基于小波变换和神经网络的水下宽带回波分类

目标识别是水下信息处理系统的主要任务之一,针对此问题,从目标识别的三个基本环节研究了水下宽带回波的分类。首先基于连续小波变换提取了实测莱蒙湖底回波的尺度———小波能量谱,以径向基函数作为分类器,得到了很好的分类效果。接着给出了三种选择特征的准则,并研究了这三种准则对分类效果的影响,结果表明,这三种方法都可以在保证分类准确度的同时有效降低特征维数。

Application of wavelet transform and neural network to the classification of underwater wideband echo

Target classification is one of the major tasks of information processing system in sonar and torpedo.Improving classification of underwater echo is studied on three basic steps of classification.Firstly, scale-wavelet power spectrum of echo is extracted according to continuous wavelet transform.Secondly,three criteria for feature selection are discussed.With one of these criteria,a goal reduced dimension of feature space is achieved. At last,radial basis function is used for classification.Compared with NN(nearest neighbor),it could improve the classification rate.The results for classification of four kinds of the bottom sediments of Geneva Lake show preliminarily that the proposed method can achieve better efficiency.