摘要
提出一种水下目标回波的特征提取方法。该方法根据平稳小波变换的冗余性和奇异值的稳健性,将湖底回波信号的平稳小波变换系数矩阵的奇异值作为特征向量。它本质上是利用平稳小波变换将信号分解到多个子空间,再采用K-L变换实现对子信号的特征压缩。实测数据分析表明,本文方法与子带能量特征法相比: (1)在相同的样本集和类内距条件下,得到的类间距大于后者, (2)不论所选测试样本和训练样本是否属于同次湖试所得,分类正确识别率均高于后者, (3)随样本集的变动,其正确识别率抖动程度远小于后者。因此,该方法能得到更加稳健、有效的特征以及更好的分类效果。
A feature extraction method of underwater echo, which takes advantage of the redundancy property of SWT （the Stationary Wavelet Transform） and the steady property of SVD （Singular Value Decomposition） was propostd. Since uses singular values of SWT coefficients matrix as feature vectors, it is a feature compress method with the K-L transform to multi-subspace signal obtained from SWT essentially. In contrast to the method of sub-band energy feature based on discrete orthogonal wavelet transform （DWT）, this method acquires better results of lake trial data： （1） Under the same sample and distance within kinds, distance between kinds are larger than former; （2） correct recognition rates are also higher than the ones of the former, whether the training and testing samples are chosen from the same lake trail or not; （3） Sample sets varying, variety range of correct recognition rates is far less than the former. Thus this method can obtain more robust, effective features and better correct recognition results.
出处
《声学学报》
EI
CSCD
北大核心
2006年第2期167-172,共6页
Acta Acustica
