Bidimensional Empirical Mode Decomposition for SAR Image Feature Extraction With Application to Target Recognition

Abstract

This study introduces a target recognition algorithm for synthetic aperture radar (SAR) images based on the features extracted by bidimensional empirical mode decomposition (BEMD). BEMD provides an adaptive and empirical way to process signals, which generates bidimensional intrinsic mode functions (BIMFs) to describe the details of SAR images. Therefore, the generated BIMFs are complementary to the original image and their joint use could probably improve the recognition performance. In order to fully exploit the discrimination of these components, the joint sparse representation (JSR) is employed during the classification. JSR operates as multi-task learning algorithm, which represents each component numerically while considering their inner correlations. The original image together with the generated BIMFs are simultaneously represented by JSR to determine the target label according to the output reconstruction errors. Experimental results on the Moving and Stationary Target Acquisition and Recognition (MSTAR) data set demonstrate the validity of the proposed method under different operating conditions. In comparison with some baseline algorithms, the superiority of the proposed method is furtherly validated.

Keywords

• Synthetic aperture radar (SAR)
• target recognition
• bidimensional empirical mode decomposition (BEMD)
• joint sparse representation (JSR)