Multi-View ISAR Target Classification Based on Mamba Network

Inverse synthetic aperture radar (ISAR) is able to acquire image sequences of space targets through multi-static observation or pulse integration. These multi-view images capture progressive variation information of target shapes and structures. In this paper, a multi-view ISAR space target classification method based on the Mamba network is proposed. Firstly, the coarse-grained spatial features are extracted by three-dimentional convolution blocks from the image sequence. Subsequently, the selective scan structured state spaces for sequences model (S6M) with a window partitioning strategy is employed to extract global contextual feature extraction, thereby obtaining fine-grained representations of multi-view ISAR images. Experimental results on the ISAR image data sets validate the effectiveness of the proposed method, demonstrating improved accuracy compared with existing methods. Furthermore, the proposed method exhibits robustness to target deformations including scaling and rotation.