Random Forest Feature Selection for SAR-ATR

Abstract

In this paper, a novel approach for selection of relevant features in SAR-ATR is proposed. The main concern of all studies in this filed is the accuracy. For this reason, many researchers have worked on feature extraction phase. Just a few studies focus on feature selection stage. The goal of working on feature selection is twofold. Firstly, the dimensionality of feature space can be reduced and secondary the accuracy can be further improved by eliminating the redundant features. Random Forest is the technique that can be easily implemented over the alternative algorithms such as Genetic Algorithms to SAR-ATR. The easy and fast implementation are the main advantages over the alternative methods. The experimental results show that by selecting just a few features, the accuracy is reaches to saturation.