Thin and flexible pressure sensors for classifying body signals enhanced by tiny machine learning algorithms

Abstract

Flexible pressure sensors are a trend that has attracted many scientists in recent years. Among them, pressure sensors made from fabrics or fabric components are widely used due to their accessibility and diversity in structure. However, current studies face many challenges in fabrication and signal processing to obtain helpful information and build a complete wearable system for e-healthcare. To solve the above problems, we present a system consisting of soft pressure sensors from conductive fabrics and tiny machine-learning (tiny ML) algorithms for body signal monitoring applications. The fabricated sensor achieves a small thickness of 0.75 mm, a good sensitivity of 0.16 kPa⁻¹, and a fast response/recovery time of 70 ms. Besides, the flexible sensor’s capabilities are optimized thanks to intelligent algorithms that help remove noises and track body signals in real-time. In the pharyngeal movement classification scenario, the accuracy of the machine learning models is multilayer perceptron (98.43 %), k-nearest neighbors (98.43 %), and decision tree (95.28 %), respectively. When the actions are increased, the accuracy of the multilayer perceptron model still reaches 96.78 %, and the classification time is speedy at only 24 ms. The support of tiny ML algorithms has helped improve the accuracy and feasibility of the system. The ML models are built and described in detail with a small size (<1 Mb), ensuring they can be deployed on small embedded boards with limited resources. The most important contributions of the paper include two aspects: the soft pressure sensor with good performance/easy manufacturing process and the details of intelligent embedded machine learning models that improve the sensor performance in practical tasks. The research represents a new trend in flexible healthcare technology development when data is processed at the network’s edge − directly on endpoint devices. We expect the work to become a highlight reference for more complete studies close to industrial production or commercial products.