An Enhanced Autoencoder for Lossy Compression of Auxiliary Physiological Data

Auxiliary physiological indicator data, though less critical than key physiological indicator data for daily health monitoring, are crucial in specific research and clinical contexts. This paper proposes a lossy compression method using neural network autoencoders to enhance the energy efficiency of sensors collecting such data. Utilizing a fully connected neural network autoencoder, the method achieves an ultra-high compression ratio of 256:1. Simulations show excellent reconstruction of skin temperature data but poor performance with EDA data due to its higher precision and complexity. To address this, a new neural network module (EENM) replaces the middle layers of the original autoencoder, enhancing its expressive power. Simulation results indicate that the enhanced model significantly reduces reconstruction errors for both skin temperature and EDA data while maintaining the same compression ratio. This method demonstrates potential for energy-saving applications in auxiliary medical sensors and provides a valuable reference for future research.