TARGET CLASSIFICATION NEAR COMPLEX BOUNDARIES

摘要

This paper presents a new approach for target classification in a realistic sonar context, based on a time-frequency filtering matched to a reference target response. An experimental data base is set up, containing acoustic responses of specific target (a spherical shell) among several other targets (solid spheres and solid cylinders with hemispherical endcaps, infinite cylindrical shells). The targets are immersed in a tank with different boundary configurations (free field, semi-infinite with various interfaces). The time-frequency filtering method is first applied to the whole data base in order to discriminate the reference spherical shell among all the other targets. The recognition rate of the spherical shell (resp. the other targets) is 100% (resp. 95%). The performance of the classification scheme in the presence of simulated noise is then investigated and more than 80% of good classification are obtained in all cases.