Multi-Classifier Fusion and Nonparametric Decision for Landmine Detection

Traditional techniques to detect and remove landmines are both dangerous and time consuming. Very little technology is currently employed in the real world for the detection of landmines. Metal detectors are effective against metal-based landmines, but many mines are plastic cased. Landmines are divided up into two broad classes: antitank (AT) mines, which are designed to impede the progress of or destroy vehicles, and antipersonnel (AP) mines, which are designed to kill and maim people. Landmine comes in a variety of shapes and sizes. They can be square, round, cylindrical, or bar shaped. The casing can be metal, plastic, or wood. These characteristics make the landmine detection challenging. An effective mine detection system should be capable of using all available thermal, spectral and spatial differences for discrimination. The main objective of this project is the development and implementation of a new landmine detection system with classifiers based on surface shapes, textures, comprehensive feature vectors and spectral structures. We propose to develop a set of new nonparametric hypothesis test schemes based on cluster trending analysis and randomness test. A new unsupervised neural network is proposed to cluster measurement data. All classifiers will be fused based on our LIM-based optimal fusion method.