Classifying different acoustic sources has been a challenge in underwater environment due to the difficulties of acquiring data in ocean environment and also due to the variety of background noise it poses. Self-attention mechanism has been shown to be effective in various machine learning tasks in challenging environments with sparse dataset. An audio transformer, designed for air acoustics, was modified and adapted for underwater sound classification task. We demonstrate the effectiveness of our method by applying to the shipsEar dataset [1], and show that the proposed method outperforms some of the latest classification methods. We also show its robustness in high noise environments.