Research on the enhancement method of offshore vessel characteristics based on transplanting memory LSTM

Offshore vessel acoustic characteristics are critical for target recognition, yet passive sonar signals are severely degraded by environmental noise, multi-source interference, and time-varying propagation effects. This study constructs a multi-target offshore vessel acoustic database for the Yellow Sea and proposes a mask-based signal enhancement and separation architecture. A Transplanting Memory Long Short-Term Memory (TM-LSTM) network is integrated to capture long-term contextual dependencies and improve signal clarity. The proposed encoder–separator–decoder framework extracts multi-scale nonlinear features, separates target and noise masks using memory-augmented representations, and reconstructs enhanced signals through transposed convolution. Experiments conducted on two tasks using the Yellow Sea dataset demonstrate that the proposed method improves the signal-to-noise ratio by 1.97 dB and 1.58 dB, and enhances the scale-invariant signal-to-distortion ratio by 1.88 dB and 1.49 dB, respectively.