=Paper= {{Paper |id=Vol-1152/paper29 |storemode=property |title=The Autonomous Detection of Tree Position with Different GPS Devices for the Need of Early Apple Yield Forecast |pdfUrl=https://ceur-ws.org/Vol-1152/paper29.pdf |volume=Vol-1152 |dblpUrl=https://dblp.org/rec/conf/haicta/StajnkoL11 }} ==The Autonomous Detection of Tree Position with Different GPS Devices for the Need of Early Apple Yield Forecast== https://ceur-ws.org/Vol-1152/paper29.pdf

The Autonomous Detection of Tree Position with
Different GPS Devices for the Need of Early Apple Yield
Forecast
Denis Stajnko, Miran Lakota

University of Maribor, Faculty of Agriculture and Life Sciences, Department for biosystem
engineering Pivola 10, SI-2311 Hoce, Slovenia, email: denis.stajnko@uni-mb.si

Abstract. The accuracy of different GPS devices for sampling of trees
required for the early yield forecast was studied in two years’ campaign. In our
experiment March II, PDA ASUS P565 and Nokia 5800 XpressMusic were
used. To determine the positions and their accuracy, 245 randomly trees were
selected from 245 orchards. We found out that the measurements made in two
years with MARCH II deviated in average in diagonal for 16.72 m, in ASUS
P565 for 15.41 m and in Nokia 5800 XpressMusic for 46.14 m. ASUS P565
was found to be the most accurate device because the two season’s
measurements deviated in average in diagonal only for 1.31 m from MARCH
II, but there was no significant difference. Despite the fact that discrepancies in
individual trees were minimal (0.33 m) in particular measurement, those two
devices are not precise sufficiently to identify unequivocally the position of
sample trees.

Keywords: GPS / positioning / orchard / sampling

1 Introduction

In the European Union (EU-25), about 10 million tons of apples are harvested yearly,
but with great fluctuations from year to year and from orchard to orchard.
Yield prediction is hence a pre-requisite for all partners in the food chain; orchard
owners, trade, shippers and retailers all require data on fruit quantity at different fruit
growth stages, since a tree bearing an excessive number of fruits will yield small,
undersized fruits. Thus, modelling of fruit growth with an emphasis on tree
variability is a crucial step in the management of fruit quantity and quality through
horticultural practices (Lescourret et al., 1998) with a great impact on yield
prediction per hectare in every growing region. Oriade and Dillon (1997)
investigated the variability of fruit growth by using a stochastic approach of fruit
growth rates and considering the sink strength of the fruit. However, all these models
simulate the environmental conditions in the orchard, which can significantly vary
from the real values. To overcome these disadvantages, Welte (1991) refined the
original Bavendorf model (Winter, 1986) by introducing orchard measurements in
________________________________
Copyright ©by the paper’s authors. Copying permitted only for private and academic purposes.
In: M. Salampasis, A. Matopoulos (eds.): Proceedings of the International Conference on Information
and Communication Technologies
for Sustainable Agri-production and Environment (HAICTA 2011), Skiathos, 8-11 September, 2011.

333
the mechanistic models. However, fruits were still manually counted on the tree in
the orchard by visual inspection on a few selected sample trees, which was a labour
intensive and time consuming procedure, hence leaving this method inappropriate for
modelling fruit yield in an individual orchard as required by the fruit industry
(Stajnko et al., 2004).
Due to time-consuming counting and the lack of experts a lot of inaccurate forecasts
appeared in Slovenia from 1998-2004. From these reasons the Bavendorf method
was changed by the method of image analysis introduced by Stajnko et al. (2004,
2009). However, even though the method itself was very accurate for forecasting the
yield on particular parcel, the small number of sample orchard made it rather vague
for entire country. It happened in the individual years that the estimated yield
differed from the actual production for 5 to 25%.
Therefore it was decided in 2009 to increase the sample population to 245 orchards
selected form the register of orchards. However, for accurate forecasting, in which
‘in situ’ samples are used for predictions, it is essential that they are always taken at
the same location. Then we can talk about the real data with which we can design the
organization in advance.
The main advantage of image analysis method is its possibility to capture a lot of
images in a variety of orchards in a short time in order to improve forecast accuracy.
However, it is desirable very much that each year the images are captured from the
same trees. Everyone who was already sampling the trees from the orchards is aware
that it is very difficult to locate the positions without any additional marks, as it
requires a lot of walking.
The solution of these problems should represent a global positioning system (GPS) as
the most frequent method for localization (Panzieri et al. 2007), which enables the
user to have access to selected trees for the taking of samples quickly and easily.
Since in the sampling method of image analysis we already have camera, we were
interested in researching the smart phone with built-in GPS receiver as a substitute
for professional GPS equipment.
The objective of our research was to determine whether three different GPS devices
(March II, ASUS P565 and Nokia 5800 XpressMusic) lead us to the same position of
sample tree, even one year after the first measurement was taken. Another aim was
also to determine whether the hail network that are installed on the new plantations
affect the signal reception and accuracy of measurements.

2 Material and Methods
From the Slovenian register of intensive crops, owned by the Ministry of Agriculture,
245 locations of the orchards were selected in June 2009 according to the different
apple varieties, planting year and growing form. In the same time when the samples
of tree images were taken from the particular orchard four tees were additionally
marked by the plastic label (Fig. 1). One year later in June 2010 the samples of
images were captured again and the position of particular mark tree on all three
devices was checked.

334
Fig. 1. A sample of marked tree (left), the orto-photo image of selected orchard with sampled
trees (right)

For our research three different GPS devices were used according to the price and
accuracy reference.

2.1 Description of Devices

March II is a professional hand-held GPS receiver for field data capture. It uses a
Motorola integrated 8-channel GPS receiver whose referencing provides a horizontal
accuracy of less than 3 m. Its main characteristic is a compact unit, which integrates
the receiver, antenna, computer and software that is designed to easily capture the
spatial data. Data were gathered in the field based on a pre-established list of objects
that we want to record (map) and their properties.

ASUS P565 is also called a ‘smart phone’ driven by an 800 MHz processor and
operates on Windows Mobile 6.1 environment and built-in digital camera with a
resolution of 3.0 mega pixels. It has got good sensitivity (<-159 dBm), a positioning
accuracy (<2.5 m), quick start (standby <1 s), small size (3.12 x 3.17 x 0.4 mm) and
the ability to track the frequency L1 (1575.42 MHz). For this handheld PDA in 2008
software Garmin Mobile XT was uploaded and a program FK mobile was developed
by Šinjur et al. (2008), which enables autonomous guiding to the marked trees by
integrated compass function.

Nokia 5800 XpressMusic is a music-oriented GSM phone, which boasts a modern
attractive design, but also is practical and user friendly. The phone has a touch screen
and built-in digital camera with a resolution of 3.2 mega pixels. It supports GPS for
car navigation and pedestrian navigation and Nokia Maps 2.0 Touch. The device also
supports Assisted GPS (A-GPS), which is used to provide packet data connection,
which assists in the calculation of the coordinates of the current location when your
device is receiving signals from satellites.

335
2.2 Procedure for Calculating the Differences

All devices used in the experiment operate in a coordinate system WGS84 which
refers the position in the usual geographical coordinate format (Lat. 13 ° 39’48.7’’,
Lon. 45 ° 57’ 41.1’’), however a direct subtract for calculating differences in position
is not possible. Therefore, we applied the Excel forms (Fig. 2), which first record the
geographical coordinates of degrees, minutes, and seconds in the decimal
representation (column E) and in coordinate form (column F). Once we got the
difference in decimal format we had to convert it again back in the geographical
coordinates (column G). Finally, the difference in meters was calculated from
geographical units so that the (latitude) was multiplied by 31 m and ! (longitude)
by 22 m (column H).

Fig. 2. A sample procedure for calculating the difference in m between two measurements

Statistical analysis was performed with a SPSS Statistics 17.0 for Windows ® as the
most familiar and widespread statistical programs in Slovenian education and
research field. To analyze the difference between three GPS devices a ‘paired
samples analysis’ at " <0.05 was used.

3 Results
The mean differences and standard deviations between the 2009 and the 2010
measurements for the device ASUS P565 is represented in Table 1, which shows that
the mean difference for X coordinate was 3.48 m and for Y coordinate 13.85 m, with
standard deviations of 3.35 m for X and 17.87 m for Y. So in the diagonal the mean
difference was 15.41 m with standard deviation of 17.21 m. With those results ASUS
P565 was proved to be statistically most accurate only in X measurements, whereby
Y measurements did differ significantly only from Nokia 5800 XpressMusic.
In MARCH II, the mean difference between 2009 and 2010 measurements was 11.41
m for X coordinate and 10.56 m for Y coordinate, respectively, with standard
deviations of 9.48 m for X and 10.68 m for Y. So in the diagonal the mean
calculated difference was 16.72 m with standard deviation of 12.84 m. Those

336
measurements did not differ significantly from the ASUS P565, but they were much
better from Nokia 5800 XpressMusic.

Table 1. The mean and standard deviation between two measurements for three different GPS
devices

Device X-measurements Y-measurements Diagonal
(m) (m) (m)
Mean Std. Mean Std. Mean Std.
Deviation Deviation Deviation
ASUS P565 3.48* 3.35 13.85* 17.87 15.41* 17.21
MARCH II 11.41 9.48 10.56* 10.68 16.72* 12.84
Nokia 5800
XpressMusic 11.37 15.19 39.60 49.83 46.14 47.61

*
statistically significant at p# 0.05 (Paired Samples Test).

The mean difference between all measurements done with Nokia 5800
XpressMusicfor showed 11.37 m difference in X coordinate and 39.60 m for Y
coordinate with standard deviations of 15.19 m for X and 49.83 m for Y
measurements. So in the diagonal the mean difference was 46.14 m with standard
deviation of 47.61 m, which was significantly different from both devices ASUS
P565 as well as MARCH II. Therefore, the Nokia 5800 XpressMusic device was
found to be inaccurate for precise horticulture. For example, once standing 30 m deep
in the orchard, the device would still indicate the position coordinates outside the
plantation.

The main reason for not receiving declared accuracy lay in the weather conditions
and the number of satellites, which was not ideal. Anyway, on the average we
achieve a relative accuracy of approximately ± 10 m. It turns out that the error in
measurements was most affected by the number of satellites, given in our situation.
For example, when we performed measurements of 08/26/2009, we received on the
average a signal from eight satellites, while on 07/01/2010 we accepted the signal
from just 6 satellites, which was otherwise satisfactory to take measurements, but
obviously not enough to improve accuracy.
A very important finding is also the fact that a hail network did not affect the signal
quality, since the same number of satellites was detected under and outside the
network.

4 Conclusions
Absolute accuracy of the declared facilities for MARCH II and ASUS P565 applied
in the experiment would be up to ± 10 m, except Nokia 5800, where the producer
recalls that it should not be used for very precise measurements. The manufacturer
contends, moreover, that the accuracy of the MARCH II and ASUS P565 is even
below 2.5 m, but in practice it was very difficult to achieve.

337
We can make a general conclusion that in spite of particular very precise
measurements, these devices are not sufficiently reliable to indicate the position of
sample trees with less than 0.7 m, which is usual space between trees in the lines.
They can be applied only for determining a wider space of orchard from which the
samples were taken one year before. This fact can help us or another person in
finding easily the sampling zone, which is without doubt a very useful tool for saving
time in locating the correct part of orchards.

Acknowledgement
This article was partly created as a result of the applied project "Preparation of a
register of yields of apples and pears by visualization of the volume of native trees"
(V4-0537). The project was funded by the Public Research Agency of the Republic
of Slovenia and the Ministry of Agriculture, Forestry and Food of the Republic of
Slovenia.

References

1. Lescourret, F., Mimoun, M.B. and Génard, M. (1998) A simulation model
of growth at the shoot-bearing fruit level, I. Description and
parameterization for peach, European Journal of Agronomy, 9, p. 173–188.
2. Oriade, C.A. and Dillon, C.R. (1997) Developments and biophysical and bio
economic simulation of agricultural systems: a review. Agricultural
Economics 17, 45-48.
3. Panzieri, S., Pascucci, F. and Ulivi G. (2007) An Outdoor Navigation
System Using GPS and Inertial Platform, Mechatronics, IEEE/ASME
Transactions on, Volume 7, Number 2, p. 134 -142.
4. Stajnko, D., Lakota, M. and Ho$evar, M. (2004). Estimation of number and
diameter of apple fruits in the orchard during the growing season by thermal
imaging. Computers and Electronics in Agriculture, 42(1), p. 31-42.
5. Stajnko, D., Rakun, J. and Blanke, M. (2009) Modelling apple fruit yield
using image analysis for fruit colour, shape and texture. European journal of
horticultural science, 74 (6) p. 260-267.
6. Šinjur, S., Zazula, D. and Stajnko, D. (2009). Sistem za avtomatsko
lociranje drevesnih nasadov in zajemanje podatkov o pridelku na terenu (=A
system for automated plantation location and field data acquisition on the
harvest). V: ZAJC, Baldomir (ur.), TROST, Andrej (ur.). Zbornik
Osemnajste mednarodne elektrotehniške in ra unalniške konference - ERK
2009, 21-23. september 2009, Portorož, Slovenija. Ljubljana: IEEE Region
8, Slovenska sekcija IEEE, zv. B, p. 59-62.
7. Winter F. (1986): Modelling the biological and economic development of an
apple orchard, Acta Horticulturae 160, p. 353-360.
8. Welte H. (1991): Analyse und Simulation des Fruchtwachstums von Äpfeln,
Grauer Verlag, Stuttgart, 161 p.

338