基于BP网络对模拟声呐信号分类

针对常规的主动声呐调查设备,在简单海洋分层模型的基础上,模拟了多波束类单频信号、侧扫类单频信号、Ch irp调频信号和混合信号4类声呐接收信号,并针对接收信号特征构造了3层BP网络模型,将隐藏层神经元数目设为可调节;利用时间域脉冲宽度和水深与频率域功率谱密度相结合的特征参量,成功地对模拟信号进行了分类。采用改进的BP网络模型,用训练成功的BP网络对102个检测信号进行了分类测试,结果表明,分类成功率较高,可达76%~84.6%,因而利用BP网络可以对不同类别设备的模拟声呐接收信号进行分类。

Classification of simulant sonar signals base on BP network

According to normal active sonar equipment,four types sonar signals were simulated,which are single frequency of multibeem and sidescan,Chirp and mixed signal based on a simple stratified ocean model.A backpropagation leaning algorithm network(BPN)has been constructed with three layer which are inputting,hiding and outputting layer.Especially,the neuron number of hide layer can be adjusted depend on the output.And the method of simulating and classification including feature extraction and classification system were discussed.We extract ten power spectral densities,two signal pulses and the depth of water as input parameters of BPN,which realized the combine of the features in frequency field with that in time field.Finally,classification of simulating signals was well carried out.The methods described in this paper will give a theory direction in practice.