Demining-Research-Community-logo.png.webp

Learn more at 

de-mine.com

                          Scatterable Landmine Detection Research                                       

Through synergies in multispectral, photogrammetry, thermal imaging, drone platforms and machine learning we have found a new innovative way to detect surface lain PFM-1 anti-personnel landmines. These particular devices pose a difficult challenge to conventional landmine detection methods like metal detecting, because the mines are primarily composed of plastic and only weigh 75 g. As a remnant of the Soviet-Afghan War which lasted from 1979 to 1989, there are an estimated 10 million PFM-1 devices scattered throughout Afghanistan. These mines remain emplaced in isolated locations, frequently out of reach of demining NGOs and act to thwart local economic and social development. The PFM-1s are infamously referred to as “toy mines” as children often mistake the mines for toys and set off the 5-25 kg of cumulative pressure it takes to detonate them.

We were able to successfully identify landmines through multispectral and thermal infrared imagery datasets with the Parrot Sequoia multispectral sensor, the FLIR Vue Pro R thermal camera and the DJI RGB camera at 10 meters above ground with various drones. Because the mines have different physical properties like reflectance, emissivity, and thermal conductivity, they heat and cool at different rate than the host geology. In the early morning hours when thermal inertia is greatest, the PFM-1 mines can be detected based on their differential thermal inertia. Because the mine has a statistically different temperature than background and characteristic shape, we are currently training a supervised learning algorithm to automate detection of the mines over large areas.

Currently I am working with Gabriel Steinberg, Timothy de Smet and Alex Nikulin, to develop a faster Convolution Neural Network (CNN) to completely automate this method. Gabriel and I have been training the neural network by drawing bounding boxes around PFM-1 mines from our datasets to improve the accuracy of the CNN, by adding minefield data in different environments and orientations to create a robust generalized model that can be extrapolated onto new unseen minefields. In the end, the CNN will be fed an orthophoto, and will be able to identify potential mines from the imagery data, and output a shape file with cm scale accurate coordinates where these mines may reside.

Applying our methods and technology to detect and remediate PFM-1 mines in post-conflict developing nations has the potential to save thousands of lives and significantly reduce the cost and time of mine detection, improve safety of operators and stimulate economic development in these regions.

snow too.png

Figure 9. (Baur et al. 2020)

Clipped images of orthophotos from six different bandwidths (plus normalized difference vegetation index), showing the success in identifying the plastic PFM-1 landmine and the aluminum KSF casing from the surrounding environment in grass, low vegetation and snow datasets.

Figure_4_Cover art with multispec.jpeg

Figure 4. (Baur et al. 2020)

Illustration of experimental design mid-flight in Afghani terrain, using the Parrot Sequoia multispectral sensor attached to the Matrice 600 Pro UAV (unmanned aerial vehicle). Processed multispectral images of the PFM-1 taken from 10 m height during flight.

Figure1_PFM-1.tif

Figure 1. (Baur et al. 2020)

Rendering of an Inert PFM-1 plastic anti-personal landmine considered in this study with small US coin for scale.