Citrus Orchards Monitoring based on Remote Sensing and Artiﬁcial

Intelligence Techniques: A Review of the Literature

Abdellatif Moussaid

1,2

, Sanaa El Fkihi

and Yahya Zennayi

IRDA Laboratory, ENSIAS, Mohammed V University, Rabat, Morocco

ESAI department, MAScIR, Rabat, Morocco

Keywords:

Smart agriculture, artiﬁcial intelligence, citrus monitoring, remote sensing, machine learning, deep learning.

Abstract:

In recent years, with the emergence of new technologies, in particular artiﬁcial intelligence techniques and

remote sensing data, agriculture has become intelligent. These technologies have helped us to improve the

quality and quantity of yield, and to facilitate many difﬁcult tasks for farmers. In this paper, we will present

an extensive review of the techniques and themes used in the ﬁeld of agriculture in general, and citrus crop in

particular, through the realization of a bibliometric and bibliographic study based on several published articles

over the last years. Through an in-depth analysis of several works, we have found that there are several factors

that are very interesting in this ﬁeld. In fact, we have many parameters related to trees such as detection and

counting; canopy or crown size; tree location; detection of individual trees and missing trees, etc. We have also

the effect of vegetation indices such as normalized difference vegetation index(NDVI),normalized difference

red edge index(NDRE),modiﬁed chlorophyll absorption ratio index(MCARI), etc. Which are strongly corre-

lated with fruit production. In addition, monitoring tree health and water stress is very interesting. All these

factors and more can be obtained from high-resolution spectral images, using machine learning algorithms,

remote sensing techniques, and image processing. The purpose of this study is to explain how we can control

the situation of orchards to have better yield.

1 INTRODUCTION

In recent years, most scientiﬁc researchers in sev-

eral ﬁelds use artiﬁcial intelligence techniques in their

projects to get good results and to develop new ap-

proaches based on data. In fact, the ﬁeld of agriculture

presents a lot of data every season. These data can be

external such as climatic or internal such as soil analy-

sis and tree monitoring. Thus, many of these data can

be extracted from spectral images in several bands,

which makes it possible to provide information on

vegetation, water stress, etc. In the electromagnetic

spectrum, we ﬁnd several electromagnetic waves with

frequencies of different levels. In the ﬁeld of remote

sensing, there is the visible part which contains the

RGB bands. These bands are the normal images that

we can see with the naked eye. Apart from this part,

there is the SWIR(Short-wave infrared ) part and the

LWIR(long wavelength infrared) part, which present

other spectral bands like multispectral images with 8

bands in the SWIR and 6 bands in the LWIR in addi-

tion to the RGB and panchromatic bands. There are

also the hyperspectral images that contain hundreds

of bands and the ultraspectral images with thousands

of bands. The presence of several bands allows us

to have more information and get several factors to

improve the ﬁeld of agriculture, especially the citrus

crop which we are interested in. Since multispectral

or hyperspectral images provide interesting datasets,

whether imagery or digital, the presence of artiﬁcial

intelligence, especially machine learning algorithms

can help us predict the factors that will improve and

facilitate the agriculture monitoring.

The goal of this paper is to present a global vision

about the exploitation of spectral images as input data

for machine learning algorithms to extract and pre-

dict several factors which are very important to obtain

good yield and to facilitate several tasks for farmers.

The remaining of this article is structured as fol-

lows: section II gives an overview about the use of the

scientiﬁc mapping method to select the most useful

papers, and get an idea of the keywords most corre-

lated with our goals. Section III present a global syn-

thesis with comparisons on the factors and the works

which deal with the problematic of citrus tree moni-

toring based on remote sensing and/or artiﬁcial intel-

ligence. Finally, the conclusion is drawn in section

IV.

2 SCIENCE MAPPING

Figure 1: connected papers.

The goal of science mapping is to obtain a lot of

information and to select the best papers based on

citation, keywords, abstracts,etc. There are several

open source softwares used for scientiﬁc mapping.

In our case, we used SciMAT software (Cobo et al.,

2012)(open-source software tool developed to per-

form a science mapping analysis under a longitudinal

framework) to obtain several graphs based on list of

keywords with document number by each keyword,

papers citation and h-index. In our case, after ob-

taining and analyzing these graphs, we get a clear

vision of the papers which are strongly correlated to

our problem and which have a good quality. We also

get a good idea about the trends of artiﬁcial intelli-

gence techniques and remote sensing in agriculture,

especially the citrus cultivation. We also used an-

other software to obtain more documents correlated

to our problem. This software called connectedpa-

pers(LLC, ). It gives a graph composed of circles

with several links between them. The size of cir-

cles signify the number of citations. The thickness

of lines measures the correlation between the papers

based on keywords and abstracts. Finally, the visibil-

ity of the circles’ color indicates the age of the paper.

The graph in ﬁgure 1 is an example of graphs pro-

vided by connectedpapers software. The used key-

words are: ”smart agriculture, remote sensing, satel-

lite images, UAV images, spectral images, tree seg-

mentation, tree detection, tree canopy, tree crown, cit-

rus orchard monitoring, image processing, vegetation

index, water stress index, artiﬁcial intelligence, ma-

chine learning and deep learning”. The ﬁgure 1 shows

that we obtained a reasonable number of papers that

are correlated to the input paper and we can select the

best ones based on the citation, the correlation, and

the age.

3 CITRUS ORCHARD

MONITORING

The Citrus orchards need speciﬁc monitoring to give

good yield. In fact, this monitoring starts at the be-

ginning of the agricultural year and continues until

the harvest period, and concerns several parameters

that are very important to obtain good quality and

quantity of fruits. this monitoring also depends on

the phenological cycle of citrus as shown in ﬁgure 2

(Pons et al., 2012). In fact, all citrus species start with

the initiation phase when the small branch begins to

develop, the ﬂowering period with two phases from

the beginning of ﬂowers until full, after that we have

the fruit period with 4 phases: start with the initia-

tion of fruits, fruit development, coloration, and the

maturation phase when the fruits are ready to be har-

vested. Finally, after the harvest period, we have a

winter phase when the trees need rest. So, each citrus

orchard requires speciﬁc monitoring at each phase.

we have clear and well-known factors such as ir-

rigation and nutrition. So we all agree that without

these two parameters, we cannot get yield. The culti-

vation of citrus fruits requires fertilization which con-

tains the product necessary for each soil, and it also

needs good irrigation with then appropriate dosing

and at the best moment. Other important factors must

be under control as well. One of these latter is the

phytosanitary treatment against diseases that attack

citrus. Another important factor is the pruning op-

eration which is very important in each citrus orchard

to remove dead and broken branches to keep the best

size of trees. In addition, we have many very inter-

esting factors in each citrus orchard such as the cli-

mate (temperature, humidity, light ...), ﬂowering, and

also the canopy and the size of the trees. So all these

factors and others can be extracted from multispectral

and hyperspectral images. Through the use of artiﬁ-

cial intelligence techniques that exploit this data, we

can obtain an important follow-up of citrus orchards

during the agricultural year. In the next, we will give

more details about each factors.

Figure 2: phenological cycle of citrus.

3.1 Irrigation and Nutrition

At the beginning of each crop year, citrus orchards

need a soil and tree test. These analyzes cover several

factors such as phosphorus which is very important

for the transfer of energy and the transport of the prod-

uct of photosynthesis, potassium for the regulation

of osmotic pressure, nitrogen for tree growth, mag-

nesium for fruit ripening and calcium which is also

important for the strength of the branches. All these

chemical elements are very important, and through to

laboratory analysis, we can control them at each pe-

riod of time to maximize or minimize the phytosani-

tary dose required(Obreza and Morgan, 2008). Along

with nutrition, there is irrigation, which is one of the

most important factors for producing good yield of

quality citrus. In fact, good irrigation consists of

knowing the quantity of water depending on differ-

ent areas in an orchard. Therefore, irrigation has a

direct impact on the health of trees as well as on the

yield, size and quality of fruits(Zarco-Tejada et al.,

2012). Thomas A. and al (Obreza and Morgan, 2008)

conducted several tests over different months on or-

ange trees and their soil in Florida. They discov-

ered that during the agricultural year, there are chemi-

cal compositions decrease such as potassium, magne-

sium, and nitrogen, and others increase like calcium.

So it is necessary to add other products or do what-

ever is necessary to balance these main factors. Also

in their work, they presented a recommendation on

the quantity of each chemical composition needed by

the soil and trees.

Salvatore Meli and al (Meli et al., 2002), say that

the period of good irrigation in the Mediterranean re-

gion begins in May and after a month the soil shows

an increase in microbial parameters.

However, soil and plants analysis in the labora-

tory is expensive and cannot be done several times a

year. There are other methods for determining wa-

ter stress and nutritional status based on imagery. For

example, multispectral and hyperspectral images pro-

vided by satellites or drones can give several signs

to vegetation and water stress which are very impor-

tant for analyzing the condition of the parcels and

for adding the necessary irrigation or nutrition. Cur-

rently, there are several projects for monitoring veg-

etation and water stress in orchards based on im-

agery and they are giving good results. Indeed, in

(Zhang et al., 2019) authors. It used multispectral im-

ages collected by an UAV to mapping water stress for

maize. In the last work, authors calculate a crop wa-

ter stress index (CWSI) by extracting several vegeta-

tion indices such as normalized difference vegetation

index (NDVI))(Rouse Jr, 1974) , renormalized dif-

ference vegetation index (RDVI)(Zarco-Tejada et al.,

2013), soil-adjusted vegetation index (SAVI)(Jackson

et al., 1981) , optimization of soil-adjusted vegeta-

tion index (OSAVI)(Haboudane et al., 2002) , and

transformed chlorophyll absorption in reﬂectance in-

dex (TCARI)(Haboudane et al., 2002).An example of

a mapping CWSI, is given in ﬁgure 3. It shows dis-

tribution of water stress; hence, he/she can decide the

accurate needed dosing water of each zone. Actually,

the CWSI can be calculated based on weather condi-

tions, but authors of (Zhang et al., 2019) demonstrate

that the CWSI extracted from multispectral images is

more efﬁcient and cheaper.

The formula of CWSI is:

CW SI =

T (canopy) − T (wet)

T (dry) − T (wet)

(1)

where T(canopy) is the surface temperature of the

canopy, and T(wet) and T(dry) are reference surfaces

that are completely wet or dry to simulate maximum

and minimal leaf transpiration under the exposed en-

vironmental conditions.

Figure 3: Water stress index map

Another very important work (Gonzalez-Dugo et al.,

2013) explores the UAV thermal imagery to assess the

variability in the water status. The results of this work

is illustrated in ﬁgure 3 which consists of a map that

shows the distribution of CWSI by color. Thus, this

map helps us to deﬁne the regions that need water eas-

ily.

3.2 Phytosanitary Treatment

Citrus diseases are fatal for some species. However,

there are useful treatments for these diseases. In the

normal case, there is a follow-up by periodic visits to

the orchards in order to monitor the health of trees and

to initiate the necessary treatment at the beginning of

the disease. Nonetheless, if there are large parcels, it

is difﬁcult to be controlled by humans, and sometimes

there is a delay in the detection of the diseases, which

negatively inﬂuences the fruits. For this reason, the

use of remote sensing images to detect a disease or

its limitations is very interesting. Actually, the use of

remote sensing images in agriculture is not a new ﬁeld

in the sense that many projects have been conducted

by it. In the next, we will give some examples:

Several spectroscopes and imaging techniques

have been studied to detect diseases such: (Bravo

et al., 2004; Moshou et al., 2004; Chaerle

et al., 2007) that are based on ﬂuorescence imag-

ing technique,(Sevick-Muraca and Paithankar, 1999;

Shafri and Hamdan, 2009; Qin et al., 2009) used spec-

tral imaging.while (Spinelli et al., 2004) and (Purcell

et al., 2009) use infrared spectroscop, and (Yang et al.,

2007; Delalieux et al., 2007; Chen et al., 2007) use the

visible/multiband spectroscopy.

The idea of remote sensing is to provide images

with several bands with frequencies that exceed the

visible spectrum and is able to ﬁnd some differences

in the structure of plants and trees in order to decide

if there are changes by diseases. Sometimes these

changes occur in the vegetation. through the expe-

rience of researchers in this ﬁeld, it has appeared that

there are speciﬁc changes in certain bands such as

near-infrared and infrared because of diseases. Au-

thors of(Hahn, 2009) says that the Sensors for disease

detection and food quality will evolve over the next

years with the help of nanotechnology. In fact, ﬂu-

orescence and vision sensors can detect the quality

of fruit and predict diseases more accurately than our

eyes. The spectral range of the sensor is wider than

the spectral response of the naked eye and the sensors

are capable of detecting polarized light. Nanotechnol-

ogy is able to capture gases in bubbles creating inter-

nal reactions without affecting the quality of fruits or

vegetables. Thus, with these technologies, we man-

age and minimize the use of phytosanitary products,

and with artiﬁcial intelligence algorithms, we can col-

lect historical data to predict and control the disease

before it occurs.

3.3 Vegetation and Canopy

Among the very interesting factors in the monitoring

of citrus orchards, there is the condition of the vegeta-

tion. This vegetation helps the farmer to know every

place or every tree that needs more care. In fact, the

vegetation index can be extracted from RGB images,

multispectral images, and hyperspectral images, but

the precision will be different. in fact, hyperspectral

images contain hundreds of bands which can get more

information, and if we have a good spatial resolution,

we can extract a vegetation index from each tree or

from each part of the orchard.

In table 1 we present some vegetation indexes.

Figure 4: Exemple of NDVI classiﬁcation map.

As we can see in ﬁgure 4, authors of (Robson et al.,

2017) control the situation of orchard by extracting

several vegetation indexes from worldview-3 multi-

spectral images, and select the most important index

based on PCA(principal component analysis)(Wold

et al., 1987) approach to map the vegetation of yield

in each part of orchard.

In addition, the detection, segmentation, and

counting of trees are very important in the monitor-

ing of each orchard. These factors can be produced

using deep learning and machine learning algorithms

(Koirala et al., 2019; Zou et al., 2019).

Actually with data science approaches, especially

deep learning architectures like U-net, RCNN, faster

RCNN, Yolo, retinanet.etc, the scientiﬁc researchers

improve the result of object segmentation. With these

Table 1: Vegetation indexes by formula.

Vegetation Index Formula

Normalized Difference Rededge/Red (NDVI rededge)(Gobron et al., 2000) (RE - R) / (RE + R)

Transformed Chlorophyll Absorption in Reﬂectance Index (TCARI)(Haboudane et al., 2002) 3 × ((RE - R) - 0.2 x (RE - G) × (RE/R))

Structure Insensitive Pigment Index (SIPI)(Penuelas et al., 1995) (NIR1 - B) / (NIR1 + R)

Coastal Blue Structure Insensitive Pigment Index (CB SIPI)(Penuelas et al., 1995) (NIR1 - CB) / (NIR1 + CB)

Normalized difference Red/Red-edge index (R / RENDVI(Gitelson et al., 1996b) (NIR1 - R) / (NIR1 + RE)

Normalized difference Red/NIR2 index (R / N2NDVI)(Robson et al., 2014) (NIR1 - R) / (NIR1 + NIR2)

Green normalized difference vegetation index (GNDVI)(Gitelson et al., 1996a) (NIR1 - G) / (NIR1 + G)

Modiﬁed Simple Ratio (MSR)(Chen, 1996) (NIR1/R - 1)/(SQRT((NIR1 / R) + 1))

Ratio Vegetation Index (RVI)(Jordan, 1969) NIR1/R

Normalized Difference Vegetation Index (N1NDVI)(Rouse Jr et al., 1974) (NIR1 - R) / (NIR1 + R)

Normalized Difference Vegetation Index (N2NDVI)(Rouse Jr et al., 1974) (NIR2 - R) / (NIR2 + R)

Normalized difference red edge index 1 (RENDVI1)(Fitzgerald et al., 2010) (NIR1 - RE) / (NIR1 + RE)

Normalized difference red edge index 2 (RENDVI2)(Fitzgerald et al., 2010) (NIR2 - RE) / (NIR2 + RE)

Transformed difference vegetation index (TDVI)(Bannari et al., 2002) 1.5 x ((NIR1 - R) / (SQRT(NIR1

+ R + 0.5))

Transformed difference vegetation index 2 (TDVI2)(Bannari et al., 2002) 1.5 x ((NIR2 - R)/(SQRT(NIR2

+ R + 0.5))

R: red band(631–689 nm), RE: red edge band(703–742 nm), G : green band(516–578 nm), B : blue band(407–448 nm), CB: coastal blue (407–448 nm),

NIR1: near infra-red 1 band(774–874 nm), NIR2: near infra-red 2 band(869–958 nm).

new technologies, We achieved a control of the tree

state or each small part in the orchard. In this case,

we have a lot of challenges to do some segmentation

of trees.

Three types of trees can be distinguished:(1) the

orchards that are not condensed and we can see the

space between the trees and between the parcels. (2)

is the case where we have a very condensed rows of

trees which cannot be separated, but we can see the

space between the rows. Therefore, we can do some

segmentation based on these rows. In the last type

(3) orchards are overcrowded; we cannot separate the

trees or the rows. In this case, we can make a seg-

mentation by deﬁning a ﬁxed area, and detecting the

spaces in orchards. Figure 5 give an example of the

overcrowded orchards.

Figure 5: Overcrowding of trees.

Many examples of projects use the crown or canopy

of the tree to control the state of trees, parcels or or-

chards. Authors of(Zortea et al., 2018) collect large

data that contain citrus trees images with a good res-

olution using a drone. Then they classify each part of

32 * 32 pixels to know whether it is a tree or not. In

fact, their model with 17 layers and trained to 56000

images gave a good result with 94% in accuracy.

Table II presents a list of works with results that use

trees segmentation. and as we can see, the deep learn-

ing approaches such as CNN and Mask R-CNN gives

a good score compared to the other methods. In ad-

dition, the image data provided by UAV with good

resolution, help the model to be perfect.

Table 2: Trees segmentation approaches and results

Approach Results Data

CNN(Zortea et al., 2018) 94(accuracy) UAV

VF(Gougeon, 1995) 81(accuracy) Aerial image

LMF(Nelson et al., 2005) 13.67 (Z-score) Satellite images

WS(Wang et al., 2004) 75.6(percentage) Aerial imagery

LM(Santoro et al., 2013) 0.8(RMSE) Satellite images

OBD(Ardila et al., 2012) 0.8(R

) Satellite images

PBP(Ok and Ozdarici-Ok, 2018) 91(accuracy) Satellite images

U-Net(Zhao et al., 2018) 61.2 (accuracy) UAV

MRCNN(Zhao et al., 2018) 98.5(accuracy) UAV

CNN: convolutional neural network,

VF: Valley following,

LMF : Local maximum ﬁltering,

WS : Watershed segmentation,

OBD: Object-based detection,

PBP: Pixel-based performance,

U-NET: Convolutional Networks for Biomedical Image Segmentation,

MRCNN: mask Region-based Convolutional Neural Networks.

4 CONCLUSION

The Citrus crop requires special monitoring to obtain

good performance in terms of quality and quantity.

In fact, several parameters are responsible for it, such

as climate, nutrition, irrigation, drop of ﬂowers and

fruits, rootstock operation, and the size of the tree

canopy, etc. In this case, the spectral images provide

a lot of historical data about all these parameters and

other. Indeed, with remote sensing techniques, we can

control vegetation, water stress, and diseases in cit-

rus orchards at any time of the year. Also, with data

science approaches, in particular deep learning, we

use spectral images to make several segmentation’s

of trees or rows and to do several classiﬁcations by

size, by vegetation and also by water stress. In this

sense, several works have given a good score greater

than 98% in precision. But these works are based

on UAV images that have a very high resolution and

in orchards that contain non-condensed trees. More-

over, the spectral images provided by satellite they

don’t have a very high resolution (the highest reso-

lution is 0.31m in worldview3), so when we have an

overcrowded orchard, it is difﬁcult to get a good re-

sult. In addition, the deep learning algorithms need

big data in training to give good precision. In fact,

if we compare drone and satellite images we can say

that the satellite offers historical images with a good

resolution, but to produce very high resolution im-

ages, we need to use the drone every year and wait

a few years to collect them. Also to make predic-

tions about yield and disease, machine learning al-

gorithms need the maximum amount of data. In this

case, smart agriculture needs an information system

capable to follow-up the orchards and collecting data

at any time. Finally, With all these technologies, we

can get several factors which are very important to fa-

cilitate several tasks for the farmers and to develop the

yield.

ACKNOWLEDGEMENTS

This work is part of the Multispectral satellite

imagery, data mining and agricultural applications

project, funded by the academy Hassan II of Science

and Technology.

REFERENCES

Ardila, J. P., Bijker, W., Tolpekin, V. A., and Stein, A.

(2012). Multitemporal change detection of urban trees

using localized region-based active contours in vhr

images. Remote sensing of environment, 124:413–

426.

Bannari, A., Asalhi, H., and Teillet, P. M. (2002). Trans-

formed difference vegetation index (tdvi) for vege-

tation cover mapping. In IEEE International geo-

science and remote sensing symposium, volume 5,

pages 3053–3055. IEEE.

Bravo, C., Moshou, D., Oberti, R., West, J., McCartney,

A., Bodria, L., and Ramon, H. (2004). Foliar disease

detection in the ﬁeld using optical sensor fusion.

Chaerle, L., Lenk, S., Hagenbeek, D., Buschmann, C., and

Van Der Straeten, D. (2007). Multicolor ﬂuorescence

imaging for early detection of the hypersensitive reac-

tion to tobacco mosaic virus. Journal of plant physi-

ology, 164(3):253–262.

Chen, B., Wang, K., Li, S., Wang, J., Bai, J., Xiao, C.,

and Lai, J. (2007). Spectrum characteristics of cot-

ton canopy infected with verticillium wilt and inver-

sion of severity level. In International Conference

on Computer and Computing Technologies in Agricul-

ture, pages 1169–1180. Springer.

Chen, J. M. (1996). Evaluation of vegetation indices and a

modiﬁed simple ratio for boreal applications. Cana-

dian Journal of Remote Sensing, 22(3):229–242.

Cobo, M. J., L

opez-Herrera, A. G., Herrera-Viedma, E.,

and Herrera, F. (2012). Scimat: A new science map-

ping analysis software tool. Journal of the Ameri-

can Society for Information Science and Technology,

63(8):1609–1630.

Delalieux, S., Van Aardt, J., Keulemans, W., Schrevens, E.,

and Coppin, P. (2007). Detection of biotic stress (ven-

turia inaequalis) in apple trees using hyperspectral

data: Non-parametric statistical approaches and phys-

iological implications. European Journal of Agron-

omy, 27(1):130–143.

Fitzgerald, G., Rodriguez, D., and O’Leary, G. (2010).

Measuring and predicting canopy nitrogen nutrition

in wheat using a spectral index—the canopy chloro-

phyll content index (ccci). Field Crops Research,

116(3):318–324.

Gitelson, A. A., Kaufman, Y. J., Merzlyak, M. N., et al.

(1996a). Use of a green channel in remote sensing of

global vegetation from eos-modis. Remote sensing of

Environment, 58(3):289–298.

Gitelson, A. A., Merzlyak, M. N., and Lichtenthaler, H. K.

(1996b). Detection of red edge position and chloro-

phyll content by reﬂectance measurements near 700

nm. Journal of plant physiology, 148(3-4):501–508.

Gobron, N., Pinty, B., Verstraete, M. M., and Widlowski,

J.-L. (2000). Advanced vegetation indices optimized

for up-coming sensors: Design, performance, and ap-

plications. IEEE Transactions on Geoscience and Re-

mote Sensing, 38(6):2489–2505.

Gonzalez-Dugo, V., Zarco-Tejada, P., Nicol

as, E., Nortes,

P. A., Alarc

on, J., Intrigliolo, D. S., and Fereres, E.

(2013). Using high resolution uav thermal imagery to

assess the variability in the water status of ﬁve fruit

tree species within a commercial orchard. Precision

Agriculture, 14(6):660–678.

Gougeon, F. A. (1995). A crown-following approach to

the automatic delineation of individual tree crowns in

high spatial resolution aerial images. Canadian jour-

nal of remote sensing, 21(3):274–284.

Haboudane, D., Miller, J. R., Tremblay, N., Zarco-Tejada,

P. J., and Dextraze, L. (2002). Integrated narrow-band

vegetation indices for prediction of crop chlorophyll

content for application to precision agriculture. Re-

mote sensing of environment, 81(2-3):416–426.

Hahn, F. (2009). Actual pathogen detection: sensors and

algorithms-a review. Algorithms, 2(1):301–338.

Jackson, R. D., Idso, S., Reginato, R., and Pinter Jr, P.

(1981). Canopy temperature as a crop water stress in-

dicator. Water resources research, 17(4):1133–1138.

Jordan, C. F. (1969). Derivation of leaf-area index

from quality of light on the forest ﬂoor. Ecology,

50(4):663–666.

Koirala, A., Walsh, K. B., Wang, Z., and McCarthy, C.

(2019). Deep learning–method overview and review

of use for fruit detection and yield estimation. Com-

puters and Electronics in Agriculture, 162:219–234.

LLC, M. connectedpapers.com.

Meli, S., Porto, M., Belligno, A., Bufo, S. A., Mazzatura,

A., and Scopa, A. (2002). Inﬂuence of irrigation with

lagooned urban wastewater on chemical and micro-

biological soil parameters in a citrus orchard under

mediterranean condition. Science of the total environ-

ment, 285(1-3):69–77.

Moshou, D., Bravo, C., West, J., Wahlen, S., McCartney,

A., and Ramon, H. (2004). Automatic detection of

‘yellow rust’in wheat using reﬂectance measurements

and neural networks. Computers and electronics in

agriculture, 44(3):173–188.

Nelson, T., Boots, B., and Wulder, M. A. (2005). Tech-

niques for accuracy assessment of tree locations ex-

tracted from remotely sensed imagery. Journal of en-

vironmental management, 74(3):265–271.

Obreza, T. T. and Morgan, K. T. (2008). Nutrition of ﬂorida

citrus trees. EDIS, 2008(2).

Ok, A. O. and Ozdarici-Ok, A. (2018). 2-d delineation

of individual citrus trees from uav-based dense pho-

togrammetric surface models. International Journal

of Digital Earth, 11(6):583–608.

Penuelas, J., Baret, F., and Filella, I. (1995). Semi-empirical

indices to assess carotenoids/chlorophyll a ratio from

leaf spectral reﬂectance. Photosynthetica, 31(2):221–

230.

Pons, E., Peris, J. E., and Pe

na, L. (2012). Field perfor-

mance of transgenic citrus trees: Assessment of the

long-term expression of uida and nptiitransgenes and

its impact on relevant agronomic and phenotypic char-

acteristics. BMC biotechnology, 12(1):41.

Purcell, D. E., O’Shea, M. G., Johnson, R. A., and Kokot, S.

(2009). Near-infrared spectroscopy for the prediction

of disease ratings for ﬁji leaf gall in sugarcane clones.

Applied Spectroscopy, 63(4):450–457.

Qin, J., Burks, T. F., Ritenour, M. A., and Bonn, W. G.

(2009). Detection of citrus canker using hyperspectral

reﬂectance imaging with spectral information diver-

gence. Journal of food engineering, 93(2):183–191.

Robson, A., Petty, J., Joyce, D., Marques, J., and Hofman,

P. (2014). High resolution remote sensing, gis and

google earth for avocado fruit quality mapping and

tree number auditing. In XXIX International Horti-

cultural Congress on Horticulture: Sustaining Lives,

Livelihoods and Landscapes (IHC2014): 1130, pages

589–596.

Robson, A., Rahman, M. M., and Muir, J. (2017). Using

worldview satellite imagery to map yield in avocado

(persea americana): a case study in bundaberg, aus-

tralia. Remote Sensing, 9(12):1223.

Rouse Jr, J. (1974). Monitoring the vernal advancement

and retrogradation (green wave effect) of natural veg-

etation.

Rouse Jr, J., Haas, R., Schell, J., and Deering, D. (1974).

Paper a 20. In Third Earth Resources Technology

Satellite-1 Symposium: The Proceedings of a Sympo-

sium Held by Goddard Space Flight Center at Wash-

ington, DC on December 10-14, 1973: Prepared at

Goddard Space Flight Center, volume 351, page 309.

Scientiﬁc and Technical Information Ofﬁce, National

Aeronautics and Space . . . .

Santoro, F., Tarantino, E., Figorito, B., Gualano, S., and

D’Onghia, A. M. (2013). A tree counting algorithm

for precision agriculture tasks. International Journal

of Digital Earth, 6(1):94–102.

Sevick-Muraca, E. M. and Paithankar, D. Y. (1999). Flu-

orescence imaging system and method. US Patent

5,865,754.

Shafri, H. Z. and Hamdan, N. (2009). Hyperspectral im-

agery for mapping disease infection in oil palm planta-

tionusing vegetation indices and red edge techniques.

American Journal of Applied Sciences, 6(6):1031.

Spinelli, F., Noferini, M., and Costa, G. (2004). Near in-

frared spectroscopy (nirs): perspective of ﬁre blight

detection in asymptomatic plant material. In X Inter-

national Workshop on Fire Blight 704, pages 87–90.

Wang, L., Gong, P., and Biging, G. S. (2004). Individual

tree-crown delineation and treetop detection in high-

spatial-resolution aerial imagery. Photogrammetric

Engineering & Remote Sensing, 70(3):351–357.

Wold, S., Esbensen, K., and Geladi, P. (1987). Principal

component analysis. Chemometrics and intelligent

laboratory systems, 2(1-3):37–52.

Yang, C.-M., Cheng, C.-H., and Chen, R.-K. (2007).

Changes in spectral characteristics of rice canopy in-

fested with brown planthopper and leaffolder. Crop

science, 47(1):329–335.

Zarco-Tejada, P. J., Gonz

alez-Dugo, V., and Berni, J. A.

(2012). Fluorescence, temperature and narrow-band

indices acquired from a uav platform for water stress

detection using a micro-hyperspectral imager and a

thermal camera. Remote sensing of environment,

117:322–337.

Zarco-Tejada, P. J., Gonz

alez-Dugo, V., Williams, L.,

arez, L., Berni, J. A., Goldhamer, D., and Fer-

eres, E. (2013). A pri-based water stress index com-

bining structural and chlorophyll effects: Assessment

using diurnal narrow-band airborne imagery and the

cwsi thermal index. Remote sensing of environment,

138:38–50.

Zhang, L., Zhang, H., Niu, Y., and Han, W. (2019). Map-

ping maize water stress based on uav multispectral re-

mote sensing. Remote Sensing, 11(6):605.

Zhao, T., Yang, Y., Niu, H., Wang, D., and Chen, Y. (2018).

Comparing u-net convolutional network with mask r-

cnn in the performances of pomegranate tree canopy

segmentation. In Multispectral, Hyperspectral, and

Ultraspectral Remote Sensing Technology, Techniques

and Applications VII, volume 10780, page 107801J.

International Society for Optics and Photonics.

Zortea, M., Macedo, M. M., Mattos, A. B., Ruga, B. C., and

Gemignani, B. H. (2018). Automatic citrus tree de-

tection from uav images based on convolutional neu-

ral networks. In 2018 31th SIBGRAPI Conference

on Graphics, Patterns and Images (SIBGRAPI), vol-

ume 11.

Zou, Z., Shi, Z., Guo, Y., and Ye, J. (2019). Object

detection in 20 years: A survey. arXiv preprint

arXiv:1905.05055.