Fine fault identification is of great significance to improve the efficiency of exploration and development. Traditional seismic attribute fault recognition technologies identify fractures based on data discontinuity, and there are many interfering factors, making it more and more difficult to meet the needs of fine exploration in deep area. In order to improve the fault identification accuracy, this paper proposes a high-resolution intelligent identification method of fault. Based on the deep learning method to predict fault attributes from seismic data, a high-resolution and low-resolution fault label library is established, and a deep neural network is trained. It is confirmed by the model and actual data that the method solves the problem of high-frequency loss caused by up sampling in the convolutional neural network in deep learning, which reduces the resolution of faults, and improves the resolution ability. The root mean square error of the simulated data decreased by 40.02%.Compared with traditional algorithms, the method not only detects fault features more accurately, but also has a higher resolution than common deep learning fault recognition.