2016 Fifth ICT International Student Project Conference (ICT-ISPC)

59

A Control System in an Intelligent Farming by using

Arduino Technology
Narayut Putjaikal, Sasimanee Phusael, Anupong Chen-Iml,Dr.Phond Phunchongharnl.2,
and DrKhajonpong Akkarajitsakup,2

lApplied Computer Science, Department of Mathematics Faculty of Science

2 Theoretical and Computational Science Center

King Mongkut's University of Technology Thonburi (KMUTT)

Bangkok, Thailand, 10140
Email: [email protected]�@[email protected]@maiLkmutt.ac.th.
and [email protected]

Abstract-"Internet of Things" (loT) is a technology that allows Recently, Internet has involved in people's daily activities.
things to communicate and connect with each other. This will Internet has been widely used to connect people together,
change the patterns and processes in both industry and agriculture people with devices, or devices with devices. In an electronics
towards higher efficiency. Particularly, agriculture is an important
device, it is embedded by software and sensors for using to
foundation of Thai economy. Consequently, we propose an
commutate and to exchange data with other devices and people.
intelligent farming system (IF) to improve the production process
When millions of devices are connected together through the
in planting. IF composes of two main parts which are a sensor
Internet, this is called Internet of Things (loT). loT encompasses
system and a control system. In this paper, we focus on the control
many new intelligent concepts for using in the near future such
part which are watering and roofing systems of an outdoor farm as smart home, smart city, smart transportation, and smart
based on the statistical data sensed from the sensor systems
farming.
(including temperature, humidity, moisture and light intensity
Recently, there are few research works on smart farming
sensors) Since the sensed data would not be always accurate due to
[lJ[2][3].1n [1], a wireless sensor network is used in potato fields
noises, we apply Kalman filtering to smooth the data before using
as an input in our decision making process. For the decision in Egypt. The proposed system is used to monitor the potato
making process, we do not consider only the sensed data, but also fields such as looking for diseases and harmful fungi and record
the weather information. A decision tree model is generated to useful information for improving future planting and managing
predict the weather condition. Then, a set of decision rules based resources such as water and soil. In Thailand, agriculture is also
on both the sensed data and the predicted weather condition is massively important for Thailand'S economy. Particularly, the
developed to automatically make a decision on whether watering agriculture section has contributed 8.4 percent to Thailand'S
and roofing system should be on or off. Moreover, we also provide
GDP. Then, to increase the crop yield, the smart farming
functions for users to manually control the watering and roofing
systems via our mobile application.
technology would help.
In this paper, we have proposed an intelligent farming elF)
Keywords- Intelligent Farming; Smart Farming; Internet of system. IF is the technology that uses the concepts of loT and
Things; Wireless Sensor Network; smart farming to help farmers to monitor and sense useful
information from their farms in order to help in the quality
I. INTRODUCTION improvement and product quantity. Our Intelligent farming
Agriculture is one of the important businesses that mainly system consists of two main parts. The first one is a sensor
affects the mankind life. From the ancient to the agricultural system, including temperature sensor, humidity sensor,
revolution in Great Britain England, farming is the way that moisture sensor and light intensity sensor. The second part is a
human used to harvest plants and consumed them in their daily control system. In this paper, we focus on the control system.
life. Farming has been improved by many technologies Our control system has two main subsystems which are
supporting cropping system. In addition to the technologies in watering and roofing subsystems. The system uses the statistical
the agricultural revolution era, there have been many data collected from the sensor system and the weather
technologies that have impacts on agriculture such as harvest information to make a decision to control the farm environment.
machine, seed drill machine, reaper machine, and the others that Particularly, the statistical data is collected from sensors. We
can reduce manpower and wasting time. then apply Kalman filtering theory to make the data more

978-1-5090-1125-4/16/$31.00 ©2016 IEEE

accurate since there can be noise collected from the sensors remove the noises, we apply Kalman filtering to filter out the
(errors sensed by sensors). noises and smooth the sensed data as described in Section III-B.
To support outdoor farming, not only the sensed data but also
weather information from an external source is also considered.
STEP A: ACQUIRING DATA FROM
Weather information records in a planting area are retrieved SENSORS
from a weather forecast repository. Then we generate a decision -Input: Raw data from sensors
tree model to predict the weather condition (either "no rain", -Output: Votage values

"rain", or "storm") by using historical weather data <including

temperature, humidity, pressure at sea level and wind speed).
Finally, our proposed decision rules are set up based on the
smooth sensed data and predicted weather condition. Our
watering and roofing subsystems can be automatically turn on
or off based on the decision made by the rules. However, it is
possible to let users manually control the watering and roofmg
subsystems via a mobile application.

II. SYSTEM ARCHITECTURE AND FUNCTIONS

Fig. 1 shows our proposed IF. In IF, the sensor subsystem is
a set of tools and sensors that are connected to a microcontroller
board called Arduino board. Sensors are used to measure
essential values of the planting process including temperature,
humidity, moisture and light intensity. The sensed values are STEP E: MAKING A DECISION

then uploaded to the server by using a Wi-Fi module integrated -Input: Forecasted weather from STEP
C, Predicted weather from STEP D
on the same Arduino board. On another part, the control and the sensed data from STEP B
subsystem is a set of devices used to control the roofing -Output: An optimal deCISion
subsystem and watering subsystem. The roofing and watering Fig. 2 A process framework of the control system in IF
subsystems can be automatically open or close according to the
decision from our proposed decision tree model (described in B. Applying Kalman Filtering
Section III). Also, the roofing and watering subsystems can be In this stage, Kalman filtering [4] is used to estimate real
manually controlled by a user via our mobile application. values interfered by noise generated from the imperfect sensors
Moreover, there is a notification system used to provide the user orland inaccurate sensors measurement. From Fig. 3, the main
the current farming status and also ask for the response from the process of Kalman filtering is a recursive process which uses a
user to an important event such as when the plants need water previous state and current estimated state to get the more precise
and the temperature is tending to high. If the user does not have state of the dynamic system. The algorithm works in a two-step
any response within a limited time. The system will work process as follows:
automatically.

• �.
Solar Energy
Database
i\1t'asurt'ItIt'111 Ulxl:ltt' ("Cllrrect")
" ..__ff!i\\\ Control
� � AR�INO
00 Timl' Upd:ltl' ("Predict")
(I) CompulC the KOllman gain

(I) Project the stale ahead

Kk = P;'HT(HP;'HT+ R)-'
ik ::::: A.ik_1 + Bilk
••• ;::::::t

Android Sman Phone

�1I11 � �
'ROUINO --- \l(.)1:\
Server
(2) Project lhe error covariance ahead
(2) Update estimate with measurement ;'k

Xk = xk + Kk(:k - H.r;')
P;'= APk_IAT+Q (3) p
Update the error c variancc
Sensor
Pk = (/-KkH)P;'
Fig. I A system architecture of the system

Initial estimates for xk _ I

an d Pk _ I
III. A PROCESS FRAMEWORK FOR OUR CONTROL SYSTEM IN IF
Fig. 3 A complete picture of the operation of the Kalman filtering
In Fig. 2, there are five steps for our process framework of
our control system in IF. The details of each step are as follows: 1) Prediction step: In this step, first the Kalman filtering
addresses the general problem of estimating the state by the
A. Acquiring data from sensors linear stochastic difference and with covariance Q to the time k.
From our experiment, we have found that the proper value of
This stage is to sense raw data from sensors. The raw data
the process noise is 10-8 Then, we will estimate the current value
from the sensors ranges between 0 and 1023. Each sensor has its
and the error covariance matrix at time k which is given by:
own method to convert the raw data to be a voltage value
xl; = Axl;_l (1)
However, the main problem in this stage is that the sensor node
would not be always precise due to noises from the sensors. To
and
EXAMPLE DECISION TREE
PI; = APk_1AT + Q
where PI; represents the error covariance of prediction which is
(2) I humid >= 767 I
used to estimate how much we can trust the values of the current
estimated state. In this case, we assume that the first error '\:
•
covariance L is set to be one which is a large number for
representing the error covariance.
Correction step: When the sensor sends data to the system, the
measurement Zk will be used to compute the observed value of
the measurement which is given by:

Zk = HXk + Vk (3)
The noise of the measurement Vk has to be Gaussian
distributed. Our value of the measurement noise is derived from
using "standard deviation" of the observed value during the
calibration. The diagonal values of the matrix Rk (as the
covariance) are 0.3377159, 0.0050016, 0.0098983, and Fig. 4 Our decision tree for predicting the weather condition
0.0005383 which are the covariance of humidity, light intensity,
used to make a set of decision rules that the farmer should do as
temperature, and moisture, respectively. Then, the system will
shown in the TABLE I.
compute the difference between the measurement Zk and the
priori state xl; . This is also called the innovation Yk : TABLE L DECISION RULES FOR DECISION MAKING PROCESS
Yk = Zk - HXk (4) Result from
Smooth sensed data Decision to do
and the innovation covariance SK is shown below. Decision tree
SK = HPI;HT + R (5) "no rain" Moisture 2: 70% "watering"

Moisture :s 70%,
The next step is to calculate the Kalman gain KK. The Kalman "no rain" Light :s 2,000 lux II "closing roof'
gain is used to indicate how much we can trust the innovation Temp :s 35 Celsius
which can be computed by the following equation.
"rain "or "storm" Moisture 2: 70% "wait for raining"
KK = PkHTSl;l (6)
"rain "or "storm" Moisture 2: 70% "opening roof'
Now we have to update the posteriori estimate of the current
state:
(7)
The final output Xk is the estimated value of the sensor data Note that each condition in TABLE I derives from the
which is more precise and accurate. Then the value will be optimal condition of cabbage [7]. Different plants will use
transferred to the database as a smooth values of the sensed data. different decision criteria based on the optimal condition for
each kind of plants. For example, the moisture threshold of
C. Acquiring Weather Data
cabbage is about 70% so when the system detect that the
Our control system also acquires the weather data, including
moisture is below 70% and the predicted weather condition is no
historical, current, and forecasted weather, from
rain. The system will decide to watering the plants as the
Openweathermap.com to use as the conditions for making a
"Decision to do" in TABLE I.
decision. A user can see the weather information via the mobile
application where we inquiry the next 3-day weather forecast to
be shown.

D. Predicting Weather
The historical weather data is now used to model a decision
tree [5] for predicting the weather condition as "no rain", "fain or
storm". The decision tree model is generated by using a machine
learning library for nodej s [6]. Fig. 4 shows our decision tree to
predict the weather condition.

E Making a Decision
In this step, the predicted whether condition obtained from
STEP D and the smooth sensed data obtained from STEP B are
 IV. RESULTS TABLE II shows an example of the predicted weather
A plot prototype equipped with watering, roofing systems condition compared with the real condition. We can see that the
and sensors is shown in Fig. 5. The user can manually operate results from our decision tree is accurate.
the watering and roofmg systems via our developed mobile TABLE II EXAMPLE DECISION VS REALITY
application and monitor the current status of the plot.
Most Result
from
Date Reality Input Conditions
Decision
tree

Temp � 29.35, Humid � 72.5%,

2016-02-20 "no rain" "no rain" Moist � 50.21%, Light � 1072.79
lux, Forecast �� "Clear"
Temp � 27.08, Humid � 40.84%,
2016-04-14 "no rain" "no rain" Moist � 86.96%, Light � 973.81 lux,
Forecast � "Clear"
Temp � 32.69, Humid � 59.69%,
2016-04-18 "no rain" "no rain" Moist � 74.91%, Light � 60.43 lux,
Forecast �� "Clouds"

Fig. 5 A plant plot as a prototype installed with watering and roofing

svstem and with sensors V. CONCLUSION
We have proposed a control system for an intelligent
Particularly, the mobile application has three main parts <in
farming for an outdoor farming, called IF. To make a decision,
Fig. 6) as follows: i) a user can see historical sensed data from the
the model requires two important information pieces which are
plot, ii) a user can monitor plants status data and manually
the sensed data from the sensors in the plot and the weather
operate watering and roofing systems, and iii) a user can check condition. To smooth the sensed data, we have applied Kalman
the current and forecasted weather for the next three days. filter to remove noises. Also, we have generated a decision tree
model to predict the weather condition. Based on this
information, we have set up rules for making a decision in our
control system on whether watering and roofing system should
be on or off Moreover, we have also provided functions for
users to manually control the watering and roofing systems via
our mobile application.

ACKNOWLEDGEMENT

The work was supported by the Higher Education Research

Promotion and National Research University Project of
Thailand, Office of the Higher Education Commission (NRU59
Grant No.59000399) and National Electronics and Computer
Technology Center (NECTEC) under the National Science and
Fig 6 Mobile application for users to monitor and control the farm Technology Development Agency (NSTDA), Thailand.

Raw: 21.339LUX
Kalman: 27.99SLUX REFERENCE

[1 ] Sherine M. Abd EI-kader, Basma M. Mohammad EI-Basioni, Precision

fanning solution in Egypt using the wireless sensor network
technology, Egyptian Infonnatics Journal, 14, 3 , 221-233
[2 ] Ilan H.(2014), International Dairy Nutrition Symposium,Smart
Fanning; Technological Solutions for Large Herds Retrieved from
https://www.wageningenur.nl
Fig. 7 A captured measurement from light intensity sensor test and the
[3 ] Qin Z. (2016). Precision agriculture technology for crop farming.
estimated values from the Kalman filtering
Taylor & Francis Group, LLC.
In Fig. 7, the sinusoidal line represents the measurement [4 ] Grewal, Mohinder S. Kalman filtering : Thoery and practice. 2nd ed.
New York :John Wiley, 2001
values of light intensity sensor which has the data in range
[5] Quinlan, J. R., d 986). Induction of Decision Trees. Machine Learning
between 10-40 lux due to the voltage noise. On the other hand,
I, Kluwer Academic Publishers
the almost straight line in the middle of the graph represents the [6] 'Machine Learning", [Online] Available:
estimated values from the Kalman filter which is 28 lux. Note https:/fwww.npmjs.com/package/machine-.earning [February 6, 2016].
that the real light intensity in the experiment is around 30 lux. [ 7] Paul R. Wonning. Gardeners Guide to Growing Cabbage in the
Vegetable Garden: Mossy Feet Books, 2015