Passive ship detection and classification using hybrid cepstrums and deep compound autoencoders

The acoustic noise radiated from various ships in the same class is varying due to the changing machinery regimes, the multi-path propagation effect, time-varying underwater channels, and fluctuating ambient noise. The complex underwater propagation environment causes random fluctuations in the frequency, amplitude, and phase of the signal at the receiver point. Hence, in order to overcome the aforementioned problems, this study proposes a novel deep convolutional-recurrent autoencoder, evolving by a compound cepstral lifter. In this approach, the compound autoencoders automatically extract the features without information loss and human intervention, and hybrid cepstral lifters reduce the multi-path distortion and time-varying shallow underwater channel effects. In order to evaluate the performance of the proposed model, three underwater acoustic datasets, including synthetic, ShipsEar, and real experimental datasets, are exploited. For the sake of having a comprehensive comparison, the performance of the designed model is compared with ten recently proposed benchmark models. The results approve that the designed model with an average accuracy of 96.11% and average giga-multiplier-accumulators equal to 0.019 reports the best accuracy and complexity than other benchmark models. Furthermore, the proposed model is less sensitive to SNR level compared to other benchmark models.

Automated summary

This study proposes a novel deep convolutional-recurrent autoencoder with a compound cepstral lifter to address the issues caused by varying ship noise, multi-path propagation effect, and time-varying underwater channels. Experimental results show that the designed model outperforms other benchmark models in terms of accuracy and complexity.