基于多尺度特征融合变换神经网络的心电信号重构算法研究_毕业论文.pdf
版权申诉
50 浏览量
2023-10-27
13:09:13
上传
评论
收藏 5.21MB PDF 举报
III
Research on ECG signals reconstruction algorithm based on a
Multi-scale Feature Fusion Transform Neural Network
Abstract
An ageing population seriously challenges the sustainability of society and health care
systems. Heart rate monitoring is an effective way to prevent cardiovascular disease, which
is the main cause of death in the elderly. Ballistocardiogram (BCG) signals acquisition
device based on piezoelectric film sensor can continuously record human physiological
information without disturbance or trauma. Since the collected BCG signals are highly
correlated with electrocardiogram (ECG) signals in reflecting heart activity, BCG signals are
often used for heart rate monitoring in home scenarios. However, compared with ECG, the
BCG wave shape is less clear to distinguish between heartbeats, and the contact-free
measurement method is easy to lead to distortion of waveform characteristics, which
increases the difficulty of heart rate detection and disease identification. Moreover, BCG
signals have not yet formed the standard of medical diagnosis. Therefore, the conversion of
BCG signals into ECG signals plays an important role in improving the accuracy of heart
rate detection at home, and provides a reliable technical solution for further promoting the
clinical application of BCG signals.
An Attention Mechanism Based on Encoder-Decoder Network(AMEDN) is used to
study the reconstruction of ECG signals based on BCG signals. AMEDN uses the attention
module to weight the position of each wavelet of the ECG signals, instructing the encoder-
decoder network to learn the hidden characteristics of the ECG signals from the BCG signals,
and thus reconstructs the ECG signals with the same dimension as the ECG signals by the
decoder. AMEDN is applied to subjects with different heights, genders, ages and weights.
The experimental results show that AMEDN can reconstruct the subjects' ECG signals from
the BCG signals.
In order to solve the problem of distortion and deviation of small waveform features of
reconstructed ECG signals, A Multi-scale Feature Fusion Transform Neural Network
(MFFTNN) is proposed based on the idea of sequence transform and multi-scale feature on
the basis of AMEDN. By setting different sizes of filters in each branch in the multi-scale
feature module, MFFTNN extracts multi-scale features of the BCG signals, realizing the
complementary information of different scales of the BCG signals through feature fusion.
Therefore, the detailed features of the ECG signals are enhanced, and the small waveform
剩余51页未读，继续阅读
资源评论
2201_75761617
- 粉丝: 19
- 资源: 7339
