Salt segmentation in seismic images is an essential process in the field of geophysics, particularly in the exploration and identification of subsurface resources like oil and gas. For companies looking for oil and gas, finding where the salt is matters a lot. Moreover, lands affected by salt become unsuitable for farming due to reduced plant absorption capacity, impacting growth rates. Seismic images are derived from sound waves bounced off underground structures, and offer critical insights into the Earth’s composition and the presence of valuable resources. However, these images may often contain salt formations. To identify salt-affected areas, seismic images are analyzed at the pixel level to classify them as either salt or sediment. TGS Salt Identification Challenge dataset is used which consists of images captured from different underground spots randomly selected. Each image is 101 \(\,\times \,\) 101 pixels, and every pixel in these images is labeled as either salt or sediment. In this proposed model deeplabv3, utilizes ASPP, and studies each tiny part of the picture, noticing textures and shapes to capture features at multiple scales. To figure out the model performance, we utilize metrics like Dice Similarity Coefficient and intersection over Union (IOU).