试验环境水下声信号的特征提取方法OA北大核心CSTPCD

The inversion of environmental parameters in underwater tests is an important topic in the field of hydro-acoustics.The key to the current research is how to obtain parameter information by extracting the features of un-derwater acoustic signals.With the aim of addressing the difficult problems of feature extraction and model fitting,this paper proposes a feature extraction method for underwater acoustic signals in an experimental environment.The underwater acoustic signals are processed simultaneously by the Mel-frequency cepstral coefficient(MFCC)and the linear prediction coefficient(LPC),to which the feature-weighted combination method is applied to obtain a new feature matrix.Furthermore,the mapping interpolation algorithm is used to process the feature matrix,thereby ob-taining a three-channel matrix that adapts to the input of the neural network.The residual neural network(ResNet)is also selected in this paper.Training with the laboratory recorded Duihekou reservoir dataset reveals that the pro-posed feature extraction method is generally better than the feature processing method that only uses MFCC or LPC.In addition,accuracy is increased by 2%on average by using single-frequency rectangular pulse signals(i.e.,continuous wave(CW)signals)to classify the environment in depth 5.At the same time,the accuracy is in-creased by 2.03%on average by using linear frequency modulation(LFM)signal to classify the environment in depth 5.The results treated by the proposed feature extraction method are better than those obtained by CW signals when applied to the depth classification task.