Objective: Artificial intelligence is a part of computer science that emphasizes the creation of intelligent machines in defense equipment and military equipment. Intelligent systems for automatic underwater target recognition are increasingly used in passive sonar to reduce human intervention and related challenges in accurately identifying vessels. Today, highly advanced methods of machine learning and deep learning are being used by the world's navies to identify acoustic targets.
Methodology: In this article, recent works in the field of automatic underwater acoustic target recognition are reviewed, and a new method based on deep learning algorithms is presented. In this method, first, the raw audio signals are received from the hydrophones, and after performing the necessary pre-processing, using the Short-time Fourier transform, the spectrogram images related to the passive sonar acoustic data are generated and fed to the model layers for model validation and classification.
Findings: The obtained results show that the multi-layer structures of the proposed model can automatically extract several features required for the classification of different ship classes. In this article, common deep learning algorithms are used to identify targets, which can increase identification accuracy and reduce evaluation errors by searching for the most informative features of sonar data.
Conclusion: The obtained results show that the recognition accuracy of the proposed model is more than 97%, and its validation loss is less than 3%. In this method, with the relative improvement of classification accuracy, the speed of target recognition has increased significantly.