Utilizing YOLOv8 for Waste Detection on Waste Bank Application

Waste management is a significant environmental challenge in Indonesia, particularly in urban areas where waste segregation awareness remains low. The waste bank program, regulated under the Decree of the State Minister for the Environment of the Republic of Indonesia No. 13 of 2012, aims to encourage community participation in waste management. However, its implementation remains suboptimal due to technological limitations and manual administrative processes. This study proposes the development of a mobile-based waste bank application that integrates the YOLOv8n model for real-time waste classification. The dataset consists of nine waste categories with a total of 4,500 images, annotated and split into training, validation, and testing sets with a ratio of 80:10:10. The YOLOv8n model was trained using a configuration of 70 epochs, a learning rate of 0.001, and the AdamW optimizer, achieving an [email protected] of 0.995 and an [email protected]:0.95 of 0.785. Further testing demonstrated good generalization capabilities across variations in object shape, background, distance, and multi-object scenarios, although performance decreased for shapes and colors rarely present in the dataset and for long-distance detection. The results show that YOLOv8n has strong potential for implementation in mobile-based waste bank systems to improve sorting efficiency and community participation.