Synthetic Aperture Radar (SAR), first developed in the 1950's, has become an increasingly critical technology for military applications. Because SAR sensors can be used to image ground targets at extremely high resolutions and long ranges, through clouds and in darkness, they play a key role in surveillance and reconnaissance missions. SAR Automatic Target Recognition (ATR) systems are designed to rapidly and reliably identify time critical military targets in SAR imagery.
 

A Scud missile launcher is one of the time critical targets that our ATR has been trained to locate in SAR imagery.

Sandia's Signal and Image Processing Department has designed ATR algorithms for SAR sensors since 1986. We were the first to demonstrate real-time SAR ATR capability in 1991, on board the Department of Energy's De Havilland DHC-6 Twin Otter aircraft. Since then, Sandia has been the leader in SAR ATR technology, integrating the latest hardware with innovative recognition algorithms.

The system shown above has demonstrated real-time ATR performance in a small, rugged package. (17.5"x17.5"x19.5")

ATR real-time requirements include both high throughput rate and low latency. For conventional image sizes, the latency between receipt of the SAR image and ATR results is typically less than 10 seconds. The basic configuration of our all-COTS real-time ATR has 12 PowerPC 300 MHz CPUs and can process imagery at the rate of one Megapixel per second for 10 targets of interest. The CPU requirements of our ATR system scale linearly with respect to pixel rate and number of targets. The 6U VME rack shown above can accommodate 64 CPUs, which enables us to upgrade the system to allow data rates as high as five Megapixels per second for 10 targets of interest or 50 targets of interest at one Megapixel per second without changing the 3.5 ft³ size of the ATR system. Upcoming advances in CPU performance will triple our current capabilities by the end of the year 2000.