New Approaches to Multibeam and Sidescan Sonar Data Processing

A fundamental aspect of the mission of the CCOM/JHC is developing techniques for improved hydrographic data processing. Our current research is focused on:

• CUBE
• The Nav Surface and BAG's
• GeoCoder
• FishPac LRSS Sonar

Related Publications for Data Processing

CUBE and Improved Uncertainty Management

PRIMARY CONTACT: Brian Calder

One of the major efforts of the Center has been to develop improved data processing methods that can provide hydrographers with the ability to very rapidly and accurately process the massive amounts of data collected with modern Multibeam systems. This data processing step is one of the most serious bottlenecks in the hydrographic "data processing pipeline" at NOAA, NAVO, and hydrographic agencies and survey companies worldwide. We explored a number of different approaches for automated data processing (see earlier progress reports for descriptions of these approaches) and, over the past four years, have focused our effort on a technique developed by Brian Calder that is both very fast (10's to 100's of times faster than the standard processing approaches) and statistically robust. The technique, known as CUBE (Combined Uncertainty and Bathymetric Estimator), is an error-model based, direct DTM generator that estimates the depth plus a confidence interval directly on each node point.

In doing this the approach provides a mechanism for automatically "cleaning" most of the data and, most importantly, the technique produces an estimate of uncertainty associated with each grid node. When the automated editing technique fails to make a statistically conclusive decision, it will generate multiple hypotheses, attempt to quantify the relative merit of each hypothesis, as well as present them to the operator for a subjective decision. The key is that the operator needs to interact only with that small subset of data for which there is some ambiguity rather than going through the current, very time-consuming process of subjectively examining all data points.

In 2003, CUBE was subjected to detailed verification studies in a cooperative research effort with NOAA that compared the automated output of CUBE to equivalent products (smooth sheets) produced through the standard NOAA processing pipeline. Verification studies were done in three very different environments (Snow Passage Alaska, Woods Hole, Mass., and Valdez, Alaska) involving surveys in various states of completion and comparisons done by NOAA cartographers. In each case the CUBE-processed data agreed with the NOAA processed data within IHO limits. CUBE processing took from 30 to 50 times less time than the standard NOAA procedures.

Based on these verification trials and careful evaluation, Capt. Roger Parsons, then director of NOAA's Office of Coast Survey notified NOAA employees as well as other major hydrographic organizations in the U.S. (NAVO and NGA) of NOAA's intent to implement CUBE as part of standard NOAA data processing protocols. As described by Capt. Parsons in his letter to NAVO and NGA, CUBE and its sister development, The Navigation Surface

"...promise considerable efficiencies in processing and managing large data sets that result from the use of modern surveying technologies such as multibeam sonar and bathymetric lidar. The expected efficiency gains will reduce cost, improve quality by providing processing consistency and quantification of error, and allow us to put products in the hands of our customers faster."

In light of NOAA's acceptance of CUBE, most providers of hydrographic software are now implementing CUBE into their software packages (CARIS, IVS, SAIC, Kongsberg-Simrad, Triton-Imaging, Reson, Fugro, GeoAcoustics and Sonartech Atlas, HyPack, QPS, and IFREMER). Dr. Calder continues to work with these vendors to ensure a proper implementation of the algorithms as well as working on new implementations and improvements. In particular, work is underway with GeoAcoustics to extend the principles of CUBE to phase comparison bathymetric sonars and to improve the uncertainty propagation equation for very shallow water and ultra-high resolution sonars. NOAA's use of CUBE has been supported through the development of a series of CUBE-related definitions for the NOAA Field Procedures Manual, and Specifications and Deliverables documents, and the development (with Dave Wells) of a "User Guide" for CUBE.

Inherent in the CUBE approach is the need for a robust error model for the sonar being used. This model should be provided by the manufacturer but unfortunately only a few manufacturers publish an error model for their system. In an attempt to develop approaches to extracting an error model from undocumented sonar as well as checking the manufacturer-provided models, Calder and graduate student Mashkoor Malik have been exploring field calibration methods for extracting error models directly from data. This approach has been applied to an EM3002 with POS/MV and the commercial navigation aid C-Nav., both appear to give realistic uncertainties, but further work is required. Within this context, Calder has also developed an "Uncertainty Patch Test" proposal - a methodology for capturing the data required to determine the uncertainties associated with a particular survey system. The proposal consists of a series of survey lines, much like a patch test, designed to isolate (as much as is possible) one component of uncertainty for each line or line-pair.

This year, Calder and Malik have used the acoustic test tank to capture data from the Reson 7125/400 kHz and the EM3002 sonars to confirm/extend the estimates of device-specific uncertainty for these systems and have continued the development of motion-related uncertainty models using time-series decomposition of the on-line measured motion series. Adding to the calibrations of various components of the overall uncertainty budget, Calder has also investigated the performance of the Fugro OmniStar HP/XP GPS positioning correctors and their likely uncertainty as a positioning source. This work has confirmed their advertised static position accuracy (0.03 - 0.04m rms horizontally and 0.04 - 0.05m rms vertically), although a number of problems were observed at higher output rates (5-10 kHz), including bad positions and bad NMEA messages. These problems have been reported to Fugro for further investigation.

Through our close collaboration (and shared field efforts) with NOAA hydrographers, it has become clear that many of the sonar systems (and particularly the Reson systems) used by NOAA survey vessels suffer from a problem when operated in steeply sloping environments due to limitations in the bottom detection algorithm's ability to deal with low signal to noise ratio situations. Calder has made a concerted effort to address this "Downhill Problem" resulting in several component algorithms that have helped mitigate the problem including a Turning Angles algorithm to utilize phase information in beam-to-beam offset vectors (with the ability to run with either GSF or HDCS data). An adaptive fusion system has also been developed to combine the component algorithms' results in order to improve the overall performance of the approach. The fusion algorithm is configured to automatically tune the weighting given to each of the component algorithms in real-time, and then combine them to give a probabilistic estimate of the best solution. In addition to the algorithmic approach to this problem an experiment was developed to determine the consistency of human operator performance on data affected by the downhill problem (so that the performance of the algorithm could be determined). Testing of these approaches on field data has shown that the fused solution is stronger and more robust than the individual component algorithm solutions. The University's Office of Intellectual Property Management has deemed the fusion algorithm (now called MASC'D-Multi-Algorithm Swath Consistency Detector) patentable and has filed an invention disclosure for it. The MASC'D has been packaged and released to SAIC for implementation in their software and a paper submitted to IHR for publication. This past year Calder has been working with NOAA graduate student Lynn Morgan to re-define CUBE's uncertainty propagation error equation using a more rational analysis of probable propagation distances in order to provide more stable uncertainty estimates in difficult conditions like steep slopes.

Related Publications for CUBE

The Navigation Surface and BAG's

PRIMARY CONTACT: Brian Calder

Another aspect of the data processing-related research being conducted at the Center involves rethinking of the final output products of a hydrographic survey. We strongly believe that the standard chart product of selected soundings and contours does not at all do justice to the information content of high-resolution multibeam and sidescan sonar data. We are working on a series of new products that will better serve the mariner as well as many other constituencies. In this context, Lt. Shep Smith developed the concept of the "Navigation Surface" as part of his thesis work. The Navigation Surface is a database that maintains bathymetric data sets at full resolution and a rule base for producing a series of derived products. Thus a single database can be the source for hydrographic (safety of navigation) products as well as products for fisheries habitat or other studies that require full-resolution data. CUBE-processed data can be a component of the Navigation Surface and thus the verification exercise described above also involved the navigation surface and tested (positively) the veracity of the navigation surface as a source of hydrographic products. The combination of CUBE and the Navigation Surface set the stage for a new research effort, "The Chart of the Future" described later.

In support of the new adoption of CUBE and the Navigation Surface into NOAA and NAVO processing protocols, the Center has organized and hosted two workshops with industry, government and academic researchers to define and maintain an open format for Navigation Surface products that will ensure interoperability amongst the various vendors and agencies. The first workshop (in 2004) defined the specifications for the Bathymetric Attributed Grid (BAG) Object, and obtained the commitment of many of those attending (NOAA, NAVO, CARIS, IVS, SAIC, 7C's) to provide source code and adopt the approach. Cross platform libraries for digital signatures have been built and tested and an example certificate construction and management scheme (based on HASP network dongles) has been implemented. In 2005, a DLL version of the Open Navigation Surface (ONS) library was created along with a Visual Basic support application to allow demonstration of the security principles of the ONS project in a graphical context. A second meeting of the Open Navigation Surface Working Group (ONSWG) took place in July 2005 resulting in a build-out of the source tree for the BAG and the linking of stand-alone applications to read and write BAG-format files. In addition this meeting adopted the digital signature system for BAG's (which supports simple XML files) using both private and public keys, with encryption and thus avoiding the need for hardware tokens. The BAG library was demonstrated at the Shallow Survey 2005 Conference in Plymouth, England with trial integrations into IVS Fledermaus, CARIS HIPS, and SAIC SABER software packages. ESRI has now expressed interest in integrating the BAG format into their software (and becoming a member of our Industrial Consortium - see below) and NOAA OCS has recommended the BAG as the transmission form of data from OCS to NGDC. A website for the ONSWG has been established: http://www.opennavsurf.org.

The Center, under the direction of Calder, has taken the lead in the ONS development, providing code development, documentation and a File Specification Document for the ONS project. Calder has chaired several ONS working group meetings to generate the code-base for a beta-test Candidate Release (CR) and Full Release (FR), established SubVersion software management environment, and updated a project website to support this. The CR/FR process was completed with full source-code being released to the web-site on 14 April 2006 and the initial version 1.0.1 on 2 May 2006. Calder presented work on ONS to the IHO Technical Standards Maintenance and Development (TSMAD) working group; the BAG has been proposed for adoption as part of the IHO S-100 suite of standards for hydrographic products (i.e., the successor to S-57). Calder continues to work with David Parker of the UKHO on the development of the IHO S100 Hydrographic Data Product Standard (HDPS) and on the definition of an initial metadata strawman to be used for the upcoming UKHO bathymetric database and the S100 HDPS.

As a focal point for the broad interest in CUBE and the Navigation Surface amongst government agencies and the private sector, the Center is fulfilling one of its prime mandates of serving as a national center of focus for ocean mapping activities. In bringing these groups together we facilitate efficient and collaborative research efforts aimed at solving a national need.

GeoCoder

PRIMARY CONTACT: Luciano Fonseca

While our initial data processing efforts focused on improving bathymetric processing, it has become increasingly clear that there is also a great need for improved processing of backscatter data (both from Multibeam Sonars and Sidescan Sonars). With this in mind, we began, in 2005, a new effort aimed at improving the suite of backscatter processing tools available to us and NOAA. Our aim was two-fold: 1- to develop easy to use tools that will generate "pretty" images of Sidescan sonar or Multibeam backscatter that will be suitable for small object detection as well as geologic and habitat interpretation, and 2- to develop tools that allow for the quantitative analysis of backscatter data in support of seafloor characterization and small object identification.

In an effort to meet these two objectives, we started a lab-wide effort to develop a new suite of backscatter processing tools. This effort is being led by Luciano Fonseca with input from many others. The goal is to create an integrated suite of tools that will allow us to import backscatter or Sidescan data from a number of sensors (in various forms and formats); convert these data to an internal GFS format, correct these data (where possible) for source levels, beam patterns, gains, area ensonified, attenuation, and local slope, and then either analyze and/or display these data in a georeferenced mosaic. The result of this is GeoCoder, a C++ mosaicking tool that reads Multibeam or Sidescan sonar data in GSF, XTF or a range of native formats, and applies a series of radiometric and geometric corrections to the data including corrections for beam pattern effects. Normally, the empirical beam pattern correction is calculated as the residual necessary to flatten the angular response registered by the sonar system, i.e. to normalize the backscatter at 45 degrees, (sometimes adding a Lambertian correction). The approach used by GeoCoder calculates the beam pattern as the residual to the modeled angular response of the ensonified seafloor which then reveals the actual non-linearity of the transducer angular response. Data is then geocoded in a projected coordinate system using an interpolation scheme that emulates the acquisition geometry.

A feathering algorithm smooths the transition between overlapping lines and an anti-aliasing algorithm that makes it possible to produce a lower resolution mosaic that is not degraded by aliasing. Slant-range is corrected for based on actual bathymetry, and a trend-adaptive angle-varying gain helps remove artifacts that appear when different bottom types are found along a single swath. Lines can be removed or remosaicked, and the overlap area between parallel lines can be controlled by filter parameters. GeoCoder also supports a statistical package that identifies patterns in the backscatter response that can be used in support of seafloor characterization (see below). Statistics calculated for backscatter bins include: mean, mode, range, minimum, maximum, standard deviation, variance, percentiles, quartile range, skewness, kurtosis, moments of any order, and also parameters extracted from a gray level co-occurrence matrix (contrast, homogeneity, dissimilarity, entropy and energy). Taking advantage of the corrections made to the backscatter, GeoCoder also serves as the front end for a new and very exciting approach to using Multibeam backscatter data for seafloor characterization called ARA (Angular Range Analysis - formally known as AVO). The ARA tool will be reported on in the seafloor characterization section.

Improvements to Geocoder in 2007 include approaches to better filtering of the Sidescan sonar data collected from towed vehicles. The navigation of towed sonars is normally noisy and not reliable; causing artifacts when the data is georeferenced into a mosaic. Improvements were made to the filtering and correction of navigation files. For that, the navigation is decimated and interpolated with splines, creating an optimal smooth navigation path. The interpolated path can then be used for a better estimation of the "course made good," which is the best option in the absence of reliable measurements of the tow-body heading. A bottom-detection feature has also been added that can extract an estimated tow-body height, so that proper slant-range corrections can still be made when vehicle altitude information is not provided with the data (as is often the case with towed Sidescan sonar systems). Slant range corrections can now also be done using an external bathymetric model, obviating the need to make a flat-seafloor assumption. Finally, several new approaches for removing the angular dependence of backscatter from the mosaic have been implemented (theme based, patch based, sediment type based) as well as correctors designed specifically for correcting nadir irregularities that have been identified in Kongsberg-Simrad low-frequency (12 and 30 kHz) data.

Since its development, GeoCoder has become a simple-to-use tool for generating a Sidescan sonar or backscatter "mosaic" which has been greeted with much excitement in the community. There has been tremendous interest in this software throughout NOAA, from our industrial partners, and other academic institutions. This has led to a number of licensing requests as well as requests for training. We have now offered two training short courses. A recent email from one of the attendees (from the Biogeography Team of NOAA's Center for Coastal Monitoring and Assessment) said "We are so pleased with GeoCoder! We jumped in with both feet and made some impressive mosaics. Thanks so much for all the support." An industrial partner collecting massive amounts of "awful" backscatter data in the Indian Ocean tried GeoCoder and it resolved their data quality problems. Given the high demand for use of GeoCoder, the list of systems that it supports (and the list of users) is growing quickly. Several new systems and formats (Reson 7K, Simrad 2100, EA600, Benthos C3D and Hypack HSX) were added to the support list this year. The complete list of systems and formats supported is now:

• Kongsberg/Simrad multibeam .all (beam time series and beam average)
• Simrad Sidescan
• Reson (.xtf, .s7k), snippets, beam average and sidescan
• Klein sidescan, sdf, sdf2
• XTF sidescan (various sonars)
• GSF multibeam (various sonars, beam average and snippets)
• HSX sidescan (various sonars)
• Seabeam (beam average and sidescan)
• Geoswath (.rdf)
• C3D (.xtf)

In further support of our backscatter (and other) processing efforts, Brian Calder has developed and licensed (to industrial partners SAIC and GeoAcoustics) software to convert GeoAcoustics data to GSF format; a prototype to convert the native GeoSwath format (RDF) into GSF has also been developed.

The value of GeoCoder is also demonstrated by the growing interest from our industrial sponsors.

This past year Triton, Reson, Hypack, Fugro and CARIS all negotiated licenses for Geocoder, bringing the total list of licensees to:

• IVS
• Caris
• Reson
• Fugro
• Triton
• Hypack
• Chesapeake Technology

Additionally a number of NOAA programs and academic partners are actively using GeoCoder, these include:

• NOAA SANCTUARIES
• NOAA Alaska Fisheries
• NOAA Pacific Coral Reef Program
• NOAA Ship Thomas Jefferson
• NOAA Ship Rude
• NOAA Ship Fairweather
• NOAA/JIMAR Coral Reef Ecosystem Division
• Jacobs University Bremen, School of Engineering and Science
• University of Galway
• University of Ulster, Northern Ireland
• Oregon State University
• University of Saint Andrews
• Geol Survey of Canada
• CIDCO-Le Centre Interdisciplinaire de Développement en Cartographie des
• Océans
• Stockholm University, Department of Geology and Geochemistry
• Alaska Department of Fish and Game
• University of Illinois at Urbana-Champaign , Departments of Geology, Geography and Civil Engineering

Beyond GeoCoder we have developed an analytical tool (Angular Response Analysis - ARA-formally called AVO) that uses the variations in the amplitude of the return as a function of the angle of incidence to predict the nature of the seafloor (sand, silt, clay, etc.). The Office of Naval Research initially funded this work (their interest is in remotely identifying seafloor properties for sonar propagation and mine burial models), yet the application of this technique to fisheries habitat studies is clear and there has been great interest in its use by a number of NOAA labs and researchers. ARA will be discussed further under the theme of seafloor characterization.

FishPac LRSS Sonar

PRIMARY CONTACT: Lloyd Huff

In August 2006, Huff and others participated on a cruise in the eastern Bering Sea on the NOAA ship Fairweather and collected acoustic backscatter data at 38 kHz, 40 kHz, 100 kHz, 180 kHz and 455 kHz which were loaded into a new FISHPAC server at the Center in order to provide the basis of numerous studies leading to an improved understanding of acoustic backscatter. A paper was prepared describing the backscatter characteristics of a patch of gastropods that were encountered on one of the eastern Bering Sea survey lines. The preparation of this research paper provided an opportunity to push the applications of GeoCoder. Although sufficient metadata did not exist to determine the actual angular backscatter characteristics of the gastropod patch, it was possible to demonstrate the frequency dependent change in angular backscatter characteristics of the patch, compared to a close by area without the presence of gastropod shells.

The eastern Bering Sea backscatter at 455 kHz was obtained with the Klein 5410 sidescan sonar. The backscatter at 45 deg grazing angle were extracted from the data set and subjected to analyses. The first analysis was the mean backscatter value at 20 sites where NMFS has a 15-18 year record of fish catch. The backscatter generally increased as the sites moved from the SW to NE end of the survey track lines. The second analysis was to conduct spatial spectral estimates. This is an important piece of the puzzle of what spatial resolution of acoustic data are best suited for mapping fish habitat in the eastern Bering Sea. The backscatter generally followed a power law with decreasing energy at smaller spatial scales. However, there appeared to be two break points where the slope of the spectra changed. The preliminary conclusion is that there are two spatial scales that may be important to sample. One is at 50 m and the other is at 250 m.

The backscatter at 38 kHz was processed toward the end of performing spatial spectral estimates. Since the backscatter at the two (38 and 455 kHz) different frequencies are dominated by quite different mechanisms it was thought that such comparisons may be very helpful in developing a strategy to establish a priority plan for analyses of the backscatter observed at 40, 100 and 180 kHz. The analysis of the 38 kHz data led to two important conclusions about the system that was used to collect the data. The first conclusion was the determination that 80 percent of the 38 kHz data were saturated and therefore of no use, even for relative changes in received backscatter levels. The saturation problem was tracked to a coupling between the fixed and time varying gain of the 38 kHz channel and the fixed and time varying gain for the sidescan sonar. When the sonar operator changed the gain to enhance the sidescan imagery, the backscatter data from the vertically looking single beam 38 KHz were driven into saturation. The second finding from the analysis of the 38 kHz data was that the channel was excessively noisy and that the I's & Q's did not have a zero mean. Investigations into the non-zero means led to the "undocumented" fact that the 14-bit A/D samples had been cast into the upper 14 positions of a 16-bit word, while the lower two bits were used as system status bits. The root cause of the excessive noise was traced to a design flaw in the system, whereby the backscatter signals were attenuated then passed through unshielded connectors; as opposed to being attenuated after having passed through unshielded connectors, which would have attenuated any electrical noise that was picked up in the unshielded connectors.

Related Publications for FishPac LRSS Sonar