Li, Z.Y, Wang, G.Q., Zhu, C.H. and Chen, S.Q., 2023. Enhancing seismic resolution based reservoirs are being developed for oil and gas exploration. High signal high-to-noise ratio seismic data is required for accurate reservoir resolution seismic data processing is a challenging and im on U-Net deep learning network. Journal of Seismic Exploration, 32: 315-336. Deep and ultra-deep reservoirs, unconventional hydrocarbons, and other complex -resolution with high description. Therefore, portant endeavor. The conventional high-resolution processing techniques, such as inverse Q-filtering technique based on the stratum attenuation model and deconvolution met hods, are nearly mod- el-dependent. Deep learning method based on image processing has been widely used in seismic data processing and inversion in recent years, which is a data-driven method and has p owerful abilities to extract data features. In this paper, we rebuild a deep learning network based on U-net network and proposed a high-resolution seismic data processing method and workflow. In the network, we utilized ResPath structure instead of the straightforward skip connection in the conventional Res-Unet, and employed a combina- tion loss function by using the mean absolute error and multiscal e structural similarity. Moreover, by introducing pre-training strategy into the processing workflow, the pro- posed network can reduce the loss of low- frequency components and perceive low-frequency information. Synthetic and real field seismic data are used to validate the network, the trained model by using pre-training strategy can better maintain the low-frequency components of the original data and recover the high-frequency compo- nents, and the signal-to-noise ratio and resolution of seismic data cantly enhanced. ave both been signifi-