基于正交小波包的水下宽带回波特征提取与识别

On Improving Feature Extraction and Identification of Underwater Wideband Echo with Orthogonal Wavelet Packet

水下目标的自动识别是水声设备智能化的关键技术之一。在对目标回波进行正交小波包分析的基础上 ,提取信号在各分解子空间的能量以构成回波的特征空间。给出了两种衡量各特征识别能力的准则 ,而且基于这两种准则解决了小波包分析中的频率重叠问题 ,用它们对莱蒙湖 (又名日内瓦湖 ,Geneva Lake)湖底 4类沉积物的反射回波进行了特征提取和分类 ,比较了两种准则所提取特征的分类结果 。

The application of Mallat algorithm to the feature extraction and identification of underwater wideband echo is, not satisfactory in the identification of wideband echo in high frequency range. But the wavelet packet algorithm is potentially satisfactory in both low and high frequency ranges. We used the to decompose target echoes. Then the feature vector included the energy of every subband of echo. We proposed two criteria for the feature selection. With either criterion, we could achieve two goals: the simplification of feature vector by discarding components that contain little discriminating information and solution of the block overlap problem of frequency bands. We applied the two criteria to feature the extraction and identification of four kinds of wideband echoes of the bottom sediments of Geneva Lake. The comparison results show that the proposed method can improve the feature extraction and identification of underwater wideband echo.