Online Recognition Method for Target Maneuver in UAV Autonomous Air Combat

In unmanned aerial vehicle (UAV) autonomous air combat, target maneuver online recognition is beneficial to predict the tactical intention of the target, which is of great significance to air combat situational awareness and decision-making assistance. However, the target trajectory data acquired by our airborne sensors may contain one or more maneuvers. The existing research methods mainly focus on the recognition of single maneuver trajectory which has been segmented according to maneuver segments or the segmentation of target maneuver trajectory by introducing human experience. The above methods can not meet the requirements of online recognition of the unknown continuous multi-segment maneuvers trajectory in autonomous air combat. In this paper, the online recognition problem of target maneuver is transformed into three classification problems, which are maneuver switch subsequence recognition problem, maneuver switch point localization problem and single maneuver recognition problem. The cascaded classification networks based on Long Short-Term Memory (LSTM) temporal feature extraction is used to map maneuver trajectory to maneuver category sequence. The algorithm proposed in this paper is tested by randomly selecting target trajectories, which can automatically segment the trajectories and recognize the correct categories. The average sequence similarity between the predicted maneuver category sequence and the real maneuver category sequence after segmentation and recognition on the test set is above 0.9. The feasibility and effectiveness of the proposed algorithm are verified.