Scholarlink

<p>To address the dimensional sensitivity and scale transformation challenges in intelligent radar signal sorting, this paper proposes a novel radar signal sorting method based on point cloud instance segmentation. The algorithm innovatively maps pulse description words (PDWs) into geometrically structured 3D point cloud space. Through guided optimization by a magnetic loss function, the enhanced PointMLP instance segmentation network achieves instance-level emitter separation in complex electromagnetic environments. The proposed point cloud segmentation network integrates the advantages of PointNet and PointMLP architectures, employing multi-scale feature fusion mechanisms to overcome the limitations of conventional point cloud methods in capturing local features. The magnetic loss function introduces dual constraints of inter-class repulsion and intra-class attraction via deep clustering principles, forming an optimized feature space configuration that separates heterogeneous pulses while aggregating homologous ones. Compared with traditional algorithms, our approach eliminates the need for prior knowledge to preset thresholds and avoids information distortion caused by dimensional compression in image-based sorting solutions. Simulation results demonstrate that the proposed method maintains over 95% average sorting purity under high missing-pulse and spurious-pulse scenarios, confirming strong robustness and engineering applicability in dynamic electromagnetic environments.</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

Radar Singal Sorting via PointMLP Instance Segmentation Network

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Boyi Yang,
  • Mengyu Shen,
  • Tao Chen,
  • Lei Zhan

摘要

To address the dimensional sensitivity and scale transformation challenges in intelligent radar signal sorting, this paper proposes a novel radar signal sorting method based on point cloud instance segmentation. The algorithm innovatively maps pulse description words (PDWs) into geometrically structured 3D point cloud space. Through guided optimization by a magnetic loss function, the enhanced PointMLP instance segmentation network achieves instance-level emitter separation in complex electromagnetic environments. The proposed point cloud segmentation network integrates the advantages of PointNet and PointMLP architectures, employing multi-scale feature fusion mechanisms to overcome the limitations of conventional point cloud methods in capturing local features. The magnetic loss function introduces dual constraints of inter-class repulsion and intra-class attraction via deep clustering principles, forming an optimized feature space configuration that separates heterogeneous pulses while aggregating homologous ones. Compared with traditional algorithms, our approach eliminates the need for prior knowledge to preset thresholds and avoids information distortion caused by dimensional compression in image-based sorting solutions. Simulation results demonstrate that the proposed method maintains over 95% average sorting purity under high missing-pulse and spurious-pulse scenarios, confirming strong robustness and engineering applicability in dynamic electromagnetic environments.