فهرست نشریات علمی دسترسی آزاد ایران

In response to the growing demand for efficient waste management, which is part of the Sustainable Development Goals (SDGs), automated trash classification is a pressing issue. Artificial Intelligence (AI) has found many applications in smart cities. One particular AI method, Federated Learning (FL), has been gaining popularity recently. Also, advances in technology are making high-precision sensors, such as cameras, increasingly affordable, enabling their widespread integration into smart bins and waste management systems. In this paper, we first argue that FL is well-suited to intelligent trash classification. Then, we recognize one of the main obstacles to FL in this context: heterogeneous data. To address this, we introduce T-FLash: A Transformer-based FL framework for trash classification. In this framework, the FL setup relies on Vision Transformer (ViT)-based architectures, enabling decentralized training on edge devices. Evaluating this model under different data distributions suggests that this framework is immune to data heterogeneity and, therefore, a suitable system for smart cities to classify trash images more efficiently.