Adaptive Estimation of Observation Noise Covariance using Neural Network in Maximum Correntropy Criterion Extended Kalman Filter

In traditional maximum correntropy criterion extended Kalman filter (mccEKF) algorithms, the observation noise covariance matrix, is usually set as a fixed value. However, in complex environments with non-Gaussian noise, a staticobservation noise covariance matrix may not accurately reflect the dynamic changes in noise characteristics. To address this, neural network can be employed to adaptively estimate observation noise covariance matrix for each point in real-time. By training the neural network to adjust observation noise covariance matrix based on incoming data, the filter’s robustness and accuracy in noisy and uncertain systems can be significantly improved.The findings demonstrate that the suggested approach has significantly better localization accuracy when compared with mccEKF filter.