State of the Art on Current trends on Neural Networks and Normalized Difference Vegetation Index (NDVI)

This report discusses the concept of artificial neural networks and their potential in various fields, including cardiovascular disease diagnosis. It also explores the applications of Normalized Difference Vegetation Index (NDVI) in agricultural studies. The report covers feedforward and feedback neural networks, and the differences between them. It also discusses the advantages of ANNs over statistical models. The NDVI is explained in detail, including its applications in identifying live green vegetation.

State of the Art on Current trends on Neural
Networks and Normalized Difference
Vegetation Index (NDVI)

ARTIFICIAL NEURAL NETROWK
1
Introduction
The artificial neural network is defined
as the system which is enthused by the
biological networks. It constitutes the
animal brains. It is one of the major
techniques which is used in the machine
learning and neural is the brain system
which is intended to the replicate the
path that consumer learns (Khashei, and
Bijari, 2010). It is observed that term
artificial play a significant role in the
field of neural networks which contains
input and output layers. The major goal
of this paper is to highlights the concept
of artificial neural network and
evaluates the role of artificial on current
trends of neural networks and
normalized difference vegetation index
(NDVI). The term NDVI refer to the
normalized difference vegetation index
which is the simple graphical pointer
and it is used to analyse the wireless
dimensions. The artificial neural
network uses this technique in order to
analyse the data and information and it
always ranges from the -1 to +1. This
report is divided into major two
sections such as the state of the artificial
on current trend on neural networks
and the concept of NDVI.
Artificial in the neural networks
The artificial is one of the major tools
which are utilized in the machine
learning and it has potential to
determine the overall patterns which
are more complex for the human
programmer to extract and teach the
machine to recognize (Al-Shayea, 2011).
The neural network is also called the
perceptron and it is a very common part
of artificial intelligence. It is observed
that the neural networks involve
various kinds of layers that extract
modern elements until it can recognize
what it is observing for. There are
various kinds of neural networks where
every type comes with their own
specific utilization and level of
complexity. Feedforward neural
network is the better type of network
that uses the artificial technique is used
for gathering and storing the data or
information and signal travels in only
single direction from the input terminal
to the output terminal.
It is analysed that the basic concept of
the artificial neural network is
completely based on the faith which
working of the individual intelligence by
assembly the right relations can be
imitated with the help of wires and

ARTIFICIAL NEURAL NETROWK
2
silicon (Kuo, Wang, and Tien, 2010). The
individual brain is composed of 86
billion nerve cells that known as the
neurons and stimuli from the input
terminals are accepted with the help of
dendrites. Such kind of input signals
produce the electric impulses that travel
with the help of neural networks and
artificial technique is used in order to
maintain the speed of data transmission.
Moreover, the neural network transfers
the information from one neuron to
another in order to control and manage
the issues or problems (Guresen,
Kayakutlu, and Daim, 2011).
Types of artificial neural networks
There are major two kinds of ANN
identified which are described below:
 Feed Forward network
 Feedback neural network
Feedforward neural network
In such kind of artificial neural network,
the data or information flow is
unidirectional that means the signal
travel in only one direction. Moreover, a
unit transfers relevant information to
another unit from which it does not get
any information and it does not contain
any kind of feedback loop in the
network. Mainly, such kind of artificial
neural networks is used for the pattern
generation, classifications of data and
recognition of the pattern. A recent
study evaluated that feedforward neural
networks have fixed input and outputs
terminals (Magnier, and Haghighat,
2010). A feedforward neural system
(FFNN) is an AI arrangement calculation
that made up of composed in layers that
are like human neuron preparing units.
In FFNN every unit in a layer identifies
with the various units in the layers.
These layers associations with units are
not all equivalent on the grounds that
every association can have an alternate
weight or quality. Loads of the system
associations measure the potential
measure of the information of the
system. Additionally, NN units are
known as hubs. The data handling in the
system includes information passage
from the info units and goes through the
system, spilling out of one layer to
another layer until it gets to the yield
units (Valipour, Banihabib, and
Behbahani, 2013). At the point when NN
work ordinarily, that is the point at
which it went about as a classifier, at
that point, there will be no input
between layers [30]. In FFNN, data
transmitted just one way that is from
the information hubs to the concealed
hubs, assuming any, and afterwards to

ARTIFICIAL NEURAL NETROWK
3
yield hubs. With this conduct, they are
called feedforward neural systems.
Feedback artificial neural networks
It refers to the artificial network that
contains the feedback loops in the
networks. Mainly such kinds of
networks are utilized for the content
addressable memories and artificial
technique monitor the data or signals
travel from one network to another. The
main difference between the
feedforward network and feedback
network is that feed forward ANN does
not involve any kind of feedback process
but feedback ANN contains the feedback
loop. Moreover, it is the bidirectional
process which travels in both direction
but it has fixed input and output signals.
Feed-in reverse NN can be connected to
errands like un-division, and example
acknowledgment (associated
penmanship acknowledgment). Feed-in
reverse neural system application zones
incorporate numerical evidence, seismic
information fitting, drug, science,
designing, characterization, work
estimation, and time-arrangement
expectation, and so on. In feedback NNs
or back propagation, associations
between hubs delivered a planned chart
in succession (Mellit, and Pavan, 2010).
The planned chart in grouping permits
input NNs to exhibit dynamic
earthbound conduct for a period
arrangement. Precedents are Kohonen's
self arranging map and intermittent
neural system (RNN).
State of the Art on Current trends on
Neural Networks
Man-made brainpower (AI) has existed
over numerous decades, and the field is
wide. Simulated intelligence can be seen
as a set that contains AI (ML), and
profound learning (DL). The ML is a
subset of AI, then, DL, thus, a subset of
ML (Govindaraju, and Rao, 2013). The
refinement between profound learning
and neural systems like feedforward
NNs and feed in reverse NNs lies in their
trademark (Chen, et al., 2011). Profound
learning has progressively complex
methods for associating layers,
additionally has a greater number of
neurons check than past systems to
express complex models, all the more
likewise with all the more figuring
capacity to prepare and additionally has
programmed extraction of the element.
An ANN uses information skill to
represent a biologic neural system with
the help of both structurally and
functionally process. Like its biologic
counterpart is an extremely unified

ARTIFICIAL NEURAL NETROWK
4
server that involves many processing
units. Neural systems give a way to deal
with information examination that
utilizes mathematic design
acknowledgment. Artificial neural
systems learn complex communications
among sources of info and produce a
yield as they train on a known
arrangement of information (Dagli,
2012). Counterfeit neural systems
utilize this preparation to create a yield
for new sources of info. Artificial neural
systems are accordingly fit for
distinguishing relations in information
not clear to person examination.
Moreover, an artificial system creates
"scientific decision" in light of past
learning and experience. The art of
artificial neural systems has been
connected in endeavours to take care of
a few issues in 4 universal territories of
a cardiovascular drug. These are
coronary supply route ailment,
electrocardiography, heart picture
examination and cardiovascular
medication dosing.
The counterfeit neural system reports
identifying with coronary corridor
illness might be isolated into those
tending to the universal finding and
action of coronary supply route illness
and those explicitly identified with
intense myocardial localized necrosis.
The ECG examines are worried about
general 2CG understanding, arrhythmia
distinguishing proof and arrhythmia
restriction. Few investigations (5-7)
have additionally been distributed that
relate counterfeit neural systems to
incidental territories important to the
cardiovascular expert (Basu,
Bhattacharyya, and Kim, 2010).
ANNs has noteworthy favourable
circumstances over measurable models
when both are moderately thought
about. In ANN models there are no
suppositions about information
properties or information dispersion.
Along these lines, ANNs are increasingly
helpful in the commonsense application.
Additionally, not at all like some
measurable models that require certain
speculation for testing, ANN models
don't require any theory. ANNs are truly
adaptable, information decrease models,
including nonlinear relapse models, and
discriminant models. All the more
likewise, dissimilar to the help vector
machine, outrageous learning machine,
and irregular backwoods, ANNs are
more blame tolerant. That is, they can
deal with fragmented information and
clamour, they can take care of non-
straight issues, Also, prepared ANNs,

ARTIFICIAL NEURAL NETROWK
5
can sum up at rapid and make
expectations. Moreover, ANNs are
versatile when generally contrasted
with the help vector machine,
extraordinary learning machine, and
arbitrary woods. For the most part,
displaying is the way toward
introducing a true wonder or article as a
lot of numerical articulation. Be that as
it may, the neural system demonstrating
is a procedure speaking to the manner
in which the sensory system capacities.
That is, a NN is an improved portrayal of
how the man cerebrum forms data. Its
capacity by reproducing many
interconnected preparing units that
look like thought forms of neurons. It is
the most widely recognized
enhancement worldview utilized in
upgrading a neural system. As of late
slope drop utilized in refreshing loads in
a NN show that is refreshing and
changing the model's parameters
toward a path to limit the Loss work.
Normalized Difference Vegetation
Index (NDVI)
NDVI produce plants for identifying the
contrast between close infrared and red
light. In order to identify the width of
emerald on land, scientists must watch
the unique wavelength of noticeable and
shut infrared daylight that reflected
with the help of the plants (Ticknor,
2013). The cell structure of the leaves
unequivocally reflects infrared light
mainly from 0.7 to 1.1 μm. The NDVI
refers to the arithmetical pointer which
utilizes the noticeable and closes
infrared groups of the electromagnetic
range and it is used to investigate
remote estimations and assess whether
the intention being contains live green
vegetation or not.
The NDVI has discovered a major
application in the field of vegetative
investigations where it is used to
appraise harvest yields, field execution,
and so on. It is frequently specifically
identified with other key parameters,
for example, % of earth spread,
photosynthetic action of the place,
outside irrigate, leaf region record and
the determination of biomass (Guo, et
al., 2012). This record portrays quickly
month to month NDVI Products
acknowledged by utilizing the Oceansat-
2 Ocean Shading Monitor (OCM2) Global
Area Coverage (GAC) sensor. This is
esteem included item from OCM2 whose
phantom groups are initially intended
for sea shading recovery applications.
Be that as it may, the two-day transient
goals with a wide swath of 1420 km and
high radiometric goals of 12 bits for

ARTIFICIAL NEURAL NETROWK
6
each pixel from the OCM2 sensor can
give helpful data to horticultural
applications. In this record, a concise
preparing plot for understanding the
NDVI and albedo items at spatial goals
of 8km is displayed (Lopes, and
Reynolds, 2012). Since the NDVI is a
middle item created for the VF items
acknowledgment, this information is
additionally given as an item to a given
month.
The current state of Artificial on NDVI
NDVI is broadly utilized as a measure of
plant efficiency, a pointer of the
ecological assets and a determinant of
biological system probability. As the
most broadly utilized vegetation file that
reflects vegetation properties, NDVI is
determined from the noticeable what's
more, close infrared light reflected by
vegetation. Solid vegetation assimilates
a large portion of the obvious light
achieving it, mirroring a substantial
segment of the close infrared light
(Eastman, et al., 2013). Along these
lines, NDVI values are a lot higher for
vegetation as contrasted with
uncovered soil, cloud and water.
NDVI and Splendor Temperature (BT)
portray sound or undesirable
vegetation. The Vegetation Condition
Index and the (TCI) portray the
dampness and warm states of a specific
territory, separately, got from NDVI and
BT. For guaranteed territory, the VCI
and TCI values have a solid connection
between these files and agrarian harvest
yields amid the basic time of yield
development (Roy, et al., 2016). A high
estimation of VCI compares to
unstressed and solid vegetation while a
low TCI esteem compares to
undesirable vegetation due to warm
pressure. A few distinct methods have
been accounted for in writing for
investigation of satellite picture, for
example, NDVI Artificial Network and
Satellite picture Contrast increment that
uses the (DWT) and Singular amount
Decomposition.
In the field of ANN system procedure,
the execution factors of the feed-
forward are defined as the loads. It is
observed that they fluctuate by which
the anticipated yield is near evident
yield esteem comparing to the info
esteems. This system depends on major
2 kinds of calculation: directed and
unsupervised. In last few years there are
many other methods and techniques
developed for example, DWT and SVD.
In all such technique, the target work is
highlight withdrawal from the

ARTIFICIAL NEURAL NETROWK
7
multiband information (Pettorelli, et al.,
2011).
A approach based on the NDVI
In this research study, the NDVI method
is utilized for extricating the different
highlights from a picture and mainly
many business industries uses this
technique for extricating data.
Vegetation spread is very common
biophysical pointer to soil disintegration
that can be evaluated with the help of
vegetation lists received from the
satellite pictures. Vegetation records
enable us to portray the circulation of
vegetation and soil dependent on the
trademark reflectance examples of
green vegetation (Turvey, and Mclaurin,
2012). The NDVI is the best arithmetical
pointer which can be involved in the
system in order to break down the
remote sensing estimation, from a
remote process. The NDVI is inspired
with the help of perception vegetation.
It is analysed that it is the contrast
between red band and NIR and it ought
to be larger for prominent chlorophyll
width. Mainly, such kind of process
collects the contrast from satellite and
standardizes in order to balance the
affect of uneven enlightenment. For
example, shadows of slopes. It is also
observed that a pixel by pixel identify
the estimation of the red band with the
help of evaluation process and NIR band
(Bhandari, Kumar, and Singh, 2012).
Low estimation of NDVI (0.1 and
beneath) compare to fruitless zones of
shake, sand, or snow.
Experimental Result and Discussion
In this research report, the NDVI esteem
is fluctuated from value 0.1 to 0.5 that
shows that the best highlights can be
removed from the destination by using
NDVI. The most reduced qualities are
observed on the small vegetated soils
and probably in light of the fact that
reflect from the dirt, and analysis less
amount in NIR and large qualities in red
spectrum; consequently, the NDVI
esteems are low. Vegetation stays large
NDVI values due to the accessibility of
water in the dirt. Despite what might be
expected, when soil water accessibility
diminishes, because of any ecological
reason (worry by water shortfall); the
green vegetation will in general vanish,
at that point the estimations of NDVI
diminishes. Satellite information was
exhibited by week by week Normalized
Contrast Vegetation Index (NDVI) and
Brightness Temperature (BT) gathered
from the NOAA Global Vegetation Index
(GVI) dataset from 1998 to 2011.

ARTIFICIAL NEURAL NETROWK
8
The GVI items are broadly utilized as
satellite information items for checking
the energy and thickness of green
vegetation, evaluating crop grounds,
distinguishing plant pressure and
characterizing land spread (Rhew, et al.,
2011). The GVI was created from
AVHRR-watched reflectance/outflow in
obvious (VIS, 0.58-0.68 m, Ch1), closeμ
infrared (NIR, 0.72-1.00 m, Ch2).μ
Accordingly, NDVI turns out to be
substantial and BT stays little for sound
Universal Journal of Agriculture and
Forestry 2018, 8(1): 16-25 19
vegetation; this situation turns around
for unfortunate vegetation. Vegetation
wellbeing records Vegetation Condition
Index (VCI) and Temperature Condition
Index (TCI) of a specific zone were
created from its NDVI and BT utilizing
some preparing and calculations. An
Artificial Neural Network (ANN)
recreation instrument was used to build
up an expectation show for this
examination.
A nonlinear auto backward with
exogenous (outside) input, or on the
other hand NARX time arrangement,
Neural Network was utilized for right on
time expectations and the model
advancement. An ANN alludes to an
interconnected gathering of "hubs"
known as "neurons", "neuroses",
"preparing components" or "units" - the
fundamental building square to shape a
system which works in the same way as
human focal sensory systems. In a
general sense, ANNs are alluded to as a
"black box" approach. Comparative as
'discovery' innovation, ANN has the
capacity to learn input– yield connection
via preparing the information to create
the normal yield (Yin, et al., 2012). NN
preparing is achieved so as to play out a
specific assignment by changing the
weight and predisposition esteems
between the components. As a result of
changing or preparing the system,
specific info (target) prompts a
particular yield. It exhibits that neural
system performs the alteration process
until the yield coordinates the objective
based on an examination between them.
Proposed Model
From system re-enactment, it was
discovered that the complete number of
weight and inclination esteems comes
up to 161. Of the 161, there were 9
predisposition qualities, in light of the
fact that for our proposed show we
utilized 8 neurons for the shrouded
layer and 1 (one) a neuron for the
yielding layer, with every neuron having
as it were one inclination esteem (Zhu,

ARTIFICIAL NEURAL NETROWK
9
et al., 2013). The staying 152 qualities
are weight esteems. The grids for loads
and inclinations from MATLAB
recreation are outlined beneath, where
b alludes to the predisposition, IW is the
input weight and LW indicates the line
weight. The determination of neurons
for the concealed layer is one of the
primary difficulties in planning an ANN.
The quantity of neurons in the shrouded
layer impacts the learning ability,
multifaceted nature, speculation, and
precision of the issue.
Thusly, the choice of a fitting number of
neurons for the concealed layer does the
trick to guarantee the intricacy of the
issue. Nonetheless, if this turns out to be
excessively expansive, it might prompt
the over-fitting of the preparation set
and loss of speculation capacity.
Consequently, experimentation
technique is the perfect method for
choosing concealed neurons (Park, et al.,
2010). The proposed model, which
mirrors a period arrangement
expectation demonstrate, is made out of
two kinds of information sources: (a)
week by week VCI and TCI time
arrangement information and (b) Boro
rice yield time arrangement
information. It is observed that changing
sun peak edge along the year can make a
counterfeit occasional minor departure
from the NDVI when determined with
best of environment reflectance.
Additionally, at high review points, the
NDVI turns out to be very low (blurring
impact at the edge of the MSG plate).
To keep this, an amendment dependent
on a Bidirectional Circulation
Reflectance Function (BDRF) is
connected to Rnir and Rvis to change
over them to zero surveys furthermore,
sun apex edges. The BDRF work has
been determined for each SEVIRI
noticeable channel and each surface
sort, by upsetting a Ross/Roujean part
approach. Changing sun peak point
along the year can make a counterfeit
occasional minor departure from the
NDVI when determined with best of air
reflectance. Likewise, at high review
points, the NDVI progresses toward
becoming very Various strategies have
been connected with remotely detected
information for assessing prairie
biomass, furthermore, these strategies
incorporate various relapse
investigation, K-closest neighbour,
counterfeit neural arrange (ANN), and
bolster vector machine (SVM) methods.
Artificial neural system, in light of its
capacity to deal with complex nonlinear
capacities, has been effectively utilized

ARTIFICIAL NEURAL NETROWK
10
in different land perception applications
including biomass estimations, land
characterizations, and land change
location. In any case, the ANN model
would not deliver ensured exactness at
the point when there are insufficient
examples and it is anything but difficult
to over-fit the information when there
are intemperate quantities of tests.
Bolster vector machine depends on the
basic hazard minimization standard and
isgood at taking care of commonsense
issues including little quantities of
preparing tests, nonlinearity, high
quantities of measurements, and nearby
minima; accordingly, SVM is viewed as a
decent option to ANN.
Conclusion
This report is completely based on the
artificial neural networks and students
can enhance their experience in the area
of artificial neural networks. It is
concluded that the artificial technique
play a major job in the neural networks
and feedback ANN contain the feedback
loop. Moreover, the researcher also
discussed the fundamental concept of
the NDVI technique and evaluated the
current state of artificial technique on
the NDVI. This report explained the
working principle of artificial neural
networks with their types and the
current state of the artificial on neural
network. Moreover, the researcher also
provided complete information about
the neural networks and the NDVI
technique.

ARTIFICIAL NEURAL NETROWK
11
References
Khashei, M. and Bijari, M., (2010) An
artificial neural network (p, d, q) model
for timeseries forecasting. Expert
Systems with applications, 37(1), pp.479-
489.
Al-Shayea, Q.K., (2011) Artificial neural
networks in medical
diagnosis. International Journal of
Computer Science Issues, 8(2), pp.150-
154.
Kuo, R.J., Wang, Y.C. and Tien, F.C.,
(2010) Integration of artificial neural
network and MADA methods for green
supplier selection. Journal of cleaner
production, 18(12), pp.1161-1170.
Guresen, E., Kayakutlu, G. and Daim,
T.U., (2011) Using artificial neural
network models in stock market index
prediction. Expert Systems with
Applications, 38(8), pp.10389-10397.
Magnier, L. and Haghighat, F., (2010)
Multiobjective optimization of building
the design using TRNSYS simulations,
genetic algorithm, and Artificial Neural
Network. Building and
Environment, 45(3), pp.739-746.
Valipour, M., Banihabib, M.E. and
Behbahani, S.M.R., (2013) Comparison
of the ARMA, ARIMA, and the
autoregressive artificial neural network
models in forecasting the monthly
inflow of Dez dam reservoir. Journal of
hydrology, 476, pp.433-441.
Govindaraju, R.S. and Rao, A.R. eds.,
(2013) Artificial neural networks in
hydrology (Vol. 36). Springer Science &
Business Media.
Mellit, A. and Pavan, A.M., (2010) A 24-h
forecast of solar irradiance using
artificial neural network: Application for
performance prediction of a grid-
connected PV plant at Trieste,
Italy. Solar Energy, 84(5), pp.807-821.
Chen, C., Duan, S., Cai, T. and Liu, B.,
(2011) Online 24-h solar power
forecasting based on weather type
classification using an artificial neural
network. Solar energy, 85(11), pp.2856-
2870.
Dagli, C.H. ed., 2012) Artificial neural
networks for intelligent manufacturing.
Springer Science & Business Media.
Basu, J.K., Bhattacharyya, D. and Kim,
T.H., (2010) Use of artificial neural
network in pattern
recognition. International journal of

ARTIFICIAL NEURAL NETROWK
12
software engineering and its
applications, 4(2), pp. 12-14.
Ticknor, J.L., (2013) A Bayesian
regularized artificial neural network for
stock market forecasting. Expert Systems
with Applications, 40(14), pp.5501-
5506.
Guo, Z., Zhao, W., Lu, H. and Wang, J.,
(2012) Multi-step forecasting for wind
speed using a modified EMD-based
artificial neural network
model. Renewable Energy, 37(1),
pp.241-249.
Pettorelli, N., Ryan, S., Mueller, T.,
Bunnefeld, N., Jędrzejewska, B., Lima, M.
and Kausrud, K., (2011) The Normalized
Difference Vegetation Index (NDVI):
unforeseen successes in animal
ecology. Climate Research, 46(1), pp.15-
27.
Eastman, J., Sangermano, F., Machado, E.,
Rogan, J. and Anyamba, A., (2013) Global
trends in seasonality of normalized
difference vegetation index (NDVI),
1982–2011. Remote Sensing, 5(10),
pp.4799-4818.
Bhandari, A.K., Kumar, A. and Singh, G.K.,
(2012) Feature extraction using
Normalized Difference Vegetation Index
(NDVI): A case study of Jabalpur
city. Procedia technology, 6, pp.612-621.
Zhu, Z., Bi, J., Pan, Y., Ganguly, S., Anav,
A., Xu, L., Samanta, A., Piao, S., Nemani, R.
and Myneni, R., (2013) Global data sets
of vegetation leaf area index (LAI) 3g
and fraction of photosynthetically active
radiation (FPAR) 3g derived from global
inventory modeling and mapping
studies (GIMMS) normalized difference
vegetation index (NDVI3g) for the
period 1981 to 2011. Remote
sensing, 5(2), pp.927-948.
Park, S.U., Choe, A., Lee, E.H., Park, M.S.
and Song, X., (2010) The Asian dust
aerosol model 2 (ADAM2) with the use
of normalized difference vegetation
index (NDVI) obtained from the
Spot4/vegetation data. Theoretical and
Applied Climatology, 101(1-2), pp.191-
208.
Rhew, I.C., Vander Stoep, A., Kearney, A.,
Smith, N.L. and Dunbar, M.D., (2011)
Validation of the normalized difference
vegetation index as a measure of
neighborhood greenness. Annals of
epidemiology, 21(12), pp.946-952.
Turvey, C.G. and Mclaurin, M.K., (2012)
Applicability of the Normalized
Difference Vegetation Index (NDVI) in
index-based crop insurance

ARTIFICIAL NEURAL NETROWK
13
design. Weather, Climate, and
Society, 4(4), pp.271-284.
Yin, H., Udelhoven, T., Fensholt, R.,
Pflugmacher, D. and Hostert, P., (2012)
How normalized difference vegetation
index (ndvi) trendsfrom advanced very
high resolution radiometer (AVHRR)
and système probatoire d’observation
de la terre vegetation (spot vgt) time
series differ in agricultural areas: An
inner mongolian case study. Remote
Sensing, 4(11), pp.3364-3389.
Roy, D.P., Kovalskyy, V., Zhang, H.K.,
Vermote, E.F., Yan, L., Kumar, S.S. and
Egorov, A., (2016) Characterization of
Landsat-7 to Landsat-8 reflective
wavelength and normalized difference
vegetation index continuity. Remote
Sensing of Environment, 185, pp.57-70.
Lopes, M.S. and Reynolds, M.P., (2012)
Stay-green in spring wheat can be
determined by spectral reflectance
measurements (normalized difference
vegetation index) independently from
phenology. Journal of experimental
botany, 63(10), pp.3789-3798.