Underwater passive sonar fusion detection based on sensor bias estimation and classification

Abstract

Passive sonar is crucial for underwater target detection but often experiences reduced localization accuracy due to sensor biases and positional deviations caused by environmental factors. Traditional approaches struggle to effectively address these biases and also lack an integrated analysis of bias estimation, sensor classification, and fusion algorithms. To address this issue, this paper proposes a bias estimation-classification-fusion localization (BECFL) algorithm, integrating bias estimation, sensor classification, and adaptive fusion weighting. Within the BECFL framework, dynamic stochastic models are first developed to characterize underwater target motion, sensor biases, and multipath fading channels. Sensor biases are then estimated using maximum likelihood estimation (MLE) based on multiple sensor observations. Sensors are subsequently classified through an improved expectation-maximization (EM) algorithm, iteratively refining the classification parameters. A variable step-size weighting fusion approach is applied, reducing the impact of sensor biases on localization results. Simulations demonstrate that, compared to existing methods, the BECFL algorithm improves localization accuracy and maintains robust performance even under significant sensor bias, data loss, false alarms, and missed detections. Additionally, analyses of actual underwater data confirm the practical applicability of the proposed algorithm for passive sonar detection and localization in underwater environments.