Enhanced E-Waste Classification Through Masked Generative Transformers

The rapid growth of electronic devices has resulted in a substantial rise in electronic waste (e-waste). Proper classification of e-waste components plays a vital role in sustainable recycling efforts. However, conventional deep learning approaches often struggle to accurately identify complex components, especially when processing images that contain noise or visual obstructions. This paper proposes a novel approach, Masked Generative Transformers (MGTs), for e-waste classification. MGTs combine generative and discriminative learning paradigms, enabling them to reconstruct missing parts of an image and gain a deeper understanding of visual structures. Unlike conventional vision transformers, MGTs can effectively distinguish between similar components and are robust to noise and occlusion. Our experimental findings highlight the advantages of MGTs over traditional approaches in e-waste classification. This study supports the advancement of more effective recycling techniques and fosters sustainable waste management efforts.