| 摘要: |
| 递归图是一种描述非平稳海杂波时间序列动态结构的有效工具。为此，该文提出一种基于双通道递归图的高维特征检测方法，提升海面小目标探测能力。首先，从幅度-相位双通道出发，分别构建时间序列的递归图。其次，在幅度通道，提取递归图上的对角结构特征，称为相对平均对角线长度、相对对角线分布熵；在相位通道，提取递归图上的垂直结构特征，称为相对捕获时间、相对垂直线分布熵。双通道特征的联合有效增强了对含目标回波与海杂波动态行为差异的刻画能力。然后，通过引入目标先验信息引导凸包的收缩方向并缩小虚警样本的搜索范围，大大减少计算时间，将现有凸包从三维以下拓展至高维空间，获取虚警可控的判决区域。最后，实验结果表明，相比现有基于三特征的检测器，所提出的检测器平均检测概率提升约15%，凸包分类器平均训练时间缩短90%以上。 |
| 关键词: 海杂波 目标检测 递归图 凸包检测 |
| DOI： |
| 分类号:TN957.51 |
| 基金项目:国家自然科学基金(No. 62171229)；中兴通讯研究基金资助(IA20240701009) |
|
| High-Dimensional Feature Detection of Sea-Surface Targets via Dual-Channel Recurrence Plots |
|
|
|
|
| Abstract: |
| Recurrence Plot (RP) is an effective tool for describing the dynamic structures of non-stationary sea clutter time series. To this end, this paper proposes a high-dimensional feature detection method based on dual-channel recurrence plots to enhance the detection capability of small targets on the sea surface. First, RPs are constructed separately from the dual channels of amplitude and phase. Second, diagonal structure features, specifically the Relative Average Diagonal Length and the Relative Diagonal Distributed Entropy, are extracted from the amplitude channel RP. Meanwhile, vertical structure features, namely the Relative Trapping Time and the Relative Vertical Distributed Entropy, are extracted from the phase channel RP. The integration of features from both channels effectively enhances the characterization of the dynamic behavioral differences between target echoes and sea clutter. Furthermore, by introducing target prior information to guide the shrinking direction of the convex hull and narrow the search range for false alarm samples, the proposed method significantly reduces computation time. This extends the application of the convex hull classifier from low-dimensional (below 3-D) to high-dimensional feature spaces, enabling the acquisition of a decision region with a controllable false alarm rate. Finally, experimental results demonstrate that, compared with existing tri-feature-based detectors, the proposed detector achieves an increase in average detection probability of approximately 15%, while the average training time for the convex hull classifier is reduced by more than 90%. |
| Key words: sea clutter small target detection recurrence plot convex hull classification |