By: Tom Reed, president of Oceanic Imaging Consultants, Inc.,

INTRODUCTION
The acquisition of seafloor mapping data has matured remarkably in the past decade. This article reviews some of these advances, with particular attention to how some new technologies have effectively dealt with "data-overload" made possible by sensors with greater resolution and bandwidth.

Seafloor mapping - while admittedly potentially a complex, detail-intensive task involving numerous electronic sensors, telemetry systems and logging/display devices - can be, and often is best thought of, as making pictures of the bottom of the sea. Pictures, as we all know, contain a lot of information. A typical 35mm color slide, at 9000 scan lines per inch and approximately a 4x3 aspect ratio, contains over 2.5 Giga-bits of information. Clearly, mapping the seafloor can be highly data-intensive, resulting in data overload on both operators and logging devices. "The Matrix" notwithstanding, people do not deal gracefully with scrolling alpha-numeric byte-streams. Scrolling pictorial representations (waterfalls, graphic recorders, etc.) are the usual solution, but often with compression in both physical resolution, and data dynamic range. With recording mimicking display, loss of resolution, or reduction in survey coverage and pace, was considered inevitable. These mandatory losses are now a thing of the past. Sustainable, 24-7-365 reliable logging and real-time processing of seabed mapping data at rates in excess of 10 mega-bytes per second is available in commercial, off the shelf PC-based packages today.

HISTORICAL PERSPECTIVE
At a first glance, seabed mapping systems can be categorized into two families: acoustic systems, such as echo-sounders, sidescans, multi-beams and swath-interferrometric devices, and non-acoustic devices, such as traditional cameras and videos, electronic still cameras (ESC's) and laser linescan systems (LLS's). We recognize that profiling systems such as single and multi-channel seismics, magnetometers and gravimeters are equally important mapping tools, but for the moment restrict our analysis to systems providing imaging in the traditional planimetric format.

A typical two-channel analog sidescan system actually provides a large amount of information. Fortunately, roll-paper records are quite efficient at gathering this, albeit somewhat difficult to carry around in significant quantities. When taken to the digital domain, a quantitative notion of the data available to us becomes apparent. Assume a reasonable digitization rate of 24 KHz, yielding a constant slant-range resolution of 3 centimeters. Assume 16-bits sample resolution to encompass the likely range of raw backscatter variations in amplitude. This implies a data rate from a simple two-channel sidescan of approximately 100 Kbytes per second, or 360 megabytes per hour. This will fill a Zip-disk in 15 minutes, and a 1 GB Optical platter in under 3 hours. To people planning a 30 day cruise, this begins to look expensive. In steps downsampling.

Perception is everything, they say. A high-quality roll-paper recorder can display 4096 samples across a scan, and you don't need to hire a programmer to show your boss the data. For the typical sidescan mentioned above, this would allow you to show full resolution data for any swath of 50 meters range or less. At any range greater than that, you would have to throw away some of the data. At 100 meter range, you would have to throw away half the data, for example. Furthermore, the human eye can only distinguish between 16 and 32 shades of grey. This translates to retaining at best one third of the original dynamic range. If you take this same data to an XGA computer monitor, you are limited to only 1024 samples across a line, and a depth resolution of 8-bits. Let's assume you may have 2 screens, so now, in theory, you would need to log only 1024 samples per side, 8-bits per sample to make maximum use of your digital display system. If you operate at 100 meters range or greater, and only log what you display, you have irretrievably discarded over 75% of your data. No amount of post-acquisition signal processing magic can bring this lost data back. Furthermore, any operator-induced changes to gains, contrast stretches, etc. are likely to be permanent. Logging raw data at full resolution and full dynamic range obviates these problems. It just requires media space, and through-put bandwidth.

NEW DEVELOPMENTS
In 1997, the United States Naval Oceanographic Office (NAVOCEANO) embarked upon a fleet modernization program for its T-AGS 51, T-AGS 60 and T-AGS 45 class oceanographic survey vessels and accompanying Hydrographic Survey Launches (HSL's). Mission requirements included high-resolution sea floor mapping and object location/identification in littoral areas of the world, typically under adverse environmental conditions. Specific requirements included simultaneous acquisition, display and logging of dual-frequency sidescan sonar data, single and multibeam bathymetry, and beam amplitude and backscatter. Per the specification, the maximum anticipated data rate would be up to 1 Gigabyte per hour, which the sonar data acquisition workstation would be required to log to both tape and hard-drive. Anecdotally, this prompted the requirement for at best a 9 GB hard drive, as it was mentioned that "...no government worker was ever required to work more than an 8 hour day...".

DOUBLE OR NOTHING...
In the fall of 2000, OIC began work on an interface to the Raytheon LS-4096 Laser Linescan system. The LS-4096 provides continuous 14-bit imagery from a scanning laser mounted on an underwater vehicle. Resolution is configurable from 512 to 4096 samples per scan line, at scan rates up to 4000 RPM. At maximum resolution, the LS-4096 delivers 2.2 Mbytes per second, at approximately 267 scan lines per second. This data rate doubled that which had been required of GeoDAS in the past, and exceeds conventional sonar imaging requirements by at least a factor of 100, if not more. The original top-side processor allowed saving of user-selected "snapshots", but any continuous data logging could only be accommodated on video tape, which could not take advantage of the full imaging resolution potential of this system. Any geo-coding of data required manual "cut-and-paste"operations post acquisition.

GeoDAS-LLS provides a complete control, processing and display interface to the laser linescan system, treating the data stream as a single-channel sidescan (albeit, a VERY FAST sidescan). This includes full resolution raw data logging, as well as realtime and post-acquisition processing, targeting and geo-coding. The laser data can even be merged with co-registered sidescan data, to provide an ultra-high resolution "gap-filler" directly beneath the track of the sonar. Co-registration of sidescan and laser imagery provides significant improvements for search and recovery, bottom characterization and mine-hunting operations, with the laser data offering the potential of on-the-fly target identification, augmenting the sonar's extended abilities for target detection.

DATA INTEGRATION
While the above examples demonstrate available solutions for the mechanics of multi-sensor, high-rate data acquisition, there remains the issue of operator overload, and interpretation. Simply put, someone has to look at all this. If we can now simultaneously acquire swath acoustic data, providing a half-kilometer swath of acoustic imagery and bathymetry, and optical data, providing sub-centimeter scale imagery over a patch which might barely cover the nadir footprint of the sonar, how are we to co-register and compare the two modalities?

ROVer's Eye, a real-time terrain visualization package developed by OIC under DARPA funding, provides one option. ROVer integrates a hi-speed 3-D rendering package with GeoDAS's realtime-processing and target analysis package, to provide an interactive immersive experience, wherein underwater vehicle operators can work not only with existing models, but see new data from on-board sensors evolve into the current model in realtime. ROVer accommodates simultaneous inputs from sidescan, bathymetry and navigation systems, while accessing a database of existing data, targets and as-built structure models. Operators see new data evolve in a model before them, just as headlights reveal the road ahead to night-time drivers. "Road-signs" in ROVer reveal not gratuitous advertising (nor Burma Shave ditties) but full resolution images of proximal targets, which "pop-up" as the vehicle passes by. The combination of synoptic swath data with detail-rich target imagery in a fully geo-coded environment provides a whole new level of data interpretation experience.