Application of the robust autoencoder to reduce reverberation and facilitate underwater target tracking

Abstract

Reverberation is the primary interference in active underwater target tracking, increasing the difficulty of the precise location of targets. To improve the accuracy of target detection under reverberation conditions, a novel sparse track-before-detect algorithm integrating a robust autoencoder and a particle filter (PF-RAE-TBD) is proposed in this paper. This method uses the robust autoencoder to build a sparse estimation model for matched received echoes. The actual measurements are then substituted with the sparse component of the target echo constructed by the nonlinear estimation. Subsequently, the track-before-detect based on particle filter (PF-TBD) is employed to track the movement of the target. Simulation and experimental results collectively demonstrate that the proposed algorithm significantly improves the performance of the active sonar in tracking targets under reverberation conditions. Using the same dataset collected in the field, the PF-RAE-TBD algorithm improves the probability of target detection by 52.94% and 22.35% compared with the conventional PF-TBD and PF-PSO (particle swarm optimized)-TBD algorithms. The PF-RAE-TBD can provide additional contributions to improve the performance of active sonar in tracking targets under strong reverberations.