ar ti cl e s doi: 10.1255/nirn.1369
Detecting plant diseases using visible/near infrared spectroscopy N. Abu-Khalaf* and M. Salman Technical and Applied Research Center (TARC), Palestine Technical University–Kadoorie (PTUK), PO Box 7, Tulkarm, Palestine. E-mail:
[email protected]
Left: Dr Mazen Salman; right: Dr Nawaf Abu-Khalaf
Introduction
P
alestine Technical University– Kadoorie (PTUK) is one of 13 higher educational institutes in Palestine. PTUK is the first and only technical and governmental university in the West Bank. It is located on the western side of the city of Tulkarm and serves the Palestinian community in the Palestinian governorates. PTUK was established in 1930 as an agricultural school (Figure 1) and was then developed to provide Diploma programmes in various disciplines. The university went through different phases of development until it was accredited as a university on August 2007 to provide technical programmes at various levels including Diploma and Bachelor degrees in different areas in engineering and applied technology. Currently there are about 5000 students enrolled in PTUK. PTUK is particularly interested in teaching and delivering knowledge of applied science in the fields of agricultural and evnironmental sciences. On the basis of its three units (biotechnology, water and environment, and agriculture), PTUK established the Technical and Applied Research Center (TARC) in 2009. Several research projects are being conducted at TARC including phytopathology, soil, air and water pollution as well as sensor technology for non-destructive quality assesment of agricultural commodities. PTUK is very interested in the olive sector in Palestine, since olive trees cover approximately 45% of Palestinian agricultural lands. The tree is very important in the economic and social lives of the population, as it accounts for one of their main sources of income; in a good year, it can contribute about 13% of the annual agricultural production. The olive (Olea europaea L.) is one of the oldest agricultural trees and is cultivated over large areas in Palestine, with more than 10 million olive trees (about 67.3% of all horticultural trees) being grown. Olive production contributes about 12–13% of national income.1
12
Vol. 24 No. 4 June/July 2013
Figure 1. Pictures showing the development of PTUK. The left-hand photo from the 1930s, when Kadoorie Agricultural School was established. The right-hand photo is a current view of the PTUK.
The olive tree is affected by many pests and diseases. Olive leaf spot (OLS) is a foliar disease which is widespread in all olive growing regions of the world and has been reported in Mediterranean areas2 (Figure 2). As a result of infection, yield losses occur and these can amount to up to 20% of total production.3 Olive leaf spot is chemicallycontrolled by application of copper (Cu) fungicides directly after harvest4 but chemical treatment rarely appears to be effective.5 Moreover, using chemical fungicides leads to the appearance of pathogen races which are resistant to Cu6 as well as more general imbalance of the plant metabolism following Cu accumulation in the soil.5 In Palestine, the disease is common throughout many areas of the southern and northern parts of the country.7 The Palestinian Ministry of Agriculture, NGOs and several private institutions work with the olive sector to provide guidance and information about improving olive crop production and protection. Visible/near infrared (vis/NIR) spectroscopy is a mature sensor technology that has been successfully applied to the noninvasive analysis and detection of many parameters (for example, quality, stress and presence of disease) of a wide range of agricultural commodities.8 Due to its success, vis/NIR spectroscopy was used in this feasibility study to detect the severity of OLS disease in olive trees.
Materials and methods Fifty (n = 50) olive leaves were collected from olive trees infected with olive leaf spot disease and brought to the biotechnology
laboratoy on the same day. Thirty (n = 30) had a very clear and obvious infection while the remainder (n = 20) were believed to have an invisible to latent infection. Spectra were collected from each leaf using a recentlyacquired USB2000+ miniature fibre-optic spectrometer (Ocean Optics, USA) with a Vivo light source operating in the 550– 1100 nm wavelength range at a resolution of 0.35 nm full width at half maximum (FWHM). The spectrometer has a 2 MHz analogue-todigital (A/D) converter, 2048-element CCDarray detector and a high-speed USB 2.0 port. The USB2000+ can be controlled by SpectraSuite software. The Vivo light system contains four tungsten halogen bulbs that can be turned on or off individually. Any risk of overheating samples is mitigated through active cooling which protects the sample and ensures accuracy. Using four tungsten halogen light sources makes the Vivo a high-powered vis/NIR source which allows a shorter integration time (OceanOptics, USA). After spectral acquisition from leaves, the severity of OLS infection was determined by dipping the leaves in 5% NaOH at 50°C7 and the number of lesions was counted (see Figure 2). Disease incidence was determined by recording the percentage of infected leaves per tree. For disease severity, the number of lesions per leaf was counted and graded as follows; 1 (1 lesion), 2 (2 lesions), 3 (3–5 lesions), 4 (6–10 lesions) or 5 (>11 lesions). Support vector machine (SVM) classification in The Unscrambler software (Version 10.2, Camo Software AS, Oslo, Norway) was used to classify leaf spectra according to their severity class.
Series 4000 FTNIR Spectrophotometer
Food
Pharmaceutical
Chemical
Agriculture
Figure 2. Disease symptoms on olive leaves (left) and spots of the disease before and after identification using 5% NaOH (right).
each row showing the instances in a predicted severity class and each column representing the instances in an actual severity class. The mean correct classification rate is about 75%, which is promising in an initial study such as this. It is obvious that, in general, the correct classification rate increased with increasing disease severity. This is in agreement with Rumpf et al.8 Tables 1 and 2 give a good indication about the possiblity of using vis/NIR spectroscopy for identification and quantification of OLS severity with an acceptable accuracy (greater than 65% for both accuracy and averaged classification rate, i.e. 65% and 75%, respectively). Further research is needed to investigate latent disease and a wider range of infection severity. Our vision in TARC is to carry out experiments on different crops to investigate the possibility of using vis/NIR spectroscopy for detecting latent plant pathogen infections. Olive trees were chosen as the first crop to be studied due to their importance for the agricultural sector in Palestine. With respect to our international collaboration, we are currently co-operating with the Applied Chemometrics, Applied Physics, Bioenergy and Sampling (ACABS) research group in Aalborg University, Esbjerg, Denmark. Furthermore, we are a new university looking forward to collaborating with other international researchers.
SVM classification algorithms try to find patterns in empirical data (training data) with regard to label classes. The main advantages of SVMs arise from their generalisation ability, which is achieved by using the maximum margin hyperplane for separation and the application of non-linear discriminant functions. Moroever, SVMs can handle complex discrimination problems. SVM classifications achieved in this work were developed using a cross-validation method.8
Results and discussions SVM classification was able to classify disease severity on leaves using spectra data with an acceptable level of accuracy in this feasibility study. Table 1 shows the accuracy levels of SVMs in both training and validation sample sets. A summary of the % SVM classification rate results can also be shown in a confusion matrix (Table 2). This matrix provides information about the correcly predicted and actual classifications of samples, with
Table 1. Accuracy (%) of training and validation sets using SVM classification of vis/NIR spectra of olive leaves.
Accuracy
Training set
Validation set
76%
66%
continued on page 25
The Series 4000 offers diffuse reflectance using a rotating sample d i s h o r transmission analysis at up to speeds of 5 scans per second. With e x t r em e ly h igh signal to noise ratio and laser aligned wavelength accuracy, the Series 4000 ensures the best analytical results from any NIR analyser.
Table 2. Percentage correct classification rate of olive leaf spot (OLS) severity class.
Actual severity class Predicted severity class
2
2
33%
3
67%
4 5
3
4
88%
11%
For more information
12%
78%
call: 612 9771 5444
11%
5
100%
email:
[email protected] or visit our web site: www.nextinstruments.net
my t hb u ste r s
i n
c h e m om e t ric s continued from page 13
Conclusion Vis/NIR spectroscopy has shown some promise for the detection of OLS disease. Latent disease sensing for olives and other agricultural commodities will be the next step of our investigations.
Acknowledgements The authors would like to thank Dr Saed Mallak (acting president of PTUK) for supporting these research activities. Thanks are also due to Mr Azmi Saleh (public relations at PTUK) for providing information about the university.
References 1. PCBS, Palestinian Central Bureau of Statistics. Ramallah, Palestine, http://www.pcbs. gov.ps (2012). 2. F.O. Obanor, M. Walter, E.E. Jones and M.V. Jaspers, “Sources of variation in a field evaluation of the incidence and severity of olive leaf spot,” NZ Plant Prot. Soc. 58, 273–277 (2005). 3. T. Azeri, “Research on olive leaf spot, olive knot and Verticillium wilt of olive in Turkey”, Bull. OEPP/EPPO Bull. 23, 437–440 (1993). doi: 10.1111/j.1365-2338.1993.tb01349.x 4. F. Sistani, S.S Ramezanpour and S. Nasrollanejad, “Field evaluation of different fungicides application to control olive leaf spot”, Aus. J. Basic and Appl. Sci. 3(4), 3341–3345 (2009). 5. F.O. Obanor, M.V. Jaspers, E.E. Jones and M. Walter, “Greenhouse and field evaluation of fungicides for control of olive leaf spot in New Zealand”, Crop Prot. 27, 1335–1342 (2008). Doi: 10.1016/j.cropro.2008.04.007 6. J.L. Vanneste, M.D. Voyle and S.M. Zydenbos, “Genetic basis of copper resistance in New Zealand strains of Pseudomonas syringae”, New Zealand Plant Protection 56, 109– 112 (2003) 7. M. Salman, A. Hawamda, A.A. Amarni, M. Rahil, H. Hajjeh, B. Natsheh and R. Abuamsha, “Evaluation of the incidence and severity of olive leaf spot caused by Spilocaea oleagina on olive trees in Palestine”, Am. J. Plant Sci. 3, 457–460 (2011). doi: 10.4236/ ajps.2011.23053 8. T. Rumpf, A.K. Mahlein, U. Steiner, E.C. Oerke, H.W. Dehne and L. Plumer, “Early detection and classification of plant diseases with Support Vector Machines based on hyperspectral reflectance”, Comput. Electron. Agric. 74(1), 91–99 (2010). doi: 10.1016/j.compag.2011.09.011 Vol. 24 No. 4 June/July 2013
25
