IoT-Enabled Smart Waste Management System Using Computer Vision for Automated Waste Segregation

Abstract

Efficient waste management is a critical requirement for sustainable urban development. This paper presents an IoT-enabled smart waste management framework that integrates computer vision techniques for automated waste segregation. Waste images are classified into biodegradable, recyclable, and hazardous categories using a Convolutional Neural Network (CNN). The working dataset contained approximately 30,000 labeled images organized into the three target classes; TrashNet was a confirmed public source, supplemented by publicly accessible waste images. Image preprocessing includes resizing, normalization, and augmentation, with augmentation applied to the training data. The dataset is divided into 75% training, 15% validation, and 10% testing sets. The CNN correctly classified 2,890 of 3,000 test images, giving an overall accuracy of 96.33%. Macro precision, recall, and F1-score are approximately 96.33%. The IoT component is presented as a proposed integration architecture for linking classification with bin fill-level monitoring and collection alerts. An illustrative threshold-based scheduling scenario reduces weekly trips from 20 to 13 (35%); this is a simulated/illustrative result rather than a field-trial measurement.