Image caption models using visual attention and reinforcement learning (The 4th place solution to the AIChallenger Contest, Image Caption Track by team xiaoquexing)