WasteNet: A Structured-Pruned and EMA-Augmented RepVGG Framework for Real-Time Trash Classification

Achieving efficient classification of garbage boosts recycling and helps with pollution control as well as sustainable waste management. To address this challenge,we developed RepVGG-MEM, which is a lightweight and highly accurate Deep Learning model for real-time waste classification. Built on RepVGG, the model applies EMA,Mixup data augmented training, and structured pruning to optimization to maintain a strong speed performance ratio. The model and its described metrics, performance of 95.06% accuracy and a 94.5% F1 score, even with closely resembling waste categories surpass several previous versions. Moreover, its accuracy makes it ideal for edge devices and smart mobile waste management systems.RepVGG-MEM shows better performance and generalization than earlier versions. Its lightweight design makes it easy to deploy on edge devices such as smart bins and mobile platforms. This capability positions it as a promising solution for automated waste management in real-world situations.