Scholarlink

<p>With the widespread application of unmanned swarms in tasks such as collaborative reconnaissance, electronic countermeasures, and autonomous decision-making, swarm platforms face increasingly stringent requirements for real-time perception within complex electromagnetic environments. Radar emitter recognition is crucial for reliable system operation, yet it remains challenging under low signal-to-noise ratio (SNR) conditions due to noise interference. To address this issue, we proposed M4FexNet, a radar emitter signal recognition model based on multi-modal mixed-feature extraction, which effectively exploits complementary features from different modalities to enhance recognition performance in low SNR scenarios. Firstly, we developed a radar emitter signal simulation platform to generate six modulated radar emitter IQ signal datasets, ensuring high fidelity. Then, the radar IQ signals are preprocessed to construct a multi-modal dataset comprising time–frequency representations, 2D spectral correlation function contour maps, time-domain sequences, and frequency-domain sequences. The M4FExNet model integrates a Transformer-based image processing branch with a BiTCN-BiLSTM sequence processing branch, employing residual fusion and self-attention mechanisms to dynamically adjust multi-modal feature weights for effective mixed-feature extraction. Experiments show M4FExNet achieves superior performance, with average recall, F1-score, and accuracy of 92.15%, 92.08%, and 92.13%, respectively, and a recognition accuracy of 54.5% at − 20&#xa0;dB. Confusion matrix and feature visualization confirm its robustness and effective feature representation across SNRs. The findings confirm that M4FexNet can effectively extract and fuse multimodal features, thereby providing robust support for radar emitter recognition and collaborative perception by unmanned swarms within complex electromagnetic environments, while simultaneously maintaining consistency with established research directions and technical roadmaps.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A multimodal mixed-feature extraction network for radar emitter signal recognition in unmanned swarm electromagnetic perception

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月30日

查看全文
  • Jieling Wang,
  • Yanfei Liu,
  • Chao Li,
  • Zhong Wang,
  • Yali Li

摘要

With the widespread application of unmanned swarms in tasks such as collaborative reconnaissance, electronic countermeasures, and autonomous decision-making, swarm platforms face increasingly stringent requirements for real-time perception within complex electromagnetic environments. Radar emitter recognition is crucial for reliable system operation, yet it remains challenging under low signal-to-noise ratio (SNR) conditions due to noise interference. To address this issue, we proposed M4FexNet, a radar emitter signal recognition model based on multi-modal mixed-feature extraction, which effectively exploits complementary features from different modalities to enhance recognition performance in low SNR scenarios. Firstly, we developed a radar emitter signal simulation platform to generate six modulated radar emitter IQ signal datasets, ensuring high fidelity. Then, the radar IQ signals are preprocessed to construct a multi-modal dataset comprising time–frequency representations, 2D spectral correlation function contour maps, time-domain sequences, and frequency-domain sequences. The M4FExNet model integrates a Transformer-based image processing branch with a BiTCN-BiLSTM sequence processing branch, employing residual fusion and self-attention mechanisms to dynamically adjust multi-modal feature weights for effective mixed-feature extraction. Experiments show M4FExNet achieves superior performance, with average recall, F1-score, and accuracy of 92.15%, 92.08%, and 92.13%, respectively, and a recognition accuracy of 54.5% at − 20 dB. Confusion matrix and feature visualization confirm its robustness and effective feature representation across SNRs. The findings confirm that M4FexNet can effectively extract and fuse multimodal features, thereby providing robust support for radar emitter recognition and collaborative perception by unmanned swarms within complex electromagnetic environments, while simultaneously maintaining consistency with established research directions and technical roadmaps.