A Study of BP Neural Network on Ballistic Target Tracking

Firefinder radar is the main reconnaissance equipment of artillery. It mainly accomplishes the tasks of reconnaissance and fire adjustment by tracking ballistic targets. Existing firefinder radar generally adopts the linear Kalman filter algorithm to realize the ballistic target filtering. During the filtering process, the state space linear equations are used to estimate the state and covariance of the target. Due to the nonlinear characteristics of the ballistic target trajectory, large process errors are easily introduced in the filtering process. To solve the problems above, a ballistic target back propagation(BP) neural network model is established for target estimation. In order to establish an accurate neural network model, a series of environmental arguments are used to generate data sets for BP neural network training. Combined with the characteristics of neural network prediction, the unscented Kalman filter(UKF) algorithm is selected as the basic filtering framework. The simulation results show that BP-UKF has faster convergence speed and higher filtering accuracy than UKF.