Landmine Detection Using Hyperspectral Imaging



When you see how rapidly the use of drones for scientific research has risen, you first conclude that it's all about precision agriculture and climatology. To be sure, those are trendsetting applications when it comes to using hyperspectral imaging sensors aboard UAVs. But over at the University of Bristol, two scientists are leading a team focused on 'finding a better way' to detect the presence of landmines that kill or maim thousands of people annually. With around 100 million landmines underneath the ground globally, traditional means of finding and eliminating them would take about 1,000 years and cost upwards of $30 billion according to some estimates.

Find A Better Way is a UK-based NFP founded by famed footballer Sir Bobby Charlton. Despite his heroic sporting achievements, Sir Bobby is now forging a legacy outside of football through his determination to champion the cause of landmine detection and elimination. He witnessed the destruction caused by landmines on visits to Cambodia and Bosnia as a Laureus Sport for Good Ambassador. He founded Find A Better Way after recognizing that research and development held the key to making the major changes necessary to allow humanitarian teams to rid the world of the threat of landmines.

"We want to do something that very quickly delivers a step-change in capability while reducing overall human risk involved with finding and eliminating landmines," said Dr. Tom Scott of the University of Bristol. He along with Dr. John Day are pairing their advanced UAV with small and lightweight hyperspectral imaging sensors from Headwall Photonics to 'see' with a specificity and resolution unheard of only a few short years ago. "These drones can be autonomously deployed to fly over a landmine area and provide high-resolution images that allow us to reconstruct the 3-D terrain with very high accuracy," said Dr. Scott. With all the landmines across the world, tactical deployment of numerous low-flying drones is going to win the day over expensive satellites or high-flying aircraft. In order to meet this objective, the package needs to be simultaneously affordable, light, and suited to its mission. There is a vast amount of integration and testing work involved before the first meaningful flights can be flown. Recognizing this, Headwall is assuming much of this work so that users can compress this time-to-deployment significantly. Because the use of drones for scientific research is still in its infancy, misconceptions abound. Acquiring a UAV and slapping a hyperspectral sensor to it without first considering all the variables is a recipe for disaster. This holds true whether the mission is landmine detection or precision agriculture. More commonly, other instruments such as LiDAR and GPS are part of the payload as well. The end result is a carefully balanced exercise in aerodynamics, optics, electrics, and data-collection that companies such as Headwall are able to manage.

The human eye can only respond to wavelengths between roughly 390-700nm. Many of the reflective 'signatures' given off by plants and chemicals fall outside that range. For example, the Nano-Hyperspec sensor used by the Bristol team operates in what is called the 'Visible-Near-Infrared' range of 400-1000nm (called 'VNIR' for short). In that range, the sensor is 'seeing' with an extraordinarily high degree of specificity and resolution and far beyond what a human could discern. Indeed, these sensors are collecting an astounding amount of spectral data on a per-pixel basis, resulting in 'data cubes' many Gigabytes in size. Armed with spectral libraries that faithfully characterize the specifics of the terrain below, scientists can match the known library information with the collected airborne data and make quite accurate calls on what's what.

The use of drones has accelerated this scientific effort because of two factors: First, they are affordable and easy to deploy on a tactical basis. One can be packed in a Range Rover and deployed almost anywhere in minutes whereas aircraft and satellites are, by nature, constricted, inflexible, and costly. This is not to say that aircraft and satellites will be supplanted by UAVs. There is valuable image data that can be collected using these high-flying platforms, and the overall knowledge base shared by the scientific community is made much more complete when all these assets are used in a synergistic fashion. Indeed, hyperspectral imaging was once the province of high-flying reconnaissance planes and satellites...neither of which could ever be used economically by university scientists. But the ubiquitous drone--a bane to some and a blessing to others--is the perfect platform from which to launch these exploratory efforts. "We're adding a bit more science to the UAV payload now," says Dr. Day. "We're starting to look at the spectrum of light and the colors of light that are coming off the minefield and using that data to find where the landmines are."

UAVs and drones seem to get media attention for all the wrong reasons, which is exactly why efforts by the esteemed team at the University of Bristol are to be applauded for developing a 'Better Way' to solve some of our toughest challenges. Hyperspectral imaging sensors can 'see' even beyond the VNIR range of interest to Dr. Day and Dr. Scott. The Shortwave-Infrared (SWIR) range starts near where VNIR leaves off, covering around 950-2500nm. The presence of certain chemicals, minerals and of course plant photosynthesis will become visible to sensors like these. Indeed, a broadband sensor package that covers the VNIR and SWIR range (400-2500nm) is particularly useful because it basically collects everything a scientific research effort might wish to see.

There are two key factors about hyperspectral imaging that are worth noting. The first is that the technology depends on 'reflected light.' The sensor is basically looking at how sunlight reflects off certain materials. Plant fluorescence, for example, has a particular 'spectral signature' that a sensor can understand. Obviously, this means an airborne hyperspectral sensor depends on a healthy amount of solar illumination and certainly is useless at night. But Headwall's sensors are designed to collect precise image data even under less-than-ideal solar conditions (cloud cover, or low angles, for example). The second factor is having a wide field of view. The sensor obviously can 'see' directly beneath the line of flight, but being able to do so off to the wide edges of the flight pattern makes the mission more efficient. Batteries being what they are, optimizing the flight duration by capturing a wide swath of land is obviously beneficial. This benefit is seen in the precise optical layout used by Headwall in the construction of each sensor.

Crop science, climatology, geology, and even the inspection of infrastructure such as pipelines and rail bed depend on imaging sensors like those produced by Headwall. Hyperspectral sensors depend on 'motion,' since they basically collect images slice by slice as the UAV flies over the scene. The combination of all of these high-resolution 'slices' comprise what is known as a 'data cube,' which is pored over by scientists during post-processing. Of course, the hardware capturing these images represents about half the story. The other half can be found in the software that makes sense out of reams of spectral image data that all needs to be 'geo-tagged' and orthorectified. First and foremost, scientists need answers; the data (and the sensor collecting the data) are simply means to an end. When you go to your local DIY or Lowe's or Home Depot, you really aren't buying a drill; you're going there to buy a hole.

But is all that image data really needed? Some efforts seek to cut corners by using less-capable 'multispectral' sensors that cover only a few bands rather than the hundreds of bands covered with hyperspectral. Using crop science as an example, a multispectral sensor might miss the telltale signature of an invasive disease on a tree canopy while hyperspectral will most certainly catch it. And that can mean the difference between saving a coffee bean harvest or a valuable wine-vineyard crop.