QQ客服 官方微博 反馈留言

题名	水声成像中的目标识别与舰船辐射噪声识别
作者	田杰
学位类别	博士
答辩日期	2002
授予单位	中国科学院声学研究所
授予地点	中国科学院声学研究所
关键词	声纳 水声图像 分类 识别
中文摘要	本文围绕着水声信号处理领域内的两个问题一一水声图像目标识别和舰艇辐射一噪声分类展开研究。其中水声图像目标识别以水下探雷为应用背景，由于水声信号的非平稳特性，以及水下环境的复杂，一般水声图像信噪比较差，而水雷目标相对来讲反射面积比较小，所以依据水声图像的水雷识别是个小目标检测的问题，而且由于环境状况和成像条件很难复现，它又是一个有限样本识别问题。本文同时刘一舰艇辐射噪声的分类方法进行了研究。对于侧扫方式获得的水声图像，本文首先采取模板匹配和分形滤波的方法进行探测，得到可一能存在目标的区域，对该区域提取特征构成特征矢量，并通过支持向量机分类器进行识别。对于舰艇辐射噪声的分类，本文采用连续功率谱和倒谱作为特征，连续功率谱反映了噪声的音色，而倒谱则有抗多途效应的作用。本文针对水声图像目标的探测和识别问题进行了实验，实验结果表明所采用的模板匹配和分形探测方法可以较好地完成探测任务，而采用支持向量机算法对图像特征进行识别具有较好的识别特性。对舰船辐射噪声的分类实验说明功率谱分析和倒谱分析可以提供稳定的识别特征，并且基于DS证据理论对这两种特征进行融合也可以提供满意的结果。
英文摘要	This dissertation encloses two important issues in the field of underwater acoustic signal processing, the object recognition in sonar images and the classification of ship radiant noise. Underwater mine detection is the main practical background of object recognition in sonar images. Generally speaking, underwater acoustic images have very low SNR, because that underwater acoustic signals are usually unstable, and that the underwater environment is very complicated. The mine detection based on underwater acoustic images remains a problem known as small target recognition, for the reflectable square of mines is relatively small. Furthermore, it is also a problem on the recognition of finite samples. The classification of ship radiant noise is discussed as well in this paper. To the underwater acoustic images acquired by side-scan sonar, detection is firstly done based on mask matching and fractal filtering, so that the regions likely containing targets can be found out. Then features are extracted from these regions to construct a feature vector, which is transmited to a SVM classifier. The continuous spectrum and cepstrum are adopted to be the recognition features to the classification of ship radiant noise, because the continuous spectrum reflects the tamber of noise, while the spectrum responses to the function of removing multi-routine phenomenon. Many experiments are hold to prove the efficency of object recognition algorithm. The results show that the mask matching and fractal filtering methods can preferablely detect the targets, and the SVM classifier adopted exhibits satisfying characteristics when used to image. Moreover, the experiment on classification of ship radiant noise proved that spectnim and cepstrum analysis could provide stable features to classification. Besides, the fusion of these two features based on DS evidence theory can also give approving results.
语种	中文
公开日期	2011-05-07
页码	63
内容类型	学位论文
源URL	[http://159.226.59.140/handle/311008/902]
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	田杰. 水声成像中的目标识别与舰船辐射噪声识别[D]. 中国科学院声学研究所. 中国科学院声学研究所. 2002.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

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。