As an important information transmission carrier, infrared images are corrupted by stripe noise for the difference in the response of the infrared focal plane array detection units and the non-uniformity of read-out circuit, which is harmful for the next analysis and research. In order to improve destriping results of convolutional neural network, infrared images destriping method based on wavelet dilated residual U-Net is proposed. The network first decomposes noisy image into vertical component,diagonal component, horizontal component and low frequency component by Haar discrete wavelet transform; Then mappings between noisy image and stripe noise are obtained sequentially through the coarse destriping and fine destriping network; The noise component is removed from input components to gain destriped components; Finally, destriped image is reconstructed by inverse Haar discrete wavelet transform.Experimental results illustrate that the method can remove stripe noise and preserve image details.