Abstarct: Faults and fractures play an important role in creating structural reservoirs. Detecting and extracting faults in seismic data is one of the most important tasks in processing and interpreting these data to identify oil traps and subsurface structures. Manual identification of faults on seismic sections is a difficult and exhausting task and depends on the skill and experience of the expert. Therefore, the use of semi-automatic to automatic methods to detect the geometry of faults has attracted the attention of many researchers in recent years. Since the emergence of artificial intelligence, revealing the phenomena and features in the image has been an important issue, and many artificial intelligence algorithms have been designed baxsed on mathematical models in this regard. With the emergence of machine learning and the spread of image processing tools, many supervised and unsupervised algorithms have been created in this field, whose accuracy has generally been better than older mathematical models. The accuracy of identifying phenomena with new models has greatly increased so that today machines are close to human recognition and have become really intelligent. Among the deep learning algorithms, the U-Net algorithm has become popular in recent years due to its accurate response, high speed in processing and learning, no need for large data sets for learning, and no need for complex and expensive hardware. The popular algorithm for identifying image components and their segmentation has become in seismic image processing. In this thesis, the structure of this algorithm and its convolutional network as well as the proper setting of the parameters of this algorithm are used to optimize the objective function to achieve maximum accuracy in the processing of seismic data images. For this purpose, first the appropriate codes were written in the Python environment, and then to check the efficiency of the written codes; Seismic data from a synthetic fault model was used. After the correctness and accuracy of the code's performance in identifying and separating errors on artificial data was determined; it was also used to process and interpret the real seismic data of F3 located in the North Sea in the Netherlands. The results of applying this algorithm on synthetic and real seismic data show that the use of the U-net deep learning intelligent method has been very good in separating and detecting faults in seismic data. That is, the features extracted by the intelligent neural network method have performed well both for faulted and non-faulted parts.