面向可穿戴生理信号的压缩感知实时重构

Real-Time Compressed Sensing Reconstruction for Wearable Physiological Signals

传统的迭代式压缩感知重构算法由于计算复杂度高,数据处理实时性差,难以在实际的可穿戴设备中发挥作用。该文结合深度学习中的一维扩张卷积和残差网络,提出了一种适用于可穿戴健康监护的非迭代式压缩感知实时重构算法。该方法基于大量生理信号数据训练一个用于压缩感知重构的网络模型,该模型可以对生理信号进行快速精确重构。通过在两个公开的生理信号数据集上的实验表明,相比于已有的基于深度学习的生理信号压缩感知重构算法,该方法有着更高的重构精度,并且该方法在文中所使用的计算机上仅需约0.7 ms即可完成对一个2 s长度信号帧的重构,这比传统的迭代式压缩感知重构算法快了大约2~3个数量级,说明该方法有着出色的实时性能。

The traditional iterative compressed sensing reconstruction algorithm is difficult to play a role in actual wearable devices because of its high computational complexity and poor real-time data processing.In this paper,a non-iterative compressed sensing real-time reconstruction algorithm suitable for wearable health monitoring is proposed by combining one-dimensional dilated convolution and residual network in deep learning.The proposed method trains a network model for compressed sensing reconstruction based on a large number of physiological signal data,and the trained neural network model can accurately reconstruct physiological signals at a very fast speed.Experiments on 2 open physiological signal data sets show that the proposed method has higher reconstruction accuracy than the existing reconstruction algorithms based on deep learning.The proposed method can reconstruct a 2 s signal frame in only about 0.7 ms on the computer used in this paper.This is about 2~3 orders of magnitude faster than the traditional iterative compressed sensing reconstruction algorithm.Therefore,the method proposed in this paper has excellent real-time performance.