Landmine Detection and Classification Using MLP

This paper expounds on the design and the implementation of the intelligence (vision and brain) of an autonomous robot for landmines localization, specifically anti-tank mines, cluster bombs, or unexploded ordnance. The landmine sweeping technique under study utilizes state-of-the-art techniques in digital image processing for pre-processing captured images of the area being scanned. After enhancing the scanned images, data is fed into a processing unit that implements the Artificial Neural Network (ANN) in order to classify the landmines' make and model. The Back-Propagation algorithm is used for teaching the network. The system proved to be able to identify and classify different types of landmines under various conditions with a success rate of up to 90%. Various conditions include different viewpoints of the landmine such as having a rotated landmine, or a partially covered landmine.

I. Introduction

Landmines were mainly designed as area-denial weapons, and are used to create tactical barriers in order to prevent direct attack or to deny access by military and civilians from a defined area. Landmines are perfect soldiers that never eat, sleep, miss, fall ill or disobey. Moreover the landmine perfectly completes its job for much less U.S. dollars than the human soldier; in addition landmines are long-term killers, active long after a war has ended. There are two major types of landmines, Anti-Personnel (AP) mines and Anti-Tank (AT) mines. AP mines are usually placed under earth, close to the surface; while AT mines are usually placed on the surface of the earth. According to the UNICEF, around 2,000 persons are involved in monthly landmine accidents, 800 (40%) of whom are innocent civilians; that is, an average of a victim every 20 minutes dies [1]. According to the UN, even though about 100,000 mines are removed every year; two million more replace them [2], [3].