disease control in cucumber, pepper, and tomato plants. Henderson, Khalilian, Han, Greene, and Degenhardt (2010) used a commercially available E-nose (Cyranose 320) for detecting stink bugs and damage induced by stink bugs in cotton. There is no published information on the use of E-nose technology for measuring damage in rice plants. Therefore, the development of efficient E-nose technology for detecting the presence of insects or differentiating between damaged and undamaged rice plants would be highly valued and is likely to be adopted by crop consultants. The objective of the present study was to determine the feasibility of detecting rice plant damage caused by the rice striped stem borer (SSB) Chilo suppressalis, the rice brown planthopper (BPH) Niaparvata lugens, and mechanical means and to compare damaged and undamaged control plants (CP) utilising an E-nose (PEN2). The aim was to develop discrimination models of E-nose signals for different types of damage. Principal component analysis (PCA), linear discrimination analysis (LDA), cluster analysis (CA), back-propagation neural network (BPNN) and learning vector quantisation (LVQ) network were used to verify the classification capacity of the proposed methods.
2.
Materials and methods
2.1.
Electronic nose
An E-nose device (PEN2), provided by WMA Airsense Analysentechnik GmbH (Schwerin, Germany), was used. The PEN2 system consists of a sampling apparatus, a detector unit containing the array of sensors, and pattern recognition software (WinMuster v.1.6) for data recording. The sensor array is composed of 10 different metal oxide sensors (MOS) positioned in a small chamber. Each sensor has a certain degree of affinity towards specific chemicals or volatile compounds. Table 1 lists
Table 1 e Sensors used and their main applications in PEN 2. Number in array
Sensor-name
General description
S1 S2
W1C W5S
S3 S4 S5 S6
W3C W6S W5C W1S
S7
W1W
S8
W2S
S9 S10
W2W W3S
Aromatic compounds Very sensitive, broad range sensitivity, react on nitrogene oxides, very sensitive with negative signal Ammonia, used as sensor for aromatic compounds Mainly hydrogen, selectively, (breath gases) Alkanes, aromatic compounds, less polar compounds Sensitive to methane (environment) ca. 10 ppm. Broad range, similar to No. 8 Reacts on sulphur compounds, H2S 0.1 ppm. Otherwise sensitive to many terpenes and sulphur organic compounds, which are important for smell, limonene, pyrazine Detects alcohol’s, partially aromatic compounds, broad range Aromatics compounds, sulphur organic compounds Reacts on high concentrations >100 ppm, sometime very selective (methane)
Reference Toluene, 10 ppm NO2, 1 ppm Propane, 1 ppm H2, 100 ppb Propane, 1 ppm CH3, 100 ppm H2S, 1 ppm
CO, 100 ppm H2S, 1 ppm CH3, 10 CH3, 100 ppm
252
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
all the sensors used and their main applications (Gomez et al., 2007). This table contains the currently known or specified reactions. The sensor response is expressed as resistance (U). The MOS sensors rely on changes in conductivity induced by the adsorption of gas phase molecules and on subsequent surface reactions. They consist of a ceramic substrates coated with a metal oxide semiconducting film and are heated by a wire resistor. Because of the high operating temperatures (200e500 C), the organic volatile compounds transferred to the surface of the sensors are totally combusted to carbon dioxide and water, leading to changes in the resistance. The high temperature avoids water interference and results in a rapid response and recovery time. The detection limit of hot sensors is in the range of 1 ppm.
2.2.
Rice plants
The rice variety used was Zhou 903. Pre-germinated seeds were sown in a greenhouse, and after 20 d, 200 seedlings were transplanted singly into common clay pots (80 mm diameter and 100 mm high). Plants were watered daily, and each pot was supplied with 0.10 g of urea 15 d and 25 d after transplanting. All plants were placed in a controlled climate room that was maintained at 28 2 C, 70e80% relative humidity (RH)., and 12-h photophase. After 30e35 days, 60 uniform plants were selected for the experiment.
2.3.
Insects
Overwintered larvae of the SSB, Chilo suppressalis, were collected from rice fields in Hangzhou, China. The larvae were reared on rice plants in 15 cm pots until pupation. After they emerged from the pupae, the moths were fed a 10% honey solution and placed on rice plants grown in a nylon mesh oviposition cage (1 1 1.5 m). The cages were placed in a greenhouse at 27 2 C with a photoperiod of 14 h light/10 h dark (L/D) and 70e80% RH. Egg masses were collected in the greenhouse and kept in glass tubes in a refrigerator. They were allowed to hatch as needed for the tests. The rice BPH, N. lugens, culture was originally obtained from the China National Rice Research Institute, Zhejiang, and maintained on Taichung Native 1 (TN1; susceptible to BPH) rice plants in a greenhouse. Late instar nymphs of BPH were captured in the greenhouse and reared on potted Zhou 903 rice plants, which were placed in cages (110 mm diameter 40 mm high), covered with a 0.1 mm mesh nylon cloth. Newly emerged adults of BPH were collected daily and cultured on potted fresh Zhou 903 rice plants. BPH adults at a uniform age of 2 d after emergence were used in the experiment.
2.4.
Experiment protocols
2.4.1.
Plant treatment
In this study, 4 treatments were tested: (1) damage caused by the rice SSB, (2) damage caused by the rice BPH, (3) mechanically damage, and (4) undamaged. The level of damage inflicted (e.g., the density of pests or number of mechanical pricks) was as described before (Lou et al., 2006; Lu, Wang, Lou, & Cheng, 2006). For the SSB treatment, 1 larva was used per rice plant. For the BPH treatment, 20 BPH adults were used per rice plant. For the
mechanical treatment, a single plant was punched 200 times with a needle. Preliminary experiments showed that after 2 h of infestation, differences in the plant volatile profiles were observed between treated and untreated plants; therefore, the experiments were conducted at this time point after infestation. Sixty plants were selected for the experiment, and there were 15 replicates for each treatment. For the SSB-treated plants (SP), pots with single plants were used. Plants were individually infested using 1 third-instar larva of SSB that had been starved for 2 h. The larvae were left to feed on the stems for 2 h before the E-nose analysis. Each of the BPH-treated plants (BP; 1 per pot) was individually infested with 20 BPH adults. The plant hoppers were contained in parafilm bags (40 mm 80 mm, with 60 small holes made by a needle) that were fixed to the plant stems in the lower position. Two hours after infestation, just before the analysis, the insects were removed from the plants. The mechanically DP (MP; 1 per pot) were individually damaged using a needle on the lower part of the rice stems (about 20 mm long). In preliminary experiments, treatments of 10, 50, 100, and 200 needle pricks per plant were tested. Using 200 pricks, it was found that the E-nose could sufficiently differentiate between damaged and undamaged rice plants. CP were not subjected to any treatment.
2.4.2.
E-nose measurement
Fig. 1 shows a schematic diagram of the PEN2 E-nose measurements during the experiments. Each rice plant was enclosed in a cylindrical stainless steel support (400 mm high 100 mm internal diameter (ID)) covered with a plastic bag over the top, such that the volume of headspace was 3.14 l. The plastic bag was a colourless and odourless food-grade polyethylene bag (0.08 mm thick). The plant was sealed at the lower stem by 2 hermetic boards with a hole in the middle. This excluded the odour coming from the medium. The plant was kept at room temperature (28 2 C) for 20 min before static headspace sampling was begun. Before each measurement, the E-nose system was cleaned with zero-air (air filtered on active carbon). The main purpose of this was to clean the circuit and to return the sensors to their baselines. During the measurement, the headspace gas of a plant was pumped into the sensor chamber at a constant rate of 200 ml min1 through a Teflon tube (3 mm) connected to a needle. When the headspace gas that had accumulated in the bag was pumped into the sensor chamber, the ratio of conductance of each sensor changed. The response of each sensor was expressed as a ratio of conductance (G/G0; G and G0 are the respective conductivities of the sensors when the plant gas and the zero gas pass over). The measurement procedure was controlled by a computer program. The measurement time was 65 s, which was sufficient for the sensors to reach stable values. The interval for data collection was 1 s. A computer recorded the response of the E-nose every second. The flush time was set to 40 s. When the measurement was completed, the acquired data was stored for later use.
2.5.
Data analysis
The number of data points per sample was 650: 1 point per second over 65 s using 10 sensors. However, a strong
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
253
Fig. 1 e Schematic diagram of the electronic nose measurements of PEN2.
correlation was obtained between the data from the same sensor but corresponding to different points. This allowed a reduction in the number of data per sample such that only 1 point per sensor was selected at the time for 60 s. Moreover, the data was subjected to different statistical analyses, including PCA, LDA, CA, BPNN and LVQ. The data were first analysed by PCA and LDA to reduce the dimensionality and visualisation of datasets. CA was employed to examine the sensorial data and to test the relationships between various rice plant groups. Finally, BPNN and LVQ were used to classify the rice plant samples using the first 5 principal components obtained by PCA as inputs. The analyses of the data were carried out using DPS version 3.11 (Data Processing System Statistical Software package), SAS v8 (SAS Institute, Cary, NC, USA), SPSS 11.5 (SPSS Inc., Chicago IL, USA) and MINITAB v14 (Pennsylvania state University, PA, USA).
3.
Results and discussion
3.1.
Electronic nose responses to rice plant volatiles
Fig. 2 shows the typical responses of 10 sensors during measurements on a rice plant. Each curve represents the change in conductivity of each sensor over time. Changes are due to electro-valve action when the volatiles from the rice plant reach the measurement chamber. It is clear that the ratio of conductance (G/G0) of each sensor was close to 1.0 at the initial period, then increased or decreased gradually, and stabilised after about 50 s. In our experiments, we used the sensor signals at 60 s for the analyses.
3.2.
Signal analysis
Fig. 3 shows the changes in the sensor signals in response to rice plants with different types of damage. Each point represents the mean value of each sensor’s response signal to rice plants, linked to the measurements of conductance increases or decreases experienced by the sensor. The E-nose sensor response changed for rice plants during 4 types of treatments. The values of the sensor response signals differed with the different types of damage. The variation in the sensor responses in the CP was small compared to the 3 groups of damaged rice plants (Fig. 3). This result is due to the specific
changes in the chemical composition in rice plants after the damage. Previous studies have shown that BPH attack alters the volatile profiles of rice plants (Lou et al., 2006). Plants damaged by herbivores emit higher levels and a larger variety of odorous hydrocarbons into the atmosphere compared with those emitted by undamaged plants (Pare & Tumlinson, 1999). BPH infestation significantly enhanced the release of ethylene during 2e24 h after infestation (Lu et al., 2006). The sensor responses to CP were more similar to those to mechanically BP than to those to herbivore-damaged plants (BP, SP). These results are in agreement with those obtained by Henderson et al. (2010) when testing cotton plants. There was a strong correlation between the number of stink bugs in a sample and the response of E-nose sensors.
3.3.
PCA and LDA analysis
PCA and LDA are 2 commonly used techniques for data classification and dimensionality reduction. The data from 60 samples (15 samples from each group) obtained by the E-nose for 4 rice groups were used for the PCA. Fig. 4 shows the results of the PCA. The first component contributed 80.8% of the total variance; the second component, 11.1%; and the third component, only 4.5%. This added up to 96.4% in total. In Fig. 4, clusters of data were divided into 3 groups (CP/MP, BP, and SP). Samples from groups BP and SP were easily discriminated, whereas samples from groups CP and MP overlapped partially. PCA did not clearly discriminate between these different types of rice plants. Therefore, LDA was used for further separation of the rice plant samples. Other methods (BPNN and LVQ; Section 3.5) are considered useful for developing classification models from the first 5 component scores of the PCA. The results of the LDA are shown in Fig. 5. Function1 (LD1) contributed 76.2% of the total variance; Function2 (LD2), 22.3%; and Function3 (LD3), only 1.5%, adding up to a total of 100%. In the LDA plot, 4 groups of rice plant samples could be clearly distinguished. The separation of the samples was better using LDA than using PCA.
3.4.
CA analysis
CA is an exploratory statistical technique that aims to identify natural groupings among individuals and to present these
254
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
Fig. 2 e Type response curves of E-nose sensors for rice (control plant) volatiles.
groupings in the form of a hierarchical tree or dendrogram. The dendrogram shows the similarities between individuals. In this paper, the similarities are based on the Euclidean distance. The clustering method used was the ‘minimum distance method’. The result was presented as a dendrogram (Fig. 6). Using a threshold of 0.09, similar results were achieved with PCA, showing that the plant samples from the 4 treatments were grouped into 3 clusters: cluster I (CP 1e15 and MP 1e15), cluster II (BP 1e15), cluster III (SP 1e15). This can be attributed to the marked difference in the profile of volatile compounds between undamaged plants and herbivoredamaged plants. Cluster I could be further separated into 3 sub-clusters at the threshold of 0.07. The first sub-cluster consisted of 11 rice plant samples of CP; the second sub-cluster, 15 rice plant samples of MP; and the third sub-cluster, 4 rice plant samples
of CP. The existence of 2 CP sub-clusters in cluster I indicates that the profile of volatile compounds of undamaged plants was more similar to that of mechanically DP than to that of herbivore-damaged plants. These results also indicate that the mechanically DP can be distinguished from the rest by CA.
3.5.
BPNN and LVQ
In this study, BPNN and LVQ were applied using features derived from the PCA (Section 3.3). Sixty samples (15 duplicates for each of the 4 groups) were divided into 2 groups: 40 samples (10 samples of each group) for the training set and 20 samples for the testing set. For the BPNN, the chosen architecture of the artificial neural network was a 5 11 4 three-layer back-propagation according to Kolmogorov’s theorem. The 5 input neurons
Fig. 3 e Response values of the sensors to rice plants with different types of damage.
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
255
Fig. 4 e PCA plot for rice plants with different types of damage. (CP, control plants; MP, mechanically damaged plants; BP, BPH-treated plants; SP, SSB-treated plants).
correspond to the first 5 component scores from the PCA, while the 4 outputs are the different types of damage. The training function is ‘traingda’ and the training epoch is 2000. The training parameters were chosen with a maximum epoch of 1000 and a goal of 0.01. The result is shown in Table 2. The correct rate of the training set was 100%. The correct rates of the testing set for the CP, MP, BP, and SP groups were 60%, 100%, 100%, and 80%, respectively. The LVQ network also had 5 and 4 neurons in the input and output layers, respectively. The number of neurons in the Kohonen layer was 10. The LVQ analysis showed an overall classification success of 100% for the training sets and 90% for
Fig. 6 e Dendrogram of cluster analysis. (CP, control plants; MP, mechanically damaged plants; BP, BPH-treated plants; SP, SSB-treated plants).
the testing sets. The result is shown in Table 2. The results indicate that it is possible to use E-nose signals to discriminate between rice plants with different types of damage.
3.6. Fig. 5 e LDA plot for rice plants with different types of damage. (CP, control plants; MP, mechanically damaged plants; BP, BPH-treated plants; SP, SSB-treated plants).
Correlation of analysis results
Our data shows that the E-nose can discriminate between different types of damage inflicted on rice plants. The
256
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
Table 2 e Results of BPNN and LVQ analyses of data from plants with different types of damage. Network style
BPNN LVQ
Correct rate of training set
100% 100%
Correct rate of testing set CP
MP
BP
SP
3(60%) 4(80%)
5(100%) 5(100%)
5(100%) 5(100%)
4(80%) 4(80%)
variation in sensor responses in the MP, BP, and SP groups was higher than that in the CP group, and the most obvious variation was in the SP group (Fig. 3). This suggests that the emission of volatiles by rice plants changes in response to different types of damage. Previously, significant differences were detected in volatile emissions from undamaged, mechanically damaged, and infested plants using GC and GCeMS methods (Xu et al., 2002). Plants infested by BPH emitted several volatiles (e.g., linalool, (3E)-4,8-dimethyl-1,3,7nonatriene, and indole) that were not detected in undamaged CP or mechanically DP. N. lugens infestation significantly enhanced the levels of salicylic acid (SA) at 1 and 8 h compared to the corresponding controls (Wang et al., 2008). Although the E-nose technique does not provide information about the identity of the different volatile compounds, the major advantage of this technique over the standard GCeMS measurements is the shorter analysis time (Saevels et al., 2004). The control group was more similar to the mechanically damaged group than to the herbivore-damaged groups in the results of the PCA and LDA (Figs. 4 and 5). Similar results can be obtained when using CA (Fig. 6). The BPNN and LVQ provided corresponding classification results (Table 2).
4.
Conclusions
To sum up, the results prove that the E-nose PEN 2 can successfully distinguish between rice plants with different types of damage. The LDA discriminates more effectively between the different groups than the PCA. The results of the CA show obvious differentiation between the rice plant samples with different types of damage. In particular, the herbivore-damaged plants (BP, SP) are easily distinguished from the CP. The front 5 principal component values of the PCA were extracted and acted as the input of the neural network analysis. Good discrimination results are obtained using these front 5 principal component values in LVQ and BPNN.
Acknowledgements The authors acknowledge the financial support of the Chinese National Foundation of Nature and Science through Project 30771246 and 31071548, the National High Technology Research and Development Program of China through Project 2006AA10Z212, the funded by Zhejiang Provincial Natural Science Foundation Z5100155 and the supported by Science Foundation of Chinese University.
references
D’Alessandro, M., & Turlings, T. C. J. (2006). Advances and challenges in the identification of volatiles that mediate interactions among plants and arthropods. Analyst, 131, 24e32. D’Amico, A., Di Natale, C., Paolesse, R., Macagnano, A., Martinelli, E., Pennazza, G., et al. (2008). Olfactory systems for medical applications. Sensors and Actuators B-Chemical, 130, 458e465. Dudareva, N., Negre, F., Nagegowda, D. A., & Orlova, I. (2006). Plant volatiles: recent advances and future perspectives. Critical Reviews in Plant Science, 25, 417e440. Gomez, A. H., Wang, J., Hu, G. X., & Pereira, A. G. (2007). Discrimination of storage shelf-life for mandarin by electronic nose technique. Lwt-Food Science and Technology, 40, 681e689. Henderson, W. G., Khalilian, A., Han, Y. J., Greene, J. K., & Degenhardt, D. C. (2010). Detecting stink bugs/damage in cotton utilizing a portable electronic nose. Computers and Electronics in Agriculture, 70, 157e162. Kushalappa, A. C., Lui, L. H., Chen, C. R., & Lee, B. (2002). Volatile fingerprinting (SPME-GC-FID) to detect and discriminate diseases of potato tubers. Plant Disease, 86, 131e137. Laothawornkitkul, J., Moore, J. P., Taylor, J. E., Possell, M., Gibson, T. D., Hewitt, C. N., et al. (2008). Discrimination of plant volatile signatures by an electronic nose: a potential technology for plant pest and disease monitoring. Environmental Science & Technology, 42, 8433e8439. Lou, Y. G., Hua, X. Y., Turlings, T. C. J., Cheng, J. A., Chen, X. X., & Ye, G. Y. (2006). Differences in induced volatile emissions among rice varieties result in differential attraction and parasitism of Nilaparvata lugens eggs by the parasitoid Anagrus nilaparvatae in the field. Journal of Chemical Ecology, 32, 2375e2387. Lu, Y. J., Wang, X., Lou, Y. G., & Cheng, J. A. (2006). Role of ethylene signaling in the production of rice volatiles induced by the rice brown planthopper Nilaparvata lugens. Chinese Science Bulletin, 51, 2457e2465. Mccall, P. J., Turlings, T. C. J., Loughrin, J., Proveaux, A. T., & Tumlinson, J. H. (1994). Herbivore-induced volatile emissions from cotton (Gossypium-Hirsutum L) seedlings. Journal of Chemical Ecology, 20, 3039e3050. Pan, L., Yang, S. X., & DeBruyn, J. (2007). Factor analysis of downwind odours from livestock farms. Biosystems Engineering, 96, 387e397. Pare, P. W., & Tumlinson, J. H. (1999). Plant volatiles as a defense against insect herbivores. Plant Physiology, 121, 325e331. Parpinello, G. P., Fabbri, A., Domenichelli, S., Mesisca, V., Cavicchi, L., & Versari, A. (2007). Discrimination of apricot cultivars by gas multisensor array using an artificial neural network. Biosystems Engineering, 97, 371e378. Pathange, L. P., Mallikarjunan, P., Marini, R. P., O’Keefe, S., & Vaughan, D. (2006). Non-destructive evaluation of apple maturity using an electronic nose system. Journal of Food Engineering, 77, 1018e1023. Saevels, S., Lammertyn, J., Berna, A. Z., Veraverbeke, E. A., Di Natale, C., & Nicolai, B. M. (2004). An electronic nose and a mass spectrometry-based electronic nose for assessing apple quality during shelf life. Postharvest Biology and Technology, 31, 9e19. Sankaran, S., Mishra, A., Ehsani, R., & Davis, C. (2010). A review of advanced techniques for detecting plant diseases. Computers and Electronics in Agriculture, 72, 1e13. Sohn, J. H., Smith, R., Yoong, E., Leis, J., & Galvin, G. (2003). Quantification of odours from piggery effluent ponds using an electronic nose and an artificial neural network. Biosystems Engineering, 86, 399e410.
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 2 5 0 e2 5 7
Tholl, D., Boland, W., Hansel, A., Loreto, F., Rose, U. S. R., & Schnitzler, J. P. (2006). Practical approaches to plant volatile analysis. Plant Journal, 45, 540e560. Vuorinen, T., Nerg, A. M., Syrjala, L., Peltonen, P., & Holopainen, J. K. (2007). Epirrita autumnata induced VOC emission of silver birch differ from emission induced by leaf fungal pathogen. Arthropod-Plant Interactions, 1, 159e165. Wang, X., Zhou, G. X., Xiang, C. Y., Du, M. H., Cheng, J. A., Liu, S. S., et al. (2008). beta-Glucosidase treatment and infestation by the rice
257
brown planthopper Nilaparvata lugens elicit similar signaling pathways in rice plants. Chinese Science Bulletin, 53, 53e57. Wang, Y. W., Wang, J., Zhou, B., & Lu, Q. J. (2009). Monitoring storage time and quality attribute of egg based on electronic nose. Analytica Chimica Acta, 650, 183e188. Xu, T., Zhou, Q., Xia, Q., Zhang, W. Q., Zhang, G., & Gu, D. X. (2002). Effects of herbivore-induced rice volatiles on the host selection behavior of brown planthopper, Nilaparvata lugens. Chinese Science Bulletin, 47, 1355e1360.