Multi-Stage Feature Interaction Model with Abundant Semantic Information for Image Captioning