Underwater Acoustic Sensing with Rational Orthogonal Wavelet Pulse and Auditory Frequency Cepstral Coefficient-Based Feature Extraction

Active pulse design, target detection and classification play an essential role in underwater acoustic sensing. This paper addresses the system design with three kinds of pulse signals, including continuous wave (CW), linear frequency modulation (LFM) signal and rational orthogonal wavelet (ROW) signal. The detector design has an architecture of feature extraction and convolutional neural network (CNN) based classification. A geometric underwater channel model is adopted to facilitate the generation of training datasets with designated geometric underwater environment parameters. The simulated received pulse signals are converted into feature maps as the input of the classifier. This paper applies the acoustic features, Short Time Fourier Transform (STFT), Mel Frequency Cepstral Coefficients (MFCC) and Gammatone Frequency Cepstral Coefficients (GFCC) to construct different feature maps. A lightweight CNN model is used as the classifier. Experiments demonstrate the superiority of the ROW wavelet pulse signals and the proposed algorithm in target localization and underwater signal classification.