热门
声学目标检测与分类研究
文献类型：学位论文
| 作者 | 鲍明 |
| 学位类别 | 博士 |
| 答辩日期 | 2008-05-31 |
| 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 | 目标检测 加权熵最大分带分析 欧氏距离分布熵 类别可分性 |
| 其他题名 | Research on the Detection and Classification of Acoustic Target |
| 学位专业 | 信号与信息处理 |
| 中文摘要 | 声是信息交互的重要载体，声学目标检测及分类研究是具有重要现实意义研究领域。它的研究基础涉及声学理论、统计信号处理、检测判决理论、统计学习、人工智能、神经计算及现代优化理论等多学科领域。随着社会信息化进程的不断深入，无线传感技术的逐渐普及，声学目标检测与分类研究愈来愈受到研究人员的广泛关注。 论文的研究工作以声学目标的检测与分类为目的，通过灵活运用‘信息熵’的分析标准，对声学目标检测分类技术中的信号分析技术，目标特征提取，声学信号分类理论等问题进行较为深入的探讨。 论文的主要工作和创新点如下： ● 针对声源目标分类分析中训练样本集分类特征优化选择问题, 改进了基于Kullback-Leiber距离的样本可分度准则，并得到有效验证。在此基础上定义了空间分布信息度量参数，即欧氏距离分布熵（Distribution Entropy of Euclidian Distance DEED），提出了“类间互欧氏距离分布熵”（Between-Class DEED）与“类内自欧氏距离分布熵”（Within-Class DEED）的分析方法。进一步将其用于样本可分性分析，验证了两者比值愈大，特征样本集可分度愈好这一结论。同时给出了它的计算方法。 ● 针对声源目标信号分带优化问题，提出功率谱加权熵最大分带分析方法。该方法在限定分带数目的条件下，以加权熵最大为优化目标，获得信号在频域的信息量最大的分带边界。以此为基础，建立了功率谱加权熵最大分析模型，同时给出其实现算法。进而，依据功率谱加权熵最大的原则，提出功率谱加权熵最大分带倒谱系数分类特征，设计了以线性分类距离为优化标准的权系数学习算法，在声源目标识别的应用中取得了较好的效果。 ● 利用多种信号特征量分析及提取技术，对声源目标进行了分析，如Fourier谱估计，自回归参数谱估计、小波分析、高阶分析等，在此基础上提出了高阶谱熵等多种目标特征检测参数，获得了良好的声源目标远距离检测效果。 论文在对上述主要内容进行论证分析过程中，始终以典型W、T类声源目标作为分析对象，方法上注意归纳与演绎并重，力图实现理论与实践紧密结合，使论文研究在对实际工作具有指导意义的同时，能给其它类似或相关研究带来一些借鉴。 |
| 英文摘要 | Acoustic target detection and classification is an important research branch in the field of distributed sensor networks in recent years. In order to explore for more effective applied algorithms, variety of engineering and scientific disciplines have joined in, such as acoustics, statistical signal analysis, signal detection, statistical learning, artificial intelligence, neural network and modern optimization theory. In this dissertation, several issues are addressed about the theory, algorithms and applications of the detection and classification of the acoustic targets. The dissertation proposes several parameters, signal analysis methods and formulated criterion for class separability under the frame work of information criterion ‘Entropy’. The major contributions of the dissertation work are highlighted hereinafter. Firstly, an improved Kullback-Leiber distance is presented as a separability criterion for optimizing feature extraction and selection in pattern classification. A nonlinear parameter, Distributive Entropy of Euclidian Distance (DEED), is introduced, and based on it, the ratio of between-class DEED to within-class DEED (Jrd) is defined as a feature extraction and selection criterion of pattern classification. DEED is a nonlinear measure which gives the sample density information in learning space. The formulation proof and Monte Carlo experiments at Gaussian assuming demonstrate that the larger Jrd is, the better separability of learning sample set would be. Secondly, as a new signal analysis method, the power spectral sub-band analysis with the criterion of maximum weighting entropy is derived. With this method, the maximum information can be obtained by optimizing band allocation in the frequency domain. Based on this method, a novel algorithm of feature extraction for classification, MECC (Maximum weighting Entropy Cepstrum Coefficients), is proposed and applied to the classification system of acoustical targets. Thirdly, the author has presented sufficient analysis for acoustical target signal by methods such as Fourier analysis, AR model, Cepstrum, wavelet and higher order statistics analysis. To realize the distant target detection, a new parameter ‘Bispectrum Entropy’ has been designed as a statistic for acoustical signal detection. It has achieved the detection range of more than 1000m for acoustic target. In this work, the sound source of ‘W’ and ‘T’ were used as the sample of acoustic target for detection and classification analysis both in theory and application studying. This work has not only summarized some of the well-known methods using in this filed, but also provided some novel insights and algorithms on it. In addition to present the forefront of the area, Emphasis is placed on giving practical reference to today’s researcher and designer on this exciting and challenging field. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 158 |
| 源URL | [http://159.226.59.140/handle/311008/296]  |
| 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 | 鲍明. 声学目标检测与分类研究[D]. 声学研究所. 中国科学院声学研究所. 2008. |
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。