中文摘要：

目前，体震信号常用于检测人体心率和呼吸率参数，适合在日常实时监测中使用。然而，要将其最终应用于医学临床实践，首先必须实现其诊断功能。总结了一系列逆于临床的体震信号诊断参数，并针对其特征提出一种盲分帧算法，结合Kohonen自纽织神经网络的聚类功能，无监督地对独立心动周期中的特征点进行分类，最终提取体震信号的诊断特征。使用同步检测的心电信号和心率不齐病例，对所提算法进行定量和定性对比实验，仿真结果准确率较高，证明了该方法的可行性。

英文摘要：

Ballistocardiogram is usually used to detect heart rate and respiratory rate at present, which is more suitable to be applied to real-time monitoring system. However, if it is wanted to diagnose diseases with ballistocardiogram signal clinically, the diagnosis characteristics must be extracted. A series of parameters were summarized to determine heart function. Based on the diagnosis characteristics, a blind segmentation method was proposed, which combined the Kohonen network to cluster without supervision. Finally, the diagnosis characteristic of ballistoeardiogram signal was extracted. Synchronization ECG signal and arrhythmia subject were recorded to compare quantitatively and qualitatively. The results show the high accuracy of segmentation and prove the superiority and feasibility of the method proposed.