Using Mine Danger Maps Based on Bayesian Networks to Evaluate an Autonomous Hunt for Multiple Naval Mine Types

This article presents a new approach to the evaluation of autonomous naval mine hunting called the mine shape Bayes net (MSBN). MSBN models the presence of various mine shapes over a 2-D grid covering the seafloor of the mined area. The cells of the grid are small (∼4 m2) and numerous, but the approach works by attaching a Bayesian network to each of them. These networks model the effects of local mine hunting operations on the probability that a mine remains in the cell. They also learn what type of mine might be in the cell by fusing acoustic and visual classification results obtained during these operations. Such fusion accounts for the quality of classification from each source, as well as for the difficulty of navigation underwater. The quality of acoustic classification is measured by applying recent through-the-sensor assessment techniques to produce a confusion matrix. While the approach provides standard evaluation estimates, like the number of mines remaining of each type, it also offers new capabilities. It can make mine danger maps that reveal how the threat posed by different mine classes varies over the minefield area. MSBN also provides a new risk metric that applies to a proposed route through the minefield. This metric measures the probability that the first ship traversing the route would suffer critical damage from mines. It opens the door to optimization of the route taken.