You are on page 1of 6

1st International Conference on Advances in Science, Engineering and Robotics Technology 2019 (ICASERT 2019)

Design and Implementation of IoT based Ideal Fish


Farm in the Context of Bangladesh Aquaculture
System
Abdul Kadar Muhammad Masum Md. Shahin Md. Kalim Amzad Chy
Dept. of Computer Science and Dept. of Computer Science and Dept. of Computer Science and
Engineering, International Islamic Engineering, International Islamic Engineering, International Islamic
University Chittagong, Bangladesh University Chittagong, Bangladesh University Chittagong, Bangladesh
akmmasum@yahoo.com shahin.iiuc@gmail.com kalim.amzad.chy@gmail.com
Shahidul Islam Khan Ahmed Shan-A-Alahi Md. Golam Rabiul Alam
Dept. of Computer Science and Dept. of Computer Science and Dept. of Computer Science and
Engineering, International Islamic Engineering, International Islamic Engineering, BRAC University,
University Chittagong, Bangladesh University Chittagong, Bangladesh Bangladesh
nayeemkh@yahoo.com shourabh_iiuc@yahoo.com rabiul.alam@bracu.ac.bd

Abstract— This research proposes an IoT based ideal fish like pH, oil level, temperature, water level, and fish hungry
farm that will be low in cost and high in fish production level. A level by perceiving the fish behavior. To achieve an accurate
WEMOS D1 microcontroller is engaged in determining various
value several data manipulation techniques like mean, mode,
essential attributes of the water like pH, oil layer, water level,
temperature, and the fish behaviour to recognize their hungry data normalization has been followed. The gathered
level that have a great effect on the fast growth of the fish. information is stored in the cloud server for further analysis
Through our developed mobile app and a user-friendly web like forecasting the probable pH, oil layer, water level,
interface the end user will be notified this information. The required fish food etc. to take a preventive step in advance. If
ESP8266 wifi module is built in the D1 microcontroller through the parameters exceed the predefined threshold level, then
which the system transmits the data to the web and mobile app.
the system will notify the user through the mobile app. To
In case of exceeding the predefined threshold value, the system
considers it as an abnormal situation and notify the concerned build up the framework only $40 - $45 is enough. This
authority to take the necessary steps. Moreover, the prediction amount is so low that a fish farmer can easily adapt and take
of the next day required fish food and other parameters assist part in settling the world protein challenge issue.
the fish farmer in taking a precaution step makes the system This paper is structured into five sections containing the
more unique and exceptional. Through these features, a fish introduction. Section II presents related work. Section III
farmer may produce more fish, and the protein challenge for combines methodology where circuit design, smart fish
the large population will be diminished drastically. monitoring, and controlling process is explained. An
Keywords — smart fish farm, pH sensor, oil layer detection, experimental result is shown in section IV. Section V The
Support Vector Regression (SVR), Random Forest Regression conclusion and future work are deliberated in section V.
I. INTRODUCTION II. RELATED WORK
The present 7.3 billion world population is estimated to A few researchers from the genetic engineering field
be 11.2 billion in 2100. With the first growth rate of the produced genetically modified fish having more breeding
current population, at least 70% more protein will be capabilities. But their survival rate and nutrition level are
necessary in the year 2100 [1]. Oceans, a great source of lower than the natural fish [7]. Also supplying artificial fish
protein, is taken about 2.5% more fish than what ocean can feed containing more chemical, microbes, seaweeds, etc.
naturally support. WWF (World Wildlife Fund) reported that causes various disease. Alternate fish feeds are developed
since 1950, 90% of very usually captured fish have vanished having lower chemicals to control the diseases.
from the ocean [2]. We are increasing pressure on the ocean Nowadays, the Internet of Things (IoT) is all about the 4th
that's a threat to the natural balance. To meet the demand, an industrial revolution. In outcomes of this revolution, IoT
alternate source should be thought without threatening advances the fish farming system that makes up a large
nature. portion of the protein problem for the more growing
Aquaculture-based fish farming is an excellent solution population [8]. Traditional fish farming faces lots of troubles
to this critical problem. Fish farming is very economical as with respect to the change of the environment like extreme
only 1.5 pound fish food is required to grow 1 pound fish weather, heavy rainfall, high temperature, limited water on
while 8-9 pound feed and 8000L water in case of meat from the farm etc. SK telecom proposed an IoT based fish farm
cow [3]. Also, with respect to financial gain, there was a system that detects pH, dissolved oxygen level, temperature
huge difference between IoT based and non-IoT based fish and inform the fish farmer through their smartphones [9]. To
farm, found on an earlier study. The quick growth of the fish examine the water quality, the traditional system collects
relies on various factors: production factors [4], several samples from different locations and process in the
laboratory. To smart this system, Shareef et al. [10]
environmental factors [5], and biotic factors [6]. It is possible
developed a WSN (Wireless Sensor Network) technology-
to monitor and control some of those parameters. based decision support system that detect different physio-
The main purpose of this work is to build up a cost- chemical characteristics of water and notify the end-user via
effective fish farm monitoring and controlling framework smartphone. Daudi et al. also developed a system based on
that can properly determine various essential water attributes WSN technology [11]. Raspberry Pi based system

978-1-7281-3445-1/19/$31.00 ©2019 IEEE


continually determine the water quality through equipped pH_median = Middle of the sorted 5 values (2)
sensors and store the data in a cloud server for further
analysis [12]. If any water attribute surpasses its threshold Best_pH = abs(pH_median – pH_avg) > 0.5? ph_median :
value than an auto-generated text message will be sent to the ph_mean (3)
end user. To display the collected information, LabView Thick oil layer hinders the fishes from meeting the
software was utilized. However, the high cost is the primed sunlight that reduces the grown of the fish even can cause
downside of this system. Based upon multiple pairs of
death. Through IR optical sensor, the proposed system
ultrasonic sensor data, fish hungry level can be divided into
estimates the thickness of the oil layer and saves the data in
multiple category [13]. Chy et al. [14] proposed an IoT based
fish farm system that can forecast the next day fish food temporary memory. Like pH value estimation, the mean and
consumption. But they followed simple regression analysis median of the density of the oil layer is measured using
instead of considering several important features like pH, equation 4 and 5 where D is the oil density level in every
temperature, oil layer etc. 3min. To overcome the biased by outlier value, if the
absolute difference between oil layer mean and median
A three-layer architecture-based system is proposed value exceeds 10, then the median value is considered as the
where a remote layer collects the data, server layer transfer best oil layer value otherwise mean value.
and store the data. The obtained information is displayed in
the client layer [15]. The software is also integrated with the 𝑂𝑖𝑙_𝑙𝑎𝑦𝑒𝑟_𝑚𝑒𝑎𝑛 = (∑5𝑖=1 𝐷𝑖 )/5 (4)
system to monitor the sensor data. The system for fish farm
monitoring system determines similar factors instead of Oil_layer_median = Middle of the sorted 5 values (5)
considering other essential considerations like the density of Best_oil_layer = abs(oil_layer_median – oil_layer _avg)
the oil layer. Moreover, a large portion of these kinds of the >10? oil_layer_median: oil_layer_mean (6)
system doesn't distinguish which sensors they have utilized
or equipped expensive sensors that make the framework hard Typically, in a fish farm after a specific period, fish
to adapt to the fish farm. In addition, most of the framework food is provided to the fishes. But maybe continuously
provide the fish food after a specific period disregarding increasing fishes need food earlier than the given time or
nourishing when they need. Forecasting the amount of more food. We know, starve fishes grow very slowly. So,
required fish food is absent in most of the framework. feeding the fish when requires is a mandatory feature of a
In perspective of Bangladesh aquaculture system, most of model fishery. The system can understand the hungry level
the fish farm follow the conventional approaches to nurture either extreme or medium or normal hungry of the fishes so
the fish. Therefore, the expense turns out to be high, yet the that adequate fish food can be provided. The irregular fish
outcome isn't at a satisfying level. To the best of our movement can assist us in recognizing the hunger level.
knowledge, this is the main endeavor to build up an ideal fish Deployed three ultrasonic sensors (a, b, c) read six value
farm observing and understanding when fish require foods after each 10 sec. For each sensor, the mean and median
by examining their movement. Additionally, predicting the value is determined using from equation 7 to 10 where a
amount of fish food in next day will help the fish farmer to refers to the position of a fish measured by sensor a and b, c
make a preventive stride ahead of time. is like a. By comparing the mean and median value best
value is considered for every sensor. Then RMS value for
III. METHODOLOGY each sensor is measured using equation 11, 12 and 13.
Figure 1 demonstrates an overview of the proposed 𝑎_𝑟𝑚𝑠 is RMS value for sensor a, 𝑏_𝑟𝑚𝑠 and 𝑐_𝑟𝑚𝑠 are
IoT based fisheries monitoring and controlling system, how similar to 𝑎_𝑟𝑚𝑠. From these three RMS values, the system
the fisheries data will be collected and transmitted to the end can understand the fish hungry level. If three RMS value
users of the system. The system determines various exceeds the threshold level, then extreme hungry, if two
important features of the water like pH, oil layer, exceeds then medium hungry otherwise satisfied level.
temperature, water level, oil layer likewise the movement of
a = [a1, a2, … a6] b = [b1, b2, … b6] c = [c1, c2, …
the fish that have a great impact on the fast development of
c6]
a fish. At the first time launching, after supplying the power,
the system active and take a few seconds to initialize the 𝑎_𝑎𝑣𝑔 = (∑6𝑖=1 𝑎𝑖 )/6 (7)
equipped sensor. As pH sensor doesn't provide a suitable 𝑏_𝑎𝑣𝑔 = (∑6𝑖=1 𝑏𝑖 )/6 (8)
6
value at a stationary state, it needs to move slightly to 𝑐_𝑎𝑣𝑔 = (∑𝑖=1 𝑐𝑖 )/6 (9)
estimate the pH value of the water properly. A servo motor a_median = b_median = c_median = Middle of the sorted
performs this duty that turns the pH sensor 30 degrees in 5 values (10)
one way and 30 degrees in the opposite way. For more a_best = compare(a_avg, a_median)
exactness and lowering the power consumption, pH sensor b_best = compare(b_avg, b_median)
measures the mean and median of five subsequent pH value c_best = compare(c_avg, c_median)
using equation 1 and 2 where each value P is taken in every 𝑎_𝑟𝑚𝑠 = √(𝑎1 − 𝑎_𝑏𝑒𝑠𝑡)2 +. . . +(𝑎6 − 𝑎_𝑏𝑒𝑠𝑡)2 (11)
3min. As mean value can be biased by the outlier value of 2
𝑏_𝑟𝑚𝑠 = √(𝑏1 − 𝑏_𝑏𝑒𝑠𝑡) +. . . +(𝑏6 − 𝑏_𝑏𝑒𝑠𝑡) 2 (12)
the sensor, the median value is considered. If the absolute 2 2
𝑐_𝑟𝑚𝑠 = √(𝑐1 − 𝑐_𝑏𝑒𝑠𝑡) +. . . +(𝑐6 − 𝑐_𝑏𝑒𝑠𝑡) (13)
difference between these mean and median value exceeds
0.5, then the median value is considered as the best pH Extreme hot and cold temperature one of the main
value; otherwise, the mean value using equation 3. reasons lies behind the sluggish growth of a fish. Deployed
temperature sensor gives the temperature to the system in
𝑝𝐻_𝑚𝑒𝑎𝑛 = (∑5𝑖=1 𝑃𝑖 )/5 (1)
real time. Like to pH and oil layer value estimating, the
framework estimates the temperature of the fishery is used to write the code and burnt into the system that is
following equation 14 to 16 where T is the temperature in easy to understand and reliable.
each 3min. To determine the best value, the absolute
difference of mean and median is considered at 0.5.
𝑇𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒_𝑚𝑒𝑎𝑛 = (∑5𝑖=1 𝑇𝑖 )/5 (14)
temp_median = Middle of the sorted 5 values (15)
Best_temp = abs(temp_median – temp _avg) >0.5?
temp_median: temp_avg (16)
Correspondingly, the excessively high and low
water level isn’t appropriate for the fishes. A liquid level
sensor can gauge the water level; however, it is costly. To
make the system more cost-effective, a low-cost ultrasonic
sensor is deployed to determine the water level. Like prior
value estimating water_level is determined using equation
Fig. 2. The circuit design of the proposed system
17 to 19 where W is the water level in each 3min Here, the
absolute difference between the median and mean value is
B. Hardware: Following pieces of equipment are utilized to
considered as 5.
develop the hardware part.
𝑊𝑎𝑡𝑒𝑟_𝑙𝑒𝑣𝑒𝑙 = (∑5𝑖=1 𝑊𝑖 )/5 (17)
Water_level_median = Middle of the sorted 5 values (18) 1) WEMOS D1: WEMOS D1 is a WIFI development
Best_ Water_level = abs(Water_level _median – board based on ESP8266 12E. The working is similar to
Water_level _avg) >5? Water_level _median: Water_level NODEMCU, but the hardware is built resembling Arduino
_avg (19) UNO. In context of performance, WEMOS D1 is almost
two time faster than arduino.
Wemos D1 microcontroller sends the gathered data 2) pH Sensor:pH sensor measures the alkalinity or
to the end user of the system. A mobile application and web acidity of a solution that’s scale range is from 0 to 14 where
interface are integrated with the system to operate it easily.
7 is considered as neutral, below 7 acidic, above 7 alkalic.
The microcontroller transmits the data to the mobile
TOL-00163 model of pH sensor is equipped to measure the
application and web interface in real time to monitor the
pH level of the fisheries water [16].
data from anywhere in the world. In the web interface, the
data is plotted graphically and stored in the database that 3) HC-SR04 Ultrasonic Sensors: In this system,
will help the supervisor of the fishery to check the previous Ultrasonic wave based, HCSR04 ultrasonic sensors are used
data when necessary. Also, if the parameters of the water to measure water level and fish movement for detecting
quality and movement of the fish exceed the preset hungry level. At 40 KHz frequency, it can identify an object
threshold value, then the buzzer will be on, and fishery at range from 2cm to 400cm [17].
supervisor can be notified through the mobile app also. 4) IR Optical Sensor: The TCRT5000 IR optical sensor
is a reflective sensor which incorporates an infrared emitter
and phototransistor in a leaded package that prevents visible
light [18].
C. Algorithm:
In figure 3, at the first level, all sensor values are
initiated. Servo motor rotates 30 degrees right and left as pH
sensor needs to jiggle for a legitimate value. By utilizing pH
formula, pH is calculated 5 times in each 3min. Mean, and
median of the pH value is measured for a 15 min duration
from where the best value is calculated. As the ideal pH
value for the quick-growing of fishes is 7 – 8.5, the
framework turns on the alarm if it surpasses the range [19].
Fig. 1. Block Diagram Two IR optical sensor employees to determine the oil layer
of the water. The output of the IR optical sensor gives an
The proposed system is classified into the following parts: aggregate value, heat and noise value. By expelling the
noise and heat value form the total value, the system gets the
A. Circuit Design and Softwares
exact oil layer value. Like the pH sensor, the mean, median
The proposed system is developed by following the and best value of this sensor are calculated in the same
designed circuit, shown in Figure 2. The VCC and ground fashion. In the case of the oil layer value becomes less than
pin of every sensor is attached to the 5v pin and ground pin 500, it informs the end user and switches on the alarm.
of Wemos D1. pH sensor is connected to the analog pin, and
Mean, median, the best value of the temperature and water
from the analog output of the pH sensor, the value of pH is
level within 15min is measured like the pH and oil layer
estimated using the formula of pH calculation. IR optical
sensor, Servo motor, temperature sensor, ultrasonic sensor, a sensor. [20] says that the best temperature for the rapid
buzzer is connected to the Wemos D1 microcontroller board growth of the fish is 22C to 27C. So, in case of surpassing
according to the circuit diagram. Arduino IDE and language this range, the system notifies the owner of the fish farm
through developed mobile application and web interface. To
understand the fish moving rate, 6 subsequent readings of
the sensor is taken within one min duration. 4.5cm Root
Mean Square (RMS) value, is considered as a threshold
level to recognize the hungry level of the fishes. Integrated
ESP8266 WIFI module with the Wimos D1 microcontroller
transmits the data from the device to the mobile app and
web interface. Based on these important features of the
water, the system forecast how much fish feed is necessary
in real time.

Fig. 4. Machine Learning Model

Figure 5 shows the 1st phase of the back-propagation


process and figure 6 final outcomes where we select 0.005, a
threshold value for p.

Fig. 5. Backward Elimination1st phase

Fig. 3. Working process of the proposed system

E. Predicting the following day power consumption:


Forecasting various essential parameters makes the
system more unique. For simplicity, just predicting the
required fish food in the following day is considered. Like
this process, the system can forecast other features. Figure 4
is a demonstration of the machine learning process.
A well-structured dataset is a heart of any machine Fig. 6. Backward Elimination Final phase
learning model. From a local fish farm, more than 8-month
data has been collected. pH, oil layer, temperature, water Dataset is split into 80% and 20% where 20% is reserved
level and hungry level data is the main features of our to evaluate the trained model. As we need to forecast the
dataset. Raw data need to preprocess to remove the noise amount of the next day fish food, a continuous value, a
and standardization. In feature scaling z-score normalization regression model is selected. Among various regression
is applied that make the mean of data of a feature zero and model, some best regression model like Linear Regression
standard deviation 1. One hot encoding is imposed on the (LR), Polynomial Regression (PR), Support Vector
water level where 1 refers water level ok and 0 not ok. Regression (SVR), Decision Tree Regression (DTR),
Similarly, one-hot encoding is applied on hungry level data Random Forest Regression (RFR) has experimented to best
where 0 refers extreme, 1 medium and 2 satisfied level train the model. After completing the training process the
hungry. To select the best features for the predictive model, performance of the model is evaluated in respect to its
the backward elimination feature selection process is chosen accuracy, cross-validation, Mean Absolute Error (MAE),
that mainly decides the necessity of a feature based upon the Mean Squad Error (MSE), Root Mean Squad Error (RMSE).
p value. Observing these values against each regression model, SVR
gives us the best outcomes.
IV. EXPERIMENTAL RESULTS AND DISCUSSION
In figure 7, a prototype of the proposed framework
shows up where pH, temperature, oil layer, water level, the
hungry level of the fish are the important outcomes. To
monitor remotely, these essential attributes of the water are
sent to an interactive mobile application interface and to the Fig. 10. Oil Layer observation
web server through the ESP8266 WIFI module built in the
D1 microcontroller. Also, data is saved to the cloud storage
for further analytics. In case of any parameters exceeding
the predefined threshold value, the system switches on the
alarm and notifies the end user. A sample of the mobile
application interface is displayed in figure 8a and in figure
8b an abnormal situation arisen is detected and inform the
end user through mobile application. Fig. 11. Temperature observation

Fig. 12. Water Level observation


Before forecasting the following day power
consumption in real time, our trained model is examined by
testing data. The first column is the true amount of fish food
value. LR, PLR, SVR, DTR, RFR predict the amount of the
fish food value that appears in table 1.
Fig. 7. Photograph of the prototype
TABLE I. TESTING DATA
Fish Food LR PLR SVR DTR RFR
10134.51 10158.56 10142.37 10156.43 10158.56 10156.33
10126.22 10120.70 10115.18 10120.53 10113.49 10118.47
10128.79 10121.50 10125.34 10118.90 10120.24 10117.87
10090.00 10095.38 10096.95 10096.22 10090.58 10095.77
10076.12 10099.29 10099.45 10097.84 10090.58 10092.26

Based upon the prediction of the testing data, what other

(a) (b)
Fig. 8. (a) A sample of the mobile application interface (b) A sample of an
abnormal situation notification through mobile application

The system transmits the gathered data to the web interface


for graphical visualization. In figure 9, the system pictures
the pH data over time and save it into the csv datasheet of
the web server for future analytics. In resemblance of the pH
data visualization, the system also pictures the oil layer data
(figure 10), temperature data (figure 11), water level data used regression model predicts is visualize in figure 13
(figure 12) graphically for the remote user. where each regression value is separated with a different
color.

Fig. 13. Testing and Predicted Value Visualization

To judge the performance of the trained model several


techniques and tools like accuracy, cross-validation, MAE,
Fig. 9. pH observation MSE, RMSE is considered, shown in table 2. In cross-
validation testing, we have considered 5-fold. Because of
varying the parameters, the outcomes diverge from one
regression analysis to another regression analysis. Shifting
the degree of polynomial regression to three offers a better
result than linear regression. RFR gives a better result than [7] J. A. Beardmore and J. Porte, "Genetically modified organisms in
aquaculture," 2003.
decision tree regression. Overall, the performance of the
support vector regression is up to the satisfaction level. So, [8] T. S. F. A. E. T. I. Vision. (2014, 28 January). Towards Smart
it can be decided that the SVR-based model is more suitable Farming—Agriculture Embracing The IoT Vision. Available:
http://www.beechamresearch.com/files/BRL%20Smart%20Farming%
to forecast the water parameters. 20Executive%20Summary.pdf
TABLE II. MODEL EVALUATION [9] (2014). SK Telecom Starts Pilot Operation of IoT-based Fish Farm
Management System. Available:
Evolution of the Linear LR PLR SVR DTR RFR http://en.c114.com.cn/576/a855896.html
Model
Accuracy Training 86.01 86.02 85.74 98.85 97.16 [10] Z. Shareef, S. Reddy, and D. IGDTUW, "Design and development of
Testing 83.62 82.10 84.12 75.63 81.16 IOT Based Framework for Aquaculture," in National Conference on
Cross Validation Product Design (NCPD 2016), 2016.
82.67 82.67 83.76 72.88 82.89
MAR Training 13.68 12.75 13.50 2.77 5.59 [11] D. S. Simbeye and S. F. Yang, "Water quality monitoring and control
Testing 14.63 17.71 14.40 21.01 15.48 for aquaculture based on wireless sensor networks," Journal of
MSE Training 260.7 255.6 267.0 26.04 45.02 networks, vol. 9, p. 840, 2014.
Testing 307.1 452.9 305.8 638.5 342.3 [12] K. R. S. R. Raju and G. H. K. Varma, "Knowledge based real time
RMSE Training 16.15 15.99 16.34 5.10 6.71 monitoring system for aquaculture using IoT," in Advance Computing
Testing 17.52 21.28 17.49 25.27 18.50 Conference (IACC), 2017 IEEE 7th International, 2017, pp. 318-321.

V. CONCLUSION [13] A. K. M. Masum, M. K. A. Chy, I. Rahman, M. N. Uddin, and K. I.


Azam, "An Internet of Things (IoT) based Smart Traffic Management
In demand for the increased level of protein for the fast- System: A Context of Bangladesh," in 2018 International Conference
growing population of the world, the fish production needs to on Innovations in Science, Engineering and Technology (ICISET),
be raised. Conventional fish cultivation system fails to 2018, pp. 418-422.
overcome the protein demand because of the adverse [14] M. K. A. Chy, A. K. M. Masum, M. E. Hossain, M. G. R. Alam, S. I.
environment and ignoring the current behavior of the fish. Khan, and M. S. Alam, "A Low Cost Ideal Fish Farm Using IoT: In
This study proposes an IoT based smart fish monitoring and the Context of Bangladesh Aquaculture System," in International
the controlling system where various essential water Conference on Inventive Communication and Computational
Technologies (ICICCT-2019), Namakkal, Tamil Nadu, India -
parameters and fish behavior is continuously measured and 637018, 2019.
notify the end user in case of any abnormal situation arisen
like increased pH or dense oil layer etc. To make the system [15] M. Hua, D.-a. Zhao, W. Xia, Y. Zhu, and X. Liu, "The design of
more user-friendly, a web interface and mobile app are intelligent monitor and control system of aquaculture based on
wireless sensor networks," in Computer Science and Information
developed so that the user can examine all earlier data from Technology (ICCSIT), 2010 3rd IEEE International Conference on,
everywhere in the world. Moreover, the system can forecast 2010, pp. 9-12.
the water by machine learning regression algorithm that will
assist them in taking an advanced step in the fish farm. The [16] T. Bangladesh. (2019). PH Sensor with Module. Available:
https://www.techshopbd.com/product-categories/meters/2576/ph-
developed system is more cost-effective and smarter than sensor-with-module-techshop-bangladesh
many existing conventional systems as well as other smart
systems. Reducing the cost of the system and integrating [17] ElectronicWings. (2018, 23 September). Ultrasonic Module HC-
other important feature like determining the poison in the SR04. Available:
http://www.electronicwings.com/sensorsmodules/ultrasonic-module-
water is the future scope of the present system. The hc-sr04
incorporation of the proposed IoT-based smart fishery
monitoring framework supports the fish farmer to raise fish [18] T. Bangladesh. (2019). Reflective Optical Sensor - TCRT5000.
production. Available: https://www.techshopbd.com/product-categories/sensors-
98778/977/reflective-optical-sensor-tcrt5000-techshop-bangladesh
VI. REFERENCES [19] Y. E. Corporation. (2016). pH in Fish Farming. Available:
[1] I. Ullah and D. Kim, "An optimization scheme for water pump https://www.yokogawa.com/library/resources/application-notes/ph-in-
Control in smart fish farm with efficient energy consumption," fish-farming/
Processes, vol. 6, p. 65, 2018.
[20] T. F. Hobbyist. (2019). Temperature Control. Available:
[2] O. Heffernan, "No more fish in the sea," Nature, vol. 460, p. 956, http://www.tfhmagazine.com/aquarium-basics/temperature-
08/19/online 2009. control.htm

[3] W. O. Review(WOR). (2013, 28 January). WORLD OCEAN


REVIEW 2, THE FUTURE OF FISH – THE FISHERIES OF THE
FUTURE. Available: http://hdl.handle.net/20.500.11822/8484

[4] M. A. Adewolu, C. A. Adeniji, and A. B. Adejobi, "Feed utilization,


growth and survival of Clarias gariepinus (Burchell 1822) fingerlings
cultured under different photoperiods," Aquaculture, vol. 283, pp. 64-
67, 2008.

[5] V. Rubio, F. Sánchez-Vázquez, and J. Madrid, "Effects of salinity on


food intake and macronutrient selection in European sea bass,"
Physiology & behavior, vol. 85, pp. 333-339, 2005.

[6] P. Kestemont, S. Jourdan, M. Houbart, C. Mélard, M. Paspatis, P.


Fontaine, et al., "Size heterogeneity, cannibalism and competition in
cultured predatory fish larvae: biotic and abiotic influences,"
Aquaculture, vol. 227, pp. 333-356, 2003.

You might also like