Seismic salt boundary mapping is a vital reservoir exploration process that helps understand salt tectonics. The existing methods employed in mapping salt boundaries require substantial time and computational resources. As a result, much research has been dedicated to the application of deep learning in this venture. The Deep Convolutional Neural Network (DCNN) is an innovative deep learning technique that can outperform humans in image identification. Researchers have utilised DCNN to locate seismic faults due to the recent rapid growth in deep learning. However, geophysicists continue to face difficulties in quickly training an effective model on a few samples for salt boundary mapping. This study proposes a novel method for detecting salt boundaries on seismic using a pre-trained DCNN with minimal training required. The Groningen field data was utilised to conduct the salt mapping process, which involves segmenting a 3D seismic problem and constructing an encoder-decoder architecture with a Deep Residual U-net for generating a salt probability volume. Transfer Learning was applied to fine-tune the DCNN model, which involved training and validating with four and three interpreted in lines and crosslines, respectively, to adjust the model�s weights. By utilising this transfer-learned technique, significant improvements in the model�s prediction were observed, with a Dyce Similarity Index of 0.99 and 0.92 achieved in the training and validation sections, respectively. This result shows that the model and Transfer Learning technique can automatically capture subtle salt bodies from 3D seismic with minimal manual input. This method will drastically reduce the time involved in fully interpreting salt boundaries on seismic volumes because of the generalisation of the DCNN model and can accurately map salt boundaries in different volumes. © 2024 Daniel Asante Otchere.