The exponential growth of electronic waste (e-waste) has created urgent environmental and health challenges, demanding advanced solutions for efficient sorting and recycling. This study presents a novel hybrid deep learning framework that integrates EfficientNet, MobileNet, and a Sequential Neural Network (SNN) to automate e-waste classification with high accuracy and speed. The model was trained and evaluated on a diverse dataset of 3,859 images spanning 12 e-waste categories, including batteries, printed circuit boards, and household electronics. Experimental results demonstrate exceptional performance, achieving 97.8% accuracy, 98.1% precision, 97.8% recall, and a 97.8% F1 score, surpassing traditional methods and single-model approaches. The system’s lightweight design (48 MB) enables real-time processing (0.12 seconds per image) on standard CPUs, ensuring scalability for industrial applications. By automating the sorting process, the framework reduces human exposure to hazardous materials, enhances material recovery efficiency, and supports sustainable waste management practices. Its modular architecture allows seamless integration into existing recycling workflows, making it a practical solution for facilities with limited resources. The study underscores the model’s potential to advance circular economy initiatives by improving resource reuse and minimizing environmental contamination. Future research will focus on real-time IoT deployment, federated learning for decentralized training, and expanding classification capabilities to include rare and unconventional e-waste items. This work contributes a scalable, cost-effective, and environmentally responsible solution to the global e-waste crisis.
Oise et al. (Wed,) studied this question.