• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文学位 >Prefiltering for improved unknown and known source correlation detection of broadband oscillatory transients and predicting the onset of paroxysmal atrial fibrillation using feature extraction and a hamming neural network.
【24h】

Prefiltering for improved unknown and known source correlation detection of broadband oscillatory transients and predicting the onset of paroxysmal atrial fibrillation using feature extraction and a hamming neural network.

机译：进行预过滤，以改进宽带振荡瞬态的未知和已知源相关性检测，并使用特征提取和汉明神经网络预测阵发性房颤的发作。

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

The main focus of this study is to improve detection of deterministic broadband oscillatory transients in Gaussian noise using ordinary and higher order correlation detectors. Functional inputs to crosscorrelation, bicorrelation, and tricorrelation detectors are investigated assuming known, partially known, and unknown sources. Previous studies done by Pflug et al. (1999) showed that the correlation detection with functional inputs of broadband transients could not be improved as much as for narrowband transients or sinusoids. In this dissertation, a number of different functional inputs are studied, employing Monte Carlo simulations and generating signal-to-noise (SNR) detection curves. The most successful approach is bandpass filtering of the received signal input. Two suites of systematically generated broadband transients with varying center frequencies and signal widths are used to test this idea in detail.; For the Gaussian modulated chirp and the partially known source assumption, the tricorrelation detector performed best in all cases. Improvement for the partially known source over the unknown source case is up to 12.5 dB for the best filter width. For the known source case, the optimal filter for the tricorrelation detector actually improves matched filter performance for the 50 Hz wide signal. Optimal filter width does not depend on center frequency for either source case.; For the cosine tapered chirp and the partially known source assumption, improvements are up to 9.8 dB over the unknown source case. Here, the filtered autocorrelation input and tricorrelation detector perform best. However, the tricorrelation detector does not always outperform the crosscorrelation detector for the filtered signal input. In the case of the known source, the matched filter cannot be improved.; The final chapter of this dissertation presents a different type of detection problem. The challenge is to develop an automated method that will detect the onset of paroxysmal atrial fibrillations (PAF) using Electrocardiograms (ECG) recorded immediately before an episode. Subjects that do not have PAF may be healthy or sick. The use of feature extraction combined with a Hamming neural network is proposed as a possible solution. Results are very encouraging with an average of 88% correct predictions among training and testing data.
机译：这项研究的主要重点是使用普通和高阶相关检测器来改善对高斯噪声中确定性宽带振荡瞬变的检测。在假设已知，部分已知和未知来源的情况下，研究了互相关，双相关和三相关检测器的功能输入。以前的研究由Pflug等完成。 （1999年）表明，与宽带瞬变的功能输入相关检测不能像窄带瞬变或正弦波那样得到改善。本文利用蒙特卡洛模拟并产生信噪比（SNR）检测曲线，研究了许多不同的功能输入。最成功的方法是对接收信号输入进行带通滤波。使用两套系统生成的，具有不同中心频率和信号宽度的宽带瞬态信号来详细测试这一想法。对于高斯调制chi和部分已知的源假设，三相关检测器在所有情况下均表现最佳。对于部分已知的源，在未知源的情况下，其最佳滤波器宽度最多可提高12.5 dB。对于已知的源情况，用于三相关检测器的最佳滤波器实际上可提高50 Hz宽信号的匹配滤波器性能。两种情况下的最佳滤波器宽度都不取决于中心频率。对于余弦锥形chi和部分已知的源假设，与未知源情况相比，改进幅度高达9.8 dB。在此，滤波后的自相关输入和三相关检测器性能最佳。但是，对于滤波后的信号输入，三相关检测器并不总是优于互相关检测器。在已知源的情况下，不能改进匹配的滤波器。本文的最后一章提出了另一种类型的检测问题。挑战是开发一种自动方法，该方法将使用发作前立即记录的心电图（ECG）检测阵发性房颤（PAF）的发作。没有PAF的受试者可能健康或患病。建议将特征提取与汉明神经网络结合使用作为可能的解决方案。结果令人鼓舞，在训练和测试数据中平均有88％的正确预测。

著录项

相似文献

  • 外文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号