A Digital Countryside Notebook for Smart Agriculture and Oranges
Classification
T. Rotondo
1
, G. M. Farinella
1
, A. Chillemi
1
, F. Ferlito
2
and S. Battiato
1
1
Department of Mathematics and Computer Science, University of Catania, Italy
2
Consiglio per la Ricerca in Agricoltura e l’Analisi dell’Economia Agraria,
Centro di Ricerca Olivicoltura, Frutticoltura e Agrumicoltura. Acireale (CT), Italy
Keywords:
Smart Agriculture, Countryside Notebook, Fruit Classification.
Abstract:
We present a digital countryside notebook designed to help land owners to monitor the operations performed
on a cultivation. The system helps to collect and trace information over time through the developed mobile
App. This guarantees the traceability of the product. Our system has many advantages, such as the easy
collection of information and the reduction of the time in the analysis of the information acquired. To improve
and automate the process of data collection, the system uses a classifier to label images of oranges during the
plant monitoring.
1 INTRODUCTION
Agriculture is one of the most important economic
sector and source of employment. In the next forty
years, the agriculture production should be increased
by 60%. This means to duplicate the production per
hectare, but in some areas the production can’t be in-
creased over 39%. Nowadays, in some areas the pro-
duction is actually decreasing (United Nation, 2017).
Smart Agriculture is the application of ICT into agri-
culture domain. It aims to introduce important inno-
vations in the whole agricultural production sector.
One of the main goals is to increase the productiv-
ity by reducing both, economic costs and natural re-
sources (such as water).
Most of the solutions for Smart Agriculture are based
on Internet of Things (IoT) (Prathibha et al., 2017),
i.e. devices which are capable of acquiring data to
be analyzed to help the land owner a better manage-
ment of the production. Considering that in the mid-
dle of 2017 the world’s population numbered nearly
7.6 billion (United Nation, 2017), and that continues
to grow, it is important to build smart infrastructures
which can be useful to increase agriculture produc-
tion. Indeed, it is estimated that the population will be
9.8 billion in 2050 and 11.2 billion by 2100. Thanks
to IoT infrastructures, the land owner can be able to
connect devices and collect information of interest in
agriculture domain. Different devices are used to help
agriculture become smart, such as drones, sensors to
drive irrigation pumps, solar panels for energy pro-
duction, sensors to monitor humidity, etc. (Varghese
et al., 2015; Codreanu et al., 2014; Sathyadevanet al.,
2011)
In (Lipper et al., 2014) is described a Climate-Smart
Agriculture (CSA) approach useful to transform and
reorient agricultural systems to support food security
under the new reality of climate changes. In (Gond-
chawar and Kawitkar, 2016), a smart GPS based re-
mote controlled robot is employed to perform tasks
like weeding, spraying, moisture sensing, bird and an-
imal scaring, keeping vigilance, etc. The system in-
cludes automatic irrigation with smart sensors and in-
telligent decision making procedures based on accu-
rate real time field data analysis and warehouse man-
agement. In (C. Wouter Bac and Edan, 2014) a re-
view and challenges related to harvesting robots to
help crops in agriculture is presented. In (Sales et al.,
2015), Wireless Sensor Networks (WSN) are used.
The nodes of the network perform acquisition, collec-
tion and analysis of data, such as temperature and soil
moisture, that can be employed to automate the irri-
gation process in agriculture while decreasing water
consumption. This implicates in monetary and envi-
ronmental benefits.
In (Murabito et al., 2017) is presented a tool for gen-
erating knowledge-enriched visual annotations of 24
fruit varieties which they use to build a benchmark
dataset for a complex fruit classification problem.
Growers and breeders of Blood oranges were able to
Rotondo, T., Farinella, G., Chillemi, A., Ferlito, F. and Battiato, S.
A Digital Countryside Notebook for Smart Agriculture and Oranges Classification.
DOI: 10.5220/0006845303810385
In Proceedings of the 15th International Joint Conference on e-Business and Telecommunications (ICETE 2018) - Volume 1: DCNET, ICE-B, OPTICS, SIGMAP and WINSYS, pages 381-385
ISBN: 978-989-758-319-3
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
381

identify and collect many somatic mutants, which dif-
fered in their major pomological traits, such as fruit
size and firmness, pulp and peel pigmentation, and
ripening period. In (Caruso et al., 2016), a dataset of
88 varieties of blood orange is collected. Pomologi-
cal parameters, such as fruit weight, equatorial diam-
eter, polar meter, width of central axis, peel thickness,
number of seeds, juice yield, total soluble solids, acid-
ity, pH, peel color index, pulp color index, total an-
thocyanin, are analyzed trough principal component
analysis (PCA). In our experiments, we used the sub-
set of this dataset containing only the images. In par-
ticular, we have been considered three orange species,
namely, “Tarocco a Buccia Gialla”, “Tarocco Meli”
and “Tarocco Rosso” (Rapisarda and Russo, 2000).
In this paper, we propose a system useful to monitor
land property based on IoT cloud platform. It involves
the dislocation of a series of sensors on the agricul-
tural field in order to detect all the interesting param-
eters to monitor and manage the land property,such as
temperature, humidity, solar irradiation, etc. Through
a wireless network, a cloud platform receives and ex-
amines these data. The proposed system represents
a first prototype with respect to a classic countryside
notebook. The farmers usually use this document to
report every operation on the product, ensuring its
traceability. More specifically, we propose a digital-
ization of countryside notebook and develop a ”K-NN
Oranges Variety Classifier” to help the land owner for
the automatic labelling of the acquired images of an
orange plant among different varieties. The mobile
App was developed to work on an Android operating
system.
In the following sections, we first describe the build-
ing blocks of our system. Finally, we report experi-
mental results and conclusion.
2 PROPOSED SYSTEM
A countryside notebook is an official documentwhich
reports all the information collected by farmers which
are related to treatments, for instance in case of culti-
vation diseases, herbicides used, time and method of
administration of treatments. The countryside note-
book is useful to guarantee the traceability of the
product. The filling procedure of the countryside
notebook has been done by using paper documents
so far. This slows down the collection and analysis
of data and strongly limits the amount of information
that can be collected. In paper format the information
and results of possible analysis aren’t immediately
available to both the producer and the consumer. Usu-
ally, the countryside notebook contains information
Figure 1: Screenshots of our countryside notebook.
about the company, the lands, the crops, phytosani-
tary treatments, herbicides used, irrigations, parasitic
diseases, ground processing, pruning operations.
We created a digital version of the countryside note-
book, as shown in Figure 1. There are many ad-
vantages in digitalizing it. As first, it simplifies the
acquisition of data. As second aspect, the time for
the analysis of collected data decreases. More ad-
vantages come from the fact that the proposed coun-
tryside notebook is able to automatically process the
voice of the operator (speech to text) to transcribe the
text related to the different involved tasks. Finally, the
proposed system gives to the farmer the possibility to
acquire images of the cultivation (oranges in our case)
and automatically classify the variety. Hence, with
the proposed countryside notebook the full story of a
cultivation can be stored and eventually analyzed over
years.
Figure 2 shows the overall scheme of the system. The
countryside notebook has been developed as a mobile
application with a client side implemented in Java for
Android platform. It uses RESTful resources from a
back-end encoded in NodeJS where data are stored in
MongoDB.
We chose model view controller (MVC) as design
pattern. This allows to separate the logic of presenta-
tion, modification and insertion of data from business
logic. In our system, the three main roles are divided
as follows: model is the Node.js server to manage in-
teractions with MongoDB storage, view is an Android
user graphic interfaces to capture and display data and
controller is written in Java with Android SDK to al-
low interactions between the other two components.
The client side is described as follows. For the user
authentication, we created an ad-hoc login page that
requires username and password. If the company
isn’t registered, can be registered with the appropriate
form. After authentication, the user can create a new
countryside notebook or view the list of notebooks al-
ready presented. For the management of information
related to the notebook, it has been implemented a
SIGMAP 2018 - International Conference on Signal Processing and Multimedia Applications
382

Figure 2: Overall scheme of the proposed System.
Tab-menu with Expandable ListView. In particular
to each tab corresponds the state of a plant, whereas
to each field in the ListView corresponds to a more
specific sub-phase. By clicking on the sub-phase,
the ListView is expanded and it shows two clickable
fields: ”New” and ”Show All”. Touching on ”New”,
an insertion form opens to register a new cultivar card.
By clicking on ”Show all”, instead, a list is displayed
with the cultivars previously enter with the ability to
view, edit and delete them. The last Tab includes soil
tillage and cultivation operations (pruning) and it is
organized in a similar way to the others.
In order to guarantee system scalability, we imple-
mented a client server architecture. The application
server has been implemented in NodeJS and Storage
of the data was made in MongoDB. Images are cap-
tured from the smartphone ’s camera, encoded into
base64 strings and sent to the server. Similarly, the
server sends the strings encoded to the client that pro-
vides the decryption. The timestamp is generated on
the server side and sent to the client that will make
it visible every time insertion and modification oper-
ations are carried out. This approach guarantees the
authenticity of the timetables and dates of compila-
tion of the countryside notebook.
The implementation of the digital countryside note-
book facilitates the process of information manage-
ment. In order to improve and further automate the
process of analyzing the information acquired, we
insert a form called ”K-NN Oranges variety classi-
fier”. This form is related to the management of or-
ange crops. During the orange harvesting phase and
when filling out the appropriate form in the country-
side notebook, the operator can take pictures of the
product that has just picked up. These photos are at-
tached to the other information included in the coun-
tryside notebook. The pictures taken are stored on the
remote server. The server automatically runs the clas-
sification module that identifies the variety of orange
under consideration and sends the result of the classi-
fication to the application client-side.
The K-NN Oranges variety classifier module is based
on the Bag of Visual Word representation paradigm
(BoVW) (Bosch et al., 2007; Battiato et al., 2010;
Farinella et al., 2014; Farinella and Battiato, 2011)
to extract features. This approach converts the set of
local descriptor into the final image representation.
This technique was first proposed for text document
analysis, but it is applied to images by using a visual
analogue of a word. To extract the BoW feature from
images, we detect regions/points of interest, compute
local descriptors over those regions/points, quantize
the descriptors into words to form the visual vocab-
ulary and find the occurrences in the image of each
specific word in the vocabulary for constructing the
BoW features (or a histogram of word frequencies).
A k-nearest-neighbor classification algorithm (K-NN)
is used on the BoVW representation for classification
purposes (Bishop, 2006).
3 EXPERIMENTS AND RESULTS
In this section, we show and discuss the results of the
”K-NN Oranges variety classifier” module. It is an
image classification module of oranges that takes in-
put a digital image of an orange view in section (cut
in half) returns the variety of it.
3.1 Dataset and Preprocessing
The dataset used for test purpose have been provided
by the public organization called Consiglio per la
ricerca e la sperimentazione in agricoltura (C.R.E.A).
The dataset is composed by 1,391 images of three
orange species, namely, Tarocco a Buccia Gialla,
Tarocco Meli and Tarocco Rosso (see Figure 3). For
each class, the oranges were collected from four dif-
ferent points of view but, in our experiments, we con-
sider only sectional view.
We crop the images with a rectangular box of size
(height, width)=

h +
h
4
,
h
2

, where h is the distance
between the center and the border of the orange, as
shown in Figure 4.
3.2 Results
The K-NN Oranges variety classifier is based on Bag
of Visual Word Model (BoVW) algorithm for feature
extraction. In our case, we use Daisy (Tola et al.,
A Digital Countryside Notebook for Smart Agriculture and Oranges Classification
383

Figure 3: Examples of oranges varieties. From left to right: Tarocco a Buccia Gialla, Tarocco Meli and Tarocco Rosso.
Figure 4: Image cropping.
Figure 5: BoW representation.
2010) as local feature to build the visual vocabulary.
To create a visual dictionary, we use a k-means
clustering algorithm with k = 500. The images
represent as BoW are used with a K-NN classifier.
Figure 5 shows a representation of an orange image.
The system has been tested splitting training e
test randomly. The K-NN algorithm finds the three
images of training set that resemble the image query.
The figure 6 shows the obtained result where the
query image is at the top left. Note that the query
image is a Tarocco Meli orange and the other images
belong to the same class.
We evaluate the K-NN classifier for different values
of k. The results of accuracy is shown in the Table 1.
We obtain the same values of accuracy for k=5 and
k=7. In Table 2, the confusion matrix for k = 5 is
reported.
Figure 6: Query result example.
4 CONCLUSION
In this paper, we present a digitalization of country-
side notebook for oranges. We want to increase the
dataset with other varieties of oranges and we plan to
use the images together with physio/chemical data to
SIGMAP 2018 - International Conference on Signal Processing and Multimedia Applications
384

Table 1: Accuracy values for different k.
Value of k 3 5 7 9
Accuracy 0.90 0.96 0.96 0.94
Table 2: Confusion Matrix.
Meli Giallo Rosso
Meli 100% 0 0
Giallo 0 86.67% 13.33 %
Rosso 0 0 100%
increase the scientific knowledge of these products.
We also plan to extend this notebook for other crops
introducing also other analysis tools. The system has
been designed to allow the farmer to trace cultivation
over time and to collect data that can be automatically
analysed.
Future work will be developed to produce automatic
procedure which can be used for decision making and
comparing advanced methodologies for image repre-
sentation and classification (e.g. deep learning).
ACKNOWLEDGEMENTS
We thank C.R.E.A for providing the dataset. We also
thank Joint Open Lab Wave of Catania (a research
laboratory of TELECOM) for helpful comments.
REFERENCES
Battiato, S., Farinella, G., Gallo, G., and Rav´ı, D. (2010).
Exploiting textons distributions on spatial hierarchy
for scene classification. EURASIP JOURNAL ON IM-
AGE AND VIDEO PROCESSING, 2010:1–13.
Bishop, C. M. (2006). Patter Recognition and Ma-
chine Learning (Information Science and Statistics).
Springer-Verlag, Berlin, Heidelberg.
Bosch, A., Muoz, X., and Mart, R. (2007). Which is the best
way to organize/classify images by content? Image
and Vision Computing, 25(6):778 – 791.
C. Wouter Bac, Eldert J. van Henten, J. H. and Edan, Y.
(2014). Harvesting robots for high-value crops: State-
of-the-art review and challenges ahead. Journal of
Field Robotics, 31(6):888–911.
Caruso, M., Ferlito, F., Licciardello, C., Allegra, M.,
Strano, M., Di Silvestro, S., Patrizia Russo, M.,
Pietro Paolo, D., Caruso, P., Las Casas, G., Stagno, F.,
Torrisi, B., Roccuzzo, G., Reforgiato Recupero, G.,
and Russo, G. (2016). Pomological diversity of the
italian blood orange germplasm. 213.
Codreanu, N., Varzaru, G., and Ionescu, C. (2014). So-
lar powered wireless multi-sensor device for an irriga-
tion system. In Proceedings of the 2014 37th Inter-
national Spring Seminar on Electronics Technology,
pages 442–447.
Farinella, G. and Battiato, S. (2011). Scene classification in
compressed and constrained domain. IET Computer
Vision, pages 320–334.
Farinella, G., Moltisanti, M., and Battiato, S. (2014). Clas-
sifying food images represented as bag of textons. In
2014 IEEE International Conference on image pro-
cessing, ICIP 2014. IEEE (The Institute of Electrical
and Electronics).
Gondchawar, N. and Kawitkar, R. S. (2016). Iot based smart
agriculture. 5:838–842.
Lipper, L., Thornton, P., Campbell, B., Baedeker, T.,
Braimoh, A., Bwalya, M., Caron, P., Cattaneo, A.,
Garrity, D., Henry, K., Hottle, R., Jackson, L., Jarvis,
A., Kossam, F., Mann, W., McCarthy, N., Meybeck,
A., Neufeldt, H., Remington, T., and Torquebiau, E.
(2014). Climate smart agriculture for food security.
4:10681072.
Murabito, F., Palazzo, S., Spampinato, C., and Giordano, D.
(2017). Generating knowledge-enriched image anno-
tations for fine-grained visual classification. In Image
Analysis and Processing - ICIAP 2017, pages 332–
344, Cham. Springer International Publishing.
Prathibha, S. R., Hongal, A., and Jyothi, M. P. (2017). Iot
based monitoring system in smart agriculture. In 2017
International Conference on Recent Advances in Elec-
tronics and Communication Technology (ICRAECT),
pages 81–84.
Rapisarda, P. and Russo, G. (2000). Fruit quality of five
tarocco selections grown in italy. pages 1149–1153.
Sales, N., Remdios, O., and Arsenio, A. (2015). Wireless
sensor and actuator system for smart irrigation on the
cloud. In 2015 IEEE 2nd World Forum on Internet of
Things (WF-IoT), pages 693–698.
Sathyadevan, S., Kallingalthodi, H., and Hari, N. N.
(2011). Irignet: Intelligent communication network
for power-scarce rural india. In Proceedings of the
1st International Conference on Wireless Technolo-
gies for Humanitarian Relief, ACWR ’11, pages 465–
471, New York, NY, USA. ACM.
Tola, E., Lepetit, V., and Fua, P. (2010). Daisy: An effi-
cient dense descriptor applied to wide-baseline stereo.
IEEE Transactions on Pattern Analysis and Machine
Intelligence, pages 815–830.
United Nation, Department of Economic e Population Divi-
sion United Nations Social Affairs, D. (2017). World
population prospects, the 2017 revision.
Varghese, V. T., Sasidhar, K., and P, R. (2015). A status quo
of wsn systems for agriculture. In 2015 International
Conference on Advances in Computing, Communica-
tions and Informatics (ICACCI), pages 1775–1781.
A Digital Countryside Notebook for Smart Agriculture and Oranges Classification
385