Evolutionary feature selection algorithm based on information theory and its application to gesture recognition in radar system

Abstract

In this study, evolutionary algorithms with an information factor are proposed for solving the feature selection problem. To find feature subsets effectively and quickly, two evolutionary algorithms, the quantum-inspired evolutionary algorithm (QEA) and the particle swarm optimization (PSO), were incorporated with an information factor based on the minimum Redundancy Maximum Relevance (mRMR) criterion. To overcome slow convergence rate of two algorithms, we employ a novel search scheme based on an information factor. The information factor is used in an update stage in QEA and PSO, so that each element of the individual or particle has a different update degree than the others. The proposed algorithms are applied to the feature selection problem and are able to extract the relevant features from all the feature sets. To demonstrate the effectiveness of the proposed algorithms, empirical comparisons with other feature selection algorithms are carried out for benchmark functions. The proposed algorithm is applied to the FMCW radar system to perform gesture recognition. After signal processing including the 2D-FFT, clutter removal, and false alarm rate, the features suitable for gesture recognition are defined based on the generated range-Doppler map. Using the proposed algorithm, unrelevant features are removed and a set of features which consist of relevant features is obtained. As a result, we built the FMCW radar system with high gesture recognition rate. Furthermore, we compared with other machine learning method such as LSTM, and a comparative analysis were performed.