Statistical Methods for Automatic Identification of Seabed

Resumo

This work proposes the use of statistical methodologies based on unsupervised classification for the automatic identification of seabed types in coastal areas. For this purpose, acoustic data obtained by means of a simple beam echo sounder (at 200 kHz) coupled to a small ship have been used as discriminant features. Each of the resulting acoustic curves has been preprocessed through the application of time corrections (elongation of the echo with depth), power (attenuation of the wave with distance), and ping length (deformation of the echo due to distance), with the aim of eliminating its dependence with respect to depth. The experimental data have been obtained in a controlled environment in the region of Cabo de Palos (Murcia, Spain), studying three different types of bottom: sandy, sandy with sparse vegetation, and rock. The statistical techniques adapted and applied to this particular case belong to the cluster classification from time series. In fact, taking into account that in actual identification problems the existing fund classes are not known in advance, the problem of identification has been addressed through an unsupervised classification perspective based on the previous calculation of dissimilarity matrices and the application of hierarchical cluster classification methods. The results obtained, correctly identifying 93% of the total funds—with little confusion between their classes—support the use of automatic classification techniques in this area for the correct characterization of the seabed.