• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文学位 >Low complexity feature extraction for classification of harmonic signals.
【24h】

Low complexity feature extraction for classification of harmonic signals.

机译：低复杂度特征提取，用于谐波信号分类。

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

In this dissertation, feature extraction algorithms have been developed for extraction of characteristic features from harmonic signals. The common theme for all developed algorithms is the simplicity in generating a significant set of features directly from the time domain harmonic signal. The features are a time domain representation of the composite, yet sparse, harmonic signature in the spectral domain. The algorithms are adequate for low-power unattended sensors which perform sensing, feature extraction, and classification in a standalone scenario. The first algorithm generates the characteristic features using only the duration between successive zero-crossing intervals. The second algorithm estimates the harmonics' amplitudes of the harmonic structure employing a simplified least squares method without the need to estimate the true harmonic parameters of the source signal. The third algorithm, resulting from a collaborative effort with Daniel White at the DSP Lab, University of Nebraska-Lincoln, presents an analog front end approach that utilizes a multichannel analog projection and integration to extract the sparse spectral features from the analog time domain signal. Classification is performed using a multilayer feedforward neural network. Evaluation of the proposed feature extraction algorithms for classification through the processing of several acoustic and vibration data sets (including military vehicles and rotating electric machines) with comparison to spectral features shows that, for harmonic signals, time domain features are simpler to extract and provide equivalent or improved reliability over the spectral features in both the detection probabilities and false alarm rate.
机译：本文研究了特征提取算法，用于从谐波信号中提取特征。所有已开发算法的共同主题是直接从时域谐波信号生成大量特征的简便性。这些特征是频谱域中复合但稀疏的谐波特征的时域表示。该算法适用于在独立场景中执行感应，特征提取和分类的低功耗无人值守传感器。第一种算法仅使用连续的零交叉间隔之间的持续时间来生成特征特征。第二种算法使用简化的最小二乘法估算谐波结构的谐波幅度，而无需估算源信号的真实谐波参数。第三种算法是与内布拉斯加州大学林肯分校的DSP实验室的丹尼尔·怀特（Daniel White）合作开发的，提出了一种模拟前端方法，该方法利用多通道模拟投影和积分从模拟时域信号中提取稀疏频谱特征。使用多层前馈神经网络进行分类。通过处理多个声学和振动数据集（包括军用车辆和旋转电机）并与频谱特征进行比较，对提议的特征提取算法进行了评估，结果表明，对于谐波信号，时域特征更易于提取并提供等效功能或在检测概率和误报率方面都优于频谱特征。

著录项

  • 作者

    William, Peter E.;

  • 作者单位

    The University of Nebraska - Lincoln.;

  • 授予单位 The University of Nebraska - Lincoln.;
  • 学科 Engineering Electronics and Electrical.;Physics Acoustics.
  • 学位 Ph.D.
  • 年度 2011
  • 页码 225 p.
  • 总页数 225
  • 原文格式 PDF
  • 正文语种 eng
  • 中图分类
  • 关键词

相似文献

  • 外文文献
  • 中文文献
  • 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