Accurate attitude estimation of unmanned aerial vehicles (UAVs) is crucial for maximizing their capabilities. The Inertial Measurement Unit (IMU) is a commonly used sensor for attitude estimation, but its output is often affected by various factors such as vibration, temperature, and narrow-band noise. Additionally, the operation of direct current (DC) motors can introduce magnetic field distortion, further reducing attitude-solving accuracy. This study proposes a robust solution to mitigate narrow-band vibration noise and magnetic distortion from DC motors in low-cost UAVs, thereby enhancing IMU output accuracy. Using a notch filter, the proposed method first utilizes the Least Mean Square (LMS) algorithm to identify primary frequency of high-frequency noise in real-time gyroscope and accelerometer outputs. Subsequently, a stochastic model is developed to compensate for the magnetic field distortion in the magnetometer output caused by different Pulse Width Modulation (PWM) settings of the DC motor. Validation experiments conducted on an actual UAV platform demonstrate that the proposed algorithm effectively suppresses narrow-band noise, mitigates magnetic distortion, and improves IMU data quality. By integrating the enhanced IMU outputs into two typical attitude estimation algorithms - the multivariate extend Kalman filter (MEKF) and complementary filter (CF), the resolution accuracy is enhanced by 25.7% to 41.2% and 22.7% to 30.5%, respectively. This solution not only strengthens the robustness of attitude estimation algorithms but also paves the way for enhancing the reliability of navigation systems in UAVs.