Wrapper approach for feature subset selection using GA

In this paper, wrapper based approach is presented to feature subset selection of Ground Penetrating Radar (GPR) echo signal using genetic algorithm in conjunction with constructive learning algorithm. GPR echo signal inherently has the nonstationary characteristic and the target echo signal is deteriorated by the clutter. Firstly, WP-based preprocessing algorithm is used to clutter reduction and feature extraction. Then wrapper based approach is adopted to find optimal feature subset, and at meanwhile obtain the result of classification. Experiment result based on GPR landmine data shows the feasibility and advantage of the presented algorithm.