Archives—February 2016

Computational Intelligence Aids in Explosive Hazard Detection

Havens, TimTo detect buried explosive hazards in places like Afghanistan and to save the lives 
of civilians and US soldiers, Michigan Tech researcher Tim Havens realizes it requires
 a team: a team of sensors.

A new $983,000 research project, “Heterogeneous Multisensor Buried Target Detection Using Spatiotemporal Feature Learning,” will look at how forward-looking ground- penetrating radar, LiDAR, and video sensors can be combined synergistically to see into the ground, capture high-quality images, and then automatically notify the operator of threats. With funding from the US Army Research Laboratory’s Army Research Office (ARO), Havens and Tim Schulz, professor of electrical and computer engineering, will work with three Michigan Tech PhD students to create a high probability-of-detection/low false-alarm rate solution.

“It’s a very difficult problem
 to solve because most of the radar energy bounces right off the surface of the earth.” says Havens, the William and Gloria Jackson Assistant Professor of Computer Systems
 and ICC Center for Data Sciences Director. “ It’s hard enough finding the targets, but coupled with that is the amount of data that these sensors produce is massive. It is the perfect project to combine Tim’s (Schulz) statistical signal processing background and my machine learning in big data expertise. This technology has the potential to not only save lives, but also to advance the basic science of how to combine large amounts of sensor data and information together to get a whole that is better than the sum of its parts,” Havens explains.

This new project builds upon a previous sensor-related work Havens and collaborators completed between 2013-2015 and a current project on which Havens and Joe Burns, a Senior Research Scientist at Michigan Tech Research Institute, collaborate. These projects, also funded by the US Army, study signal processing and computer-aided detection and classification using both vehicle-mounted forward-looking and handheld downward-looking ground-penetrating radars. In total, Havens and his collaborators have secured over $2.5 million in funding to develop solutions for this problem.

Figure 1. High-level overview of multi-sensor feature learning and fusion for forward-looking explosive hazard detection. iECO: Improved evolution-constructed (features). DBN: Deep belief network. CNN: Convolutional neural network. GAMKLp: ℓp norm-based genetic algorithm for multiple kernel learning (feature-level fusion method). DeFIMKL: Decision level fuzzy integral multiple kernel learning.

 

The Army currently fields ground-penetrating radars in its fleet. The problem is they cannot detect hazards until they’re right above them, putting a multi-million dollar radar—and Soldiers—directly in the path of danger.

“The big ideas here are to process big data to obtain better images, see into the ground in a high-fidelity manner, and to develop algorithms that automatically find buried threats—even notifying operators of w
hat those threats might be—all while keeping the Soldiers and equipment as far away as possible,” Havens adds.

Havens has partnered with the Army since he was a PhD student in 2008.