As a rule, wood species identification is performed via macroscopic and microscopic methods, when the anatomical differences in the species in question offer typical markers to this purpose. Though the methods of anatomical and histochemical analysis are well established (Widenhoeft and Regis 2005), these approaches fail in the case of closely related wood species, particularly if the anatomical differences are not unambiguous. In such situations, differentiation is possible by means of chemotaxonomical and genetic methods, which are especially useful in the protection of endangered and similar wood species (Cordeiro et al. 2012; Wu et al. 2017) and in the control of illegal logging of woods from protected areas (Hermanson and Wiedenhoeft 2011). DNA-based methods for wood identification have gained momentum in the past decade (Jiao et al. 2014, 2015; Sandak et al. 2015; Yu et al. 2016; Hung et al. 2017; Wu et al. 2017). Rapid methods have been developed for the evaluation of the physical and chemical properties and the possible economic value of woods (Garneau et al. 2004; Zhao et al. 2014). However, wood species identification based on macroscopic features is difficult in the industrial routine, when non experts in wood science have to take rapid decisions, and when the wood is already converted to sawdust or to veneers.
Methods for chemotaxonomical wood identification are mostly based on the chemical analysis of extractives (Sandermann 1962) and spectroscopic techniques involving diffuse reflectance infrared (IR) (Nuopponen et al. 2006), fluorescence (Dickson et al. 2017), Raman scattering (Nguyen et al. 2017), near infrared (NIR) absorption (Nisgoski et al. 2016, 2017), and Fourier transform infrared (FTIR) absorption (Rana et al. 2008; Emmanuel et al. 2015; Traoré et al. 2016; Wang et al. 2017). DNA analysis has also been employed (Yu et al. 2016). Most of these techniques are tedious and time consuming, and cannot be performed on the sampling site and require specially-trained personnel.
Almost all conventional methods for wood analysis are applied on solid samples. A simple alternative approach involves headspace analysis which measures the distinctive volatile compounds (VOCs) emitted by wood species. The headspace of wood samples can be characterized by means of gas chromatography/mass spectroscopy (GC/MS) (Rinne et al. 2002; Muller et al. 2006; Fedele et al. 2007).
Headspace analysis can also be carried out through “electronic nose” technologies that are based on conducting polymers (Wilson et al. 2005; Cordeiro et al. 2012, 2016) and graphite/polymer composites (Garneau et al. 2004). The described materials are efficient, but newer sensing materials, such as carbon nanotubes composites (CNT) could bring about more advantages. Composites of CNT with polymers are innovative materials, which exhibit remarkable electrical conductive behavior, extraordinary mechanical properties and good thermal stability (Moniruzzaman and Winey 2006). The CNT imparts to the composites high electrical and thermal conductivities and large surface areas (Moniruzzaman and Winey 2006). The insulating polymers induce chemical selectivity, because the polymer interacts in a selective manner with the vapor molecules to be analyzed.
Chemiresistors with CNT/polymer composites based on cellulose (Qi et al. 2015), polymethyl methacrylate (Abraham et al. 2004), polyethyleneglycol (Niu et al. 2007), and polystyrene (Zhang et al. 2005), have been applied for the analysis of the chemicals toluene, methanol, water, chloroform, isopropanol, tetrahydrofuran, dichloromethane, ethanol, heptane and cyclohexane.
The present study aimed at the development of an array of chemiresistor gas sensors based on CNT/polymer composites for a rapid, non-destructive, and cost-effective discrimination of wood species. Five Philippine wood species will be analyzed for the first time through headspace analysis in combination with chemometric techniques, such as principal component analysis (PCA) and hierarchical cluster analysis (HCA).
Materials and methods
Multiwalled carbon nanotubes (MWCNT, purity >95%; length 10–50 μm; diameter 10–30 nm) were obtained from Iljin Nanotech, Seoul, South Korea and used as received without any further purification. The polymers investigated were polyethylene glycol (PEG) (Ave. MW 3,350, Sigma Chemical Co., St. Louis, Missouri, USA), polyvinyl chloride (PVC) (low MW 100 000, BDH Limited, Poole, Dorset, England, UK), polyvinylpyrrolidone (PVP) (Ave. MW 10 000, Sigma-Aldrich, St. Louis, Missouri, USA), poly(methyl methacrylate) (PMMA) (commercial grade, Republic Chemical Industries, Quezon City, The Philippines), poly(ethylene-co-vinyl acetate) (PEVA) (Ave. MW 150 000, Aldrich Chemical Co., St. Louis, Missouri, USA), epoxy (commercial grade, Republic Chemical Industries, Quezon City, The Philippines), poly(styrene) (PS) (Ave. MW 280 000, Aldrich Chemical Co., St. Louis, Missouri, USA), and polydimethyl siloxane (PDMS) (commercial grade, Republic Chemical Industries, Quezon City, The Philippines).
Five important Philippine wood species were in focus: Pterocarpus indicus Willd. (narra), Gmelina arborea Roxb. (gmelina), Vitex parviflora Juss. (molave), Diospyros philippinensis (Desr.) Gürke (kamagong) and Acacia auriculiformis A. Cunn. (acacia). According to the Red List of Threatened Species (IUCN 2015), D. philippinensis is endangered, P. indicus and V. parviflora are vulnerable, while A. auriculiformis is of least concern. Gmelina arborea has not been assessed for the IUCN Red List. Diospyros philippinensis, P. indicus and V. parviflora are considered to be premium wood species due to their excellent hardness and durability (Gazal et al. 2004; Bankoff 2007; Pobar 2013), while A. auriculiformis and G. arborea are commercial wood species used for sawn timber and pulpwood (Sayre 2004; Hai et al. 2008).
Fresh stems of the five wood species were collected from healthy and mature trees on the grounds of the University of Santo Tomas, Manila (14° 36′ 33″ North and 120° 59′ 24″ East), and authenticated by the UST Herbarium, Research Center for the Natural and Applied Sciences, University of Santo Tomas. The samples were chopped to small pieces, stored in air-tight plastic containers and frozen (<−10°C) until use. Prior to each analysis, the wood samples were thawed and milled after debarking.
The MWCNT/polymer composites were prepared via solution blending technique (Moniruzzaman and Winey 2006). The polymer (0.5 mg) was dissolved in 10 ml of tetrahydrofuran (THF, RCI Labscan), and varying amounts of MWCNT were dispersed in this solution under ultrasonication for 30 min. This mixture led to a CNT/polymer solution with an optimized concentration of 0.1 mg μl−1. PVP is not soluble in THF, so dimethylformamide (DMF, J.T. Baker) was the solvent. The CNT/polymer solution (2.5 μl) was deposited in the gap between the gold electrodes in a laboratory-constructed chemiresistor by means of a spin-coating technique at 25°C (speed of 2000 rpm for 30 s). The chemiresistor consisted of two gold circular planar electrodes, formed by inserting gold wires (18 karat, Ø 0.05 cm, L 0.5 cm) into a Teflon block (5.0 mm×5.0 mm×5.0 mm) with a gap of 200±20 μm between the wires.
Sensor array instrumentation:
The sensor array consisted of eight fabricated chemiresistors, each with a composite film bridging the two gold electrodes. The composites in the chemiresistors are listed in Table 1 together with other information. Figure 1 illustrates the sensor array instrumentation. It is composed of a sample container with a sensing chamber, and a data acquisition system (Picolog 1216 data logger, Pico Technology, Saint Neots, Cambridgeshire, UK) coupled to a personal computer. The sensing chamber housed the sensor array, with each sensor connected to a voltage divider circuit coupled to the data acquisition system, which also serves as the DC power source (2.5 V) and the electrical conductance measuring device. The voltage measurement circuit of the data acquisition system depicts each sensor as R1. Each sensor was coupled with a corresponding resistor (R2) that has a resistance close to that of the initial sensor. This ensures that the resistor will not affect the measured sensitivity of the sensor.
The measurement involved static headspace analysis of the wood samples, which were placed in the sensing chamber. The empty chamber was purged with N2 to remove any impurities adsorbed in the sensing materials and to stabilize the output resistance baseline. For each headspace analysis cycle, the sensors were exposed to the headspace of the wood samples for 10 min and the resistance of the sensors was monitored. After the measurement, the empty sensing chamber was flushed with N2 gas again for 10 min to restore the resistance baseline. Three cycles were done for each measurement on each wood sample.
The relative response (R) of each sensor served for comparison of the response patterns as well as for the chemometric analysis:(1)
where R0 and Rsample are the values of the resistance of the sensors before and after exposure to the headspace a wood sample, respectively. The R data of the array are essential due to the baseline resistance differences of the polymer composites. The average R data for the three cycles are presented. Four replicate measurements were done on the wood samples.
A scanning electron microscope (SEM, Hitachi TM3000) was available (100–3000× normal view) for the observation of the composite film surfaces.
PCA and HCA were performed based on the data by the XLSTAT software package (Addinsoft, New York, USA).
Results and discussion
Electrical and morphological characteristics
The electrical resistance of the CNT/polymer composites varied with the amount of CNT dispersed in the polymer matrix and the values decreased dramatically with increasing concentration of CNT up to 15%. Beyond this value, the resistance change became smaller. Thus a 15% CNT concentration can be considered as an optimum. The relevant data in this context are listed in Table 1, which can be interpreted as a manifestation of the formation of a conductive network of the carbon nanotube filler in the non-conductive polymer matrix with 15% CNT content (Bauhofer and Kovacs 2009). The SEM micrographs in Figure 2a and b provide evidence for the formation of a conductive network in the CNT/PS composites with 5 and 12.5% CNT. At lower concentrations, the CNTs (black spots) were organized as separate aggregates in the polymer matrix (transparent). At higher concentration, the conductive CNT fillers in the composite are highly inter-connected.
In preliminary experiments, the electrical resistance of the CNT/polymer composites was found to be sensitive to the components of (VOCs), such as aromatic hydrocarbons, alcohols and ketones (Abraham et al. 2004; Zhang et al. 2005; Niu et al. 2007; Qi et al. 2015). Figure 3 shows the response of the various sensors in the headspace of the A. auriculiformis. Similar behavior was seen with other wood samples and the chemiresistors in the sensor array responded rapidly. The resistance of most sensors decreased in the presence of wood. However, in case of CNT/PEG and CNT/PVP based sensors, an elevated sensor resistance was observed. This might be due to the high polarity of the polymer components. The sensor responses reached a steady state within 113–240 s and the responses exhibit a good reversibility and repeatability, with a mean relative standard deviation (SD) of 6.9%.
The sensor responses are wood specific as illustrated on the corresponding radar plots of the normalized sensor responses in Figure 4. Distinctive patterns can be recognized for each sample, which have a high fingerprinting potential for the headspace identification of wood species. Two possible mechanisms are conceivable. One could be the result from the electronic charge transfer from adsorbed gas molecules to the CNTs in the polymer matrices (Zhang and Zhang 2009). Another one could also involve the reversible adsorption of gas molecules to the polymer matrix through specific intermolecular interactions. The interactions cause the polymer composite to swell or shrink, and consequently disrupt the conduction network of the dispersed CNTs in the polymer matrix (Zhang and Zhang 2009). It is likely that polar VOCs would interact to a higher degree with highly polar polymers such as PEG and PVP through dipole-dipole interaction or hydrogen-bonding, while low polarity and non-polar compounds would interact with the non-polar/low-polarity polymers such as PS, PVC, PEVA, epoxy, PDMS and PMMA through van der Waals or hydrophobic interactions. The SEM micrographs in Figure 2c and d reveal a difference in the sensing surface with and without a sample in the headspace vessel. Initially, small holes are present on the surface, but after exposure to the headspace, the holes sizes increased, which causes a lessened interconnectivity of the conducting filler and leads to a conductance decrement.
Optimization of sensor response
The factors sample mass, sample matrix and volume of the sample container affect the sensor array response towards the samples. With increasing sample masses the magnitude of the response was also increasing and the response time was decreasing. This effect is certainly related to the greater amount of the VOCs in the headspace in the case of higher sample amounts. It was observed that milled wood (MW) samples gave the same relative responses as chips with a similar mass, but the response time was shorter. This is probably related to the higher diffusion rate of VOCs in the case of MW, compared to unmilled chips. An increase of the sample container volume caused a significantly larger response time but did not affect the relative response of the sensor array. Clearly, the slower diffusion of the VOCs in a larger volume towards the sensing surface is the reason for this observation. An optimum sensor response can be obtained from 3.0 g of a MW sample in a 75-ml container.
Agitation of the sample during measurement resulted in a longer response time of the sensor array, which may be due to convection drifts inside the sample container that is interfering with the diffusion movement of gas molecules. Application of heat had an adverse effect on the response of the composites, therefore, measurements were done at room temperature. The moisture content (MC) had a positive influence on VOCs, which was observable from the very low responses of air-dried samples compared to that of fresh ones. This effect was also reported in an earlier paper (Wilson 2012), in which aged-dried wood samples had to be rehydrated through soaking in water to produce a sensor response. In view of these effects, the measurement of the sensor response was carried out on fresh MW samples without heating and agitation.
Principal component analysis (PCA)
PCA was applied for response quantification of the chemoresistive sensors. The resulting PCA scatter plot (Figure 5a) revealed a clear data clustering in terms of the wood species with a total variance of 76.8%. The first component F1 contributed to the differentiation of G. aborea and D. philippinensis. The second component F2 can be credited for the differentiation of V. parviflora and P. indicus. The clusters exhibited by the responses to P. indicus and A. auriculiformis were close to each other. This may be due to their common family (Fabaceae). On the other hand, the VOC emissions of special trees can be very different within the same taxonomical family. The loading plot presented in Figure 5b illustrates the relatedness of the variables (responses of the sensors) to each other. It reveals that sensors based on PDMS, PMMA, PS and epoxy composites, exhibit similar significant positive loadings in the positive side of the first principal component (F1) and, therefore, contributed distinctively to the total variation of the data sets. These sensors are based on nonpolar polymers. On the other hand, the sensors based on PEVA and PEG show significant positive loading contributions to the second principal component (F2), while the sensor based on PVP shows negative loading on F1.
A three-dimensional (3D)-PCA plot yielded an increased variability of the score plot from 76.8% to 87.2% through the addition of another PCA axis (Figure 6). The 3D-PCA plots offer advantage as a better clustering of the wood species can be observed, and the clusters of P. indicus and A. auriculiformis are well separated from each other.
Hierarchical cluster analysis (HCA)
HCA yielded dendrograms (computed by the agglomerative Ward’s method) are presented in Figure 7, where dissimilarity relationships between each wood species are visible. The horizontal axis is the Euclidean distance among the groups and the vertical axis indicates the wood samples similarity. The relatedness of the wood groups of P. indicus and A. auriculiformis in the dendrogram is again evident. The results of HCA complement those of PCA demonstrate that clear discrimination of five Philippine wood species could be achieved through the analysis in a chemiresistor array headspace.
A chemiresistor sensor array was developed for the discrimination of five Philippine wood species through headspace analysis. Unlike other sensors for the identification of wood species, the array is based on carbon nanocomposites, which display remarkable electrical conductive properties. These chemiresistive sensors can be easily fabricated and assembled based on easily available and low-cost materials, which have an estimated total cost of about 1 USD per sensor. The sensor assembly provided data in a straightforward, rapid and non-destructive way, which could be evaluated and visualized by chemometry for wood species identification. The suggested approach has a high development and application potential to various applied purposes of wood trade and technology.
One of the authors (J.M. Kalaw) gratefully acknowledges the scholarship and thesis grant provided by the Department of Science and Technology – Science Education Institute, Philippines. Wood samples were obtained through the generous help of the Facilities Management Office of the University of Santo Tomas.
Abraham, J.K., Philip, B., Witchurch, A., Varadan, V.K., Reddy, C.C. (2004) A compact wireless gas sensor using a carbon nanotube/PMMA thin film chemiresistor. Smart Mater. Struct. 13:1045–1049.Google Scholar
Bankoff, G. (2007) One island too many: reappraising the extent of deforestation in the Philippines prior to 1946. J. Hist. Geogr. 33:314–334.Google Scholar
Bauhofer, W., Kovacs, J.Z. (2009) A review and analysis of electrical percolation in carbon nanotube polymer composites. Compos. Sci. Technol. 69:1486–1498.Google Scholar
Cordeiro, J.R., Martinez, M.I., Li, R.W., Cardoso, A.P., Nunes, L.C., Krug, F.J., Paixao, T., Nomura, C., Gruber, J. (2012) Identification of four wood species by an electronic nose and by LIBS. Int. J. Electrochem. 2012:1–5.Google Scholar
Cordeiro, J.R., Li, R.W., Takahashi, E.S., Rehder, G.P., Ceccantini, G. Gruber, J. (2016) Wood identification by a portable low-cost polymer-based electronic nose. RSC Adv. 6:109945–109949.Google Scholar
Dickson, A., Nanayakkara, B., Sellier, D., Meason, D., Donaldson, L. Brownlie, R. (2017) Fluorescence imaging of cambial zones to study wood formation in Pinus radiata D. Don. Trees 31:479–490.Google Scholar
Emmanuel, V., Odile, B., Céline, R. (2015) FTIR spectroscopy of woods: a new approach to study the weathering of the carving face of a sculpture. Spectrochim. Acta Mol. Biomol. Spectrosc. 136:1255–1259.Google Scholar
Fedele, R., Galbally, I.E., Porter, N., Week, I.A. (2007) Biogenic VOC emissions from fresh leaf mulch and wood chips of Grevillea robusta (Australian Silky Oak). Atmos. Environ. 41:8736–8746.Google Scholar
Garneau, F.-X., Riedl, B., Hobbs, S., Pichette, A., Gagnon, H. (2004) The use of sensor array technology for rapid differentiation of the sapwood and heartwood of Eastern Canadian spruce, fir and pine. Holz Roh Werkst. 62:470–473.Google Scholar
Gazal, R.M., Blanche, C.A., Carandang, W.M. (2004) Root growth potential and seedling morphological attributes of narra (Pterocarpus indicus Willd.) transplants. For. Ecol. Manage. 195:259–266.Google Scholar
Hai, P.H., Jansson, G., Harwood, C., Hannrup, B., Thinh, H.H. (2008) Genetic variation in growth, stem straightness and branch thickness in clonal trials of Acacia auriculiformis at three contrasting sites in Vietnam. For. Ecol. Manage. 255:156–167.Google Scholar
Hermanson, J.C., Wiedenhoeft, A.C. (2011) A brief review of machine vision in the context of automated wood identification systems. IAWA J. 32:233–250.Google Scholar
Hung, K.-H., Lin, C.-H., Ju, L.-P. (2017) Tracking the geographical origin of timber by DNA fingerprinting: a study of the endangered species Cinnamomum kanehirae in Taiwan. Holzforschung 71:853–862.Google Scholar
IUCN (2015) International Union for Conservation of Nature and Natural Resources, Red List of Threatened Species Version 2015-4. Available at <www.iucnredlist.org>. Downloaded on 10 April 2016.Google Scholar
Jiao, L., Yin, Y., Cheng, Y., Jiang, X. (2014) DNA barcoding for identification of the endangered species Aquilaria sinensis: comparison of data from heated or aged wood samples. Holzforschung 68:487–494.Google Scholar
Jiao, L., Liu, X., Jiang, X., Yin, Y. (2015) Extraction and amplification of DNA from aged and archaeological Populus euphratica wood for species identification. Holzforschung 69:925–931.Google Scholar
Moniruzzaman, M., Winey, K.I. (2006) Polymer nanocomposites containing carbon nanotubes. Macromolecules 39:5194–5205.Google Scholar
Muller, K., Haferkorn, S., Grabmer, W., Wisthaler, A., Hansel, A., Kreuzwieser, J., Cojocariu, C., Rennenberg, H., Herrmann, H. (2006) Biogenic carbonyl compounds within and above a coniferous forest in Germany. Atmos. Environ. 40:S81–S91.Google Scholar
Nguyen, H., Lagarde, F., Louarn, G., Daniel, P. (2017) A new way to discriminate polluted wood by vibrational spectroscopies. Talanta 167: 436–441.Google Scholar
Nisgoski, S., Schardosin, F.Z., Batista, F.R., Muñiz, G.I., Carneiro, M.E. (2016) Potential use of NIR spectroscopy to identify Cryptomeria japonica varieties from southern Brazil. Wood Sci. Technol. 50:71–80.Google Scholar
Nisgoski, S., de Oliveira, A.A., de Muñiz, G.I. (2017) Artificial neural network and SIMCA classification in some wood discrimination based on near-infrared spectra. Wood Sci. Technol. 51:929.Google Scholar
Niu, L., Luo, Y., Li, Z. (2007) A highly selective chemical gas sensor based on functionalization of multi-walled carbon nanotubes with poly(ethylene glycol). Sens. Actuator B Chem. 126: 361–367.Google Scholar
Nuopponen, M.H., Birch, G.M., Sykes, R.J., Lee, S.J., Stewart, D. (2006) Estimation of wood density and chemical composition by means of diffuse reflectance mid infrared Fourier transform (DRIFT-MIR) spectroscopy. J. Agric. Food Chem. 541:34–40.Google Scholar
Pobar, R.A. (2013) Enhancing the use of value-added products from underutilized fruit of the endangered mabolo (Diospyros blancoi) tree. IJERD 4:100–105.Google Scholar
Qi, H., Liu, J., Pionteck, J., Pötschke, P., Mäder, E. (2015) Carbon nanotube–cellulose composite aerogels for vapour sensing. Sens. Actuator B Chem. 213:20–26.Google Scholar
Rana, R., Müller, G., Naumann, A., Polle, A. (2008) FTIR spectroscopy in combination with principal component analysis or cluster analysis as a tool to distinguish beech (Fagus sylvatica L.) trees grown at different sites. Holzforschung 62:530–538.Google Scholar
Rinne, H.J., Guenther, A.B., Greenberg, J.P., Harley, P.C. (2002) Isoprene and monoterpene fluxes measured above Amazonian rainforest and their dependence on light and temperature. Atmos. Environ. 36:2421–2426.Google Scholar
Sandak, J., Sandak, A., Cantini, C., Autino, A. (2015) Differences in wood properties of Picea abies L. Karst. in relation to site of provenance and population genetics. Holzforschung 69: 385–397.Google Scholar
Sandermann, W. (1962) Biosynthetische Untersuchungen an verschiedenen Kiefernarten. Holzforschung 16:65–74.Google Scholar
Sayre, E. (2004) Gmelina arborea promotion, planting, wood processing and product marketing: documentation of best management practices among smallholder farmers in Mindanao, the Philippines. In: The First World Agroforestry Congress, Book of Abstracts, Orlando, Florida, USA. June 27, 2004–July 2, 2004. p.396.Google Scholar
Traoré, M., Kaal, J., Cortizas, A.M. (2016) Application of FTIR spectroscopy to the characterization of archeological wood. Spectrochim. Acta Mol. Biomol. Spectrosc. 153:63–70.Google Scholar
Wang, S.-N., Zhang, F.-D., Huang, A.-M., Zhou, Q. (2017) Distinction of four Dalbergia species by FTIR, 2nd derivative IR, and 2D-IR spectroscopy of their ethanol-benzene extractives. Holzforschung 70:503–510.Google Scholar
Widenhoeft, A.C., Regis, B.M. (2005) Structure and function of wood. In: Handbook of Wood Chemistry and Wood Composites. Ed. Rowell, R.M. CRC Press, Boca Raton, FL. pp. 9–32.Google Scholar
Wilson, A.D. (2012) Application of a conductive polymer electronic nose device to identify aged woody samples. In: Proceedings of the 3rd International IARIA Conference on Sensor Device Technologies and Applications, Rome, Italy, Ed. S. Yurish, I. Chilibon, V. Carvalho, S. Gervais-Ducouret, Xpert Publishing Services, Wilmington, DE, USA. pp. 77–82.Google Scholar
Wilson, A.D., Lester, D.G., Oberle, C.S. (2005) Application of conductive polymer analysis for wood and woody plant identifications. Forest Ecol. Manage. 209:207–224.Google Scholar
Wu, C.-C., Chu, F.-H., Ho, C.-K., Sung, C.-H., Chang, S.-H. (2017) Comparative analysis of the complete chloroplast genomic sequence and chemical components of Cinnamomum micranthum and Cinnamomum kanehirae. Holzforschung 71:189–197.Google Scholar
Yu, M., Liu, K., Zhou, L., Zhao, L., Liu, S. (2016) Testing three proposed DNA barcodes for the wood identification of Dalbergia odorifera T. Chen and Dalbergia tonkinensis Prain. Holzforschung 70:127–136.Google Scholar
Zhang, W.-D., Zhang, W.-H. (2009) Carbon nanotubes as active components for gas sensors. J. Sens. 2009:1–16.Google Scholar
Zhang, B., Fu, R.W., Zhang, M.Q., Dong, X.M., Lan, P.L., and Qiu, J.S. (2005) Preparation and characterization of gas-sensitive composites from multi-walled carbon nanotubes/polystyrene. Sens. Actuator B Chem. 109:323–328.Google Scholar
Zhao, P., Duo, G., Chen, G.-S. (2014) Wood species identification using feature-level fusion scheme. Optik 125: 1144–1148.Google Scholar