Use of seafloor stereo-images to validate automatic classification of benthic habitats


Here we present a technique for automatic classification of seafloor data collected during the 2012 HABCAMV4 cruises led by NOAA, UNH and WHOI a federally funded long term project part of the annual NOAA seascallop’s survey.

This project will analyze a unique data set that includes simultaneously collected data such as:

  • Hi-resolution multi-beam (digital bathymetry and backscatter intensity)
  • Seafloor stereo image data (e.g. species and substrate)
  • Environmental parameters (e.g. temperature, salinity, water turbidity)

The analysis will be based on an unsupervised spatial clustering (K-means) of a combination of several predictors like morphological features (curvature, rugosity, fractal index, surface area) and backscatter intensity.

The final results will be validated by analyzing the identified classes with a randomly selected subset of underwater photographs for each class.

The seafloor classification map produced is then used as a preliminary "habitat classification" for further classification. It can be reused to define selection-criteria for the underwater images used by automatic classifier or by manual image annotator tools.

Results from this project will also help to define new survey track-lines prior to and during HABCAM surveys.


Center for Coastal and Ocean Mapping

Publication Date


Journal or Conference Title


Conference Date

May 5 - May 9, 2014

Publisher Place

Lorne, Victoria, Australia

Document Type

Conference Proceeding

This document is currently not available here.