Chinese Image Captioning via Fuzzy Attention-based DenseNet-BiLSTM