A Framework Combining YOLOv2 and Motion-Adaptive Inference with Multiple Data Splits for Waste Management in Smart Sustainable City

Sustainability is a key goal of the United Nations Sustainable Development Goals. Smart sustainable cities are futuristic urban centers expected to dominate the world in the future. Effective waste management is one of the critical indicators of a smart sustainable city. Previous studies have developed waste management models using AI algorithms, particularly convolutional neural networks. However, these studies often struggled with balancing detection speed and accuracy. This article proposes a framework that combines an optimized YOLOv2 model with motion-adaptive inference to achieve a balance between speed and accuracy in detecting organic and recycle materials. The proposed framework has been applied to detect organic and recycle materials alongside baseline algorithms, and it demonstrates improved performance in balancing speed and accuracy compared to the baselines. Making it suitable for adoption in smart sustainable cities. The proposed framework can be integrated into real-time systems to enhance waste disposal management in smart sustainable cities.