Design and Construction of a Smart Bin Using IoT

Devices and ML for Automatic Segregation of

Biomedical Waste
Aba Aboawah Addo1 , Emmanuel Paa Kwesi Opare2 ,
Dr. Isaac Nyantakyi3 , Newman Quarshie Phinehas4 ,
Dr. Samuel Danso5*

1 aboawah18@[Link]
2 opareemmanuelpk@[Link]
3 inyantakyi@[Link]
4 phinehasnewman@[Link]
* sdanso@[Link]
* Supervisor – Corresponding author
August 2025

Abstract learning model, trained and fine-tuned on a

biomedical waste dataset via the Roboflow
Background - Biomedical waste misman- platform. Classified outputs trigger servo
agement remains a critical challenge in de- motors to actuate the appropriate bin lids.
veloping countries such as Ghana, where up Data is synchronized with Firebase Realtime
to 83% of healthcare facilities do not segre- Database, while a web application (HTML,
gate waste properly. This leads to increased CSS, AngularJS) provides real-time moni-
risks of infection, environmental pollution, toring, logs, and notifications.
and higher treatment costs. Conventional Results - The RF-DETR model achieved
waste management systems rely heavily on a mean Average Precision (mAP@50:95) of
manual segregation, which is prone to hu- 93.6%, with precision and recall of 91.2%
man error and unsafe handling practices. and 89.0%, respectively. Hardware testing
Methods - This study presents an IoT and confirmed reliable servo actuation and fill-
AI-enabled Smart Bin System designed to level detection within 2 cm accuracy. The
automate biomedical waste segregation and web application successfully displayed live
monitoring. The system integrates a Rasp- bin statuses and triggered alerts with data
berry Pi as the central controller, a camera synchronization latency under 5 seconds.
module for capturing waste images, and ul- These results demonstrate that the Smart Bin
trasonic sensors for monitoring bin fill lev- System is both feasible and effective, offer-
els. Waste items are classified into four ing a scalable solution to improve biomedical
categories(general, infectious, sharps, and waste management in healthcare facilities.
pharmaceutical)using the RF-DETR deep Conclusion - This innovation automates

1
GCTU Smart Bin Biomedical Waste System

segregation, reduces human error and con- 2 Literature Review

tamination risks, and promotes sustainable
healthcare practices. While effective for cur-
A. IoT-Enabled Smart Bins for Biomed-
rent needs, further high-quality research is
ical Waste
required to enhance dataset size, model ro-
Recent studies have explored how IoT
bustness, and deployment in diverse settings
technologies can improve medical waste
for long-term waste management solutions.
management. Mohamed et al. (2023)
KEYWORDS — Smart Bin, Biomedical demonstrated that smart bins equipped
Waste, IoT, Machine Learning, RF-DETR, with IoT sensors, RFID tags, and cloud
Raspberry Pi, Firebase. storage could track waste from gener-
ation to disposal in hospitals. Their
system enabled real-time monitoring,
1 Introduction predictive fill-level estimation, and au-
tomatic audit reporting, which signif-
Biomedical waste is rapidly increasing in icantly reduced waste overflow inci-
both quantity and diversity across the globe. dents and improved regulatory com-
A report by the National Bureau of Statis- pliance. However, their study high-
tics revealed that medical waste production lighted challenges such as high imple-
rose from 1,222 kilo tons in 2008 to 2,235 mentation costs, dependency on inter-
kilo tons in 2017, yet recycling efforts remain net connectivity, and poor interoperabil-
low. Although healthcare waste is produced ity across different waste management
in small amounts, its mishandling poses se- phases. In another study, Okubanjo et
vere health and environmental risks. About al. (2024) proposed integrating eco-
85% of such waste is non-hazardous, while nomic modeling with IoT-enabled smart
the remaining 15% is infectious, toxic, or ra- bins to optimize waste collection sched-
dioactive. Common terms like biomedical, ules. Using sensor data and cost mod-
clinical, and healthcare waste reflect its var- els, their system reduced operational
ied classifications. costs by up to 30% in hospitals while
Historically, improper disposal practices lowering labor-intensive manual han-
such as open burning and uncontrolled dling. Despite these advantages, the
dumping have worsened pollution and en- approach was limited by high initial
dangered public health. In response, modern setup costs and its inability to segre-
regulations and technological solutions are gate different waste types, such as infec-
being adopted. tious, sharps, or pharmaceutical waste.
This project introduces a Smart Bin Sys- This gap is significant for biomedical
tem for automated biomedical waste segre- waste, where segregation is critical for
gation. It utilizes a camera and RF-DETR safety and compliance. Similarly, Ram-
deep learning model for real-time classifica- chandar Rao et al. (2017) designed
tion into four waste types(general, infectious, a GSM-based waste monitoring sys-
sharps, and pharmaceutical). Integrated IoT tem using ultrasonic sensors to measure
modules (ultrasonic sensors and servo mo- bin fill levels and SMS alerts to notify
tors) enable fill-level detection and bin actua- waste management teams. While effec-
tion. The system includes a Firebase backend tive and affordable, particularly for low-
for data handling and an AngularJS web ap- budget hospitals in Ghana, the system
plication for real-time monitoring and alerts. lacked automation and waste classifica-
Overall, it reduces human error, supports tion capabilities. Manual sorting was
safer waste handling, and provides a scalable, still required, exposing healthcare work-
intelligent solution for healthcare facilities. ers to contamination risks. Moreover,

2
GCTU Smart Bin Biomedical Waste System

reliance on GSM networks raised con- marily on urban hospitals, leaving ru-
cerns about system reliability in rural or ral healthcare facilities largely unad-
underdeveloped regions. By integrating dressed. Madihlaba et al. (2022) devel-
IoT devices, the Smart waste bin Sys- oped an IoT-enabled smart bin system
tem aims to provide a scalable, secure, equipped with load sensors, GPS, and
and cost-efficient solution for managing ultrasonic sensors to keep tabs on waste
hospital waste operations. levels and alert hospital staff. While
it doesn’t facilitate community engage-
B. AI-Based Waste Segregation in ment, this approach does help reduce
Healthcare waste overflow.
The study by Sarkar et al. (2023) ex-
plored AI-driven biomedical waste cat-
egorization using image recognition and
machine learning. The system success- 3 Methodology
fully classified infectious, chemical, and
pharmaceutical waste, improving dis-
The Smart Bin System for Biomedical Waste
posal accuracy. However, it requires
Management combines both hardware and
trained professionals and large datasets.
software components, leveraging IoT de-
Moreover, Kulkarni et al. (2024) ap-
vices, machine learning, and cloud integra-
plied machine learning for waste classi-
tion to automate waste segregation and pro-
fication, reducing landfill burden. How-
vide real-time monitoring. The following
ever, the system struggles with mixed
methodology outlines the design, training,
waste segregation and requires contin-
testing, and integration processes of the sys-
uous retraining. Mishra et al. (2022)
tem, ensuring accurate waste classification,
also used the Random Forest Algorithm
efficient bin actuation, and reliable data syn-
to predict waste accumulation rates,
chronization for healthcare waste manage-
optimizing waste collection in hospi-
ment.
tals. The system minimizes overflow
risks but demands high computational
System Architecture Overview
power. Most hospitals manually segre-
gate waste, increasing risks of contam- This shows the workflow of the Smart Bin
ination. Integrating AI-powered smart System for automated biomedical waste seg-
bins can automate segregation, improv- regation. The process begins with waste be-
ing worker safety. ing hovered over the camera, which captures
an image for analysis. The Raspberry Pi pro-
C. Real-Time Monitoring and Compli- cesses this image using the RF-DETR model
ance in Healthcare Waste Manage- to predict the waste category and triggers the
ment corresponding servo motor to open one of the
Concerns about inconsistent handling four compartments: Pharmaceutical Waste,
of biomedical waste were highlighted Infectious Waste, Sharp Waste, or General
by Kanyal et al. (2021). Their re- Waste. Ultrasonic sensors monitor the fill
search stressed the importance of real- levels of each compartment, and if a bin, such
time tracking through IoT and AI to as the general waste compartment, reaches
prevent improper disposal. Addition- capacity, the system sends a notification sig-
ally, to boost compliance, Sosunova & nal to a connected phone. This integrates
Porras (2022) explored data exchange IoT capabilities for real-time management,
methods by reviewing 173 studies on with Firebase handling data synchronization
IoT-based smart waste management sys- and an AngularJS web application providing
tems. However, their focus was pri- monitoring and alerts.

3
GCTU Smart Bin Biomedical Waste System

2. Camera module: The camera module

is responsible for capturing images of waste
items placed near the smart bin. According
to Patel et al. (2022), camera-based imaging
is crucial in AI-driven classification systems
as it provides the raw visual data required
for machine learning models. In this project,
Figure 1: the camera module transmits images to the
Illustrates the architecture of the smart bin Raspberry Pi, where the RF-DETR model
system processes them for classification. The use of
a camera ensures accurate detection and dif-
ferentiation between biomedical waste cate-
gories such as infectious items, sharps and
A. Hardware Design
pharmaceutical waste. This real-time image
The hardware components of the Smart
acquisition is a key component in enabling
Bin System include the Raspberry Pi, the
automated and contactless waste segregation.
camera module, servo motors, and ultrasonic
sensors. A regulated power supply supports
all components to ensure stable and unin- 3. Servo motors: The servo motors are used
terrupted operation. Together, these com- to automate the opening and closing of the
ponents enable real-time biomedical waste bin lids. According to Singh and Verma
classification, segregation, and monitoring. (2020), servo motors are widely applied in
automation systems because of their preci-
1. Raspberry Pi: The Raspberry Pi is a pow- sion control over angular position. In the
erful single-board computer widely used in Smart Bin System, once the Raspberry Pi
IoT and AI-driven applications due to its en- classifies the waste, it sends a control sig-
hanced processing capabilities and connec- nal to the corresponding servo motor, which
tivity features. According to Sharma and actuates the lid of the correct bin (general,
Gupta (2021), the Raspberry Pi offers suf- infectious, sharps, or pharmaceutical). This
ficient computational power to handle real- ensures that each type of waste is deposited
time image processing and machine learning into its designated container with minimal
tasks, making it suitable for embedded smart human involvement, thereby reducing risks
systems. In this project, the Raspberry Pi of cross-contamination and improving com-
serves as the central controller of the Smart pliance with biomedical waste management
Bin System. It receives image inputs from practices.
the camera module and runs the RF-DETR
deep learning model to classify waste into 4. Ultrasonic sensors: The ultrasonic sen-
four categories: general, infectious, sharps, sors are used to monitor the fill levels of
and pharmaceutical. Based on the classi- each bin. These sensors work by emitting
fication, the Raspberry Pi 4 sends signals ultrasonic waves and measuring the time it
to the servo motors to actuate the appropri- takes for the waves to bounce back from the
ate bin lid. At the same time, it processes surface of the waste. According to Khan
data from the ultrasonic sensors to monitor et al. (2019), ultrasonic sensing is a re-
bin fill levels and synchronizes this informa- liable and low-cost method for non-contact
tion with the Firebase backend. By linking distance measurement in smart waste man-
all hardware and software components, the agement systems. In this project, the sensors
Raspberry Pi 4 enables efficient real-time provide real-time data on bin capacity to the
biomedical waste segregation and monitor- Raspberry Pi 4, which then updates the Fire-
ing. base backend. When a bin reaches a pre-

4
GCTU Smart Bin Biomedical Waste System

defined threshold, notifications are triggered fine-tuned through transfer learning. Pre-
through the web application to alert waste trained weights from the COCO dataset were
management personnel. This feature helps adapted to the biomedical waste dataset to ac-
prevent bin overflows and supports efficient celerate convergence and improve accuracy.
scheduling of waste collection. A total of 4,871 images with 6,773 annota-
tions were collected from publicly available
biomedical waste datasets and manually la-
beled using Roboflow.

Figure 2:
Shows the back view and front view of the Figure 3:
Smart Bin prototype, highlighting the Overview of the number of annotations for
integration of hardware components such as each class in the dataset
the camera, sensors, and actuators.

B. Software Design
The software architecture of the Smart
Bin System integrates multiple components
to ensure seamless communication between
the hardware, backend, and end-users. At
the core is the RF-DETR deep learning
model, deployed on the Raspberry Pi for real- Figure 4:
time biomedical waste classification. The shows the dataset distribution across train,
backend is powered by Firebase Realtime validation and test splits
Database, which manages secure data syn-
chronization, storage, and push notifications. The dataset was split into 87% training,
On the frontend, a web application built with 9% validation, and 4% testing.
HTML, CSS, and AngularJS provides users
with an intuitive interface to monitor bin sta-
tuses, view classification logs, and receive
alerts when bins are full or hazardous waste
is detected. Together, these software com-
ponents maintain the reliability, scalability,
and usability of the system, enabling effi-
cient biomedical waste management across
healthcare facilities.

1. Model Training: The core of the Smart Figure 5:

Bin System is the RF-DETR (Region-based shows the dataset split into 87% training,
Detection Transformer) model, which was 9% validation and 4% testing
trained to classify biomedical waste into
four categories: general, infectious, sharps, Preprocessing included resizing all im-
and pharmaceutical. The model was imple- ages to 640 × 640 pixels and data augmen-
mented using the PyTorch framework and tation (rotation, flipping, zooming) to im-

5
GCTU Smart Bin Biomedical Waste System

prove robustness against variations in light- 4 Mathematical Equations

ing and orientation. Training was con-
ducted on Roboflow. The RF-DETR model 1. Data preprocessing: Mathematically,
was initialized with pre-trained weights from normalization can be expressed as:
the COCO dataset and fine-tuned on the 𝑥
biomedical waste dataset. Roboflow man- 𝑥′ =
255
aged key hyperparameters such as learning
2. Performance metrics: Model perfor-
rate scheduling, batch sizes, and epochs, op-
mance was evaluated using:
timizing training efficiency. The Adam opti-
mizer was applied with weight regularization Precision · Recall
to prevent overfitting. 𝐹1 = 2 ·
Precision + Recall
𝑇𝑃 𝑇𝑃
Precision = , Recall =
𝑇 𝑃 + 𝐹𝑃 𝑇𝑃 + 𝐹𝑁
2. Backend development (Firebase Real- where: TP (True positives):correctly pre-
time Database): The backend was imple-
dicted positive cases.
mented using Firebase Realtime Database,
which serves as the core data management FP (False positives):incorrectly predicted as
service. The Raspberry Pi transmits waste positive.
classification results and bin fill-level data
to Firebase, where it is stored securely in a FN (False Negatives): incorrectly predicted
NoSQL format. Firebase also supports real- as negative
time synchronization, ensuring that data up- Mean Average Precision:
dates are reflected instantly on the web inter-
𝑛
face. In addition, Firebase Cloud Messag- 1 ∑︁
ing (FCM) was used to implement the noti- 𝑚 𝐴𝑃 = 𝐴𝑃𝑖
𝑁 𝑖=1
fication system, providing alerts when bins
reached capacity or when hazardous waste where APi is the average precision for class
such as sharps was detected. The backend i.
design ensures scalability, secure data han- 3. Total Loss Function:
dling, and minimal latency in communica-
tion. 𝐿 𝑡𝑜𝑡𝑎𝑙 = 𝜆 𝑐𝑙𝑠 𝐿 𝑐𝑙𝑠 + 𝜆 𝑏𝑜𝑥 𝐿 𝑏𝑜𝑥 + 𝜆 𝑜𝑏 𝑗 𝐿 𝑜𝑏 𝑗

where:
𝐿 𝑐𝑙𝑠
3. Frontend development (AngularJS Web = classification loss (cross-entropy between
Application): The frontend was developed predicted and true labels)
using HTML, CSS, and AngularJS, provid-
ing a lightweight and responsive user inter- 𝐿 𝑏 𝑜𝑥
face. The application retrieves data dynam-
= bounding box regression loss (using L1
ically from Firebase, presenting healthcare
loss or GIoU loss)
staff with a dashboard that displays bin sta-
tuses, waste classification logs, and real-time 𝐿𝑜 𝑏 𝑗
fill levels. The interface also includes alert
notifications triggered by the backend, en- = objectness loss, ensuring correct object-
abling waste management personnel to take background assignment
immediate action when bins are full or haz-
ardous materials are identified. 𝜆

6
GCTU Smart Bin Biomedical Waste System

= values are weights balancing the contribu-

tions of each term
4. Bipartite Matching for Object Detec-
tion: Each predicted box must correspond
to exactly one ground-truth box. This is
achieved through bipartite matching using
the Hungarian algorithm. The matching min-
imizes the cost:

𝑁
∑︁
𝑦ˆ = arg min L (𝑦𝑖 , 𝑦ˆ 𝜎(𝑖) )
𝑖=1

where:
𝑦𝑖

= ground truth labels

𝑦ˆ 𝜎 (𝑖)

= prediction matched to ground truth

= matching loss (combining class and box

regression terms)

System workflow
The workflow of the Smart Bin System
begins when a waste item is placed near the
bin. The camera module captures an im-
age of the item and sends it to the Rasp-
berry Pi, which runs the RF-DETR model
to classify the waste into one of four cate- Figure 6:
gories: general, infectious, sharps, or phar- System Workflow illustration
maceutical. Once classification is complete,
the Raspberry Pi triggers the corresponding
servo motor to open the lid of the designated
bin, allowing the waste to be disposed of
correctly. At the same time, the ultrasonic
sensors measure the fill level of each bin and
transmit this data to the Raspberry Pi. Both
the classification result and fill-level data are
then uploaded to the Firebase backend, where
they are stored and synchronized in real time.
The AngularJS web application retrieves this
information and displays it on the dashboard.

7
GCTU Smart Bin Biomedical Waste System

6 Results and Analysis

Model Results:
The RF-DETR model achieved:

Figure 7:
System workflow diagram showing how
waste is detected, classified, and segregated

5 Testing

A. Unit testing: Each component of the

Smart Bin System underwent rigorous unit
testing. The servo motors were tested using
the Arduino mega to get to know the various
angles at which the bins will open and so that
we can tune them to the right angles at which
the bins will open. The sensors were tested to
verify the accuracy and reliability of the ul-
trasonic sensors under various environmen-
tal conditions, ensuring proper bin fill-level
measurements. The communication mod-
ule was tested to ensure stable and consistent
data transmission from the Raspberry Pi to
the cloud with minimal latency and errors.
Software testing was conducted to debug and
validate the functionality of the embedded
software (Raspberry Pi scripts) and the web
application, ensuring smooth data flow, real-
time updates, and user interactions.

B. Integration testing: Integration testing

was performed to ensure that all system com-
ponents worked together seamlessly: End-
to-End Data Flow:Testing the complete data
flow from sensor data collection (e.g., ultra-
sonic sensors) to real-time data visualization
on the web interface, ensuring smooth data
synchronization.

8
GCTU Smart Bin Biomedical Waste System

Table 1: Class-wise Performance of RF-

DETR Model

Class Precision (%) Recall (%) F1-Score

General waste 92 87 0.89
Infectious waste 93 91 0.92
Sharp waste 90 88 0.89
Pharmaceutical waste 91 89 0.90

mAP@50: 93.6% precision calculated at IoU thresholds rang-

Precision: 91.2% ing from 0.50 to 0.95 in increments of 0.05.
Recall: 89.0% This gives a robust assessment of the model’s
F1-Score: Balanced trade-off between ability to detect objects at varying levels of
precision and recall across all [Link] overlap between predicted bounding boxes
results indicate that the RF-DETR model can and ground truth labels. As shown in Figure
accurately detect biomedical waste items in 6, the mAP values increased steadily during
images with high confidence. training, reaching a plateau between epochs
• The infectious class achieved the best 12–15.
performance, while sharps showed
Training Loss Curves:
slightly lower recall due to small ob-
ject sizes and occlusions. This can be
improved with dataset expansion.
• Misclassifications were most common
between general vs. pharmaceutical
waste, reflecting dataset overlap.
Figure 9:
Model evaluation metrics: Training loss curves for Box Location Loss,
Class Loss, and Box Overlap Loss over 40
epochs

During the training phase of the RF-DETR

model, three primary loss functions were
monitored: Box Location Loss, Class Loss,
and Box Overlap Loss. These loss metrics
Figure 8: provide insights into the model’s learning be-
Training performance showing mAP havior across epochs and its ability to accu-
progression rately detect and classify biomedical waste
items.
The mAP (mean average precision) is
a widely used metric for object detection • Box Location Loss: This measures
tasks, measuring the trade-off between pre- the spatial discrepancy between pre-
cision and recall across multiple Intersec- dicted and actual bounding box coordi-
tion over Union (IoU) thresholds. Specif- nates. As shown in the leftmost plot, the
ically, mAP@50:95 represents the average loss exhibits some fluctuations across

9
GCTU Smart Bin Biomedical Waste System

epochs, indicating that the model peri- occurred in servo response during repeated
odically refined its object localization operations, which can be linked to the Rasp-
but faced challenges with consistent ac- berry Pi’s limited processing power. Overall,
curacy. the system achieved its objective of real-time
biomedical waste segregation. Compared to
• Class Loss: This reflects the model’s manual sorting, it reduces human error and
ability to correctly classify waste cat- improves compliance with healthcare waste
egories (general, infectious, sharps, standards. With dataset expansion, faster
pharmaceutical). The middle plot hardware, and improved calibration, perfor-
demonstrates a significant and consis- mance could be enhanced further.
tent downward trend, particularly in the
first 10 epochs, indicating that the model
Bin lid actuations:
quickly learned to differentiate between
waste classes and achieved stable per-
formance.

• Box Overlap Loss: This quantifies

the error in predicted bounding box
overlaps using Intersection over Union
(IoU). The rightmost plot shows mod-
erate fluctuation, which may be at-
tributed to the varying shapes and sizes
of biomedical waste in the dataset.

Overall, the Class Loss reduced rapidly Figure 10:

and stabilized, suggesting that the RF-DETR shows actuation of infectious waste bin and
model was highly effective in learning waste its after detection of a latex glove
classification. Despite some variability in
the box-related losses, the model maintained
good bounding box precision throughout
training.
Discussion: The model classified the glove
Hardware Integration results: These results as infectious waste with very high confidence
confirm that the integration between the ma- (0.99), and the bounding box correctly cap-
chine learning model, Raspberry Pi con- tured the object showing strong model relia-
troller, and servo motors was successful. The bility and effective differentiation from gen-
bin lids responded in real time (within 2–3 eral waste, which is crucial in medical waste
seconds) after classification, ensuring that management. Nonetheless, the bounding
each waste item was segregated into the cor- box was slightly oversized, incorporating ex-
rect category. The actuation tests showed tra background that could potentially confuse
that the Smart Bin System can automatically the model in scenarios with multiple objects.
segregate waste with high accuracy. Each Moreover, the current detection works well
test object (glove, syringe, plastic wrapper, for gloves but may not generalize to other
and pill pack) was correctly classified by types of infectious waste, such as masks,
the RF-DETR model, with confidence scores gowns, or tissues. Expanding the dataset to
above 90%. The Raspberry Pi then triggered cover these categories is necessary. Finally,
the servo motor to open the correct bin lid gloves placed near non-infectious items (e.g.
within 2–3 seconds, which is acceptable for plastic bags) may increase the risk of false
real-time operation. Occasional delays also positives.

10
GCTU Smart Bin Biomedical Waste System

• Controls to remotely open each bin or

toggle each bin lid. There is also a
global control option to Open All or
Close All bins..

• A motivational banner emphasizing on

the importance of proper waste segre-
Figure 11:
gation (“Sanitation is the foundation of
shows actuation of sharp waste bin and its
health, and health is the foundation of
algorithm after detection of a syringe
life”). The quotes change consistently.

Discussion: The model successfully de- 7 Conclusion and Recom-

tected and classified the syringe as sharps
with a confidence score of 0.95. The bound- mendations
ing box placement was reasonably accurate,
The Smart Bin System lays strong ground-
covering most of the syringe and trigger-
work for future innovations in automated
ing the correct bin action, which demon-
biomedical waste management. By tack-
strates reliable system response. However,
ling its current shortcomings and putting for-
the bounding box failed to fully enclose the
ward the suggested improvements, this sys-
needle portion, which could reduce accuracy
tem could greatly enhance waste segrega-
when handling larger or differently oriented
tion practices, lower contamination risks, and
syringes. Additionally, while the confidence
boost adherence to healthcare waste manage-
score was high, there remains a risk of mis-
ment standards. In the end, it marks a sig-
classification with objects of similar shape,
nificant leap in using technology to protect
such as pens or markers. This highlights
both public health and the environment.
the need for stronger class separation in the
training dataset. A. Recommendations: To tackle the limita-
tions, we’ve spotted and boost the perfor-
Web application results:
mance of the Smart Bin System, here are
some recommendations to consider:

I. Model Optimization and Retraining: By

broadening the training dataset to include a
wider variety of waste items, especially those
pesky small, overlapping objects, we can en-
hance the model’s recall and classification
(c) Waste accuracy. Plus, fine-tuning the RF-DETR
(a) landing (b) login management model and trying out other detection archi-
page page dashboard tectures like YOLOv5 or YOLOv11 could
really take performance to the next level.
Figure 12:
Web app Diagrams II. Mobile App Optimization: Revamping the
user interface with custom CSS and tweak-
Displays: ing the layout for better mobile responsive-
ness will make sure the web app is user-
• Bin status showing all four bins corre- friendly and fully functional on smartphones
sponding to the waste types: Infectious, and tablets, enabling staff to manage the sys-
General, Sharps, and Pharmaceutical. tem while on the move.

11
GCTU Smart Bin Biomedical Waste System

III. Servo Reliability: The servo failure was a [7] Mishra, S., Jena, L., Tripathy, H. K.,
hardware-specific issue and did not compro- \& Gaber, T. (2022). Prioritized and pre-
mise the overall functionality of the Smart dictive intelligence of things enabled waste
Bin System. The system continued to clas- management model in smart and sustainable
sify waste correctly and operate the remain- environment. PLoS ONE, 17(8 August).
ing bins. By addressing the faulty com- [Link]
ponent through replacement and adopting [8] Okubanjo, A., Odufuwa, B., Okandeji,
stronger actuator options, the reliability of A., \& Daniel, E. (2024). Smart Bin and
the bin actuation mechanism can be fully en- IoT: A Sustainable Future for Waste Man-
sured for long-term deployment. agement System in Nigeria. Gazi Univer-
sity Journal of Science , 37(1), 222–235.
[Link]
8 References [9] Ravi, V. R., Hema, M., SreePrashan-
thini, S., \& Sruthi, V. (2021). Smart bins
[1] Aman Valera KJSCE, I., Rahul Punjabi for garbage monitoring in smart cities using
KJSCE, I., \& Shweta Dhawan Chachra, I. IoT system. IOP Conference Series: Ma-
(2019). Smart Dustbins-Automatic Segre- terials Science and Engineering, 1055(1),
gation \& Efficient Solid Waste Manage- 012078. [Link]
ment using IoT Solutions for Smart Cities 899x/1055/1/012078
Viral Rambhia. International Journal of En- [10] Sosunova, I., \& Porras, J. (2022).
gineering Research \& Technology (IJERT) IoT-Enabled Smart Waste Management
[Link], 8. [Link] Systems for Smart Cities: A Systematic
[2] Ardiyan, A., Gunardi, Y., \& Hamid, Review. IEEE Access, 10, 73326–73363.
A. (2024). Design internet of things for [Link]
smart waste bin management with wemos [11] Madihlaba, E. N., Malele, V., \&
based and firebase application. Journal of Owolawi, P. A. (2022). The Design of the
Soft Computing Exploration, 4(4), 276–283. IoT-based Smart Bin System. International
[Link] Conference on Intelligent and Innovative
[3] GOUD, E. G., P, M., R, V., B, Computing Applications, 2022, 183–189.
B., \& BHARGAVA, DR. S. (2023). [Link]
SMART WASTE MANAGEMENT [12] Abba, S., \& Light, C. I. (2020). IoT-
SYSTEM USING ARDUINO. INTER- Based Framework for Smart Waste Monitor-
ANTIONAL JOURNAL OF SCIEN- ing and Control System, A Case Study of
TIFIC RESEARCH IN ENGINEER- the Smart Cities † (Vol. 15). Available.
ING AND MANAGEMENT, 07(03). [Link]/journal/proceedings
[Link] [13] Ziouzios, D., Baras, N., Dasy-
[4] Kulkarni, R., Mujtaba Hussaini, S., genis, M., \& Tsanaktsidis, C. (2021).
Ansari, M. S., Alina, S. N., \& Summaiyya, Envisioning IoT applications in a smart
U. (2024). Smart Trash Bin Using Machine city to underpin an effective munic-
Learning ipal strategy: The smartbin project.
and Internet of Things. [Link] SHS Web of Conferences, 102, 04020.
[5] Gimonkar, R. M. (2021). AUTO- [Link]
MATED WASTE SEGREGATOR. In Jour- [14] Fei, X., He, P., Ma, H., \& Qiu,
nal of Research in Engineering and Applied Y. (2020). Enhancing Urban Waste Man-
Sciences ISSN (Vol. 06). agement: Development and Application of
[6] Rao, P. R., Sanjay Kumar, S., \& Ra- Smart Garbage Bin Technologies.
jendra Prasad, C. (2017). Garbage Monitor- [15] Bano, A., Ud Din, I., \& Al-Huqail,
ing System using Arduino. [Link] A. A. (2020). AIoT-Based Smart Bin for

12
GCTU Smart Bin Biomedical Waste System

Real-Time Monitoring and Management of ICCMC 2017, 2018-January, 487–490.

Solid Waste. Scientific Programming, 2020. [Link]
[Link] [24] Ishaq, A., Mohammad, S. J.,
[16] Mohamed, N. H., Khan, S., Bello, A.-A. D., Wada, S. A., Adebayo,
\& Jagtap, S. (2023). Moderniz- A., \& Jagun, Z. T. (2023). Smart
ing Medical Waste Management: Un- waste bin monitoring using IoT for sustain-
leashing the Power of the Internet of able biomedical waste management. En-
Things (IoT). In Sustainability (Switzer- vironmental Science and Pollution Re-
land) (Vol. 15, Issue 13). Multidisci- search. [Link]
plinary Digital Publishing Institute (MDPI). 30240-1
[Link] [25] P, N. J., Sridhar, S., Professor, A.,
[17] Abuga, D., \& Raghava, N. S. \& Scholar, U. (2023). Automated Segrega-
(2021). Real-time smart garbage bin mech- tion of Bio-Medical Waste (Vol. 8, Issue 5).
anism for solid waste management in smart [Link]
cities. Sustainable Cities and Society, 75.
[Link]
[18] Sarkar, O., Kumar Dey, A., \& Malik,
T. (2023). 69 with Artificial Intelligence. Int
J Biomed Clin Anal, 3(2), 70.
[19] Deepika Kanyal, Lata Kanyal
Butola, \& Ranjit Ambad. (2021).
Biomedical Waste Management in India-
A Review. Indian Journal of Forensic
Medicine \& Toxicology, 15(2), 108–113.
[Link]
[20] Adu, R. O., Gyasi, S. F., Essumang,
D. K., \& Otabil, K. B. (2020). Med-
ical Waste-Sorting and Management Prac-
tices in Five Hospitals in Ghana. Journal
of Environmental and Public Health, 2020.
[Link]
[21] Atieno Othigo, E., \& Mo-
ronge, J. (2020). Hazardous Biomedical
Waste Management in a Level Four Pri-
vate Hospital in Nairobi County, Kenya.
[Link]
[22] Odonkor, S. T., \& Mahami, T.
(2020). Healthcare waste management
in Ghanaian hospitals: Associated public
health and environmental challenges. Waste
Management and Research, 38(8), 831–839.
[Link]
[23] Raundale, P., Gadagi, S., \&
Acharya, C. (2017). IoT based biomed-
ical waste classification, quantification
and management. Proceedings of the
International Conference on Comput-
ing Methodologies and Communication,

13