Genetic algorithm for feature extraction in landmine detection

A feature extraction method based on automatic feature generation and gradual reduction of the feature space using the best search and genetic algorithm is described. Knowledge of the problem domain collected during reduction of the initial set of features is included into the implementation of the genetic algorithm. The fitness function for evaluation of the feature subsets is based on the Bayes classifier. The classifier is constructed for the particular feature subset using the training set of samples, and the classification ability of the particular feature subset is evaluated on samples from the test set. All generated feature subsets are stored, which allows further analysis and extraction of the best feature subsets with a different number of features. The proposed algorithm is tested on acoustic signatures of real landmines and other objects that can be found in a minefield.