In this study, the extended Kalman filter (EKF) and singular value decomposition (SVD) methods are integrated into the nontraditional attitude filtering algorithm to estimate a small satellite’s attitude. It is shown that the process noise bias and process noise increment-type system changes will cause a change in the statistical characteristics of the innovation sequence of EKF. The influence of these types of changes on the innovation of EKF is investigated. It is proved that the bias-type process noise change may be converted to the mean square of innovation of EKF and such type of changes can be compensated using the covariance scaling techniques.