In order to suppress the drift of inertial navigation system and improve the seamless navigation ability of Inertial Navigation System/Magnetism Navigation System (INS/MNS) in geomagnetic lockout environment，a hybrid seamless INS/MNS strategy combining Adaptive Cubature Kalman Filter with Deep Self-Learning (DSL-Adaptive-CKF) is proposed. It mainly includes two innovative steps. Firstly，an adaptive optimization auxiliary method based on residuals and innovation is combined to enhance the robustness of the initial error of measurement noise and process noise. The heading RMSE of the Adaptive-CKF method is 2.78°，which improves the heading accuracy by 89.51% compared with that of the traditional single INS，and greatly improves the robustness of the combined navigation system and the accuracy of the heading measurement. Secondly，by introducing the Nonlinear Autoregressive model with Exogenous inputs (NARX) neural network，the Adaptive-CKF can learn deeply，which means that it can realize continuous high-precision navigation estimation even during lockout period，and its heading RMSE reaches 3.08°，thus the heading accuray is improved by 88.38% compared with that of the single INS.