This paper presents a recurrent convolutional neural network-based estimation method of guidance parameters of the pursuer under augmented proportional navigation (APN) guidance law with a time-varying switching navigation ratio. For the highly maneuvering pursuit-evasion process, realistic factors in the guidance law estimation are considered, such as the pursuer's estimation error and delay of the evader's acceleration in the APN guidance law. In view of the enormous measurement data, time dependency, transient change and unknown factors' disturbance in switching guidance law estimation, a novel neural network structure is built. 1-D CNN layer is used to extract features from enormous data obtained by the multiple previous measurements. The features extracted are processed by the recurrent cell to exploit the time dependency and eliminate the error caused by unknown factors. The result of ablation test shows the proposed RCNN's improved performance over single CNN or RNN. Compared to the multiple model guidance law estimation method, the proposed method can simplify the design of guidance law estimation system and reduce calculation load. The estimation result for switching guidance law shows the proposed method has higher accuracy and faster convergence rate than traditional interactive multiple model methods. INDEX TERMS Guidance law estimation, pursuit-evasion process, recurrent convolutional neural network, switching navigation ratio, time dependency.

Based on direct sequence spread spectrum code and varies frequencies, the terrestrial pseudo ranging system have high precision and strong anti-jamming capability and could enhance the navigation accuracy of aircraft by fusion with Integrated Navigation System (INS). A fusion method of pseudo-range measurements and inertial sensor data in INS/Pseudo-Range navigation system was proposed in this paper. Extended Kalman Filter (EKF) which is commonly used in integrated navigation inherits Kalman Filter (KF)'s advantage in having good computational efficiency, but often leads to unstable estimations. Unscented Kalman Filter (UKF)is recently suggested for stability improvements, and it is more suitable for nonlinear filter. The comparison between theses two nonlinear filter in the INS/Pseudo-Range integrated navigation was also conducted.. Simulation results show that the UKF approach offers better performances over EKF.

No abstract