In order to effectively solve the problem of image noise pollution, this paper takes the noisy image as the input of the network and the noiseless image as the output, learns the mapping relationship between the two noisy images, and proposes an image denoising algorithm based on deep convolution neural network to realize the process of transforming from noisy image to noiseless image. In this paper, the denoising algorithm based on deep learning uses the residual network (ResNet) model and combines the Inception structure in the residual block to solve the problems of gradient disappearance, over-fitting and strong difficulty in training. Residual network model can save computing resources and reduce training time while deepening the depth of network model. Taking BSD300 data set as training set, the network model is trained by GPU in colaboratory environment. The simulation results show that, when there are enough samples, the denoising model trained by noise pictures in this paper is similar to the network model trained by noise pictures in PSNR (Peak signal to noise ratio) and image detail restoration, so this model has good practicability.