A novel demosaicing algorithm based on deep learning is proposed to address the problems of zippers and artifacts that often occur in the traditional Bayer demosaicing algorithm. First, the proposed algorithm decomposes, removes, and combines pixels in the red, green, and blue channels of mosaic images to obtain two color images. The two color images are then inputted into a designed convolutional neural network to reconstruct the complete color image. The network can make the full use of the feature information generated by the convolutional layer. The experimental results show that the quality of the whole color image reconstructed by the proposed algorithm is relatively high, and the zippers and artifacts are relieved to a certain extent. The objective index and subjective evaluation of the proposed algorithm are better than the contrast algorithms.