Passive Underwater Target Classification using Sparse Models

Underwater audio signal classification has shown a wide range of applications in real-life scenarios, both for civilian and defence purposes. Passive sonars are ubiquitously used for underwater target recognition based on the sound waves emitted by the target. The radiated waves are classified to identify their source and to classify them as a threat or friend in a defence scenario. Machine learning algorithms are reported to impart good performance on audio classification problems. The paper proposes a sparse model-based algorithm as a means to classify the acoustic data received by passive sonar. A comparative study is conducted on an open dataset, namely the DeepShip dataset, a publicly available sonar audio dataset. Results show that the use of the proposed sparse model based algorithms for underwater passive target recognition is a promising direction for further research and applications.