Professional Documents
Culture Documents
2016, Ashourloo
2016, Ashourloo
Abstract—Spectral data have been widely used to estimate the Considering the importance of WLR disease and DS levels,
disease severity (DS) levels of different plants. However, such several attempts have been made over the past decade to detect
data have not been evaluated to estimate the disease stages of DS of WLR using spectral data. For example, some studies
the plant. This study aimed at developing a spectral disease index
(SDI) that is able to identify the stages of wheat leaf rust disease employed classification and regression techniques for WLR
at various DS levels. To meet the aim of the study, the reflectance disease monitoring at the canopy scale [6], [7]. Another study
spectra (350–2500 nm) of infected leaves with different symptom showed the potential of three spectral vegetation indices to
fractions and DS levels were measured with a spectroradiometer. detect WLR from other types of rust disease at the leaf scale
Then, pure spectra of the different disease symptoms at the leaf [6]. These indices have been ranked on their accuracy to predict
scale were analyzed, and a new function was developed to find
the wavelengths most sensitive to disease symptom fraction. The DS levels of WLR [8]. In addition to previous approaches,
reflectance spectra with highest sensitivity were found at 675 and Ashourloo et al. [9] proposed two spectral disease indices
775 nm. Finally, the normalized difference of DS and the ratio (SDIs) to detect WLR based on the analyses of spectra from
ρ675 /ρ775 was used as a new SDI to discriminate three different different disease symptoms using Fisher function and RGB
levels of the disease stage at the canopy level. The suggested SDI digital photographs. However, less attention has been made
showed a promising performance to improve the detection disease
stages in precision plant protection. toward estimation of the disease stages, and this has been a
neglected research area of disease detection using spectral data.
Index Terms—Disease stages, disease symptoms, hyperspectral Therefore, the potential of spectral bands for estimating WLR
reflectance, leaf rust, spectral disease index (SDI), wheat.
disease stages across DS levels needs to be studied.
In this research, we estimate disease stages by using pure
I. I NTRODUCTION spectra of disease symptoms at various disease severities. The
disease severity (DS) of the WLR disease was investigated in
S PECTRAL data in the visible and near-infrared regions
have been utilized to estimate the disease severity (DS)
levels of infected leaves in precision crop protection [1]–[3].
order to achieve two goals: 1) development of a new function
to find the most sensitive wavelengths to retrieve the disease
Wheat leaf rust (WLR) is an important crop disease caused symptom fraction and 2) development of an index for the
by the Puccinia triticina fungus. This disease has five sporing precise determination of disease stages.
stages (uredospores, teliospores, basidiospores, spermatia, and This letter is organized as follows. In Section I, a review of
aeciospores) that uredospores, teliospores, and basidiospores the disease detection methods and the objective of this letter
develop on the leaf surface [4]. In infected leaves, pustules are presented. The next section presents the experimental setup
appear in a random scatter distribution as disease develops [4]. and data collection. In Section III, the development procedure
The WLR disease exhibits different symptoms [5], manifesting for the new index is explained. In Section IV, the results based
themselves in yellow, orange, and dark brown colors. Finally, on the developed index are explained in detail, and in Section V,
the disease symptoms accrue and culminate in a dry leaf. the conclusion is given.
Yellow and orange colors show the initial stages (chlorosis),
while dark brown and dry leaves represent the final stages of
the WLR disease (necrosis) [5]. II. E XPERIMENTAL S ETUP
In order to obtain robust and independent results, two cul-
tivars of wheat, namely, Bolani and Roshan, were selected.
Manuscript received December 2, 2015; revised March 13, 2016; accepted
March 23, 2016. These cultivars are vulnerable to WLR disease and are also
D. Ashourloo, A. A. Matkan, and H. Aghighi are with the Remote Sensing extensively planted in Iran. The experiment was performed in a
and GIS Center, Faculty of Earth Sciences, Shahid Beheshti University, Tehran greenhouse located at the Iranian Research Institute of Plant
19839 63113, Iran (e-mail: d_ashourloo@sbu.ac.ir; a-matkan@sbu.ac.ir;
h_aghighi@sbu.ac.ir). Protection, under fully controllable conditions. Spectral data
A. Huete is with the Plant Functional Biology and Climate Change Cluster, from healthy and infected samples were collected in order to
University of Technology Sydney, Sydney, NSW 2007, Australia (e-mail: identify three different stages of WLR disease. In stage 1, the
Alfredo.Huete@uts.edu.au).
M. R. Mobasheri is with the Remote Sensing Group, Faculty of Geodesy and leaf begins to show symptoms with yellow and orange colors
Geomatics Engineering, K. N. Toosi University of Technology, Tehran 19697 predominant. In stage 2, all of the symptoms are fully developed
64499, Iran (e-mail: mobasheri@kntu.ac.ir). and can be seen with similar proportion of colors. Stage 3
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. includes dry leaves in which the predominant symptoms show
Digital Object Identifier 10.1109/LGRS.2016.2550529 as brown color.
1545-598X © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
ASHOURLOO et al.: DEVELOPING AN INDEX FOR DETECTION AND IDENTIFICATION OF DISEASE STAGES 3
775
. (4)
normalize ρ675 /ρ775 , the maximum and minimum values of DS + ρ775 ρ675
ρ675 /ρ775 are needed. The values of ρ675 /ρ775 for the green
and dry symptoms are 0.71 and 0.13, respectively The potential of the presented index in disease symptom
development was evaluated under two different conditions as
ρ675
− min ρ675
ρ675 ρ775 ρ775 follows.
N =
. (3)
ρ775 max ρ775 − min ρρ675
ρ675 First, it is the ability to estimate different disease stages
775
in two infected leaves under similar DS levels. Second, it is
In the remainder of this research, N (ρ675 /ρ775 ) denotes the the potential to estimate similar disease stages in two infected
normalized value of ρ675 /ρ775 . Based on the disease progress, leaves, while they have different DS levels. So as to show
DS ≥ N (ρ675 /ρ775 ). the potential of NLRHI, six samples of infected leaves with
When the entire leaf is affected and dried by the disease a variety of DS levels and disease stages were chosen. The
(DS = 1 and N (ρ675 /ρ775 ) = 1), then DS = N (ρ675 /ρ775 ) spectra and disease symptom fractions of these samples are
and DS − N (ρ675 /ρ775 ) = 0. shown in Fig. 5 and Table I, respectively. Samples a and b
The high value of the difference shows the beginning symp- have similar values of DS levels (around 0.2), while their
toms of the disease. For example, when the leaf is yellow disease stages are different, and dry and yellow symptoms are
(DS = 1 and N (ρ675 /ρ775 ) = 0.25), then the difference value predominant symptoms in the a and b samples, respectively. As
of DS = 1 and N (ρ675 /ρ775 ) reaches 0.75. As disease symp- stated before, yellow is the initial, and dry is the final stages of
toms develop, the difference value decreases. Then, two other the WLR disease. As Table I shows, NLRHI distinguishes this
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
ASHOURLOO et al.: DEVELOPING AN INDEX FOR DETECTION AND IDENTIFICATION OF DISEASE STAGES 5
TABLE I V. C ONCLUSION
F RACTION OF D ISEASE S YMPTOMS FOR D IFFERENT I NFECTED L EAVES
The proposed SDI showed high accuracy and sensitivity in
discriminating disease stages at a certain DS. This investiga-
tion, carried out with a developed new function applied over
the 450–1000-nm range, has revealed little spectral distinction
of disease symptoms between the beginning and end of the red-
edge region. Furthermore, there are large differences between
the reflectance spectra of the disease symptoms at 675 and
TABLE II 775 nm. Disease estimation is faced with much difficulty in
C ONFUSION M ATRIX FOR E STIMATION OF D ISEASE S TAGES
the early stages due to the slight development of the disease
and the spectral similarity between the yellow and noninfected
areas. Nevertheless, the proposed SDI has high accuracy in the
early stage detection of infected areas. In addition, suggested
SDI reduces data dimensionality and speeds up the disease
estimation rate. It must be noted that the proposed index needs
to be tested on various sensors and different varieties of wheat
in order to be used in precision plant protection.
ACKNOWLEDGMENT
difference between the two samples. The NLRHI values of a The authors would like to thank the anonymous reviewers for
and b samples are 0.59 and 0.23, respectively. The index value their comments which helped in improving this letter.
of 0.59 for sample a shows the initial stages of the disease in
comparison with 0.23 for sample b. Also, e and f are the same R EFERENCES
as a and b in having similar DS levels and different disease [1] W. Huang et al., “New optimized spectral indices for identifying and mon-
stages. NLRHI properly shows the disease stages, and the index itoring winter wheat diseases,” IEEE J. Sel. Topics Appl. Earth Observ.
Remote Sens., vol. 7, no. 6, pp. 2516–2524, Jun. 2014.
values are 0.21 and 0.59 for e and f samples, respectively. [2] B. J. Vigier, E. Pattey, and I. B. Strachan, “Narrowband vegetation indexes
Although the DS level of sample c is one and a half of sample d, and detection of disease damage in soybeans,” IEEE Geosci. Remote Sens.
they have similar disease stages. Based on Table I, the NLRHI Lett., vol. 1, no. 4, pp. 255–259, Oct. 2004.
[3] J. Zhang et al., “Integrating remotely sensed and meteorological observa-
values are 0.4 and 0.37 for c and d samples, respectively. On tions to forecast wheat powdery mildew at a regional scale,” IEEE J. Sel.
the other hand, the DS level of sample f is triple of b, and Topics Appl. Earth Observ. Remote Sens., vol. 7, no. 11, pp. 4328–4339,
dry symptoms are predominant in both samples. The NLRHI Nov. 2014.
[4] M. D. Bolton, J. A. Kolmer, and D. F. Garvin, “Wheat leaf rust caused by
values for both of them are close to each other, b is 0.23, and Puccinia triticina,” Molecular Plant Pathology, vol. 9, no. 5, pp. 563–575,
f is 0.21. These values for NLRHI demonstrated the ability 2008.
of the presented index to detect disease stages. The afore- [5] H. Wang et al., “Identification and disease index inversion of wheat stripe
rust and wheat leaf rust based on hyperspectral data at canopy level,”
mentioned examples clarified that NLRHI not only is able to J. Spectroscopy, vol. 2015, 2015, Art. no. 651810.
distinguish the disease stage at similar DS levels, but also it is [6] R. Devadas, D. Lamb, S. Simpfendorfer, and D. Backhouse, “Evaluating
possible to estimate the same disease stage at different DS ten spectral vegetation indices for identifying rust infection in individual
wheat leaves,” Precision Agriculture, vol. 10, no. 6, pp. 459–470, 2009.
levels. Therefore, NLRHI can be used as a new SDI in precision [7] T. Rumpf et al., “Early detection and classification of plant diseases with
plant protection. Finally, accuracy assessment of the index support vector machines based on hyperspectral reflectance,” Comput.
values in the identification of disease stages was determined Electron. Agriculture, vol. 74, no. 1, pp. 91–99, 2010.
[8] D. Ashourloo, M. R. Mobasheri, and A. Huete, “Evaluating the effect
at three levels based on the defined thresholds in the confusion of different wheat rust disease symptoms on vegetation indices using
matrix shown in Table II, and the kappa and overall accuracy hyperspectral measurements,” Remote Sens., vol. 6, no. 6, pp. 5107–5123,
were 0.78 and 0.68, respectively. 2014.
[9] D. Ashourloo, M. R. Mobasheri, and A. Huete, “Developing two spec-
It has been reported that each disease symptom of WLR has tral disease indices for detection of wheat leaf rust (Puccinia triticina),”
its own structure [12]. Based on the results obtained in this Remote Sens., vol. 6, no. 6, pp. 4723–4740, 2014.
research, the visible and near-infrared regions are suitable in [10] A. Savitzky and M. J. Golay, “Smoothing and differentiation of data by
simplified least squares procedures,” Analytical Chem., vol. 36, no. 8,
detecting the disease stages at canopy scales. This study has pp. 1627–1639, 1964.
shown the potential of SDIs, developed at the leaf scale, to be [11] H. Al-Hiary, S. Bani-Ahmad, M. Reyalat, M. Braik, and Z. ALRahamneh,
used at the canopy scale [13]. Leaf optical properties, plant “Fast and accurate detection and classification of plant diseases,” Mach.
Learn., vol. 14, p. 5, 2011.
structure, leaf area index, canopy gaps, litter, and soil back- [12] C. Robert, M.-O. Bancal, B. Ney, and C. Lannou, “Wheat leaf photosyn-
ground are important parameters which affected the collected thesis loss due to leaf rust, with respect to lesion development and leaf
spectra at leaf and canopy scales [14]. Thus, the next step is to nitrogen status,” New Phytologist, vol. 165, no. 1, pp. 227–241, 2005.
[13] A.-K. Mahlein et al., “Development of spectral indices for detecting and
consider the effect of the atmosphere and imaging systems on identifying plant diseases,” Remote Sens. Environ., vol. 128, pp. 21–30,
the index accuracy. Moreover, it must be noted that independent 2013.
data sets from the other cultivar of wheat and fungus disease [14] D. Rogge et al., “Potential discrimination of toxic industrial chemical
effects on poplar, canola and wheat, detectable in optical wavelengths
with different symptoms are required to evaluate and improve 400–2450 nm,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens.,
the suggested index. vol. 5, no. 2, pp. 563–573, Apr. 2012.