Multiview SAR Target Recognition Based on Adaptive Semantic-View Feature Embedding Fusion Network

In synthetic aperture radar (SAR) automatic target recognition (ATR) tasks, due to the SAR imaging mechanism and the characteristics of target scattering structures, the image features of a target under a single observation angle may be insufficiently expressed, and the image features under different observation angles may vary significantly, which makes target recognition using single-view SAR images challenging. Therefore, acquiring multiview images of the target for joint recognition has become a promising method. However, existing multiview recognition models suffer from problems, such as insufficient learning of multiview features, inapplicability under incomplete multiview conditions, and scarcity of multiview training data, restricting the performance and practicality. In this article, we establish a new multiview SAR target recognition framework to uniformly address the above issues. First, we construct a semantic-view dual-encoder feature embedding fusion network (SeVi). By performing classification modeling on the observation views of images to introduce view prior information, the model is guided to acquire and fuse view-aware features. Second, we propose a prototype-based missing view completion strategy, enabling the model to adapt to complex incomplete multiview scenarios while maintaining high performance. Third, we put forward a nonstrict multiview recognition data construction method based on semi-supervised classification, which efficiently acquires data to support the training of multiview recognition models. Experiments are conducted on multiple vehicle and aircraft target recognition datasets, including ATRNet-STAR, MSTAR, and SARAircraft. The results demonstrate that the proposed model can effectively capture and fuse view-aware features, achieving leading performance compared with existing SAR multiview target recognition methods. Moreover, it maintains high performance even under incomplete multiview complex conditions, which verifies the robustness and flexibility of the proposed multiview recognition framework.