Heterogeneous Information-Guided Deep Structured Dictionary Pair Learning for SAR Target Recognition

Feature learning is a critical component in SAR target recognition. Deep dictionary learning has been widely used in SAR target recognition tasks due to its excellent feature representation learning capability. However, existing deep dictionary learning methods usually use only a single modal information, resulting in insufficient dictionary discrimination and feature representation learning capabilities. To solve this problem, this paper proposes Heterogeneous Information-Guided Deep Structured Dictionary Pair Learning (HIG-DSDPL) method. Specifically, method tries to introduce different heterogeneous information (different data sources or features) to enhance the feature representation ability. Second, HIG-DSDPL firstly combines deep dictionary learning and deep canonical correlation analysis for maximizing the similarity of different heterogeneous information. Finally, HIG-DSDPL constructs a structured dictionary pair learning strategy to enhance the discriminative performance of synthesis dictionary. The experimental results show that the recognition performance of HIG-DSDPL significantly outperforms the existing state-of-the-art SAR target recognition methods on five SAR target recognition datasets.