An Underwater Moving Target Recognition Method Based on Inter-Frame Motion Feature Cyclic Reconstruction

In complex underwater acoustic environments, it is challenging for active sonar to accurately identify under-water moving targets with similar geometric structures and materials by feature extracted from single frame echo. In this paper proposes an underwater moving target recognition method, which effectively extracts the time-frequency variation characteristics of different frame echoes. Initially, multiple frames of echoes undergo preprocessing including reverberation suppression, matched filtering, time alignment, framing, labeling, noise down-sampling, time-frequency transformation, and channel stacking. Then, the inter-frame motion of the echo segments is extracted by MotionSqueeze module and cyclically reconstructed with the noise segments. Finally, combined with the Conv Next v2 classifier model, the method achieves better recognition performance compared to methods based on single-frame echo data in sea trial data.