| 引用本文: | 孙延鹏， 宁秋月， 屈乐乐. 基于半监督学习的低空小型无人机雷达识别[J]. 雷达科学与技术, 2025, 23(6): 591-602.[点击复制] |
| SUN Yanpeng, NING Qiuyue, QU Lele. Radar Recognition of Low Altitude Small Unmanned Aerial Vehicle Based on Semi-Supervised Learning[J]. Radar Science and Technology, 2025, 23(6): 591-602.[点击复制] |
|
| 摘要: |
| 为了提高低空小型无人机雷达识别的效果，本文针对基于雷达的低空小型无人机雷达识别问题提出优化的半监督学习。半监督学习方法能够结合有标签的数据和无标签的数据，改善有限标注样本条件下模型的识别性能。本文主要针对基于分歧的半监督方法进行改进，搭建了一种基于协同的半监督神经网络（Co-training Semi-supervised Neural Networks，CSNN）的无人机识别框架，该框架引入空间注意力机制和通道注意力机制以及Ghost瓶颈层，以达到增加特征聚焦和减少参数量的目的，在开源数据集和使用连续波雷达自测的无人机数据集中都做了相应的实验，实验结果表明，该框架下的半监督学习方法旋翼无人机的分类准确率分别为97.33%和95.78%，都表现出了较大的优越性，相较于有监督学习分别提高了3.01%和2.66%。 |
| 关键词: 无人机分类 半监督学习 注意力机制 Ghost瓶颈层 连续波雷达 |
| DOI：DOI:10.3969/j.issn.1672-2337.2025.06.001 |
| 分类号:TN957.5；TN958.94 |
| 基金项目:国家自然科学基金（No.61671310）； 航空科学基金（No.2019ZC054004）； 辽宁省高校基本科研业务费资助项目（No.LJ222410143071） |
|
| Radar Recognition of Low Altitude Small Unmanned Aerial Vehicle Based on Semi-Supervised Learning |
|
SUN Yanpeng, NING Qiuyue, QU Lele
|
|
College of Electronic Information Engineering, Shenyang Aerospace University, Shenyang 110136, China
|
| Abstract: |
| To enhance radar-based detection of low-altitude small unmanned aerial vehicle （UAV）, an optimized semi-supervised learning approach for radar-based low-altitude small UAV identification is proposed in this paper. The semi-supervised learning method can combine labeled and unlabeled data to improve the recognition performance of the model under the condition of limited labeled samples. In this paper, we focus on improving the disagreement-based semi-supervised approach, and build a UAV recognition framework based on co-training semi-supervised neural networks （CSNN）, which incorporates spatial attention mechanisms, channel attention mechanisms and Ghost bottleneck layers to enhance feature discriminability while reducing parameter complexity. Comprehensive experiments were conducted on an open-source dataset and a self-collected continuous wave radar dataset. The experimental results demonstrate that the proposed framework achieves classification accuracies for rotary-wing UAV of 97.33% and 95.78% on the two data-sets respectively, outperforming supervised learning baselines by 3.01% and 2.66% accuracy improvements. This validates the framework’s robustness in low-altitude complex environments with limited labeled data. |
| Key words: UAV classification semi-supervised learning attention mechanism Ghost bottleneck layer conti-nuous wave radar |
