Edge computing based on intelligent techniques provides powerful processing abilities in cyber‐physical systems (CPS), thereby enabling real‐time insights, and analyses for applications. To decrease network redundancy and improve the speed of convolutional neural networks (CNNs) when performing high‐resolution stylization in image style transfer tasks, a resolution enhancement scheme is proposed based on an improved CNN in CPS. A style transfer network (RESTN) model and a two‐stage neural network training method are presented. In the first stage, the resolution enhancement ability of RESTN is trained with high‐resolution and low‐resolution image pairs, and in the second stage, the style transfer ability of RESTN is supplemented. The parameters of the feature extraction phase are shared for learning different styles, and the octave convolution operation is adopted to reduce network redundancy. Experimental results show that compared with the traditional model‐based iterative method, the proposed RESTN model can conduct training three times faster and with half of the memory cost.