基于连续小波变换的水下目标特征提取与分类

Featrure Extraction and Classification of an Underwater Target Based on CWT

在连续小波变换的基础上,提出了信号的尺度-小波能量谱和时间-小波能量谱的概念 ,同时定义了选择特征的一种判别函数。利用信号的尺度-小波能量谱、时间-小波能量谱提取了菜蒙湖底沉积物反射回波的特征,依据判别函数给出了回波的特征向量,并用最近距离分类器和近邻分类器对反射回波进行了分类。结果表明,尺度-小波能量谱和时间-小波能量谱相比有较好的分类效果和较低的特征维数。最后研究了所提取特征的鲁棒性。

Based on continuous wavelet transform,scale-wavelet power spectrum and time-wavelet power spectrum of the signals are proposed.Moreover,a discriminating function is defined for feature selection with the aim of reducing feature's dimension.For four kinds of wideband echoes of the bottom sediments of Geneva Lake ,the scale-wavelet power spectrum and time-wavelet power spectrum of the signals are computed.According to discriminationg function,feature vector of the echoes is formed and the echoes are classified by the nearest distance classifier and K-NN classifier.The result shows that,compared with the time-wavelet power spectrum,the scale-wavelet power spectrum has a higher recognition accuracy and smaller dimension of feature vector.At the end,the robustness of the extracted feature is studied.