| 引用本文: | 汪 晋，苏洪涛，汪圣利，田 强. 一种基于判别投影特征提取的杂波点迹滤除方法[J]. 雷达科学与技术, 2026, 24(1): 52-61.[点击复制] |
| WANG Jin, SU Hongtao, WANG Shengli, TIAN Qiang. A Feature Extraction Method Based on Discriminant Projections for Clutter Plot Filtering[J]. Radar Science and Technology, 2026, 24(1): 52-61.[点击复制] |
|
| 摘要: |
| 在复杂环境下雷达面临大量杂波点迹，严重影响雷达对目标的有效跟踪。而杂波点迹滤除技术可有效抑制杂波引起的虚假点迹，减小系统处理负担，提升目标跟踪效能。用于雷达点迹过滤的特征越丰富，点迹分类判别的正确率越高，点迹过滤性能越好，因此，如何提取有效的特征是雷达杂波点迹过滤面临的一个重要问题。传统基于正则化最小二乘的特征提取算法在构造代价函数时忽略了投影空间中样本的类间重构关系和类内聚集性，导致降维后特征损失，不利于点迹属性的正确判别。为此，本文提出了一种基于正则化最小二乘加权判别投影的特征提取算法。该方法在重构样本时利用类间表示系数，实现重构误差的最小化；在构造代价函数时利用样本的表示系数生成类内加权矩阵，提升同类样本的紧致性，使降维后样本保留更多有效特征。为验证本文所提方法的通用性和有效性，我们采用公开人脸数据集和雷达实测数据分别进行了验证。在公开人脸数据集上的实验结果表明，所提方法的分类正确率至少提升4%，雷达实测数据处理结果表明，本文所提方法对杂波点迹抑制率达90%以上，同时目标点迹损失率保持在1%以下。 |
| 关键词: 点迹滤除 特征提取 维数简约 判别投影 |
| DOI：DOI:10.3969/j.issn.1672-2337.2026.01.006 |
| 分类号:TN953 |
| 基金项目:国家自然科学基金（62201418，62192714） |
|
| A Feature Extraction Method Based on Discriminant Projections for Clutter Plot Filtering |
|
WANG Jin, SU Hongtao, WANG Shengli, TIAN Qiang
|
|
1. National Key Laboratory of Radar Signal Processing, Xidian University, Xi’an 710071, China;2. Nanjing Research Institute of Electronics Technology, Nanjing 210039, China
|
| Abstract: |
| In complex environments, radars face a large number of clutter plots, which severely affect the effective tracking of targets. The clutter plot filtering technology can effectively suppress the false plots caused by clutter, reduce the system processing burden, and improve the target tracking performance. The richer the features used for radar plot filtering, the higher the accuracy of plot classification and discrimination, and the better the plot filtering performance. Therefore, how to extract effective features is an important issue for radar clutter plot filtering. The traditional feature extraction algorithm based on regularized least squares ignores the inter?class reconstruction relationship and intra?class clustering characteristics of samples in the projection space when constructing the cost function, resulting in feature loss after dimensionality reduction. To address this issue, this paper proposes a feature extraction algorithm based on regularized least squares weighted discriminant projection. The proposed method utilizes inter?class representation coefficients to reconstruct samples, thereby minimizing the reconstruction error. Furthermore, when constructing the cost function, the representation coefficients of the samples are used to generate an intra?class weighting matrix, which improves the compactness of similar samples, and preserves more effective features after dimensionality reduction. To verify the universality and effectiveness of the proposed method, a public facial dataset and radar measured data are used to conduct validation respectively. Experimental results on the public facial dataset show that the proposed method achieves at least 4% improvement in recognition accuracy. Processing results of the radar measured data demonstrate that the proposed method achieves a clutter plot suppression rate of over 90%, while the target plot loss rate remains below 1%. |
| Key words: plot filtering feature extraction dimensionality reduction discriminant projections |
