Abstract: Aiming to identify and process outliers based on micro-electro-mechanical systems unmanned aerial vehicle (UAV), we design an outlier removal algorithm based on fuzzy prediction and extended Kalman filter (EKF). First, we introduce the length of angular velocity into the gradient descent method and improve the accuracy of the fuzzy prediction by dynamically adjusting the gradient step. This effectively avoids the high dependence on the threshold value in the fuzzy prediction and the loss of data points after the outliers are removed. Second, we consider innovation as the criterion for the outlier and adjust the innovation threshold by using the accelerometer measurement trust factor to eliminate the outliers of the observation sensor. Third, we use the conjugate gradient method to convert the fuzzy prediction value into quaternion and recorrect the state estimation of the EKF. Finally, based on the constructed coaxial twin-propeller UAV experimental platform system, we verify the effectiveness of the proposed algorithm. The experimental results show that the proposed algorithm can ensure the stable flight of the coaxial twin-propeller UAV under hovering and strong maneuver and effectively improve the attitude tracking accuracy and stability.