EDA Denoising Framework for Wearable IoT Under Harsh Conditions
A research study introduces a strong EDA denoising framework tailored for wearable Internet of Medical Things (IoMT) devices, tackling issues of motion artifacts and environmental noise in extreme settings, such as underwater. This framework integrates a hybrid CNN-Transformer teacher model with a compact depth-wise separable CNN student model through knowledge distillation. To mimic various distortions, a realistic data augmentation strategy is employed. The student model achieves a size reduction from 7.87 MB to 0.51 MB and cuts computational costs from 105.1M to 11.61M FLOPs, all while preserving its denoising effectiveness.
Key facts
- EDA signals are vulnerable to motion artifacts and environmental noise.
- Framework integrates CNN-Transformer teacher and lightweight CNN student via knowledge distillation.
- Data augmentation simulates motion artifacts and environmental distortions.
- Student model size reduced from 7.87 MB to 0.51 MB.
- Computational cost reduced from 105.1M to 11.61M FLOPs.
- Denoising performance is maintained after compression.
- Targets wearable IoMT for continuous health monitoring.
- Generalizes across multiple measurement locations and harsh environments.
Entities
—