Graduate Program in Electrical Engineering, Federal University of Espírito Santo (UFES), Fernando Ferrari Avenue, Vitória, 29075-910, Brazil.
Neurophysics Group, "Gleb Wataghin" Institute of Physics, University of Campinas, Campinas, Brazil.
Sci Rep. 2022 Mar 8;12(1):4067. doi: 10.1038/s41598-022-08048-9.
Smart textiles are novel solutions for remote healthcare monitoring which involve non-invasive sensors-integrated clothing. Polymer optical fiber (POF) sensors have attractive features for smart textile technology, and combined with Artificial Intelligence (AI) algorithms increase the potential of intelligent decision-making. This paper presents the development of a fully portable photonic smart garment with 30 multiplexed POF sensors combined with AI algorithms to evaluate the system ability on the activity classification of multiple subjects. Six daily activities are evaluated: standing, sitting, squatting, up-and-down arms, walking and running. A k-nearest neighbors classifier is employed and results from 10 trials of all volunteers presented an accuracy of 94.00 (0.14)%. To achieve an optimal amount of sensors, the principal component analysis is used for one volunteer and results showed an accuracy of 98.14 (0.31)% using 10 sensors, 1.82% lower than using 30 sensors. Cadence and breathing rate were estimated and compared to the data from an inertial measurement unit located on the garment back and the highest error was 2.22%. Shoulder flexion/extension was also evaluated. The proposed approach presented feasibility for activity recognition and movement-related parameters extraction, leading to a system fully optimized, including the number of sensors and wireless communication, for Healthcare 4.0.
智能纺织品是远程医疗监测的新颖解决方案，涉及非侵入式传感器集成服装。聚合物光纤（POF）传感器具有智能纺织技术的吸引力特征，并结合人工智能（AI）算法，增加了智能决策的潜力。本文提出了一种完全便携式光子智能服装的开发，该服装具有 30 个多路复用 POF 传感器，并结合人工智能算法，以评估该系统在多个主体活动分类方面的能力。评估了六种日常活动：站立、坐下、蹲下、上下手臂、行走和跑步。采用 K-最近邻分类器，所有志愿者的 10 次试验结果的准确率为 94.00（0.14）%。为了实现最佳传感器数量，对一名志愿者使用主成分分析，结果表明使用 10 个传感器的准确率为 98.14（0.31）%，比使用 30 个传感器低 1.82%。还估计了节奏和呼吸率，并与位于服装背面的惯性测量单元的数据进行了比较，最高误差为 2.22%。还评估了肩部的屈伸。所提出的方法在活动识别和与运动相关的参数提取方面具有可行性，为医疗保健 4.0 系统的全面优化提供了方向，包括传感器数量和无线通信。
