Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

International Journal of Computer Science Trends and Technology (IJCST) – Volume 10 Issue 4, Jul-Aug 2022

A comprehensive study of AI/ML in intelligent agriculture


Saumya Awasthi [1], Ajeet Singh [2], Nikita gupta [3]
[1], [2],[3]
computer science & Engineering department, Kanpur institute of technology - Kanpur

ABSTRACT
This study explores the equipment and technologies utilized in intelligent agriculture. Artificial Intelligence and system gaining
knowledge of strategies, consisting of basic block models which are used to do clever agriculture. How are we able to use fuzzy
logic and artificial Neural network, is likewise protected on this paper. We have explored a number of the IOT primarily based
irrigation systems including crop prediction structures. The vital hardware, software program and sensors that can be used to
make precision agriculture also are covered. The principle motto of this paper is to get an in depth literature assessment that is
required for clever agriculture.
Keywords: Artificial Intelligence, Machine Learning, fuzzy logic, Artificial Neural Network, IOT, Precision Farming

I. INTRODUCTION II. LITERATURE REVIEW


Techniques which might be used for agricultural paintings Plenty of things have been evolved within the field of AI
traditionally are no longer a good deal powerful and during the last many years due to its soundness in the software
accessible. It calls for masses of workforces and everyday and on account that it's miles ubiquitous in almost every
tracking. So, to reduce the exertions and increase subject. An instance of it's far agriculture which we've got
productiveness, we want to change TO smart agriculture over explored in this paper. Agriculture is the spine of India and it
traditional farming techniques .A farmer in conventional additionally faces a variety of demanding situations every day.
techniques of agriculture wants to recognition at the Elements that save you the crop from developing are-
parameters which includes soil moisture, temperature tracking, • negative water availability
humidity, and so forth. So glaringly hard paintings fee is extra • hot temperature
and non-stop monitoring isn't feasible all of the time .We can • opposition among flora for sunlight, nutrients, water
use the internet of factors to restriction the cost. IOT is the or area
mixing of diverse such things as sensors, community devices,
• out of control use of chemical substances
and devices that we use to automate statistics recording.
• Fungal, bacterial or viral contamination
Through the use of IOT gear we're able to advocate
movements required .Artificial Intelligence has also been • attack from insects or other organisms above or
extensively utilized in masses of fields, be it scientific below floor
technological expertise, schooling, finance ,agriculture, • So our objective is to-
industry, protection, and so forth. AI trains machines a terrific • Automate tracking and analysis assignment off
manner to take choices inside the same way people armers
take .Device getting to know is the sub area of AI. The • identify plant disorder
primary venture involved in the approach of system gaining • Forecasting plants
knowledge of is that we positioned the information in the tool • Detection of larva activities
that has come from beyond outcomes and this facts, is used by • Maximizing yield using minimum resources
the device to make similarly picks. In fields like voice
reputation, photograph identity, weather forecasting ,and 2.1 Simple block models for smart agriculture
many others, professional information is analyzed ML is a
mathematical approach to construct intelligent machines. With 2.1.1
the improvements in AI, numerous extraordinary ideas and An event driven WSN for monitoring plant health and Larva
strategies have been invented and discovered which simplified actives [1]
the trouble-solving assignment .Few methods are: ARM Cortex A4 processor-based WSN for facts collection
1. Fuzzy logic (inclusive of Larva sports) from sensors has been used. This
2. Artificial neural networks (ANN) has been awoke by the bottom station. Sensing nodes transmit
3. Neuro-fuzzy logic4. Expert systems statistics only in case when there's a whole lot difference in
values that is going to automatically lessen the power
consumption. Extracted statistics from a set of sensing nodes

ISSN: 2347-8578 www.ijcstjournal.org Page 28


International Journal of Computer Science Trends and Technology (IJCST) – Volume 10 Issue 4, Jul-Aug 2022

is transmitted to the base station with the aid of the usage of e) Integration level- this degree is in charge to be a part
wireless Zig bee protocol.The bottom station is attached to the of all the elements to shape the required architecture.
CPU through USB Liaison between CPU and Base station, f) Application-level- consequences will be displayed on
and right here we can see all of the gathered statistics which the application.
we obtained from the sensors in the form of an ACI evaluation
and this evaluation is carried out the use of Matlab. 2.2 Irrigation System using Internet of Things:
In 2014, GPRS characteristic of a phone cellular
2.1.2 telephone used to be used to raise an irrigation machine
UAV Routing protocol for crop health management [18] via Ali, T. A. A. et al. In this microcontroller-based rain
gun irrigation system, water furnish will be there
completely when there is an acute water shortage. The
sole intent is to hold water, alternatively, this laptop is
no longer financially affordable.
Internet of Things In 2017, Arif Gory et al. [3] proposed
a mesh-based approach for putting sensor nodes in
agricultural areas. The statistics from these sensors are
accumulated with the aid of the base station (which is
stationary) (they are mobile). Following the gathering of
all facts base stations, the information is processed. The
pH sensor used to be used to decide whether or no
longer the dirt used to be acidic or antacid in nature. The
Fig 1. UAV routing protocol discovered attributes from the pH sensor had been
relayed to the Android application, and as appropriate,
picks had been made.
• All sensors hubs are heterogeneous and might be
exceptional long. Among those hubs, one hub is chosen
as a group head (CH). Bea, J. G., & Palaoag et al. [4] used formal good
• To choose CH, an arrangement of rules is characterized judgment in 2017 to classify rice and maize soil wetness
and the chance of each and every hub teaming up in CH system readings inside the lessons as "too wet," "wet,"
is determined through the Bayesian classifier set of "normal," "warm and "to heat." Such readings are
rules. frequently accustomed to evaluate the facet charge for
these two plants which would possibly be accustomed
• all of the part hubs convey realities to CH the utilization
flip the motor pump on / off.
of TDMA (Time division more than one access)
In 2018, Chaudhry, S., & Garg, S. et al. [5]
conspire.
deliberate a device that used to be divided into three
• CH will supply data to UAV (automated ethereal
sections-sensing a section of humidness, area of
vehicle) (drone).
management and part of the production. Soil humidness
• Reenactment is completed in OMNET ++ and the way is evaluated by means of using a YL-69 sensing element,
of a robot is given ahead of time. and also the management unit is that the Arduino
2.1.3 platform is primarily based on the whole ATMega328
Accuracy agribusiness checking framework utilizing microcontroller. The sensing factor sends the perceived
Green IOT [2] expertise to the small controller, which then works with
the aid of flipping the motor pump on and off
a) Requirement comparison - all the purposeful and non- consequently.
functional necessities are gathered.
b) Specification stage specifies the required gadgets in 2.3 Fuzzy logic in smart agriculture
particular primarily based on necessities gathered and
solely these sensors will get activated, this will saves Fuzzy operators carry out like a human driver and
value automatically. therefore regulate the interaction among entering and
c) Architectural degree become aware of slave node and output similar to the formerly described collection of
hold close node. All the slave nodes are accountable to club functions and guidelines. A fuzzy skeleton is a
furnish the statistics to the corresponding grasp node. linguistic-based control tool, which tries to imitate a
d) Component level- its duty is to operate all the human professional’s moves. Six inputs And outputs
elements in the architectural stage. Hardware seem in AGR Says [6]. They have got used the multiple
components- gadgets and sensors Software components- input to more than one output ( MIMO)type the use of
BBB Debian, cloud9 IDE for programming editing, the fuzzy good judgment library beneath Lab VIEW.
debugging python code

ISSN: 2347-8578 www.ijcstjournal.org Page 29


International Journal of Computer Science Trends and Technology (IJCST) – Volume 10 Issue 4, Jul-Aug 2022

This device is deployed on a web server to permit emphasis on applications distributed over the most recent
faraway get admission to and managed or discovered years, in order to provide a comprehensive image of DL to
thru any smart tool. scientists in agribusiness fields and upgrade current day
shrewd farming turn of events. Song (2005) [8] used a
• This system uses five inputs, each with its own combined expert system and an artificial neural network to
minimum and maximum values: humidity, temperature, forecast the level of nutrition in the crop. The entire system is
PH, soil moisture, and thermocouple. based on a single chip computer.
Second, Stone and Toman (1989) [9] developed the
• The following are the specifications for its function: COTFLEX framework for cotton. The framework was created
• If the sensor value of a Rectangular Humidity as a stab at the Pyramid 90 PC using UNIX as the operating
Membership Function is less than 55, it is considered system. The architecture combined the field and homestead
dry; otherwise, it is considered wet. databases to provide the rancher with important data on the
• The pH membership function of a trapezoidal shape is cotton yield, making it easier for the cultivator to make
classified as acidic, normal, or alkaline. fundamental and strategic decisions. The framework was
• Triangular Situation Cold, Warm, and Hot are the three developed in Texas, and it included recreation models and
temperature membership functions. databases in a standard-based master framework to help Texan
• The membership function for triangular soil ranchers make financially sound decisions. COTFLEX was
temperature is Cold, Warm, and Hot. then loaded onto an IBM microcomputer and made available
• The Unsaturated and Saturated membership functions for usage.
for Trapezoidal Soil Moisture (Water Level) are In Rajasthan, India, a specialised framework PRITHVI by
studied. Batchelor, W. D. et al. [7](2013) based on fuzzy and
implemented on Matlab was developed for Soybeanscrop. It
2.4 Artificial neural networks in agriculture gathers information from rural officers, disseminated writing,
and soybean crop professionals. When considering the
The ANN method was created by recalling a comparable complete framework and encouraging the rancher to be an
concept of how the human cerebrum works. ANN is a project- expert, fuzzy rules were explored. The system was divided
based method that instructs the framework to work based on into five modules. The major goal of constructing this master
some inbuilterr rather than a standard computational modified structure was to assist ranchers in increasing production.
assignment. Three layers make up the ANN engineering: A forecasting technique that advocates using ANN
1 Input layer computations for crop forecasts in smart phones was
2 Hidden (middle) layer successfully tested in 2016 [10]. There were three layers to
3 Output layer this architecture. The model's efficiency was determined by
The ANN model is made up of several neurons, each of which the number of hidden layers. To determine the number of
delivers a grouping of true valued activations. Initial neurons shrouded layers, an experimenting technique was used.
will be actuated when sensors detect changes in condition, and Because the precision of the expectation framework is
different neurons will be activated through weights linkages dependent on the number of hidden layers, this can be a
from already active neurons. Depending on the specific precise technique to investigating the choice of shrouded
challenges and topology used, these procedures may layers. It was also discovered that increasing the number of
necessitate some computational work, as each stage affects the shrouded layers will result in a more precise forecast.
entire actuation of the system. Deep learning allows
computational models made up of numerous handling layers Furthermore, using image analysis and neural systems, a
to communicate with data with varying degrees of reflection. technique was developed to separate weeds from crops with a
Many areas of analysis have seen significant improvements in precision of over 75% with no prior plant data taken into
the strategy. Some DL calculations are driven by the concept account in the framework. Aitkenhead, M. J. et al. (2016) [11]
of a Back Propagation Neural Network. Extremely high up developed masterframeworks based on a strategic horticulture
grades have recently been achieved in the DL profession, framework. The purpose of IoT in this framework was to
thanks to a lot of enthusiasm. Horticulture has gotten a lot of transmit information to the server so that field actuators may
attention because of DL. Picture recognition is one of its make appropriate decisions. For this, the server must be
agribusiness applications, and it has overcome a slew of intelligent enough to make independent decisions.
roadblocks that have stymied rapid advancement in
mechanical and automated agriculture. Plant illness detection, 2.4.1 Neural Network in smart agriculture
weed control, and plant inspection are just a few of the areas
where these advancements have been noticed. When it comes to plant ailment detection, it's time-consuming.
This work summarises BP and normal DL calculations Plant disease recognition can now be cultivated through IP
(Convolutional Neural Networks (CNN), Recurrent Neural thanks to the growth of man-made consciousness. The most
Networks (RNN), and Generative Adversarial Networks reliant on leaf image categorization and pattern recognition
(GAN))and their applications in horticulture, with an are plant illness recognition systems [13]. A plant disease

ISSN: 2347-8578 www.ijcstjournal.org Page 30


International Journal of Computer Science Trends and Technology (IJCST) – Volume 10 Issue 4, Jul-Aug 2022

identification structure was constructed using an epic DL 99.8% accuracy in crops and 90.7 percent accuracy in grass is
structure. This structure has the ability to recognise plant available. It has also enhanced performance and is
leaves based on environmental parameters and can detect 13 significantly superior to existing methods[17].
different types of plant ailments.
In a study of using Deep Learning to detect plant diseases, the 2.5 Prediction System
overall precision was 95.7 percent after 100 minutes of
preparation and 96.4 percent after more tuning. In actuality, Attempted to describe soil based on its nutrient content using
the results are better than manual location. All of this showed a decision tree algorithm . Crops appropriate for various soil
that DL is capable of identifying plant diseases. Yalcin, H [14] types were also predicted. Classification and Regression Trees
is a pretrained convolutional neural network that is used for (CART) and C4.5 are two decision tree techniques that are
plant classification. According to the results of this model in commonly utilised. The CART's main disadvantage is that it
2017, CNNengineering outperforms AI algorithms that rely on makes mistakes when the selected parameter region is larger,
available produced highlights for phenological stage whereas the C4.5 method favours higher value characteristics.
separation. They offered a modified method to the decision tree in order
Self-arranging Kohonen maps (SOMs) were used for optical to solve these drawbacks. Instead of using the CART
image segmentation and the reconstruction of missing algorithm to calculate for each pair, they calculated the Gini
information in a periodarrangement of satellite symbols in index for several sets of parameter values and then used ratios
another study. This technique comprised a post-preparation of these measured Gini indices to reduce the bias introduced
step that included a few filtering computations based on the by knowledge gain in the C4.5 approach. Employed decision
available data as well as geographical analysis. The tree induction to investigate the impact of climate variables on
classification of major harvests was done with an accuracy of soybean productivity in order to help researchers and farmers
84.9 percent [12]. anticipate market dynamics. Agro-climatic data was collected
In any event, the use of CNN for plant grouping has been in the Bhopal district from 1984 to 2003
hampered by issues. For yield forecasting, organic product
inspection is critical. Manual tallying or checking with a 2.6. Uesing of Hardwares
handheld camera can't produce satisfactory results and is time
consuming. Blob detection has been shown to be useful [10], 2.6.1 ESP Units
leading to the recommendation of using a full CNN. The
structure's main goal is to build man-made names out of a The ESP Internet of Things network is a low-cost, high-
large number of organic product images. The fully performance network. It's used to send real-time data from the
convolution network (FCN) approach was ready to do picture field to the internet of things. It has two functions, and it may
division at that time. The check convolution organizer was be utilised with a self-contained Wi-Fi networking solution to
then ready to accept the segmented image and calculate the transport and drive the entire programme. It can also be used
number of grapes. as a Wi-Fi adaptor to power other microcontrollers that can
Our final goal is to use a linear regression equation to convert run more complex applications than ESP. In this project, the
the number of fruits counted initially to the final number of Arduino UNO will be the primary microcontroller, with the
fruits counted manually. The accuracy and productivity of ESP 8266 acting as a Wi-Fi adaptor. Multiple general-purpose
checking were both increased using this Deep Learning with I/O pins of the ESP are utilised to connect sensors to the
blobdetection approach. Satellites and unmanned aerial microcontroller directly. It has a 32-bit microprocessor on
vehicles (UAVs) are rarely used to investigate various assets board that controls communication, power modes, and GPIO.
[15], which rely on deep convolutional neural networks Within 3.0 to 3.6 volt power range.
(DCNN) and transfer learning (DTCLE). To discover
cultivated land data, a characteristic identification technique 2.6.2 Motor Operator
based on DCNN was used. Finally, for cultivated land
information extraction, the DTCLE and e-Cognition were Although the Arduino UNO runs on 5 volts, the engine runs
used (ECLE). Although both are approximately 90% accurate, on 9 volts. In addition, the engine requires more current than
DTCLE outperforms ECLE in terms of integrity and the Arduino provides. For this, a motor driver controller
consistency. The usage of UAVs provides high-resolution circuit can be used. The motor driver circuit connects the
images. Arduino UNO to the motor. It supplies the necessary power to
The detection of any item, or finding any thing, is a rapidly run the motor.
growing domain in Deep Learning. Detection is also
important for ranchers in the horticultural industry, especially 2.6.3 Odroid XU4
since that extremely self-ruling equipment are being
increasingly used. These computers should be highly reliable The Odroid XU4 Model B improves on the features of its
with automatic real-time risk identification in order to work predecessors by including a second, faster processor. It also
these tools securely but without oversight [16]. When picture combines WiFi and low-vitality Bluetooth capabilities to
classification is done with AlexNet and DCNN, a system with boost the network and USB port capacity, allowing it to

ISSN: 2347-8578 www.ijcstjournal.org Page 31


International Journal of Computer Science Trends and Technology (IJCST) – Volume 10 Issue 4, Jul-Aug 2022

control more cell phones. Upgraded micro USB power source The water level sensor's specifications are as follows: outputs:
to 2.5A. Odroid XU4 is a single-board computer designed to 4.0-20.0 mA or 0.50 to 2.50 VDC, supply voltage: 3.30 to 5.0
support and encourage programming and computing education. VDC, dimensions: 60.0 x 20.0 mm, contacts 45.0mm, when
It's also an excellent starting place for ongoing Internet of the water level is low, the digital output is high, and the
Things (IoT) projects. Because of its inexpensive cost and sensitivity is adjustable.
'plug-and-play' nature, the Odroid XU4 may be used by
anyone and provides a wide range of networking choices. 2.7.4 Data collection sensor
Whether you want to use it as a workstation, media centre,
server, or monitoring/security device in your house, Pi is the Routing of static sinks Data acquisition from mobile sinks by
ideal experimental instrument. There are none. Linux-based direct interaction Data collection based on rendezvous
operating systems operate on the XU4, and there is plenty of Routing using hash tables.
free software available.
2.8 Decision Trees
2.6.4 Relay
In data mining, decision trees are commonly used for
A relay is a device that uses electricity to work. To accurately classification. A decision tree is a flowchart-like tree structure
operate a switch, many relays use an electromagnet, however in which an interior hub represents a trait test, each branch
alternative working standards, such as strong state relays, are represents the test's result, and the leaf hub represents the
also used. When a circuit needs to be confined by a different classes. The root hub is the first node in the network. A
low-power signal, or if several circuits need to be constrained strange example is set up to test its attribute values along the
by the same signal, relays are used. It's referred to as a decision tree, following a path from the root node to a leaf
contractor type of transfer since it can handle the tremendous node that contains the forecasting class label.
force needed to control an electric engine with or without
loads. Decision trees are of 2 types:
Relays are commonly used in safety-critical logic because
they are substantially more resistant to radioactive radiation When the expected output is a class name, the Decision tree is
than transistors. To distinguish over-burden and other defects referred to as a characterisation tree.
in onelectrical lines, electromechanical protection relays are The decision tree is called a regression tree when the projected
used to open and shut off circuit breakers. result is a real number.
2.7 Sensors
2.7.1 Sensor for Temperature And Humidity
III. CONCLUSION
The DHT-11 temperature and humidity sensor is used. A
thermistor is used to detect the temperature of the surrounding An artificial intelligence has been developed to help ranchers
air. This technology is energy efficient and cost-conscious. A in choosing the best composts for their crops. Field
temperature-proportional output is provided by the DHT 22 IC monitoring is obviously a required lessening of people
temperature sensor. The DHT22 sensor is more precise than intercessions. The machine talks among itself to pick which
the thermostat in measuring temperature. The temperature yield is sensible forgathering and moreover the composts
range for operation is -40 to 120 degrees Celsius. Each 0C which advance the greatest turn of events. DL has a wide
rise/fall in ambient temperature causes a 10mV variation in range of applications, and its use in the company has
output voltage. advanced significantly. It is an even better situation over AI
2.7.2 Sensor for Humidity with the use of DL, and it gives depth to AI. This, in turn, aids
the overall progress of the country, as food is the most basic
Sensors and probes YL-69 The moisture content of the soil is demand of each individual. IOT intended for its magnificence
measured with the YL-38. Two copper strips, coated with to aid in constant data perception. The Internet of Things is
acrylic paint to increase conductivity, make up the YL-69 mostly used in a smart watering system. Because practical use
probes. One of the sensor samples is made of aluminium, of open freshwater is significant, and with advancements in
while the other is coated in polyaniline nanoparticles to invention and application of innovation, the water emergency
improve conductivity. can be addressed. Farmers that use conventional methods
The humidity sensor consists of a tiny condenser made of a confront problems such as water scarcity and flooding. We
dielectric substance that is positioned between two electrodes. have provided a complete literature evaluation of factors
0 RH at 100% Every two seconds is the average sensing related to smart agriculture in this research.
interval. 1.3–2.1 mA current 25.1 X 15.1 X 7.7 mm in size.

2.7.3 Sensor of Water Level

ISSN: 2347-8578 www.ijcstjournal.org Page 32

You might also like