Abstarct: Seismic data interpretation and diagnosing subsurface structures, such as salts, diapirs and mud volcanoes play a vital role in detecting oil and gas potentials which is economically important. So far, various methods including seismic attributes with advantages and disadvantages have been used to identify various structures such as, salt domes, diapirs and mud volcanos. To minimize insufficiencies such as low speed in processing, advanced yet unsolved complex calculations, recognition of structures in a correct way, accuracy and also manual interpretation problems by humans, Artificial Intelligence, specifically Machine Learning and in-depth Deep Learning use, has emerged in geophysics in recent years. The proposed method in this Thesis, the identification of mud diapirs and gas chimneys in seismic data using U-net in deep learning, which in short is called The U-Net algorithm, is a fully convolutional neural network. This method, classified as a subset of deep learning, follows a supervised learning way and is greatly good at semantic segmentation. Using fewer data would be another characteristic of this algorithm. Identifying structures, seismic data processing, inversion, seismic interpretation is the application of the U-net algorithm. The contraction and expansion path applied in the U-Net algorithm has made it capable of extracting more accurate details about the input data. To use the U-Net algorithm in this thesis, real data is applied. The real data taken from Gorgan, North of Iran is a picture, as the U-Net model needs a picture as an input. The final result shows high accuracy and validation in processed data for the output. As a final thought, the outcome of this method resulted in diagnosing 79% true positive (mud volcano), 21% false positive (No mud volcano), 93% true negative (No mud volcano), and 7% false negative (mud volcano). It reaches an F1-score of 86%, which shows a reliable value to obtain this method as a good way to identify a structure such as diapirs, salt domes, and so on.