Intelligent Waste Manager - Deep learning-based Waste Management System

Abstract

"The concerns in the waste management domain are ever-increasing concerns that demand innovative solutions. Adequate waste segregation is a crucial step in the waste management process that can generate various issues if not done correctly. Although various experiments and research are taking place to solve the problem of waste classification, significant gaps still need to be bridged. Accordingly, through this study, the author discussed leveraging state-of-the-art technologies to bridge some of the existing gaps. Throughout the study, the challenges of identifying different types of visible waste that are not neatly separated, as well as the impracticality of existing methods, are discussed. The study uses the Yolov8 model, which utilizes advanced YOLO techniques to accurately detect objects that overlap with each other. It also employs mechanisms such as center distance and data augmentation techniques such as mixup, mosaic, and overlay augmentation. The author's goal is to develop a mobile application that can perform real-time detection of waste in a live camera feed. The application should have the ability to detect objects in complex environments such as overlapping objects. The performance metrics of the Yolov8 model, which has been trained for waste detection, demonstrate its effectiveness in accurately distinguishing between various classes. With a mean Average Precision (mAP) of 0.913 at an Intersection over Union (IoU) threshold of 0.5, the model exhibits a strong capability in classifying and precisely localizing objects within images. Moreover, the model achieves an optimal balance between precision and recall with an F1 score of approximately 0.87 at a confidence threshold of about 0.495, indicating a robust classification performance across different classes."