You are on page 1of 6

2022 25th International Conference on Computer and Information Technology (ICCIT)

17-19 December 2022, Cox’s Bazar, Bangladesh

Design and Implementation of a Smart Bin using


IOT for an Efficient Waste Management System
Nafisa Anjum Antora1, Md. Ashiqur Rahman1, Abdullah Al Mosharraf1, Mehrab Ibn Ehsan1, Md. Alve1 and
2022 25th International Conference on Computer and Information Technology (ICCIT) | 979-8-3503-4602-2/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICCIT57492.2022.10055998

Mohammad Mamun Elahi2


1
Department of Electrical and Electronic Engineering
2
Department of Computer Science and Engineering
United International University, United City, Dhaka, Bangladesh
Email- nantora173008@bseee.uiu.ac.bd, mrahman172016@bseee.uiu.ac.bd, anishan171086@bseee.uiu.ac.bd,
mehsan173018@bseee.uiu.ac.bd, malve173022@bseee.uiu.ac.bd, mmelahi@cse.uiu.ac.bd

Abstract— Waste management is a challenging task in this In 2025, the daily waste generation in Bangladesh will be
modern era and different approaches are still being discovered around 47,064 tons. Dhaka alone generates 37% of the
to make the separation of waste more efficient using the state- county’s waste. This happens due to the poor waste
of-the-art technologies like Internet of Things (IoT), Edge-cloud management system. Chief Waste Manager of Dhaka South
integration, machine learning etc. In this work, we have City Corporation Md. Bakhtiyar Ahmed said - “We have a
designed and implemented an IoT-based smart bin that will use plan to produce gas, electricity and fertilizer from the waste,
a machine learning algorithm to separate different types of then we can reduce waste and waste will be used as the raw
wastes and send the data to the cloud server. It will sort organic material.” According to the government, it is possible to
and inorganic waste materials in an efficient manner using three
generate 60 MW of electricity from the amount of waste
different collection boxes built in. The wastes will be scanned
through several sensors and image classification was used to
generated daily in Dhaka alone [2].
separate the wastes in the designated collection boxes using the The target of the proposed work is to build a cleaner city
sensors. The sensor will also detect the level of the waste all around the world by developing a smart waste management
collected in the boxes and inform the personnel of the waste system, which will also contribute to the regeneration of
collection. The sensor will send a signal when it detects 70-80% energy. This project will help separate different types of waste
filled boxes, which will give them enough time to get to the effectively using IoT.
collection point. Each bin will have its own GPS tracking system
to locate its location and also to avoid the hassle of being stolen. Rest of the paper is organized as follows. Section II
Although this smart bin will be battery powered, to make it eco- discusses the background and related past and recent work.
friendlier and economical, solar power will be used to recharge Section III includes the detailed methodology of the proposed
the battery. A working prototype has been developed as a proof- system, the architecture and the development in the project.
of-concept and preliminary results prove the efficiency of the Section IV discusses the implementation details. Section V
proposed smart bin. discusses the details. Lastly, Section VI summaries the project
and shades some future remarks.
Keywords— Waste management system, Recycling, Raspberry
pi, Image processing, CNN, WasteNet. II. RELATED WORK
I. INTRODUCTION In [3], authors have proposed and implemented a smart
garbage management system to provide an easy collection
Waste management is an important area of research in this
system by providing direction. With the help of a mobile
modern era but introduction to technology and smart cities can
application, the garbage collector can get alert when a bin is
bring in different processes to make waste management
full and collects the garbage in time. The designed smart bin
convenient [1][2]. Smart cities enhance lifestyles, along with
did not consider segregating different types of wastes and only
reducing the harm on the environment. The idea of smart cities
generates alerts when the bin is full. In [4], a project was
aims to reduce urban waste and to operate efficiently via the
established using Arduino Mega along with various sensors
modern technologies available to date.
like IR sensors, a moisture sensor and a metal detector. In this
By 2050, global waste is expected to grow up to 3.40 project, they tried separating the waste in 3 different
billion tons, which is more than double population growth compartments. One compartment was to collect the metal and
over the same period. Overall, there is a positive correlation the other two was used to separately collect the wet and dry
between waste generation and income level. Daily per capita waste depending on its permittivity. Higher permittivity
waste generation in high-income countries is projected to means the waste is wet and it will be collected in the 2nd
increase by 19% by 2050, compared to low- and middle- compartment. If not, it will be collected in the 3rd
income countries where it is expected to increase by compartment. The compartments are placed on a rotating
approximately 40% or more. More than half of waste is plate; it rotates according to the type of waste detected through
currently openly dumped, and the trajectories of waste growth the sensors. The placement of the bins for collection is
will have vast implications for the environment, health, and programmed using delay/time. NODEMCU is also used to
prosperity, thus requiring urgent action. Lower-income inform the filling of the bins and to help send information to
countries generally rely on open dumping; 93% of waste is the collection office. The literature states, the smart bin is
dumped in low-income countries and only 2% in high-income functioning as per the simulation. But when implemented
countries [1].

Page 774
979-8-3503-4602-2/22/$31.00 ©2022 IEEE

Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on April 04,2023 at 17:40:58 UTC from IEEE Xplore. Restrictions apply.
problems have been faced in identifying the wastes effectively The block diagram in Fig. 1 below shows an overview
and proper calibration of the sensors were suggested. of the proposed system architecture:
Convolutional Neural Network (CNN) is widely used to
classify images, especially low-level features such as edges
and curves and then builds up to more abstract concepts
through a series of convolutional layers. This feature allows it
to classify images precisely such as cardboard, glass, metal,
paper, plastic, or trash. A literature from 2019, states the
process of CNN and how it can be used to classify different
types of wastes [5][10]. Resnet34 CNN, a residual neural
network with 34 pre-trained layers on the ImageNet database
has been used [6]. Classification of images is better when the
CNN is pre-trained as it has already learned some visual
features and can transfer that information easily. It has been
stated that a 92.1% accuracy was achieved by the neural Fig. 1. Proposed system architecture
network. The efficiency of the method was justified through
The interfacing will be combined with all the sensors,
some demonstration as well. Although it was an effective way
of classification, they did not really separate the wastes in modules and mechanisms with the Arduino Mega which will
different bins or different categories and they plan to work on execute commands sent from Raspberry Pi, the main
it in the future. processing unit. A Raspberry Pi camera module v2 will be
connected to the Raspberry pi to collect real time images for
WasteNet, a waste classification model has also been used image classification. For this we used a pre-trained model
very recently to separate wastes accurately. According to our from the TensorFlow object detection model. TensorFlow
citing from the literature of 10th June, 2020, they managed to allows us to make custom datasets for image classification as
train images and identify them accordingly and separate them a result we choose this platform for training our model. As
in designated boxes [8]. The training of the images and tuning TensorFlow itself is bulky and it cannot function properly in
them is quite complex but freezing the initial layers of the
Raspberry pi we are using TensorFlow Lite instead. Before
tuning and allowing the last layers with more information to
that we had to set a custom dataset for all the varieties of
pass through gives the maximum output. Through this paper,
it can be learnt that a 97% accuracy can be achieved using this waste that would be collected accordingly to its category and
process. Although the level of accuracy is huge, this project is after that the images had to be labeled and have been trained
limited to inorganic waste only. using TensorFlow 2 detection model zoo’s SSD mobileNet
V2 FPNLite 320x320 pre-trained model.This specific model
In [9], the authors have proposed an intelligent waste has been used because of his fast detection property for
management system by predicting the possibility of waste mobile devices like Raspberry pi and also it has a low
things to monitor hazardous and nonhazardous waste by processing power. After the training with TensorFlow object
sensor-based dustbins and effectively collect and predict the detection model we converted our custom trained model to
waste. But the proposed method did not consider classifying
TensorFlow Lite for further use in Raspberry pi. After that
the waste based on its image and analyze it with machine
the trained model was executed through a python script to
learning algorithms.
evaluate it using Raspberry pi. We had up to 71% detection
Using machine learning to differentiate among different level from our training and in future we want to broaden our
types of waste things to detect and classify waste materials is dataset for more accuracy in detection level.
an important feature of a smart bin as part of an intelligent
waste management system [11]. Considering both waste The Arduino will execute commands in a sequential
segregation as well as machine learning to detect them is a manner where it will first check whether the waste is organic
challenging task, which is missing in most of the above works. or not using image classification for more accurate results. If
it is organic, then it will directly be sorted to the organic waste
III. PROPOSED SYSTEM ARCHITECTURE
container. IR sensors will be used to sense the intake of
A smart bin will be a fundamental part of this project and wastes. A humidity sensor will also be used to measure the
intake of wastes will be through this bin. This smart bin will humidity of the organic waste and separate it more precisely.
help sort organic and inorganic waste materials in an efficient The waste will then be sorted to its distinct container using
manner using the modern technologies available. There will the waste routing system which will be custom made using
be 3 different collection boxes built in. One to collect organic motors. A conveyor belt will be made using the motors where
waste, second to collect metal waste and the third to collect the waste will be dropped and the humidity sensor on the side
non-metal waste. The wastes will be scanned through several and after that the metal detector underneath will one by one
sensors and if someone dumps something in the bin, the bin sort the waste. After sorting using a tilting mechanism (the
will scan and separate it in the designated collection boxes seesaw) the waste will be dropped into its designated boxes.
using the sensors. Sensors like sonar sensors will send data The seesaw will tilt towards the right if metal, tilts towards
when it detects 70-80% filled boxes, which will give them the left if non-metal, or using servo motors it will open the
enough time to get to the collection point. Each bin will have middle door to drop the organic into the middle collection
its own GPS tracking system to locate its location as well. box. The sonar sensor x3 will detect the level of the waste
Although this smart bin will be battery powered, to make it collected in the boxes and inform the personnel of the waste
eco-friendlier and economical, solar power will be used to collection. The sensor will send a signal when it detects 70%-
recharge the battery. 80% filled boxes, which will give them enough time to get to

Page 775

Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on April 04,2023 at 17:40:58 UTC from IEEE Xplore. Restrictions apply.
the collection point. Each bin will have its own GPS (Neo-
7MV2) tracking system for effective waste collection and
also to avoid the hassle of being stolen. 6V, 4.5Ah
rechargeable SUNCA lead acid battery would be used to
power the bin. To make the bin eco-friendly it will be solar
powered using a 9v solar panel connected to the power supply
to charge its batteries.

A central cloud based server will also be developed by


interfacing Raspberry Pi with MySQL which will generate a
table in real time letting the authority know the bin number,
location & threshold. This would be done using a Wi-Fi Fig. 3. Front View of the Structure
module (DP06) to make the waste unloading process more
smooth and efficient.
IV. IMPLEMENTATION DETAILS
Fig. 2 below gives a 3D representation of the Smart Bin
that would be used in this project. The figures give a top, side
and front view of the Smart Bin. From the representation we
can see that there is an opening for the intake of waste. When
the waste is sensed by the IR sensor the system will be awake
and the waste will fall on the conveyor belt. The waste will
be classified using the sensors and image processing and once
it is done with the help of the servo motor it will be deposited
into its designated box. The 3 boxes will be removable so
when it is 70% filled, the sonar sensor will detect it and the
information will be sent to the cloud system and the bin will
Fig. 4. Front View of the Structure
be tracked using its GPS for the smart bin to be emptied.

Fig. 2. 3D View of the Structure

A. Physical Prototype
Fig. 3 shows the front view of the physical prototype of
the smart bin and fig. 4 shows the side view of the bin from Fig. 5. The funnel and the seesaw
which the intake of wastes will take place. We had to develop
a mechanical system which is complex engineering as we had B. Methodology
to combine a lot of things together keeping in mind the Image classification: The interfacing will be combined
position, measurements of where to install all the sensors for with all the sensors, modules and mechanisms with the
better detection. The whole prototype is made from scratch. Arduino Mega which will execute commands sent from
The waste will be detected through the IR sensors of the Raspberry Pi, the main processing unit. A Raspberry Pi
intake passage and which will turn on the entire system. camera module v2 will be connected to the Raspberry pi to
Underneath the belt there is a metal detector, on the side there collect real time images for image classification as shown in
is a humidity sensor and the waste will be detected in a few Fig. 6. For this we used a pre-trained model from the
steps while on the conveyor belt which is driven by a motor TensorFlow object detection model. TensorFlow allows us to
and it will finally be delivered into the funnel underneath make custom datasets for image classification as a result we
(Fig. 5). As soon as the detection is complete, using the tilting choose this platform for training our model. As TensorFlow
mechanism (the seesaw) the waste will be dropped into its itself is bulky and it cannot function properly in Raspberry pi
designated boxes. The seesaw will tilt towards the right if
we are using TensorFlow Lite instead. Before that we had to
metal, tilts towards the left if non-metal, or using servo
set a custom dataset for all the varieties of waste that would
motors it will open the middle door to drop the organic into
the collection box. Underneath the seesaw, there are 3 sonar be collected accordingly to its category and after that the
sensors attached to monitor the threshold of the bins to avoid images had to be labeled and have been trained using
overflowing. TensorFlow 2 detection model zoo’s SSD mobileNet V2
FPNLite 320x320 pre-trained model. This specific model has

Page 776

Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on April 04,2023 at 17:40:58 UTC from IEEE Xplore. Restrictions apply.
been used because of its fast detection property for mobile Lite for further use in Raspberry pi. As shown in Fig. 7, the
devices like Raspberry pi and also because it has a low trained model was executed through a python script to
processing power. evaluate it using Raspberry pi. We had up to 71% detection
level from our training and in future we want to broaden our
dataset for more accuracy in detection level.
We have trained our custom dataset with a pretrained
object detection model from TensorFlow object detection
model zoo and we have achieved a precision level up to 95%
with our custom dataset. In Fig. 8, we can see our
classification loss which is getting lower with increasing the
steps of training. Fig. 9 and Fig. 10 show the detection level
and live evaluation, respectively.

Fig. 6. Primary setup for image collection

Fig. 8. Classification loss of the custom dataset training

Fig. 9. Detection level in Tensor board

Fig. 7. Training with custom dataset Fig. 10. Live image detection evaluation

After the training with TensorFlow object detection


model we converted our custom trained model to TensorFlow

Page 777

Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on April 04,2023 at 17:40:58 UTC from IEEE Xplore. Restrictions apply.
Cloud interfacing: The Arduino will execute Sensor connection & communication system: In the
commands in a sequential manner where it will first check schematic above, we are using a humidity sensor, IR sensor,
whether the waste is organic or not using image classification metal detector, GPS module, servo motor, sonar sensor,
for more accurate results. If it is organic, then it will directly raspberry pi camera module, conveyor system, LED panel
be sorted to the organic waste container. IR sensors will be (for illumination) etc. We will be connecting all the sensors
used to sense the intake of wastes. A humidity sensor will including the GPS module with the Arduino Mega via I2C
also be used to measure the humidity of the organic waste and protocol. We will interface the Arduino and the raspberry pi
separate it more precisely. If it is inorganic, a metal detector as a single processing unit so that the system can be more
will check whether it’s metal or non-metal. The waste will robust. To send data over the cloud for monitoring we will be
then be sorted to its distinct container using the waste routing using the built-in Wi-Fi or LAN of the Raspberry Pi. The
system which will be custom made using motors. A conveyor schematic diagram of the corresponding setup is shown in
belt will be made using the motors where the waste will be Fig. 13.
dropped after sorting and a tilting mechanism will be made
using servo motors for the wastes to fall into the designated
bins.

Fig. 13. Schematic of sensor connection and Communication process

Fig. 11. Raspberry Pi interface with mySQL C. System Flow Chart

Fig. 12. Data table using MySQL

A central cloud-based server will also be developed by


interfacing Raspberry pi with MySQL which will generate a
table in real time letting the authority know the bin number,
location & threshold as shown in Fig. 11 & 12. This would
be done using a Wi-Fi module (DP06) to make the waste
unloading process smoother and more efficient. 6V, 4.5Ah Fig. 14. System flow chart
rechargeable SUNCA lead acid battery would be used to
power the bin. To make the bin eco-friendly it will be solar From the system flowchart as shown in Fig. 14, there are
powered using a 9v solar panel connected to the power supply two separate portions of sensing, one is to detect the waste
to charge its batteries. level in the bins via sonar sensor and the other portion is for
sorting the waste. The sorting part begins with an IR sensor
at the intake of the smart bin where the system launches right
The sonar sensor x3 will detect the level of the waste
after any waste is dumped through the intake. After that the
collected in the boxes and inform the personnel of the waste waste goes through image classification and there it will be
collection. The sensor will send a signal when it detects 70%- detected as either Organic or Inorganic. If detected as
80% filled boxes, which will give them enough time to get to organic, then in the next stage by using a humidity sensor we
the collection point. Each bin will have its own GPS (Neo- confirm it as organic. If detected as inorganic then it goes
7MV2) tracking system for effective waste collection and through metal detector and gets confirmed if it's metal waste
or other kind of inorganic waste such as plastic. For the left
also to avoid the hassle of being stolen.
part of the flowchart, we constantly check whether the waste

Page 778

Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on April 04,2023 at 17:40:58 UTC from IEEE Xplore. Restrictions apply.
level is beyond 70% or not. If it crosses 70% then the system of recyclable waste are being dumped into the landfills each
sends a notification to the central monitoring system. day which leads to pollution and disrupts the ability to reuse
recyclable materials and regenerate energy. This smart bin
D. Attributes of complex engineering
will help sort these waste and lead to a more efficient waste
Machine learning itself is a complex engineering management system for the future and help create job
problem which will be used in this project for image opportunities in the waste management sector.
classification. Thousands of images have to be trained for
sorting the wastes accurately and effectively. This process is
Although this smart bin showed positive results through
quite time consuming and complex. Once the module is
the sensors and the pre-trained model through TensorFlow
trained properly it can give approximately 98% accuracy in
sorting. In our case, the dataset gave 90% accuracy but if we object detection accuracy up to 90%, in future we want to
can train more objects, we can increase the accuracy. Also, broaden our dataset for more accuracy in detection level and
the mechanical system we developed is complex engineering interface it with our entire system. We also want to set up a
as the entire system was accurately measured, planned and complete cloud system using MySQL which would generate
implemented for all the sensors and the mechanism to work a table in real time so that actions can be taken accordingly.
accordingly.
REFERENCES
V. PRILIMINARY RESULTS [1] Oatman-Stanford, Hunter. "A Filthy History: When New Yorkers Lived
Knee-Deep in Trash". Collectors Weekly. Retrieved 2021-01-08.
We have successfully built a prototype and the entire
[2] Waste Management Indicators. Accessed: May 2021. [Online].
system showed promising results by separating metal non- Available: https://ec.europa.eu/eurostat/statistics-
metal and Organic wastes in its distinct boxes. We separately explained/index.php?title=Waste_ management_indicators#Overview.
tested our image processing dataset and it showed 90% [3] M. J. Siddique, M. Aynul Islam, F. N. Nur, N. N. Moon and M.
accuracy as shown in Table I below. Saifuzzaman, "BREATHE SAFE: A Smart Garbage Collection System
for Dhaka City," 2018 10th International Conference on Electrical and
Computer Engineering (ICECE), 2018, pp. 401-404, doi:
TABLE I. WASTE DETECTION ACCURACY 10.1109/ICECE.2018.8636767.
[4] Silva, B.N., Khan, M., Han, K.: Towards sustainable smart cities: a
review of trends, architectures, components, and open chal-lenges in
smart cities. Sustain. Cities Soc. (2018). https://doi.org/
10.1016/j.scs.2018.01.053
[5] D. Ziouzios and M. Dasygenis, "A Smart Recycling Bin for Waste
Classification," 2019 Panhellenic Conference on Electronics &
Telecommunications (PACET), 2019, pp. 1-4, doi:
10.1109/PACET48583.2019.8956270.
[6] D. Misra, G. Das, T. Chakrabortty, and D. Das, ‘‘An IoT-based waste
We have tested 10 waste samples for every segment of our management system monitored by cloud,’’ J. Mater. Cycles Waste
Manage., vol. 20, no. 3, pp. 1574–1582, Jul. 2018.
targeted waste. From the table, we can see that detection
[7] T. J. Sheng, M. S. Islam, N. Misran, M. H. Baharuddin, H. Arshad, M.
through sensors gave promising results for the separation of R. Islam, M. E. H. Chowdhury, H. Rmili, and M. T. Islam, ‘‘An
metal and non-metal but for organic waste the accuracy was Internet of Things based smart waste management system using LoRa
not so high. Image classification test results were high with and tensorflow deep learning model,’’ IEEE Access, vol. 8, pp.
respect to the humidity sensor detection for organic waste. 148793–148811, 2020.
[8] G. White, C. Cabrera, A. Palade, F. Li, and S. Clarke, ‘‘WasteNet:
Therefore, image classification can improve the organic Waste classification at the edge for smart bins,’’ 2020,
waste detection and give higher accuracy. For data collection arXiv:2006.05873.
of our cloud system, we have successfully interfaced [9] G. Uganya, D. Rajalakshmi, Yuvaraja Teekaraman, Ramya
raspberry pi with MySQL and we have manually generated a Kuppusamy, Arun Radhakrishnan, "A Novel Strategy for Waste
table for monitoring which can be accessed through the Prediction Using Machine Learning Algorithm with IoT Based
Intelligent Waste Management System", Wireless Communications
Internet. and Mobile Computing, vol. 2022, Article ID 2063372, 15 pages, 2022.
https://doi.org/10.1155/2022/2063372.
VI. CONCLUSIONS AND FUTURE WORKS [10] W.-L. Mao, W.-C. Chen, C.-T. Wang, and Y.-H. Lin, ‘‘Recycling
A lot of the waste around the world is not recyclable due waste classification using optimized convolutional neural network,’’
Resour., Conservation Recycling, vol. 164, Jan. 2021, Art. no. 105132.
to the lack of the separation techniques and the hassle of
[11] S. Zavare, R. Parashare, S. Patil, P. Rathod, and V. Babanne, ‘‘Smart
sorting different types of wastes. Waste management is a very City waste management system using GSM,’’ Int. J. Comput. Sci.
complex issue and it needs to be addressed. Large quantities Trends Technol., vol. 5, no. 3, pp. 74–78, 2017.

Page 779

Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on April 04,2023 at 17:40:58 UTC from IEEE Xplore. Restrictions apply.

You might also like