结合多尺度学习的WBAN数据处理机制

无线体域网进行生理信号长期非间断观测过程中,为捕捉生理数据动态特征以提升信号压缩率,增强网络能量有效性,综合考虑压缩感知和多尺度在线字典学习,提出了一种能量有效数据处理机制。首先依据各尺度系数能量占比设计观测矩阵,以提高观测数据有效性；其次,节点对多尺度小波系数进行随机投影,以深入削减数据量；最后,结合多尺度鲁棒字典学习,高质量恢复原始数据。数值结果表明,所提机制在重构精度、数据压缩率方面表现尤佳,有助于提升网络能量有效性。当压缩率高达85时重构信号仍满足医疗诊断要求。

Multi-scale learning based data processing mechanism for WBAN

In the process of long-term continuous biomedical monitoring with wireless body area network, an energy efficient data processing mechanism, incorporating both compressed sensing and multi-scale online dictionary learning, is proposed, which aims at capturing the dynamic attributes of physiological data to improve signal sparsity and enhance network energy efficiency. To begin with, sensing matrixes, relying on energy proportion of each scale, are constructed to improve the effectiveness of measured data. Then multi-scale wavelet coefficients are projected randomly on sensor nodes and the amount of data decreases deeply. Finally, data is recovered through multi-scale robust dictionary learning in remote terminal. Numerical results show that the proposed mechanism has better performance, especially in terms of reconstruction fidelity and data compression ratio. Even if compression ratio is up to 85, the reconstruction signal still satisfies medical diagnostic requirements.