无线传感器网络中多声源分离与识别研究
文献类型：学位论文
| 作者 | 苗浩 |
| 学位类别 | 博士 |
| 答辩日期 | 2007-07-24 |
| 授予单位 | 中国科学院声学研究所 |
| 授予地点 | 声学研究所 |
| 关键词 | 传感器网络 声源识别 声源分离 小波分析 分类器 |
| 其他题名 | Research on multiple source signal separation and classification in wireless sensor networks |
| 学位专业 | 声学 |
| 中文摘要 | 随着微机电系统（MEMS）技术、微芯片技术、无线通信技术和数字电子技术的发展，出现了一种由大量的具有一定计算能力和通信能力的无线接入传感器节点所组成的传感器网络，可以协同感知与处理特定区域内发生的事件。传感器网络在工业监控、环境监测等许多领域具有广阔的重要应用和发展价值。 特定环境中的声源识别作为传感器网络的一个重要组成部分，是传感器网络在现实世界的一个直接应用，而多源识别又是研究重点。本文以分布式传感器网络中的地面车辆声源识别为背景展开研究，重点研究了多传感器多声源的分离、地面声源的特征提取和分类问题，论文主要工作和创新点如下： （1）提出了基于熵最大化原理的改进多声源EME（Enhanced Maximum Entropy）盲分离算法。针对地面车辆声信号概率密度函数的分布，选取参数可调节的广义指数函数作为非线性函数的倒数，更好地与源信号相匹配，仿真表明与原来的熵最大化算法相比，基于广义指数函数的改进EME算法使地面车辆信号分离的性能得到明显的提高。 （2）研究了真实环境下地面车辆卷积混合的多声源分离问题。通过把基于信息论的熵最大化算法推广到频率域，使得时域的卷积混合问题转变为频率域的瞬时混合问题，进而可以在每个频率段分别进行独立分量分析，分离效果有明显改进，算法收敛性也得到提高。为了克服在频域中实现盲分离所固有的位序不确定性和比例缩放问题而严重影响分离性能，采用聚类的方法对每个频率段的分离结果进行排序，取得了很好的效果。 （3）针对分布式传感器网络的声源识别问题，将小波包分析等时频分析技术应用于车辆声信号特征的提取及分析中。利用小波包对频域能够更加精细划分的特性，从中提取出声信号的特征向量。实验结果表明:这些特征提取技术从频率和能量的角度考虑问题，能够很好地体现不同地面声源之间的差异，提取的特征量较为稳健，分类结果准确率较高，能很好的完成传感器网络中地面车辆的分类。 |
| 英文摘要 | Together with the technological development of Micro-electrical-mechanical- systems (MEMS), very-large-scale-integration (VLSI) and wireless communication, a new category of wireless sensor networks with multifunctional sensor nodes that integrate sensing, processing, and communication capabilities, emerged to sense the environment independently. The sensor networks can collaboratively achieve complex information gathering and dissemination, and can therefore be applied to industrial and environmental monitoring with a promising technological development. As an important branch of sensor networks, sound source recognition in a given field is one of its direct applications in the real world. Multi-sensor multi-source blind signal separation of ground vehicles, feature extraction and source classification are investigated under this background. The main works and contributions of this paper are as follows: (1) An improved multi-source blind signal separation algorithm is proposed based on an Enhanced Maximum Entropy（EME）.Subject to the probability density function of the ground vehicles, a generalized parameter adjustable exponential function is used to match the non-linearity characteristics of the source signal. Simulation demonstrates that comparing with the original ME algorithm, the performance of the new EME algorithm for ground vehicle separation could be improved significantly. (2) A multi-source blind separation of convolved mixture in practical application was investigated. With the extension of the ME algorithm to the frequency domain, the convolutive problem can be converted into an instantaneous one, where independent component analysis (ICA) can be performed separately in each sub-band of frequency domain. Both the separation performance and the algorithm convergence can be improved. To overcome the uncertainties led by permutation and scaling in frequency domain blind source separation, clustering method is introduced to sort the results in each sub-band. (3) Several time-frequency analysis techniques, such as short time-frequency spectrum and wavelet analysis, were applied to extract the features of the vehicles in a distributed sensor network. Experiments showed that the wavelet characteristics could well reveal the differences among vehicle sources in both frequency and energy distributions, these methods could improve the performance of source recognition and classification for vehicles both in robustness and accuracy. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 101 |
| 源URL | [http://159.226.59.140/handle/311008/22]  |
| 专题 | 声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 苗浩. 无线传感器网络中多声源分离与识别研究[D]. 声学研究所. 中国科学院声学研究所. 2007. |
入库方式： OAI收割
来源：声学研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。