AI Powered Waste Management System

Abstract

Waste mismanagement remains a significant environmental and public health challenge, particularly in rural and semi-urban regions where structured waste segregation systems are limited. The increasing diversity of waste materials further complicates manual segregation, leading to inefficient recycling, environmental pollution, and health risks. To address these challenges, this paper presents an AI-powered automated waste classification system based on MobileNetV2 and transfer learning for fine-grained multi-class waste categorization. The proposed system classifies waste into thirteen categories, including recyclable, organic, hazardous, and non-waste classes. A multi-source dataset is constructed by combining a publicly available garbage classification dataset with non-waste images from the CIFAR-10 dataset to improve robustness under real world conditions. Image preprocessing and data augmentation techniques are employed to enhance generalization across varying environmental conditions. Experimental evaluation demonstrates that the proposed MobileNetV2-based model achieves a test accuracy of 94.62%, along with strong precision, recall, and F1-score across most classes. The lightweight architecture and efficient training strategy make the system suitable for deployment on low-cost and resource constrained devices, enabling scalable and sustainable waste management solutions aligned with national initiatives such as Swachh Bharat Abhiyan and Digital India.