Available via license: CC BY 3.0
Content may be subject to copyright.
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd
CISAI 2020
Journal of Physics: Conference Series 1693 (2020) 012058
IOP Publishing
doi:10.1088/1742-6596/1693/1/012058
1
Artificial Intelligence in Agriculture
Jiali Zha1
1Moses Brown School, Providence, 02906, United States
Abstract. The application of AI in agriculture has been widely considered as one of the most
viable solutions to address food inadequacy and to adapt to the need of a growing population.
This review provides an overview of AI’s application in agronomic areas and progress in
research labs. The review first presents two fields that AI can potentially play an important role
in, which are soil management and weed management, and then Internet of Things (IoT) a
technology that shows great potential in future usage is mentioned. Three challenges that need
to be addressed in order for AI-based technology to be popularized in markets are uneven
distribution of mechanization, the ability of algorithms to process large sets of data accurately
and quickly, and the security and privacy of data, as well as the devices. Agricultural robots
targeted at diverse aspects in agricultural industry have been developed and improved greatly in
the past years, and although pointing out the hardship of applying machines and algorithms tested
in experimental environment to real environments, the review highlights an already prosperous
development and a promising prospect of application.
1. Introduction
The term “Artificial Intelligence” was first introduced in the 1955 Dartmouth Conference, in which John
McCarthy proposed a study to be carried out grounded on the hypothesis that “every aspect of learning
or any other feature of intelligence can in principle be so precisely described that a machine can be made
to simulate it” [1]. Nowadays, AI, one of the essential areas in computer science, has penetrated a variety
of domains, such as education, healthcare, finance and manufacturing, because of its nature to tackle
problems that cannot be solved well by humans [2]. Humans continue to be shocked by AI’s capacities.
One example is IBM’s Deep Blue’s historical victory over world chess champion Garry Kasparov in
1997 and the triumph of AlphaGo over the world Go champion Lee Sedol in 2016, which proves that
deep learning, the principle that AlphaGo is based on, enables AI to surpass the most human brainpower
[3].
Agriculture, an essential consideration of any country, is still one of the major challenges currently.
It is approximated that over 820 million people are in hunger today [4]. Furthermore, with the global
expected to reach 9.1 billion in 2050, 70 percent more food needs to be produced. In addition to the
projected investments in agriculture, further investment will be needed, otherwise about 370 million
people would be in hunger in 2050 [5]. In addition, an expanding gap between a growing water demand
and the available water supply is anticipated, and it is likely that over three billion people would
experience water stress by 2025 [6].
Except for traditional measures, scientists and the government recognize the important role played
by AI, despite its relatively short history of development. The application of AI in agriculture was first
attempted by McKinion and Lemmon in 1985 to create GOSSYM, a cotton crop simulation model using
Expert System to optimize cotton production under the influence of irrigation, fertilization, weed
control-cultivation, climate and other factors [7]-[8].
CISAI 2020
Journal of Physics: Conference Series 1693 (2020) 012058
IOP Publishing
doi:10.1088/1742-6596/1693/1/012058
2
This review aims to present the current situation of artificial intelligence in agriculture by
highlighting three important considerations and achievements-soil management, weed management and
the use of Internet of Things. It also evaluates the pressing challenges that are confronted in this field,
like the predictable uneven distribution of mechanization in different areas, security and privacy issues,
and the flexibility of algorithms in practical applications, when plants are physically heterogeneous and
large data sets and additional factors need to be processed. Finally, this review emphasizes on the
development of agricultural robots, providing the background of this specific field, giving particular
examples and then pointing out major challenges. identifies the future prospects of application and also
takes into considerations diverse circumstances in different countries.
2. Status of AI applications in agriculture
2.1 The definition of Artificial Intelligence (AI)
The definition of Artificial Intelligence changed over time because of its rapid development, and a
unified definition does not exist even in current days. However, the definitions around can be generally
classified into four categories: AI is a system that thinks like a human, acts like a human, thinks
rationally or acts rationally [9]. Alan Turing wrote a paper in the 1950s, in which he proposed a game
to answer the question “Can a machine think?” and the game is known as the Turing Test [10]. To pass
the Turing test, a computer must possess four skills - natural language processing, knowledge
representation, automated reasoning, and machine learning [9]. In this case, Turing gave the most widely
spread definition of AI, but it had the problem of not distinguishing between the knowledge from the
intellect, just like separating software from hardware when defining a computer [11]. AI was also
defined as “such a program which in an arbitrary world will cope not worse than a human,” which means
that AI is a set of programs, has inputs and outputs and also exists in an environment [11]. Some
applications of AI include intelligent retrieval from databases, expert consulting systems, theorem
proving, robotics, automatic programming and scheduling problems, perception problems, etc. [12].
2.2 Current status of AI application in Agriculture
2.2.1 Soil Management
Soil is one of the most important factors of successful agriculture, and as the original source of nutrition,
soil stores water, nitrogen, phosphorus, potassium, and proteins that are crucial for proper crop growth
and development [13]. Soil condition can be enhanced with compost and manure, which improves soil
porosity and aggregation, and with an alternative tillage system to inhibit soil physical degradation. With
soil management, for example, negative factors, such as soil-borne pathogens and pollutants, could be
minimized [13]. Another example is that AI can be used to make soil maps, which helps to show soil-
landscape relationships and various layers and proportions of soil underground [14].
2.2.2 Weed Management
Weed is one of the aspects that reduces a farmer’s expected profit most: for example, if weed invasion
is not under control, a 50% loss in yield can occur for dried beans and corn crops, and weed competition
can cause a 48% reduction in wheat yield. Weeds compete with crops for resources, like water, nutrients
and sunlight, regardless of some being poisonous and even threatening public health [13]. While spray
is often used to inhibit weeds, it has a potentially negative impact on public health and the excess use
can pollute the environment. Therefore, artificial intelligence weed detection systems have been tested
in laboratories to calculate the precise amount of spray to be used and to spray on the target location
accurately, which also lower costs and the risk of damaging crops [15].
2.2.3 The Use of Internet of Things Technology
The Internet of Things (IoT) is a system consisted of computing devices, mechanical machines and
various objects that are interrelated, and each is provided with a unique identifier and possesses the
CISAI 2020
Journal of Physics: Conference Series 1693 (2020) 012058
IOP Publishing
doi:10.1088/1742-6596/1693/1/012058
3
capability of data transfer. Therefore, human-to-human or human-to-computer interactions can be
avoided. IoT is an advancement built on several existing technology, such as wireless sensor networks
(WSNs), cloud computing and RF identification. IoT can be applied in manifold fields, such as
monitoring, precision agriculture, tracking and tracing, greenhouse production and agricultural
machinery. For example, the tracking and tracing of agricultural product chain include information input
(the complete life cycle of the product, the transportation process, etc.), the ability to store the
information for a period of time, and to transfer, process and output the data. The tracking and tracing
of the product chain can be used for commercial reasons, especially forming trust between the seller and
buyer – by seeing the entire history of the product, the agricultural companies can make better decisions,
find business partners wisely, and save time and money. The IoT applies data analysis in a variety of
ways, and the data are in various forms, such as sensor data, audio, image and video. Areas that data
analysis is vital to includes prediction, storage management, decision, farm management, precise
application, insurance, etc. [14].
3. Challenges of practical application of AI-based techniques in agriculture
3.1 Possible uneven future distribution of mechanization
From the projection of robot shipments during the period 2011-2013, an average 9% increase each year
in the U.S.s, a 12% increase in Asia-Australia countries and an 8% increase in Europe are anticipated.
According to this trend, it is estimated that the penetration rate of robots by 2030 will be 15% and will
be 75% by 2045. However, the distribution of mechanization can possibly be unevenly distributed with
some areas lacking access to resources and having situations which can’t not be changed with science
discoveries and technological development [6]. For example, since most AI systems are based on the
Internet, their utilization may be restricted in remote or rural areas with the absence of a web service
and familiarity with handling AI operations [13]. Therefore, a slower and unequally distributed adoption
process of AI in agriculture should be expected, and meanwhile, whether the adoption would increase
food production beyond certain natural limits of land or not remains uncertain [6].
3.2 Discrepancies between control experiments and actual implementation
The fact that images taken when applied differ from the images used in control environments because
of factors such as lighting variability, the complexities in the background, the angle when capturing, etc.
In addition, grains cultivated in the field, even at the same location, are physically heterogeneous as a
result of the impact of other elements, like insects, soil and inert matter. In that case, the physiological
characteristics of individuals increases the complexity of variables to be considered when processing
images, and therefore, a larger and more diverse set of control data was required to improve the current
classification accuracy. Nevertheless, with the help of computer vision, algorithms like DBN (Deep
Belief Networks) and CNN (Convolution Neural Network), regardless of the small number of case
studies, indicate promising applications in the future for processing large sets of complicated data [16].
Moreover, in order to shorten the response time of a system, data processed should be the most relevant
ones. A system’s capability of executing tasks precisely in a short period of time is critical in deciding
its commercial value, affecting users’ selection greatly - what customers consider most is the minimized
effort required for them and the maximized accuracy [13].
3.3 Security and privacy
Many physical devices, such as the IoT, are first vulnerable to attacks on the hardware because the
device can be placed in an open space for long periods of time without supervision. Typical security
counter measurements are data encryption, tag frequency modification, tag destruction policy, use of
blocker tags, etc. Location-based services are also exposed to device capture attack, which means after
capturing the device, the attacker can extract cryptographic implementations and therefore have
unlimited access to data stored in the device. Data can also be attacked when transferring from the device
to the gateway, where the data is then uploaded to other infrastructures, like the cloud. The cloud servers
CISAI 2020
Journal of Physics: Conference Series 1693 (2020) 012058
IOP Publishing
doi:10.1088/1742-6596/1693/1/012058
4
are vulnerable to data tampering, which can unauthorizedly interfere the automated operations in the
farm. Means such as session hijacking, logon abuse and denial of service (DoS) can also interfere cloud
infrastructures. The corresponding security policies include cryptographic algorithms, data flow control
policies, identity authentication mechanisms etc. Therefore, security issues are causing serious problems
and should be addressed in different levels [14]+[17].
4. Development of Agricultural Robots
4.1 Background and Examples
One field of applications that AI plays an important role in is the robotics system, and to incorporate
robotics into agriculture and to improve the efficiency, reliability and precision have been attempted for
years, which would dramatically replace manual labor needed with automatic labor- intensive work.
Automation are keys to pressing social phenomena such as aging population and decreasing population,
but to be able to accomplish the accurate and complicated operations that were traditionally done by
farmers to maintain the good quality always remains as a great challenge.
The study of robots for agricultural purposes began as early as in the 1980s, and Japan first developed
a robot that can spray pesticide [18]. Acknowledging that to navigate in real agricultural environments
is a hardship, a research team in 1996 designed an autonomous mobile robot called AURORA that was
able to navigate autonomously or to be controlled remotely in greenhouses while performing specific
task that conventionally required considerable manual labor. In fact, the initial motivation for designing
robots specific to greenhouse environment was that human operators are vulnerable to pesticides,
fungicides and other chemical products especially in the warm and poor ventilation greenhouse
environment, which caused them skin diseases, chronic diseases and even mortality [19]. One early
example of agricultural robot the tractors obtain an input, or more specifically, a program indicating the
travelling path, from the global positioning system (GPS), and using machine vision, the device can
operate along with crop line [20]. In an experiment to estimate apple fruit location for manipulation in
2000, robots designed for picking apples used a Cartesian coordinate system to determine the position
of the apples. The non-linear least square method was used to store the distribution of the apples in the
horizontal and vertical directions, which can be applied for designing the manipulator of the apple
harvesting robot [21]. Regarding to the weed management problem discussed above, a 2003 design
aimed to test a robotic platform for mapping weed populations and focused on the mobility and user
friendliness of the four wheel system, of which its functionality is predominately carried out with
embedded controllers and standard communication protocols [22]. To expand on the idea of weed
management, another study conducted in 2003 emphasized on distinguishing between crops and weeds
to locate precise spots for herbicide. Image recognition of species focusing on plant morphology is one
of the most reliable methods: if characteristics such as leaf edge, border patterns and overall shape is
determined, then the plant type should be interpreted. However, because of the variability of
measurements, such as lighting conditions, distortion of the shape of the leaf and position, and the fact
that young plants vary significantly due to different burgeon dates, growth rate and variation in growing
environments, like temperature and moisture, to distinguish between weeds and crops remains
challenging. It also shows that the device needs to learn important features for itself based on neural
network (NN) approaches in order to attain desired functions. Furthermore, the selectivity of herbicides
being used in fields reduces the total quantity used and therefore can reduce herbicide pollution in water
[23].
An Autonomous Fruit Picking Machine (AFPM) for harvesting apples published in 2008 focused on
designing a flexible gripper, which ensured the accuracy that was crucial for picking apple by apple
instead of harvesting many in one go and therefore minimized economic loss due to damages of apples’
qualities [24]. A fruit picking robot published in 2013 has an automatic extraction method applied to
varying agricultural background for vision system, and the method is based on features in OHTA color
space and an improving Otsu threshold algorithm. The OHTA color space has color features that
transform color extraction in one-dimension rather than three-dimension. A new color feature in OHTH
CISAI 2020
Journal of Physics: Conference Series 1693 (2020) 012058
IOP Publishing
doi:10.1088/1742-6596/1693/1/012058
5
color space is first defined, and then an Otsu threshold algorithm extracts the fruit objects based on
properties in OHTA color space. The distinguish of colors serves the need to recognize ripe fruits, and
the extract rate is more than 95%, indicating its accuracy and effectiveness [18].
4.2 Challenge of agricultural robots
Although the study of agricultural robots has made tremendous progress, robots that are applicable to
work in complex agricultural environment are still not available in the market. The main reason was that
algorithms that can cope with the uncontrolled and unpredictable real agricultural environment have not
been developed yet, and other factors, such as the seasonality of agriculture, also marks the difference
between real environment and experimental environment in laboratories. The dynamic and rapid
changing in time and space of agricultural environment are almost unavoidable, no matter in
unstructured environments, such as military and space environments, or in environment where
atmospheric conditions have uncertainty inherently, like rugged terrain, visibility and illumination.
Nevertheless, partial autonomy will still benefit the production with technology. The Pareto principle
applies to many tasks, and it basically means that automation can be applied in 80% of a task, leaving
the remaining 20% very difficult. In another word, 80% of required manual work can be reduced with
automation. Furthermore, the 80% automation can serve as a transition from traditional farming systems
to completely autonomous farming systems and more experience will be learnt by experimenting with
software and hardware elements [20].
5.Conclusion
This review presents an overview of the application of AI technology in agriculture. Corresponding to
the current social situation of decreasing manual labor, limited usable agronomic land and a greater gap
between total food produced and the world population, AI has been regarded as one of the most feasible
solution to those problems and has been developed and improved for years by scientists worldwide. In
this review, the definitions of AI are first introduced, in which the highlight is the Turing Test. Then
two sub fields that AI has been playing an important role in are demonstrated, which are in soil
management, weed management, and Internet of Things (IoT), a useful data analysis and storing
technology that has wide application in agriculture, is introduced. This review also points out three
major practical challenges of AI in agriculture: first, due to certain geographical, social or political
reasons, the distribution of modern technology is uneven, which foreshadows that the application of AI
will have its limitation in certain areas; secondly, despite significant improvements made in the past
years, to transfer AI-based machines and algorithms from control experiments to real agricultural
environment requires much more studies and research, and to be able to handle large sets of data and to
interpret them accurately and quickly are two main challenges that need to be addressed in order to
enable the application; finally, the security of devices used in open spaces of agricultural environment
and the privacy of data collected are also problems to address. Then this review specifically introduces
the development of agricultural robots. First, a couple of examples of robots designed to tackle different
tasks in the agricultural industry are listed. There are autonomous mobile robots that can spray pesticides
in greenhouses, tractors that use GPS and machine vision and have a travelling path pre-programmed,
apple picking robots that use a Cartesian coordinate system to locate objects, two types of robots that
manage weed problems and innovate in several directions, such as physical mobility and the ability to
distinguish between crops and weeds, an apple harvesting machine that has an innovative flexible
gripper, etc. Then the review indicates challenges of applying agricultural robots, basically circulating
around the question of the unpredictability in real environments, but underscores the considerable
development and a promising prospect in this field.
References
[1] McCarthy J, Minsky ML, Rochester N, Shannon CE. A Proposal for the Dartmouth Summer
Research Project on Artificial Intelligence, 1955, 27(4):12-12.
CISAI 2020
Journal of Physics: Conference Series 1693 (2020) 012058
IOP Publishing
doi:10.1088/1742-6596/1693/1/012058
6
[2] G. Banerjee, U. Sarkar, S. Das, I. Ghosh. Artificial Intelligence in Agriculture: A Literature
Survey[J]. International Journal of Scientific Research in Computer Science Applications and
Management Studies, 2018, 7(3):1-6.
[3] Michael, C, Fuetc. Google DeepMind's AlphaGo[J]. Or/ms Today, 2016.
[4] “Hunger and Food Insecurity.” Food and Agriculture Organization of the United Nations, Food and
Agriculture Organization of the United Nations, 2020, www.fao.org/hunger/en/.
[5] “2050: A Third More Mouths to Feed.” Food and Agriculture Organization of the United Nations,
Food and Agriculture Organization of the United Nations, 2020,
www.fao.org/news/story/en/item/35571/icode/.
[6] Popa, Cosmin. Adoption of Artificial Intelligence in Agriculture[J]. Bulletin of the University of
Agricultural Sciences & Veterinary, 2011.
[7] Mckinion J M, Lemmon H. E. Expert systems for agriculture[J]. Computers & Electronics in
Agriculture, 1985, 1(1):31-40.
[8] Gertsis AC, Galanopoulou-Sendouca S, Papathanasiou G, Symeonakis A. USE OF GOSSYM-A
COTTON GROWTH SIMULATION MODEL-TO MANAGE A LOW INPUT COTTON
PRODUCTION SYSTEM IN GREECE. In First European Conference for Information
Technology in Agriculture, 1997.
[9] Kok J N, Boers E J W, Kosters W A, et al. ARTIFICIAL INTELLIGENCE: DEFINITION,
TRENDS, TECHNIQUES, AND CASES[C]. International Symposium on Southeast Asia
Water environment. 2002.
[10] Kevin Warwick. Artificial Intelligence: The Basics[M]. Routledge, 2011.
[11] Dobrev, D. (2012). A Definition of Artificial Intelligence. arXiv preprint arXiv:1210.1568, 2012.
[12] Nilsson, Nils J. Principles of artificial intelligence. Morgan Kaufmann, 2014.
[13] Eli-Chukwu N, Ogwugwam E C. Applications of Artificial Intelligence in Agriculture: A Review[J].
Engineering, Technology and Applied ence Research, 2019, 9(4):4377-4383.
[14] Elijah O, Member S, IEEE, et al. An Overview of Internet of Things (IoT) and Data Analytics in
Agriculture: Benefits and Challenges[J]. IEEE Internet of Things Journal, 2018, PP(99):1-1.
[15] Partel V, Charan Kakarla S , Ampatzidis Y . Development and evaluation of a low-cost and smart
technology for precision weed management utilizing artificial intelligence[J]. Computers and
Electronics in Agriculture, 2019, 157:339-350.
[16] Diego Inácio Patrícioa, Riederb R. Computer vision and artificial intelligence in precision
agriculture for grain crops: A systematic review[J]. Computers and Electronics in Agriculture,
2018, 153:69-81.
[17] Tzounis A, Katsoulas N, Bartzanas T, et al. Internet of Things in agriculture, recent advances and
future challenges[J]. Biosystems Engineering, 2017, 164:31-48.
[18] B X W A., C K J., D J L., et al. Automatic method of fruit object extraction under complex
agricultural background for vision system of fruit picking robot[J]. Optik, 2014, 125(19):5684-
5689.
[19] Mandow A , Gomez-De-Gabriel J M , Martinez J L , et al. The autonomous mobile robot AURORA
for greenhouse operation[J]. IEEE Robotics & Automation Magazine, 1996, 3(4): P.18-28.
[20] 28 Bechar A, Vigneault, Clément. Agricultural robots for field operations: Concepts and
components[J]. Biosystems Engineering, 2016, 149:94-111.
[21] Kataoka T, Murakami A, Bulanon D M, et al. Estimating Apple Fruit Locations for Manipulation
by Apple Harvesting Robot[J]. Ifac Proceedings Volumes, 2000, 33(29):67-72.
[22] Bak T, Jakobsen H. Agricultural Robotic Platform with Four Wheel Steering for Weed Detection[J].
biosystems engineering, 2004, 87(2):125-136.
[23] Aitkenhead M J, Dalgetty I A, Mullins C E, et al. Weed and crop discrimination using image
analysis and artificial intelligence methods[J]. Computers & Electronics in Agriculture, 2003,
39(3):157-171.
[24] Baeten J, Kevin Donné, Boedrij S, et al. Autonomous Fruit Picking Machine: A Robotic Apple
Harvester[J]. Springer Tracts in Advanced Robotics, 2007, 42.
