MAENet: A novel multi-head association attention enhancement network for completing intra-modal interaction in image captioning