Portland, Ore. -- The revelation of a terrorist plot to smuggle liquid-explosive precursors onto airliners headed for the United States, then mix them together in the plane's bathroom, has sent the U.S. Transportation Security Administration and other anti-terrorism task forces worldwide scouring for systems that can detect liquid threats.
The problem is that traditional, "sniffer" type sensors depend on detecting traces of explosives left on the outsides of the containers in which they are packed. Carefully sealed liquid containers, possibly fitting an unusual form factor or even shielded by metal, might get by a sniffer. And conventional X-ray detectors, which depend on human screeners to read the X-rays and identify weapons via shape recognition, cannot screen for liquids.
Now Guardian Technologies International Inc. (Herndon, Va.) says it has devised PC-based software that can be used in tandem with conventional X-ray detectors in the field, acting as "a second set of eyes" to identify liquid-based compounds for explosives in scanned baggage.
"Usually the threats [anti-terrorism officials] are trying to discover are guns, knives and things like plastic explosives, but as of yesterday there is suddenly a high priority on identifying liquid-based compounds for explosives, which we can do," Steve Lancaster, vice president of Guardian, told EE Times after British authorities exposed the foiled plot. Lancaster claimed that Guardian Technologies' PinPoint software can detect liquid explosives in scanned baggage, regardless of metal shielding or the containers' orientation in the bag.
Since "you can't do shape recognition" for liquid explosives, Lancaster said, "we use iterative algorithms that perform spatial and spectral analysis down to the pixel level and build a unique response or 'signature' from each kind of object." The PinPoint software can detect the telltale signature of any type of threat that its neural network has been previously trained to detect, including the liquid-explosive pre- cursors the British terrorists had planned to use.
Examining a bag as it is passed through an X-ray scanner, PinPoint takes several seconds to analyze the contents and identify any target items in the bag for which it has been previously trained. The company claims better than 90 percent accuracy in identifying specific threat objects and a less than 9 percent false-alarm rate. The tool is being beta-tested at U.S. airports, according to Guardian.
Using neural-network learning algorithms combined with other image segmenting, feature extraction and classification algorithms, PinPoint is preprogrammed at Guardian's laboratory to identify specific compounds. Guardian claims the pixel-level information obtained via the standard, dual-energy X-ray machines at airports already harbors the unique signatures for any type of weapon or concealed explosives, regardless of their orientation or shape. The characteristic relationships among pixels that correspond to the density, color and geometry of specific explosives each have a unique signature, according to the company.
After the learning step, Guardian Technologies installs the algorithm into the PinPoint software, which runs on a PC in the field. At the airport, the PinPoint algorithm segments each X-ray image into objects (even though some occlude each other in the image), classifying them as "not a threat" or "a threat," and explaining why.
"We run thousands of images of differently packed luggage with whatever types of threats we want to detect, which could include liquid explosives. Then, through our learning algorithms, we generate the signature that we come to expect from that specific threat," said Lancaster. "When PinPoint gets that signature from an X-ray image of a carry-on bag, we put a big red box around it on the operator's screen" that alerts the operator to hand-search the bag. As long as the threat's signature has already been programmed into PinPoint by Guardian, the red box around suspicious objects and the corresponding messages to operators can be configured by the user to keep them current with changing policies.
Threats cannot be concealed by objects that are not threats, the company claims, because overlaying a threat with a nonthreatening object changes the signature of the threat in predictable ways.
Thus, PinPoint can "see through" nonthreatening ob- jects to perceive threats. Both the spatial and the spectral properties are used during the generation of rules for pattern classification, for later use in the field.
"We use many machine-learning algorithms as well as classical image-processing techniques," Lancaster said. "Iterative transformational divergence follows image features down to the pixel level, where we detect divergence and bifurcation and develop a plane of data. We then find the statistically significant differences in this hypercube of data planes to build a signature that a vector processor can later use to classify X-rays."
Currently, the signatures of newly emerged threats, such as the recent plot's liquid-explosive precursors, can only be programmed into PinPoint at Guardian Technologies. But the company is experimenting with adding learning capabilities to its installations in the field, "so that if an airport got a specific threat for a certain type of item, they could quickly just run that item through the X-ray machine and learn its signature," Lancaster said. "That way, any airport could evaluate specific new threats without having to wait for upgrades from us."