The extraction of line-of-sight (LOS) rate is a key question for strapdown image guidance. The filtering estimation method is a common method, but the method requires relative distance and its rate between ammunition and target. It’s difficult for strapdown seekers to get the information. For this reason, we propose a LOS rate extraction method based on LSTM-UKF. By leveraging the nonlinear mapping capability of LSTM to capture the functional relationship between state equation inputs and outputs, we can fit a state-space model, thereby replacing the process in the Kalman filter where the state equation is used to calculate state prediction values. The specific implementation involves the following steps: First, we construct a state transition network based on LSTM, which takes historical LOS angle and LOS rate as input to perform one-step prediction. Subsequently, this network is integrated into the UKF filtering framework. By combining observations, the framework executes temporal updates and observational updates iteratively, ultimately estimating LOS angle and LOS rate. This approach achieves LOS rate extraction without relying on prior knowledge of relative positional information between the ammunition and target. Finally, we carry on a simulation experiment to contrast the behavior of the proposed method with the UKF (which assumes ammunition-target relative distance and its rate are known in simulation). Simulation results demonstrate that our method achieves higher accuracy.