Integrated Data Analysis Methods for Explosives Detection
Dr. Kristen Jarman, a researcher in statistics at Pacific Northwest National Laboratory, is working on developing integrated data analysis methods for explosives detection.
Transcript
00:00:04 I'm Dr Kristen Jarman and I'm a researcher in statistics at Pacific Northwest National Laboratory I'm working on developing integrated data analysis methods for explosives detection we're developing a mathematical method using a B net that scores the likelihood that some item of interest is an explosives threat the Bas net is constructed from nodes which
00:00:22 contain different types of relevant information the first type of node represents the explosives we might be interested in finding the second type type of node represents physical and chemical characteristics of these explosives the last type of node represents sensor measurements for those characteristics so given an item of Interest we take the sensor measurements
00:00:42 and input them into the bayet those measurements then indicate the physical and chemical characteristics we're looking for which in turn get mathematically combined to score the likelihood of a threat one of the things many government clients are interested in for security applications would be combining systems in airports train stations at the border or for monitoring
00:01:03 large events with a lot of people so if we think of an example where we're looking for explosives in an airport we may have a metal detector and then we might also have some sort of a chemical sensor the metal detector in and of itself wouldn't necessarily tell you if the explosives were present it would just tell you if metal was present chemical sensors in and of themselves
00:01:22 are designed to detect explosives but certain items such as medicines or fertilizers could also set off these sensors and on top of the metal and chemical sensors we could integrate an infrared camera that would detect Heat or the shielding of heat that you might see from a gun or metal that is next to the body and so you can take all of these three different pieces of
00:01:42 information and you may get answers for each individual sensor that are maybe close to a threshold or mildly indicative of a thread but then when you combine them together they will give you a very definite indicator of a threat our method will be more robust and easier to calibrate to specific scenarios because we're actually incorporating the physics in chemistry
00:02:01 instead of just taking the data and crunching it through a blackbox algorithm this kind of integration is characteristic of the work we do at pnnl where we have teams of scientists mathematicians and Engineers to make this work possible