Ship Radiated Noise Line Spectrum Enhancement Based on Adaptive Filtering-Deep Learning Fusion

To address the challenge of identifying line spectrum features of ship radiated noise in complex underwater acoustic environments, a line spectrum enhancement model combining adaptive filtering and deep learning (DL) is proposed. This model integrates a double-layer collaborative adaptive filtering module with a deep learning module to form a multi-level feature enhancement framework. The adaptive filtering module combines frequency-domain adaptive filtering(FDAF) with the maximum correntropy criterion (MCC) to enhance the suppression of non-stationary noise while effectively reducing broadband background noise. The deep learning module employs a bidirectional long short-term memory (BiLSTM) to extract the local temporal dependencies of line spectra. It also incorporates an attention residual mechanism to focus on weak line spectra and utilizes a Transformer encoder to capture long-range correlations in the time-frequency domain. The model effectively combines the advantages of filtering and deep learning, both suppressing noise and enhancing the detection of weak spectral lines. Simulation results demonstrate that the proposed method outperforms single adaptive filtering or deep learning models in terms of overall line spectrum enhancement and weak line spectrum enhancement. The effectiveness of this method is further validated through lake test data.