Text matching is one of the fundamental tasks in natural language processing. Most of the existing models focus only on encoding the text itself but ignore other semantic information which may further improve matching accuracy. In this paper, we propose a novel model for text matching with double-channel multi-layer information fusion. It treats text and part-of-speech information of words in a sentence as double-channel information which is fused by multi-layer interactions. Meanwhile, our model uses a Siamese network structure to learn common and unique features of two sentences, which can improve its ability to learn the relationship between two sentences while reducing the parameter size and complexity. Experimental results on SNLI dataset show that our model can achieve better performance than baseline methods.