Mine Classification based on a Multiview Characterisation

In the context of mine warfare, detected mines can be classified from their cast shadow. A standard solution is to perform image segmentation first, and then to extract a set of features from the shape allowing classification in a final step. In this paper, we extend this procedure to a sequence of images obtained along a part of a circular trajectory of the sonar. For a given ground mine except mine with radial symmetry, cast shadow appearance generally depends on the point of view. Consequently, different features values can describe the same object. Whereas this often entails misclassification when a single view is used, we propose to use feature values computed over a sequence of images, especially its evolution, to characterise objects from multiple views. Our supervised classification scheme is based on the correlation, for each feature, between the sequence of values obtained from the unknown object and typical values related to each class.