According to the United Nations, more than 100 million land mines are buried in 68 countries around the world. The task of finding these land mines is not easy - the deadly devices are small, varied in shape and material, and commonly found with harmless objects that can confuse current detection techniques.

The Georgia Tech Research Institute is leading a team of researchers from the University of Hawaii, University of Florida, University of Maryland, Rochester Institute of Technology and Clark Atlanta University to develop better methods for detecting land mines from the air. This research is supported by the U.S. Army Research Office under a Multidisciplinary University Research Initiative (MURI) grant.

"Land mine detection is never 100 percent, but if the military has an indication of where land mines are, they can direct their troops around them and send explosive ordinance disposal experts to that area," explained J. Michael Cathcart, project director and GTRI principal research scientist.

Cathcart and GTRI research scientist Alan Thomas are developing computer algorithms to detect minefields from aerial hyperspectral long-wave infrared images.

Disturbed Ground and Geometric Patterns
"The key we're looking for in these images is disturbed ground," said Cathcart. "Once someone digs up the ground and puts something into it, the character of the soil below is changed and this disturbance shows up in thermal images."

To test their algorithms, images at 70 different wavelengths between eight and 12 micrometers were collected using an airborne hyperspectral imager developed by the University of Hawaii. Mines were buried at a depth of four inches, inserted flush with the surface or laid on the surface of the ground.

Once the researchers removed all vegetation anomalies from the images and found a signal they believed to be a land mine, their computer program searched in specific patterns to locate other land mines nearby. Since land mines are typically laid out in geometric patterns, the researchers developed methods for determining the orientation and size of the land mine pattern within the image.

"Our program has been successful in locating buried minefields in the images," noted Cathcart. "We are still examining which spectral signatures show land mine signals best because they vary based on moisture content and the size of the particles on the surface of the ground."

In the future, the research team plans to examine ways to identify nonlinear minefield patterns - such as skewed grids, radially symmetric patterns and mines laid in relation to naturally occurring contours like roads - using physics-based detection algorithms.

From the 2008 Georgia Tech Research Institute annual report, Problem. Solved.