To address the issue that complex target motion patterns in underwater bearing-only target tracking lead to cumulative estimation errors with single-model approaches, this paper utilizes the Interacting Multiple Model (IMM) framework combined with the Cubature Kalman Filter (CKF) to enhance tracking performance under nonlinear measurement conditions. Aiming at the problems of subjectivity and poor adaptability in traditional IMM methods which employ fixed model probability transition matrices, an Exponentially Weighted Moving Average (EWMA) strategy is introduced to achieve online adaptive updates of the model probability transition matrix. This method dynamically adjusts the trend of model switching by incorporating historical model probability information, effectively improving tracking accuracy. Simulation comparisons between the proposed method and both standard CKF and traditional Interacting Multiple Model Cubature Kalman Filter (IMMCKF) indicate superior performance of the proposed method in terms of response speed and tracking accuracy, the proposed algorithm achieves a 55.59% improvement in tracking accuracy compared to CKF, and a 40.53% improvement compared to IMMCKF.