An Automatic Prototype Segregation Waste Bin Based on Machine Learning

Abstract

A great number of waste to recycle, such as bottles and cans, are mixed in the trash.  The segregation of wastes using a sensor has limitations when wastes are deformed, causing inaccurate results. This research proposed an automatic prototype waste segregation bin based on machine learning to detect the shape of waste using the YOLOv8 algorithm with a training dataset of images, and was used in a convolutional layer neural network architecture. It can extract multi-level features from the input image, which will have the function for object prediction.  To create a bounding box, it can use a confidence score for an object. The classified processing data of waste uses the Raspberry Pi board model 5, and controls the operation of the board drive model L298N. It supplies electricity to the stepper motor, which rotates the waste into the bin. The notification system will send an SMS message to the phone of the attendant when the trash reaches a set point by receiving values from the HC-RSO4 ultrasonic sensor. An automatic prototype waste segregation bin based on machine learning has test cases for four types of segregating waste. The average precision of regular-shaped bottles and cans were 86.7% and 96.7%. For bottles and cans of deformed shape were 70.0% and 86.7%, respectively. The results of regular-shaped bottles and cans were found to be better than deformed bottles and cans. Because shaped bottles and cans have complete annotations in the bounding box, which enables the trash bin to segregate waste efficiently. However, an automatic prototype waste segregation bin based on machine learning has a good segregating potential.