Feature-Level Fusion Recognition of Radar Targets Based on Regularized Sequential Representation Network

Radar target recognition has become a key technology in space surveillance and security due to its capability to utilize the characteristics for effectively judge and recognize. However, current single-domain data-based recognition methods suffer from limited recognition accuracy due to paucity of inherent information, while existing multi-domain fusion methods demonstrate constrained representational capacity as they predominantly perform basic data fusion rather than achieving effective cross-domain feature representation within compressed feature spaces. To address these issues, this paper proposed a sequential hybrid neural network based on feature-level regularized representation that utilizes multi-domain radar information and prevents the characterization of a large amount of redundant information during the fusion process. The proposed network sequentially processes radar cross-section (RCS) time series, high-resolution range profile (HRRP) sequences, and timefrequency (TF) spectrograms through three distinct branches, leveraging their complementary characteristics. Additionally, a dual-branch pixel attention fusion unit is incorporated into the feature-level fusion module to adaptively fuse the temporal-spatial features of multi-domain data, effectively avoiding the repetitive characterization. Consequently, it captures both discriminative and complementary representations, significantly enhancing the accuracy of radar target recognition. The recognition results of electromagnetic simulated data under various conditions have demonstrated the effectiveness and robustness of the proposed method.