Optic-Acoustic Seabed Classification

The Use of Information Criteria in Quantifying Sea Floor Classification Data

Cooperative Research
Optical and acoustical data processing are two powerful remote sensing techniques for sea floor classification. Following a number of visits to the University of Miami Rosenstiel School of Marine and Atmospheric Science (RSMAS) it became clear that there was potential of a marriage between the QTC approach to acoustic classification and the RSMAS scientists processing of hyperspectral data for seabed classification.

The result was a collaborative research project to explore the potential for the fusion of these two data sets to extend classification capabilities beyond traditional methods.

Acoustic Seabed Classification
The amplitude and shape of an acoustic signal reflected from the sea floor are determined mainly by the sea floor roughness, the impedance mismatch between water and the sediment, and reverberation within the substrate. Remote classification of the sea bottom requires an acoustic data acquisition system and a set of algorithms that analyze the data to determine the seabed acoustic class.

Subsequent processing relies on the extraction of characteristic features from the seabed echo. Using the first echo-return, the window surrounding the seabed is analyzed by a series of algorithms that derive 166 feature descriptors. Some of these features are based on echo shape, and others on spectral characteristics. Classification implies some kind of ordination technique to group echoes with similar features. Non-acoustic data, from direct sampling or observation, is usually used to relate the acoustic classes to the physical properties of the substrate.

Principal components analysis is used to reduce the 166 features to three "Q" values representing linear combinations of the features most useful in distinguishing seabed types. Points defined by the triad of Q coordinates are plotted in 3-dimensional space for visual inspection of clustering. Class assignments are based on multivariate distances between echoes to be classified and clusters representing the acoustic properties of selected seabed types.

Optical Seabed Classification
Until recently, optical classification of the sea floor has used aerial photography and multiband sensors such as Landsat™ to discriminate major bottom types such as sediments, seagrass, and corals. New techniques for collecting hyperspectral imagery greatly extend the potential for more detailed discrimination of bottom types.

Recent work in the Office of Naval Research's ( ONR ) program on Coastal Benthic Optical Properties has shown that distinct bottom types have unique reflectance signatures. Sediment reflectance is, for example, affected by mineralogy, grain size, grain roughness, microalgal pigments and polymer concentrations. In addition, hyperspectral data can be used to estimate densities of sea grass.

Techniques for analyzing hyperspectral data and extracting bottom information are, however, still in early phases of development.

Early Collaboration
In a novel approach to analyzing hyperspectral optical data, RSMAS and Quester Tangent collaborated to use QTC IMPACT ™ acoustic seabed classification software to process hyperspectral data collected using a tethered spectral radiometer buoy (TSRB). The classification was based on feature extraction and multivariate statistical analysis of spectral shape. Preliminary results using HTSRB data from May 1999 indicate the potential usefulness of this approach.

Preliminary Test Results
In the first preliminary test indicated here, QTC IMPACT ™ clustered Lu/Ed ratios from 10 records of HTSRB data into nine classes.

These classes are 100% site specific.

Some class distinctions no doubt correspond to changes in spectral shape due to differences in water depth. Where water depths are equivalent, however, data clusters show an exact correlation with sediment type measured at the site.

Data Acquisition
In the summer of 2001 a QTC VIEW™ Series V research system was used to acquire acoustic seabed classification data from the seabed off Lee Stocking Island in the Bahamas. A TSRB was used to acquire hyperspectral data from the sea floor simultaneously with the acoustic system. Survey lines were set to cross as many features as possible. All data were logged, with quality assurance performed in the field. Ground truthing information was captured by videography.

Data Processing
The next phase in the project is processing the data. There will be a number of steps in this process:

• The optical and acoustic data will be processed individually. Regular unsupervised classification will be performed using QTC IMPACT ™ to develop a classification of both the optical and acoustic diversity of the surveyed area. The data will be plotted to form maps of optical and acoustic diversity.
• Next, integrated "opti-acoustical" processing will investigated. This requires concatenation of the individual feature sets to generate a combined feature set. Multivariate statistical processing will be performed on the integrated data. An "opti-acoustic" diversity plot will be made from the results and a correlation analysis made with the individual plots.
• A third component of the processing will be an evaluation of results against ground-truth data. Ground-truth information will be compiled and presented in a GIS. The "opti-acoustic" data will form layers in the database. All information will be spatially modeled with the following goals:
  • Evaluating the utility of optical and acoustic mapping for shallow water bottom mapping using feature extraction and statistical processing methods.
  • Evaluate the utility of integrated "opti-acoustic" processing in relation to individual optical and acoustic methods.

Partners
This work is a collaborative project between Quester Tangent and Dr. Pam Reid of the University of Miami Rosenstiel School of Marine and Atmospheric Science. It is sponsored by the Office of Naval Research ( ONR ).

Publications
Shape Analysis of Hyperspectral Bottom Reference Data: Application to Remote Sensing of Benthic Habitats
Louchard, E.M., Gleason, A.C., Reid, R.P., Collins, W.T., and Mobley, C.D.
American Geophysical Union Spring Meeting, Washington, D.C., May 28-31, 2002

Contacts
For more information contact Bill Collins at Quester Tangent or Dr. Pamela Reid at RSMAS .