Estimating the flying attitude of the missile body during flight is vitally important for the intelligent ammunition. The earlier and more accurate the posture information is obtained, the more conducive it is to the control process of the intelligent ammunition. The magnetometer is commonly used to measure the roll attitude of the missile. However, the cumbersome calibration steps and noise interference of the missile usually cause the delay in the attitude calculation and inaccurate measurement. To solve these issues, an adaptive reconfigurable unscented Kalman filter (ARUKF) method based on energy estimation is proposed. The magnetic field measured by the magnetometer is analyzed in real time based on the roll characteristic of the intelligent ammunition to achieve the initial calibration of the magnetometer parameters. With the obtained calibrated parameters, the filter state variables are estimated by establishing the residual estimation function in the UKF, which guides the reconstruction of the UKF to minimize the computational cost of the attitude estimation process. Simultaneously, the on-board computer system feeds back the action of the rudder system in real time to guide the configuration of the noise parameters in the UKF process, which can effectively reduce the interference of the magnetic field generated by the rudder system on the attitude estimation. The numerical simulation and experiments show that the proposed method performs better in terms of parameter convergence speed and estimation accuracy, compared with the traditional UKF method, which has great application prospect in the field of intelligent ammunition.