Achieving High-Accuracy Target Recognition Using Few ISAR Images via Multi-Prototype Network with Attention Mechanism DOI Open Access
Linbo Zhang, Xiuting Zou,

Shaofu Xu

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(23), P. 4703 - 4703

Published: Nov. 28, 2024

Inverse synthetic aperture radar (ISAR) is a significant means of detection in space non-cooperative targets, which that the imaging geometry and associated parameters between ISAR platform targets are unknown. In this way, large number images for high-accuracy target recognition difficult to obtain. Recently, prototypical networks (PNs) have gained considerable attention as an effective method few-shot learning. However, due specificity mechanism, often unknown range azimuth distortions, resulting poor effect. Therefore, condition poses challenge PN represent class through prototype. To address issue, we use multi-prototype network (MPN) with mechanism image recognition. The multiple prototypes eases uncertainty fixed structure single prototype, enabling capture more comprehensive information. Furthermore, maximize feature extraction capability MPN images, introduces classical convolutional block module (CBAM) attentional where CBAM generates maps along channel spatial dimensions generate robust prototypes. Experimental results demonstrate outperforms state-of-the-art methods. four-class classification task, it achieved accuracy 95.08%, representing improvement 9.94–17.49% over several other approaches.

Language: Английский

Achieving High-Accuracy Target Recognition Using Few ISAR Images via Multi-Prototype Network with Attention Mechanism DOI Open Access
Linbo Zhang, Xiuting Zou,

Shaofu Xu

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(23), P. 4703 - 4703

Published: Nov. 28, 2024

Inverse synthetic aperture radar (ISAR) is a significant means of detection in space non-cooperative targets, which that the imaging geometry and associated parameters between ISAR platform targets are unknown. In this way, large number images for high-accuracy target recognition difficult to obtain. Recently, prototypical networks (PNs) have gained considerable attention as an effective method few-shot learning. However, due specificity mechanism, often unknown range azimuth distortions, resulting poor effect. Therefore, condition poses challenge PN represent class through prototype. To address issue, we use multi-prototype network (MPN) with mechanism image recognition. The multiple prototypes eases uncertainty fixed structure single prototype, enabling capture more comprehensive information. Furthermore, maximize feature extraction capability MPN images, introduces classical convolutional block module (CBAM) attentional where CBAM generates maps along channel spatial dimensions generate robust prototypes. Experimental results demonstrate outperforms state-of-the-art methods. four-class classification task, it achieved accuracy 95.08%, representing improvement 9.94–17.49% over several other approaches.

Language: Английский

Citations

0