Seabed sediment classification of side scan sonar data using histogram approach

Abstract

Side scan sonar (SSS) delivered an advantage for sediment classification studies. In particular, the availability of backscatter intensity offers an alternative method to study seafloor hardness and softness. Numerous seabed mapping processes used many kinds of classification techniques can produce sediment maps from backscatter images, ranging from simple clustering to machine learning approaches. This study aims to perform a pixel grouping method for backscatter images from SSS using the histogram generated from the backscatter intensities. The aim will be achieved through three (3) objectives; to classify the seabed characteristic using histogram classification, to produce a sediment map through a histogram classification created and, lastly to test the model's validity using ground-truthing data ground-truthing data such as sediment distribution and coral video transect. Acoustic data from the SSS data acquired in Labuan Marine Park was used in this study. The 900 kHz side scan data was processed and corrected using SonarWiz 7 software. The data were then categorised the pixel intensities based on the histogram shape. A few data classify techniques were tested to produce classification maps using equal intervals, quantile methods, natural breaks, and, geometrical intervals. Classification maps derived from these methods were then validated with ground truth samples collected using underwater videos and sediment grabs to assess their accuracies via qualitative assessment. The result shows that geometrical interval was the only method that relatively complemented the ground truth data and works reasonably well. Therefore, this can be a good tool in designing management programs for the marine park to know the general view of the sediment distribution in that area. It creates a simple and straightforward but statistically robust objective of general overview based on geophysical data provided.