Abstract:
Waste classification before collection is essential for the better management of household
waste by municipalities. However, very little attention is paid to domestic waste
classification at household level in Sri Lanka. To overcome the inefficient manual waste
classification method, we have proposed a smart waste classifier model that can categorize
dry, wet, and metal garbage in order to solve the existing waste classification issue.
This system works as when a person puts waste into a dustbin, three sensors detect
it: an infrared sensor to detect the waste, an inductive sensor to detect metallic nature
and categorize it as metal waste, and a soil moisture sensor to check the moisture level
in the waste thrown inside and classify it as wet waste if the humidity level exceeds
a threshold value. To make a better decision on waste classification, along with the
sensors, Convolutional Neural Network (CNN) is used. An ESP32 camera is fixed in
the trash can which will capture the image of the waste and let CNN to classify the
waste. After the CNN identifies the waste, it is thrown into the proper bin using a servo
motor that spins at 180 degrees, and a step motor that revolves at 360 degrees with the
three bins attached and rotates in response to the waste detected. The Arduino UNO
is used to connect all of the sensors and motors. By using this model, we were able to
classify the waste as dry, wet, and metal. The CNN yielded an accuracy of 99% in waste
identification under different sets of images taken at different angles. The model allows
people to discard the waste into a dustbin, regardless of its nature. The model itself
then classifies and transfers the waste into the appropriate dustbin. Moreover, the three
dustbins are monitored by a sensor to check whether the dustbin is filled or not.
