Abstract: In underwater multi-station passive sonar systems, fusing bearing-only measurements from multiple stations for multi-target joint tracking and localization presents significant challenges, particularly due to unknown target priors and the high complexity of data association. To address these issues, this paper proposes a measurement-driven and efficient multi-station joint observation algorithm for passive underwater multi-target tracking task. Based on the generalized labeled multi-Bernoulli framework for fusing multi-source measurements, the proposed algorithm introduces a measurement-driven mechanism to dynamically generate newborn target components. To further reduce computational complexity, submodular optimization theory is employed to select key measurements and predicted components for data association. Experimental results demonstrate that the proposed algorithm maintains robust performance under challenging conditions involving missed detections, false alarms, and measurement noise. Compared with existing methods, it achieves a 28.36% improvement in the average optimal sub-pattern assignment metric while significantly reducing computational cost, thereby providing foundation for the development of efficient multi-station underwater sensing systems.