State-split deep reinforcement learning approach to weapon target assignment problem