Baseline Classification of Acoustical Signatures of Mine-Like Objects

Classification de Base des Signatures Acoustiques des Objets Semblables aux Mines


  1. Sildam, J
  2. Fawcett, J.A.
Corporate Authors
Defence R&D Canada - Atlantic, Dartmouth NS (CAN)
We investigate the problem of the detection of mines and the sensitivity of a binary classifier to mine types using a broadband signal incident upon an elastic object. In the course of the sensitivity analysis a realistic situation is simulated when a binary classifier that is optimal for a part of collected- or modeled data turns out to be sub optimal for new, previously unseen data. In general, such a problem corresponds to a generalization problem when for the purpose of classification analysis, the data is set is divided into three parts: training data, test data used for classifiers’ fine tuning via cross-validation, and test data not used in the cross-validation. However, in addition to such division, we ensure that one or more types of mines’ acoustical signatures are left out of training data, as well as out of test data used for cross-validation purposes. We call such approach a Baseline Classification. To create a data set of mine-like and non mine-like objects that includes the required diversity of mine types, we use analytical models for the elastic shelled spheres and for infinite elastic shelled cylinders. We use a total of 9600 spectra. The spectra can be divided according to their shape, shell material, filling, relative thickness, and surrounding environment into eight types of mines and nonmines. Boundary between mines and non-mines is defined by the relative thickness of shells and it is found that a threshold of 3.5% is optimal. The shells thinner or

Il y a un résumé en français ici.

pattern analysis;classification;wideband sonar signal;mine-like objects
Report Number
DRDC-ATLANTIC-TM-2005-058 — Technical Memorandum
Date of publication
13 Jul 2005
Number of Pages
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: