INTELLIGENT DETECTION OF LAND MINE

Abstract

Detection and de-mining of a buried land mine are among the most difficult problems faced during and after the war. Specially, anti-personal mines are still scattered in many countries, preventing the land to agricultural use, and many of civilians are injured from these dreadful threats. The detection problem has become extremely hard due to the large variety of landmine types with minimum sizes and non metal content, different soil conditions, temperature and weather conditions, and the highly clutter environment. Consequently, the detection problem can be tackled in a broad context of efficient sensors, and image and signal processing for the data associated with these sensors. In the development of sensors technologies, Infrared (IR) is an effective

approach based on' the thermal properties for objects. Over the last few years IR polarization filter was introduced into IR sensors for improving the low target to clutter ratio in infrared image. Currently Ground Penetrating Radar (GPR) sensor offers the promise of detecting any object with little or no metal content, and gives information on both the existence and the location of objects. On the other hand, concerning of image and signal processing, lot of research for noise reduction, segmentation, and pattern recognition has appeared regarding the pre-processing and the decision mine or non mine in detection applications. In this thesis, we handle the problem of mine detection using new efficient techniques of image and signal processing for the data associated with IR, IR polarization and GPR sensors. The work is divided into two parts: the first part introduces the mine detection in IR, and IR polarization images in context of pre­ processing and segmentation techniques. Principle Component Analysis (PCA) as a dynamic pre-processing is used to extract the whole dynamic information contained in a sequence of images. Also, we propose two new different segmentation techniques for discriminating land mine from background clutter, and focus on evaluating the suitable technique for mine detection. The first one, a new hierarchical segmentation based on watershed is proposed for mine detection application.