This research introduces a semantic segmentation framework employing the U-Net model. The U-Net model is utilized in this study to classify individual pixels within synthetic aperture radar images. The model has been adapted to semantically segment the image pixels into oil spills, look-alikes, sea-surface, land and ships. The semantic segmentation network was utilized using four different backbones to observe their performance. The U-net model that utilizes an Efficient-net-B3 backbone performs the best among all the four backbones with a mIoU score of 76.53% and intersection over union (IoU) value of 62.08% for oil spills, 54.20% for look-alikes, 97.02% for sea surface, 72.06% for ships and 97.30% for land class. The results indicate that U-Net models with appropriate backbones effectively segment oil spill areas in SAR images.