Reactive Data Gathering for Underwater Mine Countermeasuresy
Jordi Barrz, Mark Williams, David Nicholson · 2011
We demonstrate how mathematical techniques for Reactive Data Gathering (RDG) can be used to increase the probability of correctly classifying seabed threats. The problem is one of separating underwater objects into mine or non-minelike bottom object (NoMBO) classes using noisy and incomplete sequences of Sonar data. The incorporation of new individual sensor observations is undertaken at two levels. We operate on sensor output directly using Sequential Importance Sampling (otherwise known as the Particle Filter). We also undertake inference at the decision level using confusion matrix output from an Automatic Target Recognition (ATR) algorithm. Predicted observations are used as the basis for our RDG scheme, which seeks to maximise the one-step-ahead information gain. RDG is evaluated against current operating procedure and standard benchmarks, e.g. rosette patterns or random views of the targets. We use a Sonar simulator developed by the Ocean Systems Laboratory at Heriot-Watt University and, where appropriate, a BAE Systems-developed ATR to conduct a series of experiments to test the benets