Automated Biomedical Waste Segregation with Residual Neural Networks

The rapid growth in Biomedical Waste (BMW) generation poses significant challenges to effective BMW management, impacting the health and safety of healthcare and sanitation workers, patients, and the public.Segregating BMW at the source is critical to the BMW management system.However, in developing countries like India, less than half of the BMW is segregated at the source due to logistical challenges associated with manual segregation.This study addresses these challenges by developing a deep learning model for automated BMW segregation based on the Indian Government's Biomedical Waste Management Rules, 2016.To achieve this, the BMW-Seg-India dataset was created with images collected from a multi-specialty hospital in India and other publicly available datasets and organized into four classes: yellow, red, white, and blue, in line with the guidelines.A novel deep-learning model, BMWNet52, employing a modified ResNet50 architecture, was trained with transfer learning on the dataset.The model achieved an accuracy of 97.99% in classifying the four categories, demonstrating that this deeplearning classifier is an effective solution for automated BMW segregation.The findings could lead to a fully autonomous BMW management system, integrating automated segregation and disposal, minimizing exposure to hazardous waste, and enhancing efficiency worldwide.