A Self-Adaptive Waste Management System Employing Convolutional Neural Network for Real-Time Categorization

The exponential and dramatic rise in urbanization and industrialization has made waste management a critical global issue. As the global population reaches unprecedented levels, so has the production of waste reached degrees where it has become uncontrollable. In this study, we present a self-adaptive waste management system leveraging the power of Convolutional Neural Networks (CNNs) for real-time categorization of waste. The proposed system employs a robust CNN model trained on a diverse dataset of waste images, enabling it to accurately identify and segregate various types of waste, including plastics, metals, organic materials, and paper. By utilizing real-time image processing and adaptive learning capabilities, the system continually improves its classification accuracy, adapting to new waste types and environmental conditions. The system was tested using Yolov8 module configured with a specially tailored dataset designed to exclusively detect and classify metallic, plastic, and paper waste. The custom-trained model demonstrated high accuracy in identifying metallic and paper waste, achieving 220 and 140 true positives during the validation phase. However, the model struggled to reliably detect plastics, registering only 56 true positives. This suggests that object detection algorithms may encounter challenges when identifying objects with clear and transparent properties, as seen with plastics and polymers. This difficulty is further evidenced by the false positive score of 60, indicating that the model frequently misidentified plastics as background. Overall, the model implemented in the proposed system achieved precision scores of $90.7 \%, 69.1 \%$ and $82.8 \%$ for metals, plastics, and paper, respectively.