A Method for Few-Shot Radar Target Recognition Based on Multimodal Feature Fusion.

Area of Science:

• Radar target recognition
• Machine learning
• Signal processing

Background:

• Limited sample sizes and dataset quality hinder radar target recognition generalization.
• Developing robust models with less data is crucial for practical applications.

Purpose of the Study:

• Introduce a few-shot learning framework for radar target recognition.
• Enhance generalization and robustness using multimodal feature fusion.
• Reduce reliance on extensive, high-quality datasets.

Main Methods:

• Developed a cross-modal representation optimization mechanism using natural resonance frequency features.
• Established a multimodal fusion classification network integrating bi-directional long short-term memory and residual neural network architectures.
• Proposed a cross-modal equilibrium loss function for optimizing metric spatial discrimination and classification balance.

Main Results:

• Achieved 95.36% recognition accuracy in a 5-way 1-shot task on simulated datasets.
• Outperformed unimodal image and concatenation fusion methods by 2.26% and 8.73%, respectively.
• Improved inter-class feature separation by 18.37%.

Conclusions:

• The proposed few-shot learning framework effectively enhances radar target recognition capabilities.
• Multimodal feature fusion and a novel loss function significantly improve accuracy and feature separation.
• The method demonstrates robustness and generalization even with limited data.