UC Irvine

In this PhD thesis, we explore the development and enhancement of wearable technology for real-time health monitoring. Wearable devices have become transformative tools in healthcare, enabling continuous monitoring, diagnosis, and management of health conditions. This thesis leverages advanced machine learning (ML) techniques, artificial intelligence (AI), and innovative hardware solutions to address challenges in the accuracy and power efficiency of health monitoring systems.

We introduce a novel ML-based model for accurate tracking of in-mouth nutrient sensors. These sensors face challenges due to movement, which affects measurement accuracy. By employing a wide range of frequencies and advanced ML algorithms, our approach effectively predicts sensor positions, thereby enhancing the reliability of nutrient tracking.

To address motion artifacts in photoplethysmography (PPG) signals, which are critical for heart rate monitoring in wearable devices, we propose an innovative use of Cycle Generative Adversarial Networks (CycleGAN). This technique transforms noisy PPG signals into clean, artifact-free data without relying on accelerometers, thus conserving power and significantly improving signal quality.

Our research extends to the development of smart clothing systems embedded with passive sensors and copper coils. These garments utilize Near Field Communication (NFC) and energy harvesting technologies for wireless connectivity and power. This design ensures flexibility, durability, and functionality in various environments, including underwater, overcoming limitations of current wireless communication technologies.

Additionally, we present AI-Coach, a low-maintenance, full-body, personalized wearable system for real-time workout monitoring. Integrating multiple sensors and employing a BERT model, AI-Coach captures detailed movement data and provides personalized activity monitoring and coaching. This system adapts to individual users' needs, enhancing the accuracy and effectiveness of activity tracking.

Overall, this PhD thesis demonstrates a comprehensive approach to advancing wearable health monitoring technology. By integrating ML, AI, and innovative hardware solutions, we address key challenges in accuracy and power efficiency. The proposed methods and systems have significant implications for improving individual health outcomes and public health monitoring, paving the way for more reliable and efficient wearable health devices.