Simulator Driven Machine Learning on Embedded Systems for GPR Navigation and IED Detection, Akronym: MEDICI-LIBERTAD

点击翻译按钮获取中文摘要

Decades of guerilla warfare in Colombia have left behind a deadly legacy of thousands of landmines. In what could be a new start for Colombia, in 2016 the bloody civil war which raged intermittently for five decades was finally brought to an end. A long hoped-for peace agreement was signed between the government in Bogotá and guerilla group FARC. Prior to this, over 220,000 people were killed in conflicts between state security forces, left-wing rebels, and right-wing paramilitaries. Millions of people were also driven from their homes. Colombia now has the chance to improve its economy, political culture, and human rights and perhaps also rehabilitate its murky image abroad. But for the people of Colombia, everyday life is still marred by danger. The country is still peppered with landmines from the armed conflict with the guerilleros.The applied project “MEDICI-LIBERTAD” is one of two follow-up projects of “MEDICI”, which was funded by DFG (MU 3507/3-1, RO 2493/4-1, SA 1035/6-1 ) and Colciencias between 2014 and 2017. In order to face the great diversity in Colombian demining operations, due to altering surroundings and IED construction, the main objective of “MEDICI-LIBERTAD” is to investigate hardware embedded artificial neuronal networks in combination with sensor fusion approaches. This includes sensor fusion concepts for the positioning of a GPR device on the one hand and pattern recognition supported detection algorithms on the other hand. Regarding the required training data for the machine learning, massive and adequate data of numerous demining scenarios must be available. Here, the suitability of a self-written 2D-FDFD simulator will be investigated, which allows for randomized GPR-IED simulations including combined simulations with real and artificial data. The essential parts of the aimed objectives are described in the following.In the view of the collaboration partners, the proposed project has potential to improve the state of the art regarding both, hardware and software for neuronal network supported pattern recognition in GPR systems. Here, an optimized GPR-IED simulator is applied to generate the required teaching data for the machine learning.Since humanitarian demining operations are an important task within the peace process in Colombia, governmental and non-governmental organizations like HALO-Trust will support this project by evaluating the investigated procedures in actual mine–contaminated areas. Moreover, findings of the MEDICI projects will be directly included in the RADAR lecture, which is hold every year by the German partners in Colombia in order to guarantee a sustainable education of Colombia’s next generation engineers.