Automated Waste Classification with Tensorflow and Mobilenet: A Step Towards Sustainable Development Goal 12

Abstract

Waste mismanagement is a critical environmental concern, with manual sorting often inefficient
and prone to error. This paper presents an automated waste classification system using
Convolutional Neural Networks (CNNs) and transfer learning to automatically categorize waste
into classes such as plastic, glass, organic, metal, and paper. The model was trained on a combined
dataset sourced from the TrashNet repository and locally collected images, with data augmentation
applied to improve robustness under varying conditions. Experimental results demonstrate high
classification accuracy, outperforming traditional machine learning baselines. The system was
further optimized for deployment on resource-constrained devices, enabling real-time operation
on platforms such as Raspberry Pi. By improving waste segregation at the source, this approach
supports Sustainable Development Goal 12 (Responsible Consumption and Production) and
promotes circular economy practices. Future work will focus on expanding classification
categories and integrating IoT-enabled smart bin technology.