=Paper= {{Paper |id=Vol-1498/HAICTA_2015_paper12 |storemode=property |title=An Image Processing Based Approach for Detection of Nitrogen Status in Winter Wheat Under Mild Drought Stress |pdfUrl=https://ceur-ws.org/Vol-1498/HAICTA_2015_paper12.pdf |volume=Vol-1498 |dblpUrl=https://dblp.org/rec/conf/haicta/TavakoliMG15 }} ==An Image Processing Based Approach for Detection of Nitrogen Status in Winter Wheat Under Mild Drought Stress== https://ceur-ws.org/Vol-1498/HAICTA_2015_paper12.pdf

An Image Processing Based Approach for Detection of
Nitrogen Status in Winter Wheat Under Mild Drought
Stress

Hamed Tavakoli1, Seyed Saeid Mohtasebi2, Robin Gebbers3
1
Department of Mechanical Engineering of Biosystems, Faculty of Agriculture, Arak
University, Arak 38156-88349, Iran, e-mail: htavakoli1985@gmail.com
2
Department of Agricultural Machinery Engineering, Faculty of Agricultural Engineering &
Technology, University of Tehran, P.O. Box 4111, Karaj 31587-77871, Iran
3
Leibniz-Institute for Agricultural Engineering, Max-Eyth-Allee 100, 14469 Potsdam,
Germany

Abstract. Nitrogen is one of the most important agricultural inputs affecting
crop growth, yield and quality in rain-fed cereal production. In this study an
image processing based approach was used for detection of nitrogen status in
winter wheat. Four N fertilization rates (0, 60, 120 and 240 kg N ha-1, in total)
and two water regimes (irrigated and non-irrigated) were applied to winter
wheat. Digital images of the plant canopy were acquired using a Canon camera
during the growing season 2012. Different indices were extracted by
processing of the images. According to the statistical analyses, all the indices
were affected by both N and water supplies. However, Rm, RMB, NRMB,
Hue and INT were less sensitive to water supply. Among the indices, crop
coverage (CC) showed better results for detection of nitrogen status of the
plant. We conclude that digital cameras can be used to assess nitrogen status of
winter wheat.

Keywords: Precision agriculture, nitrogen, drought stress, wheat, digital
camera.

1 Introduction

Nitrogen (N) is one of the most important agricultural inputs affecting crop growth,
yield and quality in rain-fed cereal production. A mismatch between N supply and
crop requirement can potentially hamper crop growth or harm the environment,
resulting in poor N use efficiency (NUE) and economic losses (Tremblay et al.,
2009). Thus, considerable efforts have been done to develop crop sensors that
provide instant information as a basis for decision making on nitrogen fertilization.
Examples are: Spectral-optical spot-sensors: like Yara N-Sensor (Tremblay et al.,
2009), Acoustic sensors (Sui and Thomasson, 2006), Chlorophyll fluorescence
sensors (Limbrunner and Maidl, 2007), Laser distance sensors (Ehlert et al., 2008),
and Cameras (Lee and Lee, 2013).

74
Since digital cameras precisely record the appearance of photographic subjects in
a non-destructive manner, they can be considered as proximal sensing devices to
provide an alternative and inexpensive method for characterizing crop stand
parameters (Sakamoto et al., 2012). There have been some efforts to use digital
cameras for estimation of nitrogen status in wheat (Li et al., 2010), corn (Rorie et al.,
2011), and rice (Lee and Lee, 2013).
The basic assumption in using the crop sensors for nitrogen fertilization is that
nitrogen of a crop has the strongest effect on the crop attributes assessed by the
respective sensor. However, there is an evidence that these crop attributes can be
affected by other environmental factors such as water supply (Clay et al., 2006).
By reviewing the literature, information about effect of drought stress on the
measurements of nitrogen status using digital cameras was not found. Therefore, the
objective of this study was to investigate the possibility of using a RGB camera for
assessing nitrogen and water supply in winter wheat.

2 Material and Methods

During the 2012 growing season, a field experiment was conducted at the
Bundessortenamt Marquardt experimental station, Potsdam, Germany (52°27' N,
12°57' E). The experiment was designed as a randomized split block design with two
replications. Treatments on winter wheat (cv. Cubus) consisted of four N fertilization
rates (0, 60, 120 and 240 kg N ha-1, in total) and two water regimes (irrigated (Irr)
and non-irrigated (NIrr)). During the growing season, the non-irrigated plots received
272 mm of precipitation, while the irrigated plots received an additional 20 mm of
irrigation on two dates (18 April and 29 May).
Soil moisture was assessed by TDR soil moisture probes (ECH2O, Decagon
Devices, Inc., Pullman, WA, USA). The sensors were positioned at a depth of 15 cm
in irrigated and non-irrigated soils.
Above ground biomass sampling was performed three times (at weeks 19, 21 and
23 of the year 2012). The fresh biomass was put into plastic bags, immediately
weighed, and then oven dried at 75 ºC for 24 h. The shoot fresh biomass (FB) and the
shoot dry biomass (DB) (g m-2) were recorded. The plant samples were chopped and
the N content (% dry weight) was measured by the standard Kjeldahl method in
laboratory. Crop yield and final biomass were also recorded during the harvesting
time.
Digital images of winter wheat canopy were acquired by a Canon camera model
EOS 550D with a resolution of 18.0 megapixels. Medium resolution of the camera
was used. The resulting images had a size of 3456 × 2304 pixels at Program AE
shooting mode of the camera. The camera was set to automatically adjustment f-stop
and shutter speed, however, focus was set manually. The colour images were
recorded in JPEG format and downloaded to a desktop computer for subsequent
processing.
The images were taken looking vertically downward from a height of 1.8 m,
which resulted in a rectangular area of 1.5 × 1.0 m on the ground. The photos were

75
recorded at different plant growth stages including: stem elongation, booting stage,
Inflorescence emergence, heading, flowering, and development of fruit.
To extract crop coverage and colour indices from the digital images, image
processing was performed using MATLAB software (Version 7.13, R2011b,
Mathworks Company). For segmentation of the green plant against background, a by
a mask (M) (binary image) was derived from the difference between green (G) and
the red (R) band of each image together with the threshold t:
⎧ 1 for ( G − R ) ≥ t
M = ⎨ (1)
⎩0 for ( G − R ) < t
Crop coverage (CC) was defined as the proportion of plant pixels in an image:
∑M
CC = (2)
n⋅m
where n and m are number of rows and columns of pixels.
Various colour indices were obtained from plant part of the images defined by the
mask M:
Rm = R*M; Gm = G*M; Bm = B*M; GMR = Gm – Rm; GMB = Gm – Bm;
RMB = Rm – Bm; NGMR = (Gm – Rm)/(Gm + Rm); NGMB = (Gm – Bm)/(Gm +
Bm); NRMB = (Rm – Bm)/(Rm + Bm).
where R, G and B are the intensity levels of the red, green and blue channels,
respectively. The values were then averaged for each image.
Hue, Saturation (SAT) and Intensity (INT) were also calculated according to Tang
et al. (2003):
(3)

(4)

(5)

(6)

The data obtained from the measurements and the image processing were
analyzed using analysis of variance (ANOVA) and the means were compared at 5%
level of significance using the Tukey range test in SAS software (version 9.3, SAS
Institute, Inc., Cary, N.C., USA). Regression and correlation analysis were done
using MATLAB software (Version 7.13, R2011b, Mathworks Company).

2 Result and discussion

The statistical analysis results indicated that there were strong significant
differences among the N supply levels and between the irrigation regimes in the case
of crop yield and final straw of the crop (p<0.01). The differences of N supply levels
for fresh and dry biomass and also plant N content were highly significant (p<0.01)

76
in all three times of biomass sampling. However, the differences of irrigation regimes
for these crop properties were mostly insignificant (p>0.05).
According to the soil water retention curve for sand, a volumetric water content of
10% was considered as field capacity and about 3% as permanent wilting point
(Ehlert, 1996). Results of soil water content showed that the lowest readings (around
permanent wilting point) were observed between days 235 and 245 after sowing
(weeks 22 and 23 of the year) for non-irrigated part of the trial field. During this
period, leaf rolling was also observed in the plants of non-irrigated part. Therefore, in
spite of having a relatively wet vegetation season for the year 2012 in northeast
Germany, a mild drought stress was observed.
Statistical analyses of the data showed that the effect of nitrogen supply on all the
colour indices (except Bm) was significant at 1 % level for all the growth stages
considered. In addition, the analyses showed that among the indices, Rm, RMB,
NRMB, Hue and INT were less sensitive to water supply.
Time course of the crop coverage (as an example of the colour indices) is shown
in
Fig. 1.
Crop Coverage (CC)

Week of the year (w)

Fig. 1. Time course of crop coverage for irrigated (Irr) and non-irrigated (NIrr) wheat crops
growing under 4 levels of nitrogen supply (N0=0, N1=60, N2=120, N3=240 kg N/ha)

According to the statistical analyses and time course of the colour indices, all the
indices were affected by both N and water supplies. However, Rm, RMB, NRMB,
Hue and INT were less sensitive to water supply. Li et al. (2010), Wang et al. (2013)
and Lee and Lee (2013) showed that CC estimations obtained by digital cameras
were good indicators for detection of nitrogen status in wheat and rice.
The three above studies did not consider effect of other plant stresses such as
drought stress on the results. Based on the results obtained in the current study, the
values of CC and other indices can be affected by drought stress. Therefore, in the

77
case of drought stress, these indices can become less reliable for site-specific N
management.
The indices Rm, RMB and Hue were less affected by water supply and showed
high differences among nitrogen supply levels. Thus, they have a potential to be
good indicators for detection of nitrogen deficiency.
As it is seen in Table 1, CC, NGMR and Hue had a strong positive correlation,
while Rm, Gm, GMB, RMB, NGMB, NRMB, SAT and INT had a high negative
correlation with the direct measurements of the agronomy parameters at all three
times of sampling. Bm and GMR had a weak correlation with these parameters.
Among the colour indices, the strongest correlations with N content, FB and DB
were obtained for Rm (rho=-0.926), Rm, GMB and RMB (rho=-0.953), and GMB
and RMB (rho=-0.956), respectively (Table 1).
Table 1. Spearman’s rho for correlation of direct and indirect measurements of winter wheat

10.05.2012 25.05.2012 08.06.2012
N content FB DB N content FB DB N content FB DB
variable
(%) (g m-2) (g m-2) (%) (g m-2) (g m-2) (%) (g m-2) (g m-2)
N content 1 0.915** 0.897** 1 0.594* 0.591* – – –
** ** * **
FB 0.915 1 0.994 0.594 1 0.979 – 1 0.994**
DB 0.897** 0.994** 1 0.591* 0.979** 1 – 0.994** 1
** ** ** * ** **
Height 0.918 0.950 0.930 0.575 0.905 0.936 – 0.885** 0.872**
CC 0.912** 0.944** 0.938** 0.815** 0.891** 0.876** – 0.915** 0.929**
** ** ** ** ** ** **
Rm -0.926 -0.953 -0.950 -0.826 -0.885 -0.868 – -0.851 -0.853**
Gm -0.918** -0.935** -0.935** -0.841** -0.897** -0.882** – -0.828** -0.829**
ns * * ns ns ns ns
Bm -0.468 -0.529 -0.515 0.144 -0.026 -0.029 – 0.328 0.312ns
GMR 0.253ns 0.209ns 0.215ns 0.315ns 0.326ns 0.312ns – 0.679** 0.674**
** ** ** ** ** ** **
GMB -0.885 -0.953 -0.956 -0.812 -0.868 -0.871 – -0.835 -0.824**
RMB -0.885** -0.953** -0.956** -0.791** -0.891** -0.897** – -0.859** -0.847**
NGMR 0.915** 0.935** 0.924** 0.844** 0.862** 0.832** – 0.862** 0.876**
** ** ** ** ** ** **
NGMB -0.853 -0.909 -0.918 -0.812 -0.812 -0.821 – -0.791 -0.779**
NRMB -0.871** 0.935** -0.935** -0.809** -0.862** -0.868** – -0.797** -0.791**
** ** ** ** ** ** **
Hue 0.871 0.935 0.935 0.797 0.894 0.906 – 0.850 0.844**
SAT -0.865** 0.929** -0.932** -0.818** -0.859** -0.862** – -0.800** -0.794**
** ** ** ** ** ** **
INT -0.894 -0.935 -0.932 -0.824 -0.879 -0.859 – -0.700 -0.712**
ns
: No significant difference; **: Significant at the 0.01 level; *: Significant at the 0.05 level; –: Data not available

The colour indices were used to develop regression models for predicting plant
fresh and dry biomasses, and also N content.
Performance of the indices for predicting the plant parameters were near to each
other as demonstrated by the r2 and RSME of the equations. However, for predicting
N content, the index CC (r2=0.94) presented the best relation (exponential). The Rm
showed the weakest quality among the indices to relate the three plant parameters.

78
References

1. Clay, D.E., Kim, K.I., Chang, J., Clay, S.A. and Dalsted, K. (2006)
Characterizing water and nitrogen stress in corn using remote sensing. Agron. J.
98, 579–587.
2. Ehlert, D., H.-J. Horn, and R. Adamek. (2008) Measuring crop biomass density
by laser triangulation. Computers and Electronics in Agriculture 61(2):117-125.
3. Ehlert, W. (1996) Wasser in Boden und Pflanze. Ulmer: Stuttgart, Germany. 272
pp. (Water in soil and plant)
4. Lee, K.J. and Lee, B.W. (2013) Estimation of rice growth and nitrogen nutrition
status using color digital camera image analysis. European Journal of Agronomy,
48: 57–65.
5. Li, Y., Chen, D., Walker, C.D. and Angus, J.F. (2010) Estimating the nitrogen
status of crops using a digital camera. Field Crops Research, 118(3): 221 227.
6. Limbrunner, B. and Maidl, F.X. (2007) Non-contact measurement of the actual
nitrogen status of winter wheat canopies by laser-induced chlorophyll
fluorescence. In Proc. of the 6th European Conf. on Precision Agriculture, 173–
179. J.V.
7. Rorie, R.L., Purcell, L.C., Mozaffari, M., Karcher, D.E., King, C.A., Marsh, M.C.
and Longer, D.E. (2011) Association of “Greenness” in Corn with Yield and Leaf
Nitrogen Concentration. Agronomy Journal, 103(2): 529–535.
8. Sakamoto, T., Gitelson, A.A., Nguy-Robertson, A.L., Arkebauer, T.J., Wardlow,
B.D., Suyker, A.E., Verma, S.B. and Shibayama, M. (2012) An alternative
method using digital cameras for continuous monitoring of crop status.
Agricultural and Forest Meteorology, 154–155: 113–126.
9. Sui, R. and J.A. Thomasson. (2006) Ground-Based Sensing System for Cotton
Nitrogen Status Determination. Transactions of the ASABE 49(6):1983−1991.
10. Tang, L., Tian, L.F. and Steward, B.L. (2003) Classification of broad leaf and
grass weeds using Gabor wavelets and an artificial neural network. Transactions
of the ASAE 46, 1247–1254.
11. Tremblay, N., Wang, Z., Ma, B-L., Belec, C. and Vigneault, P. (2009) A
comparison of crop data measured by two commercial sensors for variable-rate
nitrogen application. Precision Agriculture, 10, 145–161.
12. Wang, Y., Wang, D., Zhang, G. and Wang, J. (2013) Estimating nitrogen status
of rice using the image segmentation of G-R thresholding method. Field Crops
Research 149(0):33-39.