基于改进小波能熵的水下目标识别

摘要

A method based on improved wavelet energy entropy and probabilistic neural network for underwater target recognition was studied in the paper.Firstly,the radiated noise signal of underwater target is processed by using multi-resolution wavelet decomposition and reconstruction.Then,the sliding time window is introduced and the improved wavelet energy entropy of each decomposed band in sliding time window is extracted as feature vectors for target recognition.Finally,these feature vectors are used as input vectors of probabilistic neural network for target classification.The feature of signal under different frequency domain can be reflected by multi-resolution wavelet decomposition.While the feature of signal under different time domain can be reflected by the improved wavelet energy entropy which is defined by introducing the sliding time window.The improved wavelet energy entropy can reflect the time-frequency feature of signal at the same time and it is suitable to underwater target feature extraction.The result from test and simulation shows that the method is effective.%文章研究了基于改进小波能熵和概率神经网络的水下目标识别方法。首先对水下目标辐射噪声信号进行小波变换多分辨率分解和重构,然后引入滑动时间窗,提取各分解子带在滑动时间窗内的改进小波能熵值作为目标识别的特征矢量,最后将特征矢量输入到概率神经网络中实现水下目标识别。对信号进行小波多分辨率分解可反映信号在不同频域上的特征,而引入滑动时间窗并在此基础上定义改进的小波能熵可反映信号的时域特征,因此改进小波能熵方法能同时反映信号的时频特征,更适合于水下目标特征提取。仿真结果表明了该方法的有效性。