RECYCLABLE WASTE SORTING MACHINE WITH IMAGE PROCESSING

Abstract

This research aimed to establish and test the performance of a recyclable waste sorting machine with image processing to provide control over the sorting of 4 types of recyclable waste: cans, glass bottles, carton boxes, and plastic bottles, and to give notification via LINE application when the bin was almost full. The system consisted of an IMILAB Webcam camera, receiving real-time images of objects, connected to the Raspberry Pi4 Model B to process waste classification with YOLOv5's model, which was trained with data sets of 1,321 images for 250 epochs. It had a mAP50% value of 0.995, recall 0.995, precision 0.999, and F-measure 0.997. This machine was connected to Module L298N to control the rotation position of the step motor to the output bin and control the servo motor to pour waste into each type of bin. The performance test results of the machines showed that the correct sorting of waste by type was with 98.33% accuracy, and the confidence interval results in the range of 0.50 - 0.97. Furthermore, the results of the test the performance of the hardware in separating waste to each type of output bin with step motor and servo motor and the notification system when the bin was nearly full via the LINE application could work 100% correctly.