Seabed Sediments Classification using Multifrequency MBES Bathymetry Data.
Khomsin (a,b), Mukhtashor (b), Danar Guruh Pratomo (a), Suntoyo (b)

(a) Geomatics Engineering Department ITS Surabaya

(b) Ocean Engineering Department ITS Surabaya

Abstract

Recently, most bathymetric surveys utilized multi-beam echo sounder (MBES) multi-frequency on a ping-by-ping basis, meaning that the depth is measured with different frequencies in several consecutive pings. However, the measured inter-frequency depth in a similar location may be different due to the varying capability of a frequency in penetrating the signal through water and to the seabed sediment. Nevertheless, the depth difference opens the possibility of conducting seabed sediment classification. The hypotheses are that the higher and smaller differences indicate a softer (mud) and more rigid (coral) sediment, respectively. This study attempted to perform seabed sediment classification using a neural network model for testing the hypotheses using MBES multi-frequency data in Bedford Bay of Canada in 2017 obtained from the R2Sonic Multispectral Challenge. Input to the model is the difference of measured depth from frequencies, resulting in a sediment class as output. With 19 stationery-labeled datasets, the test was conducted in a 10-fold cross-validation. The average accuracy was 98.5%, indicating a positive correlation between the thickness of the sediment and the type of sediment that was validated.

Keywords: bathymetric difference, multifrequency, sediment classification, neural network

Topic: Ocean Sciences