基于高维特征随机森林的雷达扫描方式自主识别方法

吴康徽; 郭子薰; 范一飞; 谢坚; 陶明亮

doi:10.11999/JEIT250985

基于高维特征随机森林的雷达扫描方式自主识别方法

doi: 10.11999/JEIT250985 cstr: 32379.14.JEIT250985

西北工业大学西安 710072

基金项目: 国家自然科学基金(62301435 and 62571444)

详细信息

作者简介:
吴康徽：男，硕士研究生，研究方向为电子侦察，机器学习

郭子薰：女，副教授，研究方向为目标智能探测与侦察

范一飞：男，副研究员，研究方向为雷达信号处理、雷达信号分选

谢坚：男，教授，研究方向为电子侦察定位理论及应用、阵列信号处理理论及应用

陶明亮：男，教授，主要研究方向为目标智能探测与侦察

通讯作者:
郭子薰　zxguo@nwpu.edu.cn

中图分类号: XXX
计量
- 文章访问数: 11
- HTML全文浏览量: 8
- PDF下载量: 1
- 被引次数: 0
出版历程
- 修回日期: 2025-12-30
- 录用日期: 2025-12-30
- 网络出版日期: 2026-01-08

Autonomous Radar Scan-Mode Recognition Method Based on High-Dimensional Features and Random Forest

School of Electronics and Information, Northwestern Polytechnical University, Xi’an 710072, China

Funds: Item1, Item2, Item3

摘要

摘要: 在非协作电子侦察条件(被动截获、无先验/不同步、参数随任务时变且电磁环境拥挤)下，快速、稳健地区分雷达扫描方式是实现威胁评估、资源调度与对抗策略生成的重要环节。因此，面向非协作场景下的雷达扫描方式识别，本文围绕机械扫描(机扫)与相控阵电子扫描(相扫)的物理差异，提出一套时–频–图多域自主特征体系。时域方面，建立变异系数、总变差、高斯拟合度、主瓣相对宽度，用于度量平滑/跳变和形态规整度；频域方面，采用谱平坦度刻画能量集中与分散；图结构方面，将幅度序列映射为水平可见性图，并计算全局聚类系数与归一化度熵，以捕获由序列形状诱发的全局拓扑模式。结合所提出的7个有效差异性特征，形成7维特征向量，随后结合随机森林算法，完成扫描方式识别。基于包含机扫与相扫、覆盖多信噪比条件、含到达时间抖动与多普勒的合成对照数据，实验结果表明，所提方法实现了97.59%的识别准确率，并在低信噪比条件下仍具稳健分辨力，充分验证了方案可行性。
- 雷达扫描方式 /
- 特征提取 /
- 随机森林 /
- 水平可见性图
Abstract: Objective Rapid and robust recognition of radar scanning modes under non-cooperative electronic reconnaissance conditions is a prerequisite for threat assessment, resource scheduling, and countermeasure design. Mechanical scanning (MST) and phased-array electronic scanning (EST) leave distinct physical imprints in the time-of-arrival - pulse amplitude (TOA–PA) stream, yet their separability degrades in low SNR, non-stationary dwell scheduling, and jittered timing typical of dense electromagnetic environments. This work develops a physics-grounded, multi-domain feature framework coupled with a Random Forest (RF) classifier to reliably discriminate MST from EST using only intercepted TOA–PA sequences, without synchronization or prior emitter knowledge. Methods The reconnaissance reception chain is modeled and PA formation is formalized to make explicit the link between antenna-pattern traversal and amplitude texture. From this physical view, seven complementary features are extracted spanning time, frequency, and graph structure: coefficient of variation (CV), total variation (TV), Gaussian fitting degree (GFD), relative width (RW), spectral flatness measure (SFM), global clustering coefficient on a Horizontal Visibility Graph (GCC), and normalized degree entropy (NDE). HVG construction preserves temporal order while revealing global structure induced by sequence shape. Features are computed per frame and concatenated into a 7-D vector. The random forest (RF) classifier is trained with bootstrap sampling and random-subspace splits; inference aggregates leaf-level posteriors by majority vote. The full pipeline is summarized in Fig. 10. Overall complexity remains near-linear: CV/TV/RW in O(N); SFM dominated by one FFT in O(NlogN); HVG-based features in O(NlogN), satisfying low-latency constraints. Results and Discussions The dataset is constructed using paired MST and EST frames, with TOA jitter at approximately 0.2% of the PRI, AWGN across SNR levels, and realistic beam patterns including sidelobes for both scanning schemes. Training spans 0–30 dB, and testing covers –5.5–29.5 dB in 5 dB steps. Using the proposed seven-feature vector, the Random Forest attains 97.59% average accuracy over all SNRs and surpasses an SVM baseline with the same inputs at 96.01%. The largest margins appear at low to mid SNR, as detailed in Fig. 11. Single-feature studies reveal pronounced heterogeneity and complementarity: SFM is the best single feature at 0.9161; TV and NDE follow at 0.822 and 0.8065; CV and GFD are mid-tier at about 0.66; RW and graph-similarity measures lag at about 0.56–0.57. Using joint multi-feature inputs increases accuracy to 0.9759, an absolute gain of 5.98 percentage points over the best single feature, and reduces error from 8.39% to 2.41%, a relative drop of about 71%. Table 1 and Fig. 12 summarize these improvements. Runtime measurements indicate that the dominant cost arises from the FFT and the HVG construction; nevertheless, a per-frame compute time of approximately 0.515 ms keeps the method suitable for on-orbit and embedded processing. The performance gains stem from the joint capture of four factors: smooth versus stepwise amplitude evolution captured by CV and TV; main-lobe morphology and time scale captured by GFD and RW, as illustrated in Fig. 4; spectral concentration versus dispersion captured by SFM, as illustrated in Fig. 5; and topology induced by alternating highs and lows under dwell switching captured by HVG clustering and entropy, with details in Fig. 7. These elements together stabilize the decision boundary against noise and dwell nonstationarity. Conclusions A physics-grounded, multi-domain feature framework combined with a RF discriminator is presented for radar scanning mode recognition under noncooperative conditions. The approach starts from intrinsic contrasts between mechanical scanning (continuous, smooth, quasi-periodic) and phased array electronic scanning (dwell, jump, nonstationary), and constructs a TOA–PA signal model consistent with engineering practice. Complementary features are then designed across time (CV, TV), main lobe morphology (GFD, RW), frequency (SFM), and graph structure (GCC, NDE) to jointly characterize smooth vs. stepwise evolution, concentration vs. dispersion, and topological organization. The RF classifier uses bootstrap sampling and random subspaces to reduce variance and mitigate overfitting, enabling robust decisions. Across detection scenarios from −5 dB to 30 dB, the method achieves an average accuracy of 97.59%. In addition, compared with schemes using single-domain or a small number of features, the proposed multi-domain feature framework exhibits higher recognition stability under low-SNR and other challenging disturbance conditions.
- Radar scanning mode /
- Feature extraction /
- Random forest /
- Horizontal visibility graph

HTML全文

图 1 侦收场景示意图

下载: 全尺寸图片幻灯片

图 2 机扫情况下的PA序列

下载: 全尺寸图片幻灯片

图 3 相扫情况下的PA序列

下载: 全尺寸图片幻灯片

图 4 机扫与相扫的时域特征分布

下载: 全尺寸图片幻灯片

图 5 机扫与相扫的SFM取值直方图

下载: 全尺寸图片幻灯片

图 6 序列构建水平可见性图的示意

下载: 全尺寸图片幻灯片

图 7 机扫与相扫的图特征分布

下载: 全尺寸图片幻灯片

图 8 随机森林构建示意图

下载: 全尺寸图片幻灯片

图 9 随机森林的判决过程

下载: 全尺寸图片幻灯片

图 10 面向高维自主特征随机森林的非协作雷达扫描方式智能识别方法的流程图

下载: 全尺寸图片幻灯片

图 11 所提方法与其他方法的比较

下载: 全尺寸图片幻灯片

图 12 单特征与全特征情况的比较

下载: 全尺寸图片幻灯片

图 13 单域特征、跨域特征和全特征情况比较

下载: 全尺寸图片幻灯片

图 14 两类PRI目标的识别性能对比

下载: 全尺寸图片幻灯片

表 1 单特征识别准确率

特征 CV TV GFD RW SFM GCC NDE 全特征

准确率 0.6617 0.822 0.6695 0.5595 0.9161 0.5723 0.8065 0.9759

下载: 导出CSV

参考文献(20)

[1]	何芸倩. 基于机器学习的多功能雷达辐射源识别技术研究[D]. [硕士论文], 电子科技大学, 2025. doi: 10.27005/d.cnki.gdzku.2025.001418. HE Yunqian. Research on multifunction radar emitter identification technology based on machine learning[D]. [Master dissertation], University of Electronic Science and Technology of China, 2025. doi: 10.27005/d.cnki.gdzku.2025.001418.
[2]	李燕平. 机械扫描雷达的DBS成像和动目标检测研究[D]. [硕士论文], 西安电子科技大学, 2006. doi: 10.7666/d.Y1137370. LI Yanping. Study of DBS imaging and moving targets detection in mechanical scanning radar[D]. [Master dissertation], Xidian University, 2006. doi: 10.7666/d.Y1137370.
[3]	陈舒敏, 郑文文, 杨程, 等. 机扫二维相控阵雷达自适应资源调度算法研究[J]. 舰船电子对抗, 2025, 48(1): 51–58. doi: 10.16426/j.cnki.jcdzdk.2025.01.009. CHEN Shumin, ZHENG Wenwen, YANG Cheng, et al. Research on adaptive resource scheduling algorithm for two-dimensional mechanical scanning phased array radar[J]. Shipboard Electronic Countermeasure, 2025, 48(1): 51–58. doi: 10.16426/j.cnki.jcdzdk.2025.01.009.
[4]	嵇慧明, 于昊, 宋帅, 等. 基于改进粗糙集-云模型理论的空战态势评估[J]. 战术导弹技术, 2019(4): 20–27. doi: 10.16358/j.issn.1009-1300.2019.9.038. JI Huiming, YU Hao, SONG Shuai, et al. Air combat situation assessment based on improved rough set–cloud model theory[J]. Tactical Missile Technology, 2019(4): 20–27. doi: 10.16358/j.issn.1009-1300.2019.9.038.
[5]	张文峰, 牟皓, 赵耀东, 等. 基于SVM与DNN的雷达扫描体制识别方法[J]. 电子信息对抗技术, 2022, 37(2): 33–37. doi: 10.3969/j.issn.1674-2230.2022.02.009. ZHANG Wenfeng, MU Hao, ZHAO Yaodong, et al. Intelligent recognition of mechanical scanned radar and electronically scanned array based on support vector machine and deep neural network[J]. Electronic Information Warfare Technology, 2022, 37(2): 33–37. doi: 10.3969/j.issn.1674-2230.2022.02.009.
[6]	郭国华, 何明浩, 韩俊, 等. 基于脉幅信息的相控阵体制雷达识别技术[J]. 中国电子科学研究院学报, 2009, 4(6): 589–593. doi: 10.3969/j.issn.1673-5692.2009.06.008. GUO Guohua, HE Minghao, HAN Jun, et al. Phased-array radar recognition technology based on pulse amplitude[J]. Journal of China Academy of Electronics and Information Technology, 2009, 4(6): 589–593. doi: 10.3969/j.issn.1673-5692.2009.06.008.
[7]	张玉虎, 周正. 基于信号聚集度的相控阵雷达识别技术[J]. 火力与指挥控制, 2018, 43(8): 22–24,30. doi: 10.3969/j.issn.1002-0640.2018.08.005. ZHANG Yuhu and ZHOU Zheng. Recognition of phased-array radar based on analysis of aggregation degree[J]. Fire Control & Command Control, 2018, 43(8): 22–24,30. doi: 10.3969/j.issn.1002-0640.2018.08.005.
[8]	叶巍, 牟连云, 李仙茂. 基于脉冲包络的相控阵雷达识别技术研究[J]. 航天电子对抗, 2011, 27(1): 41–44,57. doi: 10.3969/j.issn.1673-2421.2011.01.012. YE Wei, MU Lianyun, and LI Xianmao. Phased-array radar recognition technology study based on the resemble coefficient of pulse amplitude contour[J]. Aerospace Electronic Warfare, 2011, 27(1): 41–44,57. doi: 10.3969/j.issn.1673-2421.2011.01.012.
[9]	李程, 王伟, 施龙飞, 等. 雷达天线扫描方式的自动识别方法[J]. 国防科技大学学报, 2014, 36(3): 156–163. doi: 10.11887/j.cn.201403028. LI Cheng, WANG Wei, SHI Longfei, et al. Automatic recognition method of radar antenna scan type[J]. Journal of National University of Defense Technology, 2014, 36(3): 156–163. doi: 10.11887/j.cn.201403028.
[10]	GREER T H. Automatic recognition of radar scan type[P]. US, 6697007B2, 2004.
[11]	BARSHAN B and ERAVCI B. Automatic radar antenna scan type recognition in electronic warfare[J]. IEEE Transactions on Aerospace and Electronic Systems, 2012, 48(4): 2908–2931. doi: 10.1109/TAES.2012.6324669.
[12]	QUAN Wei, LI Ping, and XU Fengkai. An algorithm of signal sorting and recognition of phased array radars[C]. Proceedings of the IEEE 10th International Conference on Signal Processing, Beijing, China, 2010: 1877–1880. doi: 10.1109/ICOSP.2010.5657138.
[13]	高刚, 孙盼杰, 刘正彬. 基于脉冲幅度及频率分析的雷达扫描方式识别[J]. 电子信息对抗技术, 2016, 31(6): 12–17. doi: 10.3969/j.issn.1674-2230.2016.06.003. GAO Gang, SUN Panjie, and LIU Zhengbin. Automatic radar antenna scan type recognition based on analysis of pluses amplitude and frequency[J]. Electronic Information Countermeasure Technology, 2016, 31(6): 12–17. doi: 10.3969/j.issn.1674-2230.2016.06.003.
[14]	ZENG Jie and TANG Jinjun. Modeling dynamic traffic flow as visibility graphs: A network-scale prediction framework for lane-level traffic flow based on LPR data[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(4): 4173–4188. doi: 10.1109/TITS.2022.3231959.
[15]	李牛牛, 张云华. 基于可见图和全局相似度的时间序列预测分析[J]. 集成电路应用, 2025, 42(3): 348–351. doi: 10.19339/j.issn.1674-2583.2025.03.148. LI Niuniu and ZHANG Yunhua. Analysis of time series forecasting based on visibility graph and global similarity[J]. Application of IC, 2025, 42(3): 348–351. doi: 10.19339/j.issn.1674-2583.2025.03.148.
[16]	ROY S S and CHATTERJEE S. Partial discharge detection framework employing spectral analysis of horizontal visibility graph[J]. IEEE Sensors Journal, 2021, 21(4): 4819–4826. doi: 10.1109/JSEN.2020.3028849.
[17]	LI Jingchen, SHI Haobin, CHEN Wenbai, et al. Semi-supervised detection model based on adaptive ensemble learning for medical images[J]. IEEE Transactions on Neural Networks and Learning Systems, 2025, 36(1): 237–248. doi: 10.1109/TNNLS.2023.3282809.
[18]	陈俊英, 席月芸, 徐琳, 等. 基于MLP集成随机子空间决策树的航空发动机剩余使用寿命预测[J]. 航空发动机, 2024, 50(6): 81–87. doi: 10.13477/j.cnki.aeroengine.2024.06.012. CHEN Junying, XI Yueyun, XU Lin, et al. Remaining useful life prediction of aeroengines based on MLP integrated random subspace decision trees[J]. Aeroengine, 2024, 50(6): 81–87. doi: 10.13477/j.cnki.aeroengine.2024.06.012.
[19]	王奕森, 夏树涛. 集成学习之随机森林算法综述[J]. 信息通信技术, 2018, 12(1): 49–55. doi: 10.3969/j.issn.1674-1285.2018.01.009. WANG Yisen and XIA Shutao. A survey of random forests algorithms[J]. Information and Communications Technology, 2018, 12(1): 49–55. doi: 10.3969/j.issn.1674-1285.2018.01.009.
[20]	YANG Yi, SUN Yan, LI Feng, et al. MGCNRF: Prediction of disease-related miRNAs based on multiple graph convolutional networks and random forest[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(11): 15701–15709. doi: 10.1109/TNNLS.2023.3289182.