Acoustic detection of underwater buried targets based on transfer-learning-based full waveform inversion

Conventional full waveform inversion (CFWI) easily falls into local minima, leading to poor inversion results. This study adapts a transfer-learning-based full waveform inversion (TFWI) workflow to shallow, small-scale acoustic detection of underwater buried targets. TFWI is pre-trained and fine-tuned using 49,100 samples and 9,110 samples, respectively, and the generalization ability of TFWI is investigated on three typical structures. Finally, based on numerical simulation and field test data, the inversion performance of CFWI and TFWI is compared using three iterative schemes. The results show that the average computational time, sidelobe energy ratio (SER), and maximum sidelobe amplitude (MSA) of TFWI are all reduced compared with CFWI. Ablation experiment analysis shows that transfer learning is the dominant factor contributing to the performance improvement of TFWI. TFWI shows potential advantages and provides an effective approach for the research of acoustic detection of underwater buried targets.