Development of a Solid Waste Identification and Segregation Hub (SWISH) System with CNN

Improper waste disposal leads to significant environmental challenges, including pollution that adversely affects ecosystems. Effective waste segregation is crucial for sustainability, significantly reducing the amount of waste sent to landfills. This research leverages computer vision technology to develop an automated waste segregator that classifies waste into biodegradable, non-biodegradable, and recyclable categories. Fifteen waste object classes were specified, and model performance was found to be highly dependent on dataset quality and the choice of model. Extensive experimentation using ResNet-50, trained on 8,621 images, yielded optimal results. The final model, enhanced with extended training and data augmentations, achieved a training accuracy of 94.2% and a validation accuracy of 95.9%. Despite challenges with classifying deformable waste items, overall hardware deployment accuracy is 76.67%. This study highlights the potential of machine learning techniques to enhance automated waste segregation and improve waste management efficiency.