The UK's vast marine territories are monitored by the Royal Navy using high tech sonar equipment. Sonar systems emit and receive acoustic signals which can be analysed to detect underwater objects and map our subsea environments.
The Defence and Security Accelerator 1 (DASA), a part of the Ministry of Defence (MoD), contracted the IMI and Systems Engineering & Assessment Ltd. (SEA) to develop an AI algorithm capable of automatically classifying underwater environments directly from sonar measurements.
Developing AI algorithms to classify underwater environments
Underwater environments vary hugely in terms of water temperature, salinity and depth as well as seabed slope and composition, all of which affect sonar. IMI and SEA first analysed the many characteristics of underwater environments and classified them into different types.
We then reviewed various AI techniques to determine the most effective classification approach. The selected method (Probabilistic Generative Modelling) was then adopted as a framework to develop three different AI algorithms for identifying underwater environments.
A Probabilistic Principal Component Analysis (PPCA) model proved to be the basis for the most effective algorithm. We then developed the model further through rigorous experimentation to achieve the highest possible classification accuracy.
Classification accuracy of up to 96%
After developing the AI algorithm, we tested its performance on a wide range of simulated acoustic data representative of a broad spectrum of underwater environments.
The tests demonstrated that our PPCA algorithm can classify underwater environments from simulated sonar measurements with an average accuracy of 93%. Classification accuracy remained high even when we used a short spatial interval of the test data, which is promising for the practical use of the technique.
An alternative Latent Variable Gaussian Process (LVGP) model also showed strong performance and enabled us to achieve an even higher classification accuracy of 96%.