Feature set selection for impulse radar based landmine detection

This paper discusses feature reduction methods for classification of multi-dimensional data. The application in mind is landmine detection using a ground penetrating impulse radar. By classifying detected objects, the false alarm rate in a landmine detection system can be greatly reduced. The measured data is mapped to feature vectors that are supposed to represent the data. To keep the classifier fast and simple the dimension of the feature vectors should be as low as possible. Four different methods for feature reduction are here compared and evaluated on real data from an impulse radar system. The different performance, resulting from the different design criteria of the methods, are discussed in this paper.