Development of Covid Medical Waste Object Classification System Using YOLOv5 on Raspberry Pi

The increase of COVID-19 cases had affected the increase of medical waste, reaching 12,740 tons in Jakarta. Medical and hazardous waste are often found in the community environment, due to poor waste management and improper waste segregation before disposal. Based on these problems, studies have been carried out regarding waste classification. However, there were some limitations in the proposed method. The use of Convolutional Neural Network (CNN) method with model compression has a significant drop in accuracy. The application of Single Shot MultiBox Detector (SSD) method with TensorFlow framework also has low level of accuracy. Results of other studies using YOLOv4 method showed that the accuracy of this model was lower when running on Raspberry Pi. In the context of classifying objects in general, You Only Look Once (YOLO) is interesting and potential. Therefore, this study proposed the development of areal-timee hazardous waste classification system using YOLOv5 method. Based on the test results on the Raspberry Pi 3b+, YOLO produced 85% and 96% accuracy at image size of 128x128 and 25x320 pixels within 1-5 seconds of computational time. With the development of a new framework for sorting waste, it is expected to be an alternative method that can be applied by other researchers in waste classification.