Abstract: The disturbance torque of flight vehicles' optoelectrical image stabilization platform increases due to the mass imbalance and other internal factors of the system under the impact of strong airflow, and this leads to a serious reduction of the system's line-of-sight (LOS) stabilization accuracy. To solve this problem, we propose a disturbance torque model of a typical two-axis orthogonal stabilized platform, whereby the model considers the influence of the kinematic coupling between frames. Aiming at the influence of the multi-sensors' noise in the process of torque reconstitution, we propose an on-line estimation method of disturbance torque based on unscented Kalman filter and construct a torque feed-forward control loop to realize the disturbance's adaptive compensation. The semi-physical simulation experiment results show that the torque estimation has fast convergence speed and smooth estimation process. Compared with the control method of disturbance observer with friction torque compensation, the LOS stabilization of the system is improved by 10.9% to 29.3% at 0.5 Hz to 4 Hz carrier disturbance.