Seabed type clustering using single-beam echo sounder time series data

Acoustic seabed characterisation is known to make available very useful information for various marine related applications. Most existing single-beam echo sounder (SBES) classifications require ground truthing in order to have confidence in the final results and sometimes to relate labels to geological features. A statistical feature extraction approach, with an emphasis on quality control, is proposed as an alternative to extensive ground truthing. Survey data collected from Malin Sea during 2003 is used as a test bed. Principal component analysis (PCA) and k-means as well as quality threshold (QT) local clustering are employed to label the echo return samples with an arbitrary number of classes. Results show that proper quality control enables more geologically realistic seabed type clustering. Key-Words: single-beam; echo sounder; seabed; clustering; PCA; k-means; quality threshold