SAR-Oriented Visual Saliency Model and Directed Acyclic Graph Support Vector Metric Based Target Classification

Abstract

The performance of a synthetic aperture radar automatic (SAR) target recognition system mainly depends on feature extraction and classification. It is crucial to select discriminative features to train a classifier to achieve desired performance. In this paper, we propose an efficient feature extraction and classification algorithm based on a visual saliency model. First, an SAR-oriented graph-based visual saliency model is introduced. Second, relying on the ability of our saliency model in highlighting the most significant regions, Gabor and histogram of oriented gradients features are extracted from the processed SAR images. Third, in order to have more discriminative features, the discrimination correlation analysis algorithm is used for feature fusion and combination. Finally, a two-level directed acyclic graph (DAG) support vector metric learning is developed that seamlessly takes advantage of a two-level DAG by eliminating weak classifiers and the Mahalanobis distance-based radial basis function kernel which emphasizes relevant features and reduces the influence of noninformative features. Experiments on real SAR data from the MSTAR database are conducted and the experimental results demonstrate that the proposed method outperforms the state-of-the-art methods.