Underwater acoustic signals are sparse and non-stationary, which make them difficult to be recognized. Therefore, this paper proposes an underwater target recognition method based on supervised contrastive learning to enhance feature representation by introducing supervised contrastive learning. We propose an efficient framework that is annotation-based and architecture-compatible. The controlled trials establish α = 1.0 as the optimal hyperparameter configuration, achieving peak average performance metrics of 0.9135 accuracy, 0.9194 precision, and 0.9151 f1-score. It is improved that the proposed method significantly enhances target recognition accuracy and provides a new method for acoustic perception in complex marine environments.