Professional Documents
Culture Documents
The Future of Digital Agriculture
The Future of Digital Agriculture
net/publication/339185329
CITATIONS READS
9 2,118
5 authors, including:
Some of the authors of this publication are also working on these related projects:
Multidisciplinary Special Issue "Multi-source data fusion of big spatial-temporal data in soil, geo-engineering and environmental studies" (Science of the Total
Environment) View project
INNOSETA - H2020 Thematic network Innovative Spraying Equipment Training Advising View project
All content following this page was uploaded by Spyros Fountas on 15 May 2020.
& THE DIGITAL REVOLUTION is transforming agri- Precision agriculture management systems are
culture by using modern machinery, computer- allowing growers to benefit from this new tsunami
ized tools, and information and communication of data they can gather. These systems collect,
technologies (ICTs) to improve decision making classify, and analyze vast amounts of data to
and productivity. The spread of several cutting- detect patterns and solutions. They enable farm-
edge technologies, from GPS and remote sensing ers to observe, comprehend, and manage the vari-
to big data, artificial intelligence and machine ability in their production systems by tailoring
learning, robotics, and the Internet of Things inputs to get desired outputs. GPS-controlled trac-
(IoT), to agriculture is leading to increased tors can work around the clock, ploughing, plant-
yields, lower costs, and reduced environmental ing, and harvesting while gathering continuous
impact. Data-driven solutions are unlocking pro- “on-the-go” georeferenced data. These self-driving
duction potential in a sustainable and resource- vehicles can perform precise operations, with the
efficient way. help of GPS, Geographical Information Systems
(GIS), and Variable-Rate Technology (VRT).
Digital Object Identifier 10.1109/MITP.2019.2963412 Weather stations supply a variety of agricul-
Date of current version 12 February 2020. ture-specific weather data. Such weather data
January/February 2020
25
Authorized licensed use limited to: Agricultural University of Athens. Downloaded on February 13,2020 at 08:46:07 UTC from IEEE Xplore. Restrictions apply.
Envisioning Our Future Digital World
Figure 1. View of a digital agricultural field that has IoT devices to sense soil moisture, robotic vehicles
(ground and aerial), and antennas to receive satellite images. Copyright: Authors; published with permission.
26 IT Professional
Authorized licensed use limited to: Agricultural University of Athens. Downloaded on February 13,2020 at 08:46:07 UTC from IEEE Xplore. Restrictions apply.
and commercial platforms are limited to task-spe- deep learning, such as transfer learning or cap-
cific operations, whereas the scalability to differ- sule networks, will lead future decision making
ent crops or environments is questioned. by taking into account several factors, such as
Furthermore, there are many interoperability and environmental conditions, harvesting practices,
standardization problems when it comes to agro- financial needs, soil characteristics, or water
robotics, with the communication of platforms availability.
and safety issues posing considerable barriers. Deep learning methods, specifically convolu-
Expanding the capabilities of robots from the lab tional nets (CNNs), are gaining fast growth for
and greenhouse environment to the outdoor con- the automation of crop classification and dis-
ditions is crucial, when it comes to sensing under ease/weed identification. The main reason is
harsh environments and operating under unpre- that contrary to the conventional machine-learn-
dicting conditions. Regardless of the technical ing techniques, CNN training allows automatic
barriers, there is considerable skepticism related learning representations of data with multiple
to social, economical, and ethical challenges in levels of abstraction. Although CNNs are very
agricultural robotics. Labor shift from repetitive data demanding, a technique called transfer
tasks to high-skilled engineering jobs and the learning is used to overcome this requirement
imbalanced adoption of agritechnologies by farm- by reusing patterns learned by state-of-the-art
ers is expected to create social implications. CNNs in the related tasks. Additionally, CNNs
Advanced intelligence and decision-making capa- have also shown promising results with other
bilities of robots results in debates about moral problems, such as yield prediction, by extracting
aspects and, overall, who has the responsibility patterns from satellite imagery.
for these actions. Moreover, expensive technolo- In the future, other techniques are expected to
gies and the demand for resources and infrastruc- emerge solve open problems in agriculture.
tures to launch robotics in the field will challenge Regarding to computer vision related problems,
the effective adoption of robotics from the feasi- such as crop classification and disease/weed
bility point of view. identification, new architectures, such as capsule
Moving toward the era of intelligence systems, networks, will address current CNN limitations.4
robotic platforms will be the cornerstone of agri- Contrary to CNNs, capsules (grouped neurons)
culture operations. UAV platforms will manage consider the spatial relationships between enti-
several light weight tasks, such as spraying and ties and learn these relationships via dynamic
health inspections, as well as supervising ground routing, utilizing a nonlinear function called
robot fleets operating in the field. In the future, squashing. In regard to the use of reinforcement
robots will be able to collaborate with humans learning instead of the supervised one, architec-
without compromising safety in tasks like harvest- tures, such as imagination-augmented agents, will
ing. To this end, robot cognition is expected to also play an important role, extending intelligence
reach supremacy levels, with the systems able to through imagination by learning to interpret pre-
have full environment awareness and clear reason- dictions from a learned environment model.5
ing of the decisions taken, for example, how to On the other hand, the extraction of informa-
treat rotten fruit during harvesting or interrupt tion from natural language documents contain-
spraying when humans are in close range. ing policies and regulations could be addressed
by a recurrent neural network (RNN). Specifi-
cally, a subtype of RNNs called long short-term
AI IN DIGITAL AGRICULTURE memories have arisen in the past years as good
AI applications in agriculture are increasing. at modeling varying length sequential data,
Deep learning constitutes the state-of-the-art achieving state-of-the-art results for many prob-
method for image and language processing with lems in natural language processing (NLP), such
promising results for addressing farming prob- as machine translation, information extraction,
lems, such as weed detection, plant disease diag- and text classification. Similar to computer
nosis, crop type classification, and pesticide vision, the use of transfer learning by word
recommendations. Techniques stemming from embedding techniques based on ELMO and
January/February 2020
27
Authorized licensed use limited to: Agricultural University of Athens. Downloaded on February 13,2020 at 08:46:07 UTC from IEEE Xplore. Restrictions apply.
Envisioning Our Future Digital World
BERT will help to avoid the expensive creation of 4. S. Sabour, N. Frosst, and G. E. Hinton, “Dynamic
the necessary language model for more routing between capsules,” in Proc. 31st Int. Conf.
advanced NLP tasks. Such systems have been Neural Inf. Process., 2017, pp. 3859–3869.
used to extract expert knowledge for regulations re et al., “Imagination-augmented agents
5. S. Racanie
on pesticide use or agricultural knowledge from for deep reinforcement learning,” Adv. Neural Inf.
existing research databases. Process. Syst., 2017. arXiv:1707.06203.
Finally, another technique will be relevant in
the near future: generative adversarial networks
Spyros Fountas is an associate professor with
(GANs). This type of network could address the
the Agricultural University of Athens, Greece. He
problem of data scarcity within agricultural com-
received the M.Sc. degree in information technology
puter vision when transfer learning and tradi-
from Cranfield University, Cranfield, U.K. and the
tional data augmentation are not enough. The Ph.D. degree in precision agriculture from Copenha-
main advantage of GANs is that they can create gen University, Copenhagen, Denmark. He is the
pictures with synthetic “real” crops instead of Editor-in-Chief for Computers and Electronics in
just rotating or adding noise to the existing Agriculture (Elsevier), since 2014. He is the
ones. This synthetic data could be used to corresponding author of this article. Contact him at
improve the generalization ability of a CNN that sfountas@aua.gr.
obtains poor results due to the constraints in
Borja Espejo-Garcia is a postdoc researcher with
the size of the original dataset and the limita-
the Agricultural University of Athens, Greece. From
tions of the traditional techniques.
2015 to 2019, he worked toward the Ph.D. degree in
artificial intelligence with the University of Zaragoza,
CONCLUSION Spain, and the M.Sc. degree in computer science
Digital agriculture is developing rapidly, with a postgraduate course in systems engineering.
driven by many technological advances in the Contact him at borjaeg@aua.gr.
area of remote sensing, artificial intelligence, and
robotic systems. These systems enable farmers Aikaterini Kasimati is an agricultural engineer with
experience in precision agriculture, bioenvironmental
to produce comprehensive, accurate, and trans-
engineering, and energy production. She received the
parent crop and livestock products, both at the
M.Sc. degree in agricultural and biological engineer-
national and regional levels and to get increased
ing from the University of Illinois, Champaign, IL, USA.
yield and quality, minimizing the environmental She is currently working toward the Ph.D. degree with
impact. However, several challenges and limita- the Agricultural University of Athens, Greece, in the
tions, such as accuracy, interoperability, data field of precision agriculture and big data analytics.
storage, computation power, and farmers reluc- Contact her at akasimati@aua.gr.
tance to adoption, need to be addressed for effec-
tive use of these technologies and widespread Nikos Mylonas is a mechanical engineer and
digital transformation of agriculture. received the M.Sc. degree in biosystems engineering,
Wageningen University and Research, Wageningen,
the Netherlands, focusing on computer vision in agri-
& REFERENCES culture. His main research interests include the field of
AI and robotic applications in horticulture. Contact
1. J. Miranda, P. Ponce, A. Molina, and P. Wright,
him at nmylonas@aua.gr
“Sensing, smart and sustainable technologies for Agri-
Food 4.0,” Comput. Industry, vol. 108, pp. 21–36, 2019. Nicoleta Darra is an agricultural biotechnologist
2. A. Khanna and S. Kaur, “Evolution of Internet of Things with the Agricultural University of Athens, Greece,
(IoT) and its significant impact in the field of Precision from where she received the M.Sc. degree in bioin-
Agriculture,” Comput. Electron. Agriculture, vol. 157, formatics. For three years, she has been involved in
pp. 218–231, 2019. research on precision agriculture with emphasis
3. S. R. Oza, S. Panigrahy, and J. S. Parihar, “Concurrent on remote sensing applications. Contact her at
use of active and passive microwave remotesensing nicoletadarra@aua.gr.
data for monitoring of rice crop,” Int. J. Appl. Earth
Obst. Geoinform., vol. 10, pp. 296–304, 2008.
28 IT Professional
Authorized licensed
View publication stats use limited to: Agricultural University of Athens. Downloaded on February 13,2020 at 08:46:07 UTC from IEEE Xplore. Restrictions apply.