Empirical Mode Decomposition For Underwater Acoustic Signals Noise Reduction And Feature Extraction

Underwater acoustic signal analysis is a significant means of achieving water surface and underwater perception,which has important implications for modern naval warfare,intelligent fisheries management,marine ecological protection,and other applications.It is currently a hot topic of research.Due to the complex underwater marine environment and severe background noise interference,underwater acoustic target signals often exhibit "three non" characteristics——non-Gaussian,non-stationary,and non-linear.Therefore,how to analyze and process them in real-time effectively has been a longstanding challenge for underwater acoustic workers.In this paper,we propose an improved method based on empirical mode decomposition(EMD)theory for noise reduction and feature extraction of ship-radiated noise signals.Experimental results show that our proposed method can effectively improve the noise reduction and feature extraction performance of ship-radiated noise signals and provide new processing tools for acoustic signal analysis.The main innovations of this article include:(1)In order to address the issue of ship-radiated noise being severely interfered by ocean background noise,this paper proposes a denoising method based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and improved permutation entropy(IPE).Firstly,the ship-radiated noise is decomposed into a set of intrinsic mode functions(IMFs)using CEEMDAN,and noise IMFs are identified and removed by calculating the mutual information between adjacent IMFs.Then,the complexity of remaining IMFs is analyzed using IPE,and they are divided into signal IMFs and noise-containing IMFs.Finally,the least mean squares(LMS)algorithm is used to denoise the noise-containing IMFs,and the denoised IMFs are reconstructed with the signal IMFs to obtain the denoised result.The experimental results of ship signal denoising show that the proposed method makes the ship-radiated noise waveform smoother and the attractor trajectory more regular.(2)In response to the problem of traditional feature extraction methods failing to fully describe the complexity of underwater acoustic signals,a feature extraction method that combines CEEMDAN with multi-scale improved permutation entropy is proposed by this paper.The method analyzes the entropy of the signal in different frequency bands at multiple scales to enhance the separability of the features.By extracting features from five types of actual ship signals and inputting them into a probabilistic neural network for verification,experimental results show that compared with the traditional high-low frequency energy difference and center frequency method based on intrinsic mode function,the method proposed in this paper has improved the recognition accuracy of the five types of ship signals by 31.4% and 20.5% respectively;compared with permutation entropy and improved permutation entropy methods,the recognition accuracy was improved by 14.3% and 10.1%respectively.Therefore,the feature extraction method proposed in this paper can be used to analyze underwater acoustic signals well,and can be used as a complementary tool for underwater acoustic signal identification.