Computers and Electronics in Agriculture 143 (2017) 49–56

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

journal homepage: www.elsevier.com/locate/compag

Mapping skips in sugarcane fields using object-based analysis of unmanned MARK

aerial vehicle (UAV) images
⁎
Carlos Henrique Wachholz de Souzaa, , Rubens Augusto Camargo Lamparellib,
Jansle Vieira Rochaa, Paulo Sergio Graziano Magalhãesa,b,1
a
School of Agricultural Engineering, State University of Campinas, Campinas SP CEP 13083-875, Brazil
b
Interdisciplinary Center for Energy Planning, State University of Campinas, Campinas SP CEP 13083-896, Brazil

A R T I C L E I N F O A B S T R A C T

Keywords: The use of unmanned aerial vehicles (UAVs) as remote sensing platforms has tremendous potential for describing
Object-based image analysis detailed site-specific features of crops, especially in early post-emergence, which was not possible previously
Planting rows with satellite images. This article describes an object-based image analysis (OBIA) procedure for UAV images,
GIS designed to map and extract information about skips in sugarcane planting rows. The procedure consists of three
Skip rate
consecutive phases: (1) identification of sugarcane planting rows, (2) identification of the existent sugarcane
Decision making
within the crop rows, and (3) skip extraction and creation of field-extent crop maps. Results based on experi-
mental fields achieved skip rates of between 2.29% and 10.66%, indicating a planting operation with excellent
and good quality, respectively. The relationship of estimated versus observed skip length had a coefficient of
determination of 0.97, which was confirmed by the value of the enhanced Wilmott concordance coefficient of
0.92, indicating good agreement. The OBIA procedure allowed a high level of automation and adaptability, and
it provided useful information for decision making, agricultural monitoring, and the reduction of operational
costs.

1. Introduction of the planting and harvesting operations is essential (Lebourgeois

In agriculture, the availability of reliable and timely information on
crop condition and the possibility of problems, assists proper planning
decisions, which in turn contribute to increased profitability and re-
duced costs (Zhang et al., 2002).
Currently, because of the current demand for biofuels and sugar
production, Brazil has large areas of agricultural land covered by su-
garcane fields (Martinelli and Filoso, 2008; Nassar et al., 2008;
Martinelli et al., 2011). Thus, developments of techniques and tech-
nologies aimed at operational efficiency, cost reduction, and yield in-
crease have attracted considerable attention because of its economic
and environmental importance (Abdel-Rahman and Ahmed, 2008;
Bégué et al., 2010; Bocca et al., 2015).
The sugarcane cycle is semi-perennial with a growth cycle of ap-
proximately 12 or 18 months. After the first harvest, the ratoons are
harvested annually for a period of about 5–7 years (Xavier et al., 2006;
Rudorff et al., 2010). Therefore, to maintain the longevity and pro-
ductivity of the crop field during this period, monitoring of the quality
et al., 2010; Bocca et al., 2015). Matsuoka and Stolf (2012, p.148)
defined “Gappy fields”, in terms of either plant canes or ratoons, as “a
signal of bad crop management in most cases: bad soil preparation,
inappropriate cultivar, improper seed cane, improper season for
planting, unexpected disease or pest occurrence, herbicide damage,
salinity, bad mechanical operations, mainly during harvest in the case
of ratoons, if set aside climatic factors like freezing temperatures,
lightening and extreme drought.”
The identification and quantification of skips in sugarcane fields is
of great importance because it assesses the uniformity of germination
and tillering and the consequent formation of stalks, which in turn is
correlated directly with yield (Matsuoka and Stolf, 2012; Bocca et al.,
2015). In addition to measuring the quality of the planting operation,
quantification of skips provides support for decision making regarding
the replanting or renovation of the field, in order to guarantee its
continued profitability (Keerthipala and Dharmawardene, 2000;
Matsuoka and Stolf, 2012). Currently, the presence of skips in planting
rows is verified by in situ visual inspection. However, these inspections

⁎
Corresponding author.
E-mail addresses: carlos_hws@hotmail.com (C.H.W.d. Souza), rubens.lamparelli@gmail.com (R.A.C. Lamparelli), jansle.rocha@feagri.unicamp.br (J.V. Rocha),
graziano@g.unicamp.br (P.S.G. Magalhães).
1
CNPq Researcher (process: 307362/2014-0).

http://dx.doi.org/10.1016/j.compag.2017.10.006
Received 3 March 2016; Received in revised form 18 August 2017; Accepted 6 October 2017
0168-1699/ © 2017 Elsevier B.V. All rights reserved.
C.H.W.d. Souza et al.
Fig. 1. UAV orthoimage of the study fields, together with pictures of the sugarcane at the time of flight.
are generally conducted from the field borders and thus, heterogeneity
across the entire crop field might affect the accuracy of such estimates
(Bocca et al., 2015).
Recently, the use of remote sensing technologies for agricultural
monitoring has been gaining ground, because they can provide spatially
and temporally distributed information objectively and quickly over a
variety of scales (Zhang et al., 2002; Ahamed et al., 2011; Mulla, 2013).
The development of new technologies such as unmanned aerial vehicles
(UAVs) as platforms for the acquisition of remote sensing imagery, al-
lows some of the limitations of orbital and airborne platforms that
hinder crop monitoring in real-time to be overcome, e.g., the suitability
of revisit times, avoidance of cloud cover, costs, complexity of opera-
tion, and limitation of spatial resolution (Berni et al., 2009; Everaerts,
2009; Zhang and Kovacs, 2012; Colomina and Molina, 2014). These
characteristics make UAV platforms suitable for a number of applica-
tions, including crop monitoring (Hunt et al., 2005, 2010; Torres-
Sánchez et al., 2014, 2015; Comba et al., 2015), weed detection
(Torres-Sánchez et al., 2013; Peña et al., 2015), water stress assessment
(Berni et al., 2009; Zarco-Tejada et al., 2012), disease detection (Garcia-
Ruiz et al., 2013), and yield estimation (Swain et al., 2010), i.e., ap-
plications where time-critical management is required.
UAVs can provide imagery with very high spatial resolution of only
a few centimeters and they allow images to be acquired at optimal
moments for the desired purposes, which makes them ideal for distin-
guishing crop plants during their first stages of development (Hengl,
2006; López-Granados, 2011; Torres-Sánchez et al., 2015a). However,
very-high-resolution images require powerful image analysis proce-
dures because, unlike lower resolution images, single pixels might no
longer capture the characteristics of the classification targets. Ad-
ditionally, these images show higher intra-class spectral variability and
50
Computers and Electronics in Agriculture 143 (2017) 49–56
subsequently, a reduction in the degree of statistical separability among
the classes compared with conventional pixel-based classification
methods (Yu et al., 2006; Castillejo-González et al., 2014; Peña et al.,
2015; Torres-Sánchez et al., 2015a). To overcome this limitation and to
attain a high level of automation and adaptability, object-based image
analysis (OBIA) has been used successfully with high-resolution satellite
imagery (Novack et al., 2010; de Castro et al., 2013; Castillejo-González
et al., 2014) and UAV imagery (Laliberte and Rango, 2011; Peña et al.,
2013, 2015; Diaz-Varela et al., 2014; Qin, 2014; Torres-Sánchez et al.,
2015a,b). The OBIA approach first identifies spatially and spectrally
homogeneous units called “objects”, which are created by grouping
adjacent pixels following a segmentation process, and then using the
created “objects” as the basic elements for analysis. Thus, it is possible
to create automated and auto-adaptive classification methods by com-
bining the spectral, contextual, morphological, and hierarchical in-
formation of these elements (Blaschke, 2010).
This article presents an innovative procedure for the creation of skip
maps in sugarcane fields that combines high-resolution images from a
commercial UAV and the OBIA approach to generate useful decision-
support data. Based on UAV images, this procedure first performs the
identification of the sugarcane crop rows. It then identifies the existent
sugarcane within the crop rows and finally, performs skip extraction
and the creation of field-extent crop maps.
2. Materials and methods
2.1. Study area
The fields used in this study are located near Euclides da Cunha
Paulista in São Paulo state, Brazil (22°26′21″S, 52°35′46″W). They have
C.H.W.d. Souza et al. Computers and Electronics in Agriculture 143 (2017) 49–56

Fig. 2. Flowchart illustrating each stage of the object-based image analysis (white boxes) + geographic information system (GIS) (gray boxes) approach proposed to map skips of
sugarcane areas using unmanned aerial vehicle (UAV) images. First step: Identification of crop rows; Second step: Sugarcane classification; Third step: skip extraction and creation of the
maps. (NDVI = normalized difference vegetation index.) (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

51
C.H.W.d. Souza et al.
two predominant classes of soils: Rhodic Hapludox (Typic Hapludox)
and Quartzarenic Neosol (Typic Quartzipsamment) and they present a
moderate slope (< 12%) that allows mechanical green harvest. The
region has an average ground elevation of approximately 400 m a.s.l.
and a humid subtropical climate (Cfa) according to the Köppen classi-
fication. The average annual temperature is 22 °C, and the average
annual rainfall is 1200 mm, with much more precipitation in summer
than in winter. The landscape of the region comprises some environ-
mental conservation units with a predominance of sugarcane cultiva-
tion and livestock production.
The crop variety planted in the area was RB-86-7515. The first
specimens were planted in rows 1.5-m apart over an area of approxi-
mately 100 ha in September 2014 and they were harvested in October
2015. UAV imagery was acquired on November 3, 2014, when the
sugarcane was at the tillering stage, approximately 60 days after
planting (DAP) (Fig. 1). The ideal time for image acquisition is when
the sugarcane has grown sufficiently to fill the rows but without cov-
ering the inter-row gaps, i.e., 60–90 DAP. However, this period might
vary according to the crop variety, number of ratoons, and amount of
rainfall; therefore, the determination of the ideal time will depend on a
visual inspection of the field.
2.2. UAV system and image acquisition
The UAV system used is based on an eBee Ag drone designed and
developed by senseFly (senseFly Ltd, Cheseaux-Lausanne, Switzerland).
It is built of expanded polypropylene and it weighs about 0.69 kg, in-
cluding the payload, which means it can be hand-launched. The UAV is
battery-powered and it can fly for 50 min at 40–90 km h−1 at altitudes
of up to 900 m, within wind speeds of up to 45 km h−1. Its usual flight
height is 250 m, which provides spatial resolution of 0.10 m over an
area of up 12 km2 in a single flight. The flight is controlled auto-
matically using real-time positioning based on a GPS system that is
accurate to within a few meters. The attitude of the plane is monitored
by accelerometers. The drone is supplied as standard with two ad-
vanced software packages: eMotion 2 (senseFly, Cheseaux-Lausanne,
Switzerland) for flight planning and control and Postflight Terra 3D
(Pix4D SA, Lausanne, Switzerland), which is photogrammetry software
designed for post-flight image processing and analysis.
The coupled sensor is a small consumer camera based on a Canon
Powershot S110 compact camera with modified color filters. Instead of
the standard RGB system, the S110-NIR model uses spectral filters for
green, red, and infrared wavelengths centered at 550, 625, and 850 nm,
respectively. This separation is performed by a Bayer color filter di-
rectly on the sensor (Shortis et al., 2005). Thus, this sensor allows the
implementation of some vegetation indices derived from different
combinations of the three channels, such as the normalized difference
vegetation index (NDVI) (Rouse et al., 1974). The camera has a 24-mm
focal length and the CMOS sensor (7.6 × 5.7 mm) stores the acquired
images on a compact flash card with up to 12-bit radiometric resolu-
tion. The images were acquired using a shutter speed of 1/1200, in
order to prevent them from being affected by vibrations and motion.
For a particular mode, the camera adapts the aperture and ISO speed
according to the light conditions for every single shot to ensure the best
image exposure, which ranged from 2.0 to 4.0 and 100 to 200, re-
spectively.
The UAV imagery was acquired at flight altitudes of 150–200 m.
Thus, for our objective, the pixel size of approximately 0.10 m was
sufficient with 75% forward lap and 70% side lap. The eBee Ag UAS
allows for direct georeferencing of the images using an inertial mea-
surement unit (IMU) in combination with an on-board Global
Navigation Satellite System (GNSS). Therefore, the ground control
points can be omitted, which speeds up data collection and processing,
allowing for a fully automatic processing chain (Bendig et al., 2014;
Turner et al., 2014). The Postflight Terra 3D software (Pix4D SA,
Lausanne, Switzerland) was used for orthorectification and the
52
Computers and Electronics in Agriculture 143 (2017) 49–56
mosaicking of the 161 images acquired. The mean absolute geolocation
variance obtained in each direction were of 0.0016 m in X, −0.0008 m
in Y and 0.0042 m in Z. No radiometric correction was performed.
2.3. Skip mapping
The procedure to analyze the UAV images and develop the OBIA
procedure for skip mapping (Fig. 2) was developed using the com-
mercial software eCognition Developer 8 (Trimble GeoSpatial, Munich,
Germany). Similarly, the extraction of the vectorial data features and
production of the final maps used ArcGis software (ESRI, Redlands, CA,
USA).
The rule set of the algorithm for failure mapping of the sugarcane
ran semi-automatically and it comprised three principal consecutive
steps: (1) identification of crop rows, (2) classification of sugarcane in
the rows, and (3) skip extraction and the creation of the maps.
2.4. First step: identification of crop rows
This step is necessary if information on the crop rows is unavailable.
For example, if the auto-steering guidance system (Autopilot) was used
at planting, the planting rows could be used for this purpose. Once this
has been achieved, only the second and third steps are necessary, for
example, if the requirement is to study the increase of skips in succes-
sive ratoons, which might increase after each successive ratoon
(Matsuoka and Stolf, 2012).
For the identification of crop rows, the NDVI was used to increase
the differences between the vegetation and bare soil. NDVI was selected
because it is able to achieve greater spectral separability compared with
other vegetation indices (Torres-Sánchez et al., 2013). Then, a filter was
applied to create a layer in which each pixel represented the difference
between the Kernel max or min value and the center value (Pixel Min/
Max Filter Algorithm). In order to enhance the difference between the
crop rows and inter-row gaps in the image, the algorithm parameters of
“difference-brightest-to-center” and Kernel size of 15 × 1 were used.
A sub-rule set called “rows extraction” was created to identify the
row. First, a temporary layer called “Rows” (Layer arithmetic algo-
rithm) was created, which was used to add lines with different direc-
tions and widths. Then, a loop process was used to search lines in dif-
ferent directions (from 0° to 180° at intervals of 5°). The “Update
Variable” algorithm updated the angle parameter (angle: += 5°) and
the “Line Extraction” algorithm was used to classify the pixels ac-
cording to their line signal strength in the different directions and to
create a thematic output layer. This process was replicated three times
by presenting different values of the line width parameter (3, 6, and
12 pixels) to the “Line Extraction” algorithm. The values of the others
parameters remained the same in each of the three replicates.
Furthermore, at the end of each sub-process, each thematic output layer
was added into the layer “Rows” by the “Layer Arithmetic” algorithm
(“output layer + Rows”). This way, at the end of the rule set, the layer
“Rows” represented the sum of these three sub-processes (“Rows
Identification” result; Fig. 2).
Subsequently, the layer “Rows” was segmented into homogeneous
multi-pixel objects using the “Contrast Filter Segmentation” algorithm,
which uses pixel filters to detect and create objects based on contrast
and gradient data.
Thereafter, small objects that represented noise were removed using
the “Remove Objects” algorithm, with a threshold condition of the
number of pixels being < 100. Then, the remaining crop row objects
were assigned into a class called “Crop Rows” by the “Assign Class”
algorithm, and the main line of each “Crop Rows Class” objects were
exported as shapefile vectors. Finally, the exported crop row lines were
entered into ArcGis to check and correct for possible errors. The red
boxes in Fig. 2 shows two examples of errors that are common in the
process, which are caused mainly by long skips, late germination, or
sugarcane covering the inter-row gaps.
C.H.W.d. Souza et al. Computers and Electronics in Agriculture 143 (2017) 49–56

2.5. Second step: sugarcane classification Table 1

Global features obtained from the skip-mapping procedure.
After the crop rows lines were checked and the errors corrected
Global features Field 1 Field 2 Field 3 Field 4
using ArcGis, the data were returned to the OBIA software (eCognition)
as a thematic vector layer. Then, the image object domain was split Fields
using the “Chessboard Segmentation” algorithm based on the thematic Field area (ha) 17.53 21.15 21.97 37.40
vector layer (“Crop Rows”). Those pixels close to the rows were seg- Perimeter (km) 1.76 1.91 1.96 2.73
mented and assigned into a temporary class called “Main Rows”. The Crop rows
“Main Rows” class was expanded to cover the sugarcane with the crop Number of rows 520 365 418 746
rows. For this, a sub-rule set called “Grow Main Rows” was created, Minimum length (m) 11.38 5.57 9.51 3.98
where a loop with Conditional Quad Tree segmentation was inserted. Maximum length (m) 433.88 667.02 492.17 811.78
Average length (m) 214.12 376.92 336.95 318.46
For each loop, the condition to segment the pixels was the existence of
Total length (km) 111.34 137.57 140.84 237.57
the “Main Rows” class, i.e., the pixels close to the “Main Rows” class Average angle (°) 111 90 69 41
were segmented and then assigned into the class. The loop was exe- Skips
cuted until partially covering the width of the sugarcane canopy; in this Number of skips 2877 1467 3306 11,948
study, three cycles were used. Finally, the sugarcane was classified by Maximum length (m) 50.92 125.72 255.75 185.86
the “Classification” algorithm in the pixels belonging to the “Main Average length (m) 1.61 2.14 2.17 2.12
Rows” class. The NDVI was used to classify the sugarcane pixels Total length (km) 4.65 3.15 7.17 25.33
Skip rate (%) 4.17 2.29 5.09 10.66
through a fuzzy membership function with values > 0.5. Then, the
pixels classified as sugarcane were merged and exported as a shapefile
vector called “Sugarcane at Rows.” One of the results of this procedure is the identification of the crop
rows (1st step). Thus, it was possible to extract the characteristic fea-
2.6. Third step: skip extraction and creation of the maps tures of the crop rows within the sugarcane fields. For instance, in-
formation regarding the number of rows, average row length, and row
A process was constructed using the “ModelBuilder” in the ArcGis angle could be used in studies of optimization and operational costs of
software to extract automatically the failure rows. For this, the tools boundary maneuvers (Spekken and de Bruin, 2012; Spekken et al.,
“Erase,” “Split Line at Vertices,” and “Unsplit Line” were used se- 2015). This information could help minimize the time and energy spent
quentially. The input parameters were the two shapefiles that were on turns and thus, contribute to cost reductions and improved sus-
output from the previous steps: “Crop Rows” and “Sugarcane at Rows.” tainability of agricultural production (Spekken et al., 2015).
At the end of the process, the skip shapefile was generated. Thereafter, Table 1 presents the information obtained from the mapping of
the lengths of the skips were calculated, where only skips > 0.5 m were skips. Field 4 had the highest number of skips (11, 948), which ac-
considered (Stolf, 1986), following which they were split into classes counted for 10.66% of the crop rows, whereas there were only 1467
and a field-extent map was created. skips in Field 2, which represented 2.29% of the crop rows. The higher
skip rate in Field 4 was probably related to slope, because this field is
2.7. Validation located in the lower part of the cropping area and thus, it is more prone
to sediment deposition and flooding.
The validation of the method was performed through comparison of Stolf (1986), in his method for the assessment of the quality of su-
the skip length obtained from the generated maps (estimated) against garcane planting, suggested that a skip rate of up to 10% indicated that
ground measurements (observed). Overall, 54 samplings were collected the planting operation could be classified as excellent. Thus, all the crop
in the field using a tape measure. The analysis was performed through a fields studied here were within this classification, except for Field 4,
simple linear regression and its coefficient of determination (R2). The which was classified as normal quality. Based on a group of three ex-
mean error, root mean square error, and the enhanced Willmott con- periments using inter-row spacing of 1.3 (two) and 1.4 m (one), a linear
cordance coefficient (dr), as proposed by Willmott et al. (2012) (Eq. decrease in yield of 3.2% for each 10% increase in skips was found for
(1)), were the statistical indicators used. The dr establishes the accuracy plant cane harvested with 18 months (Stolf, 1986; Matsuoka and Stolf,
of the method and the degree of detachment of the estimated values in 2012).
relation to the values observed. This index varies between ± 1, where
positive values closer to one indicate better agreement.
3.1.1. Skip maps
n
∑i = 1 |E −O| The maps created at the end of the procedure show the spatial
dr = 1− ,
n
2 ∑i = 1 (|O−O |) (1) distribution of skips with the sugarcane fields (Figs. 3 and 4). Some
blank patterns can be seen in the maps, this due to the presence of trees
where n = number of samples, O = observed skip length, E = esti- and terraces inside fields. Because the data are georeferenced, they can
mated skip length, and O = average observed skip length. be used to direct field verification to the most relevant points, which
optimizes the effort. Furthermore, the traditional method of skip eva-
3. Results and discussion luation considers only some parts of some planting rows (Stolf, 1986).
The method proposed here presents an estimate based on all the crop
3.1. Global features rows and thus, it is much more representative. Furthermore, there is the
possibility of comparison with other maps, such as vegetation index
The procedure allows the computation of multiple datasets and maps, yield maps, and soil and slope maps, generating additional source
statistics derived from the outputs, which can help characterize the crop information to assist with decision making.
field in terms of crop rows and skip incidence. In addition, the data can It can be seen in Figs. 3 and 4 that there are concentrations of skips
be exported in different formats, e.g., vector, raster, or tables, allowing in the corners of the fields, possibly related to poor germination due to
further integration with other data sources. The global features calcu- soil compaction by excessive machine traffic. Furthermore, many skips
lated for the study area are presented in Table 1. The dimensions of the occur next to terraces, which are areas where water often concentrates.
field were obtained from the shapefiles, which indicated a total area of An excess of water might result in various problems such as the delay in
98.05 ha. germination, bud and root rot due to lack of oxygen, and increased risk

53
C.H.W.d. Souza et al. Computers and Electronics in Agriculture 143 (2017) 49–56

Fig. 3. Skip maps for the sugarcane fields (F1 and F2) within the study area.

Fig. 4. Skip maps for the sugarcane fields (F3 and F4) within the study area.

54
C.H.W.d. Souza et al.
45
n = 54
40 y = 1.0015x - 0.3512
35 R² = 0.978
Estimated (m)
30 ME = -0.33
25 RMSE = 1.29
20 dr = 0.92
15
10
Samples
5 Line 1:1
0
0 5 10 15 20 25 30 35 40
Observed (m)
Fig. 5. Comparison between the skip length observed in the field and that estimated by
the procedure (n = number of samples; ME = mean error; RMSE = root mean square
error; dr = enhanced Wilmott concordance coefficient).
of disease. Ray et al. (2009) found that the higher the flood time, the
greater the loss of sugarcane yield. Glaz and Lingle (2012) reported that
the leaves, stalks, roots, and biomass were all affected negatively by
increased flood time in recently planted sugarcane fields. Thus, those
skips related to late germination at such flood points could not be
identified on the acquired images at 60 DAP (Fig. 1). Subsequent field
checks at these flood points revealed higher skip incidence, although
fewer in number and with shorter lengths than obtained from the maps.
The sugarcane plants at these points presented reduced size and fewer
numbers of leaves and stalks per unit area compared with the re-
mainder of the field. Therefore, even if it is accepted that the method
overestimates skip numbers at such flood points, it will provide evi-
dence of where yield loss will be higher.
Finally, the maps also provide some information about the quality of
the planting operation, which can help improve the mechanical op-
erations. For example, in Field 2 (Fig. 3), two crop rows with huge skips
were detected, which were caused by bad mechanical operations at
planting. Thus, the frequency, length, and distribution of skips can help
identify the sources of the problems.
3.2. Validation
The relationship between the estimated and observed skip length
was highly satisfactory with a value of R2 = 0.97, which was confirmed
by the value of dr = 0.92, indicating good agreement (Fig. 5). Through
the mean error, we conclude that the procedure underestimated skip
length by an average of 0.33 m with a root mean square error of 1.29 m.
These validation results indicate that the use of high-resolution images
from UAV systems during the early growth season of sugarcane can
provide accurate information about skip incidence, which could help
decision-making procedures.
4. Conclusions
The use of UAV images allows the creation of skip maps of su-
garcane fields. The method presented in this study proved efficient in
the estimation of skip length when compared with information derived
in situ. Such information is useful for decision making, agricultural
monitoring, and reduction of operational costs reduction, and it can
help maintain the longevity and productivity of the crop over succes-
sive cycles. The use of UAV technology optimized the surveying of skips
in fields, previously unfeasible with satellite imagery and currently
realized by field investigation.
The OBIA method resolves complex problems related to the UAV
images through the creation of process trees, which permits a high level
Computers and Electronics in Agriculture 143 (2017) 49–56
of automation and adaptability. Moreover, its combination with other
geographic information system platforms, as demonstrated in this
study, represents an advance in the extraction and interpretation of
remote sensing data.
The time of image acquisition is important when mapping skips. The
method presented here was influenced by the presence of long skips,
late germination, or sugarcane covering the inter-row gaps. The ac-
quisition of good quality imagery at the ideal time will reduce errors in
the construction of the crop rows. Future studies will monitor succes-
sive ratoons to determine the rate of skip increase, which will allow
estimation of the optimal time for field renewal.
Acknowledgments
45
This study was supported by the São Paulo Research Foundation
FAPESP (Fundação de Amparo à Pesquisa do Estado de São Paulo) and
Odebrecht Agro-Industrial (Process Number. 12/50048-7). The first
author was supported by a PhD scholarship from CAPES (Coordenação
de Aperfeiçoamento de Pessoal de Nível Superior). We also thank Prof.
Mariana Abrantes Giannotti, coordinator of the Geoprocessing
Laboratory (LABGEO) of the Polytechnic School of University of São
Paulo, for allowing the use of the eCognition software.
References
Abdel-Rahman, E.M., Ahmed, F.B., 2008. The application of remote sensing techniques to
sugarcane (Saccharum spp. hybrid) production: a review of the literature. Int. J.
Remote Sens. 29, 3753–3767. http://dx.doi.org/10.1080/01431160701874603.
Ahamed, T., Tian, L., Zhang, Y., Ting, K.C., 2011. A review of remote sensing methods for
biomass feedstock production. Biomass Bioenerg. 35, 2455–2469. http://dx.doi.org/
10.1016/j.biombioe.2011.02.028.
Bégué, A., Lebourgeois, V., Bappel, E., Todoroff, P., Pellegrino, A., Baillarin, F., Siegmund,
B., 2010. Spatio-temporal variability of sugarcane fields and recommendations for
yield forecast using NDVI. Int. J. Remote Sens. 31, 5391–5407. http://dx.doi.org/10.
1080/01431160903349057.
Bendig, J., Bolten, A., Bennertz, S., Broscheit, J., Eichfuss, S., Bareth, G., 2014. Estimating
biomass of barley using crop surface models (CSMs) derived from UAV-based RGB
imaging. Remote Sens. 6, 10395–10412. http://dx.doi.org/10.3390/rs61110395.
Berni, J., Zarco-Tejada, P.J., Suarez, L., Fereres, E., 2009. Thermal and narrowband
multispectral remote sensing for vegetation monitoring from an unmanned aerial
vehicle. IEEE Trans. Geosci. Remote Sens. 47, 722–738. http://dx.doi.org/10.1109/
TGRS.2008.2010457.
Blaschke, T., 2010. Object based image analysis for remote sensing. ISPRS J.
Photogramm. Remote Sens. 65, 2–16. http://dx.doi.org/10.1016/j.isprsjprs.2009.06.
004.
Bocca, F.F., Rodrigues, L.H.A., Arraes, N.A.M., 2015. When do I want to know and why?
Different demands on sugarcane yield predictions. Agric. Syst. 135, 48–56. http://dx.
doi.org/10.1016/j.agsy.2014.11.008.
Castillejo-González, I.L., Peña-Barragán, J.M., Jurado-Expósito, M., Mesas-Carrascosa,
F.J., López-Granados, F., 2014. Evaluation of pixel- and object-based approaches for
mapping wild oat (Avena sterilis) weed patches in wheat fields using QuickBird
imagery for site-specific management. Eur. J. Agron. 59, 57–66. http://dx.doi.org/
10.1016/j.eja.2014.05.009.
Colomina, I., Molina, P., 2014. Unmanned aerial systems for photogrammetry and remote
sensing: a review. ISPRS J. Photogramm. Remote Sens. 92, 79–97. http://dx.doi.org/
10.1016/j.isprsjprs.2014.02.013.
Comba, L., Gay, P., Primicerio, J., Ricauda Aimonino, D., 2015. Vineyard detection from
unmanned aerial systems images. Comput. Electron. Agric. 114, 78–87. http://dx.
doi.org/10.1016/j.compag.2015.03.011.
de Castro, A.I., López-Granados, F., Jurado-Expósito, M., 2013. Broad-scale cruciferous
weed patch classification in winter wheat using QuickBird imagery for in-season site-
specific control. Precis. Agric. 14, 392–413. http://dx.doi.org/10.1007/s11119-013-
9304-y.
Diaz-Varela, R.A., Zarco-Tejada, P.J., Angileri, V., Loudjani, P., 2014. Automatic identi-
fication of agricultural terraces through object-oriented analysis of very high re-
solution DSMs and multispectral imagery obtained from an unmanned aerial vehicle.
J. Environ. Manage. 134, 117–126. http://dx.doi.org/10.1016/j.jenvman.2014.01.
006.
Everaerts, J., 2009. New Platforms - Unconventional Platforms (Unmanned Aircraft
Systems) for Remote Sensing, Frankfurt.
Garcia-Ruiz, F., Sankaran, S., Maja, J.M., Lee, W.S., Rasmussen, J., Ehsani, R., 2013.
Comparison of two aerial imaging platforms for identification of Huanglongbing-in-
fected citrus trees. Comput. Electron. Agric. 91, 106–115. http://dx.doi.org/10.
1016/j.compag.2012.12.002.
Glaz, B., Lingle, S.E., 2012. Flood duration and time of flood onset effects on recently
planted sugarcane. Agron. J. 104, 575–583. http://dx.doi.org/10.2134/agronj2011.
0351.
Hengl, T., 2006. Finding the right pixel size. Comput. Geosci. 32, 1283–1298. http://dx.
55
C.H.W.d. Souza et al.
doi.org/10.1016/j.cageo.2005.11.008.
Hunt Jr., E.R., Hively, W.D., Fujikawa, S.J., Linden, D.S., Daughtry, C.S.T., McCarty,
G.W., 2010. Acquisition of NIR-green-blue digital photographs from unmanned air-
craft for crop monitoring. Remote Sens. 2, 290–305. http://dx.doi.org/10.3390/
rs2010290.
Hunt, E.R., Cavigelli, M., Daughtry, C.S.T., Mcmurtrey, J.E., Walthall, C.L., 2005.
Evaluation of digital photography from model aircraft for remote sensing of crop
biomass and nitrogen status. Precis. Agric. 6, 359–378. http://dx.doi.org/10.1007/
s11119-005-2324-5.
Keerthipala, A., Dharmawardene, N., 2000. Determination of optimal replanting cycles
for sugarcane production in Sri Lanka. Sugar Tech 2, 9–19. http://dx.doi.org/10.
1007/BF02945751.
Laliberte, A.S., Rango, A., 2011. Image processing and classification procedures for
analysis of sub-decimeter imagery acquired with an unmanned aircraft over arid
rangelands. GISci. Remote Sens. 48, 4–23. http://dx.doi.org/10.2747/1548-1603.48.
1.4.
Lebourgeois, V., Begue, A., Degenne, P., Bappel, E., Lebourgeois, B.V., Begue, A.,
Degenne, P., Bappel, E., 2010. Improving harvest and planting monitoring for
smallholders with geospatial technology: the Reunion Island experience. Int. Sugar J.
109, 109–119.
López-Granados, F., 2011. Weed detection for site-specific weed management: mapping
and real-time approaches. Weed Res. 51, 1–11. http://dx.doi.org/10.1111/j.1365-
3180.2010.00829.x.
Martinelli, L.a., Garrett, R., Ferraz, S., Naylor, R., 2011. Sugar and ethanol production as a
rural development strategy in Brazil: evidence from the state of São Paulo. Agric.
Syst. 104, 419–428. http://dx.doi.org/10.1016/j.agsy.2011.01.006.
Martinelli, L.a., Filoso, S., 2008. Expansion of sugarcane ethanol production in Brazil:
environmental and social challenges. Ecol. Appl. 18, 885–898.
Matsuoka, S., Stolf, R., 2012. Chapter 5 - sugarcane tillering and ratooning: key factors for
a profitable cropping. In: Goncalves, J.F., Correia, K.D. (Eds.), Sugarcane: Production,
Cultivation and Uses. Nova Science Publishers, pp. 137–157.
Mulla, D.J., 2013. Twenty five years of remote sensing in precision agriculture: key ad-
vances and remaining knowledge gaps. Biosyst. Eng. 114, 358–371. http://dx.doi.
org/10.1016/j.biosystemseng.2012.08.009.
Nassar, A.M., Rudorff, B.F.T., Antoniazzi, L.B., Aguiar, D.A.De, Bacchi, M.R.P., Adami, M.,
2008. Chapter 3 Prospects of the sugarcane expansion in Brazil : impacts on direct
and indirect land use changes. In: Sugarcane Ethanol: Contributions to Climate
Change Mitigation and the Environment. Wageningen Academic Publishers, The
Netherlands, pp. 63–93.
Novack, T., Kux, H.J.H., Feitosa, R.Q., Costa, G.a, 2010. Per block urban land use inter-
pretation using optical VHR data and the knowledge-based system Interimage. Int.
Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 38, 6p.
Peña, J.M., Torres-Sánchez, J., de Castro, A.I., Kelly, M., López-Granados, F., 2013. Weed
mapping in early-season maize fields using object-based analysis of unmanned aerial
vehicle (UAV) images. PLoS One 8, 1–11. http://dx.doi.org/10.1371/journal.pone.
0077151.
Peña, J.M., Torres-Sánchez, J., Serrano-Pérez, A., de Castro, A., López-Granados, F., 2015.
Quantifying efficacy and limits of unmanned aerial vehicle (UAV) technology for
weed seedling detection as affected by sensor resolution. Sensors 15, 5609–5626.
http://dx.doi.org/10.3390/s150305609.
Qin, R., 2014. An object-based hierarchical method for change detection using unmanned
aerial vehicle images. Remote Sens. 7911–7932. http://dx.doi.org/10.3390/
rs6097911.
Ray, J.D., Sinclair, T.R., Glaz, B., 2009. Sugarcane response to high water tables and
intermittent flooding. J. Crop Improv. 24, 12–27. http://dx.doi.org/10.1080/
15427520903304269.
Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1974. Monitoring vegetation systems
56
Computers and Electronics in Agriculture 143 (2017) 49–56
in the Great Plains with ERTS. In: NASA Goddard Space Flight Center 3d ERTS-1
Symposium, United States, pp. 309–317.
Rudorff, B.F.T., de Aguiar, D.A., da Silva, W.F., Sugawara, L.M., Adami, M., Moreira,
M.A., 2010. Studies on the rapid expansion of sugarcane for ethanol production in
São Paulo State (Brazil) using Landsat data. Remote Sens. 2, 1057–1076. http://dx.
doi.org/10.3390/rs2041057.
Shortis, M.R., Seager, J.W., Harvey, E.S., Robson, S., 2005. Influence of Bayer filters on
the quality of photogrammetric measurement. Electron. Imaging 5665, 164–171.
http://dx.doi.org/10.1117/12.588217.
Spekken, M., de Bruin, S., 2012. Optimized routing on agricultural fields by minimizing
maneuvering and servicing time. Precis. Agric. 14, 224–244. http://dx.doi.org/10.
1007/s11119-012-9290-5.
Spekken, M., Molin, J.P., Romanelli, T.L., 2015. Cost of boundary manoeuvres in su-
garcane production. Biosyst. Eng. 129, 112–126. http://dx.doi.org/10.1016/j.
biosystemseng.2014.09.007.
Stolf, R., 1986. Methodology for gap evaluation on sugarcane lines. STAB 4, 12–20.
Swain, K.C., Thomson, S.J., Jayasuriya, H.P.W., 2010. Adoption of an unmanned heli-
copter for low-altitude remote sensing to estimate yield and total biomass of a rice
crop. Trans. ASABE 53, 21–27.
Torres-Sánchez, J., López-Granados, F., De Castro, A.I., Peña-Barragán, J.M., 2013.
Configuration and specifications of an unmanned aerial vehicle (UAV) for early site
specific weed management. PLoS One 8, 1–15. http://dx.doi.org/10.1371/journal.
pone.0058210.
Torres-Sánchez, J., López-Granados, F., Peña, J.M., 2015a. An automatic object-based
method for optimal thresholding in UAV images: application for vegetation detection
in herbaceous crops. Comput. Electron. Agric. 114, 43–52. http://dx.doi.org/10.
1016/j.compag.2015.03.019.
Torres-Sánchez, J., López-Granados, F., Serrano, N., Arquero, O., Peña, J.M., 2015b.
High-throughput 3-D monitoring of agricultural-tree plantations with unmanned
aerial vehicle (UAV) technology. PLoS One 10, 1–20. http://dx.doi.org/10.1371/
journal.pone.0130479.
Torres-Sánchez, J., Peña, J.M., de Castro, A.I., López-Granados, F., 2014. Multi-temporal
mapping of the vegetation fraction in early-season wheat fields using images from
UAV. Comput. Electron. Agric. 103, 104–113. http://dx.doi.org/10.1016/j.compag.
2014.02.009.
Turner, D., Lucieer, A., Wallace, L., 2014. Direct georeferencing of ultrahigh-resolution
UAV imagery. IEEE Trans. Geosci. Remote Sens. 52, 2738–2745. http://dx.doi.org/
10.1109/TGRS.2013.2265295.
Willmott, C.J., Robeson, S.M., Matsuura, K., 2012. A refined index of model performance.
Int. J. Climatol. 32, 2088–2094. http://dx.doi.org/10.1002/joc.2419.
Xavier, A.C., Rudorff, B.F.T., Shimabukuro, Y.E., Berka, L.M.S., Moreira, M.A., 2006.
Multi-temporal analysis of MODIS data to classify sugarcane crop. Int. J. Remote
Sens. 27, 755–768. http://dx.doi.org/10.1080/01431160500296735.
Yu, Q., Gong, P., Clinton, N., Biging, G., Kelly, M., Schirokauer, D., 2006. Object based
detailed vegetation classification with airborne high spatial resolution remote sensing
imagery. Photogramm. Eng. Remote Sensing 72, 799–811. http://dx.doi.org/10.
14358/PERS.72.7.799.
Zarco-Tejada, P.J., González-Dugo, V., Berni, J.a.J, 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 Sens. Environ. 117,
322–337. http://dx.doi.org/10.1016/j.rse.2011.10.007.
Zhang, C., Kovacs, J.M., 2012. The application of small unmanned aerial systems for
precision agriculture: a review. Precis. Agric. 13, 693–712. http://dx.doi.org/10.
1007/s11119-012-9274-5.
Zhang, N., Wang, M., Wang, N., 2002. Precision agriculture—a worldwide overview.
Comput. Electron. Agric. 36, 113–132. http://dx.doi.org/10.1016/S0168-1699(02)
00096-0.