Enhancing Waste Categorization Using Ensemble of Transfer Learning and Light-Weight Convolutional Neural Network

Proper waste management is essential to reduce the negative impacts of waste on human health and the environment. Waste management can play a significant role in achieving several Sustainable Development Goals (SDGs), including ‘Sustainable Cities and Communities’, ‘Responsible Consumption and Production’, ‘Climate Action’, and ‘Life Below Water’. Automated waste sorting using machine learning improves waste management in smart cities by increasing efficiency and accuracy, leading to reduced disposal costs and increased recycling rates. However, limited research has been conducted on automatic waste categorization. In this study, a proposed ensembled architecture for waste type recognition using transfer learning techniques and a custom light-weight convolutional neural network (CNN) achieved an overall accuracy of 95.54% on the Flickr Material Database, outperforming previous approaches. The proposed light-weight CNN architecture utilized filters of different sizes to extract deeper and more complex features, while the ensemble technique of features from different models specialized in identifying different aspects of the input data, resulting in improved classification accuracy. The proposed approach exhibited only nine misclassifications, indicating its proficiency in accurately distinguishing between the different waste classes with comparable levels of accuracy, precision, recall, and f1-score.